5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? e.g. locale */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
125 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
126 if (!CAT2(PL_utf8_,class)) { \
128 ENTER; save_re_context(); \
129 ok=CAT2(is_utf8_,class)((const U8*)str); \
130 assert(ok); LEAVE; } } STMT_END
132 /* Doesn't do an assert to verify that is correct */
133 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
134 if (!CAT2(PL_utf8_,class)) { \
135 bool throw_away __attribute__unused__; \
136 ENTER; save_re_context(); \
137 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
140 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
141 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
142 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
144 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
145 LOAD_UTF8_CHARCLASS(X_begin, " "); \
146 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
147 /* These are utf8 constants, and not utf-ebcdic constants, so the \
148 * assert should likely and hopefully fail on an EBCDIC machine */ \
149 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
151 /* No asserts are done for these, in case called on an early \
152 * Unicode version in which they map to nothing */ \
153 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
154 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
155 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
156 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
157 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
158 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
159 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
161 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
163 /* The actual code for CCC_TRY, which uses several variables from the routine
164 * it's callable from. It is designed to be the bulk of a case statement.
165 * FUNC is the macro or function to call on non-utf8 targets that indicate if
166 * nextchr matches the class.
167 * UTF8_TEST is the whole test string to use for utf8 targets
168 * LOAD is what to use to test, and if not present to load in the swash for the
170 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
172 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
173 * utf8 and a variant, load the swash if necessary and test using the utf8
174 * test. Advance to the next character if test is ok, otherwise fail; If not
175 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
176 * fails, or advance to the next character */
178 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
179 if (locinput >= PL_regeol) { \
182 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
183 LOAD_UTF8_CHARCLASS(CLASS, STR); \
184 if (POS_OR_NEG (UTF8_TEST)) { \
187 locinput += PL_utf8skip[nextchr]; \
188 nextchr = UCHARAT(locinput); \
191 if (POS_OR_NEG (FUNC(nextchr))) { \
194 nextchr = UCHARAT(++locinput); \
197 /* Handle the non-locale cases for a character class and its complement. It
198 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
199 * This is because that code fails when the test succeeds, so we want to have
200 * the test fail so that the code succeeds. The swash is stored in a
201 * predictable PL_ place */
202 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
205 _CCC_TRY_CODE( !, FUNC, \
206 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
207 (U8*)locinput, TRUE)), \
210 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
211 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
212 (U8*)locinput, TRUE)), \
215 /* Generate the case statements for both locale and non-locale character
216 * classes in regmatch for classes that don't have special unicode semantics.
217 * Locales don't use an immediate swash, but an intermediary special locale
218 * function that is called on the pointer to the current place in the input
219 * string. That function will resolve to needing the same swash. One might
220 * think that because we don't know what the locale will match, we shouldn't
221 * check with the swash loading function that it loaded properly; ie, that we
222 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
223 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
225 #define CCC_TRY(NAME, NNAME, FUNC, \
226 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
227 NAMEA, NNAMEA, FUNCA, \
230 PL_reg_flags |= RF_tainted; \
231 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
233 PL_reg_flags |= RF_tainted; \
234 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
237 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
240 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
241 nextchr = UCHARAT(++locinput); \
244 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
248 locinput += PL_utf8skip[nextchr]; \
249 nextchr = UCHARAT(locinput); \
252 nextchr = UCHARAT(++locinput); \
255 /* Generate the non-locale cases */ \
256 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
258 /* This is like CCC_TRY, but has an extra set of parameters for generating case
259 * statements to handle separate Unicode semantics nodes */
260 #define CCC_TRY_U(NAME, NNAME, FUNC, \
261 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
262 NAMEU, NNAMEU, FUNCU, \
263 NAMEA, NNAMEA, FUNCA, \
265 CCC_TRY(NAME, NNAME, FUNC, \
266 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
267 NAMEA, NNAMEA, FUNCA, \
269 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
271 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
273 /* for use after a quantifier and before an EXACT-like node -- japhy */
274 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
276 * NOTE that *nothing* that affects backtracking should be in here, specifically
277 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
278 * node that is in between two EXACT like nodes when ascertaining what the required
279 * "follow" character is. This should probably be moved to regex compile time
280 * although it may be done at run time beause of the REF possibility - more
281 * investigation required. -- demerphq
283 #define JUMPABLE(rn) ( \
285 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
287 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
288 OP(rn) == PLUS || OP(rn) == MINMOD || \
290 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
292 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
294 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
297 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
298 we don't need this definition. */
299 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
300 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
301 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
304 /* ... so we use this as its faster. */
305 #define IS_TEXT(rn) ( OP(rn)==EXACT )
306 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
307 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
308 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
313 Search for mandatory following text node; for lookahead, the text must
314 follow but for lookbehind (rn->flags != 0) we skip to the next step.
316 #define FIND_NEXT_IMPT(rn) STMT_START { \
317 while (JUMPABLE(rn)) { \
318 const OPCODE type = OP(rn); \
319 if (type == SUSPEND || PL_regkind[type] == CURLY) \
320 rn = NEXTOPER(NEXTOPER(rn)); \
321 else if (type == PLUS) \
323 else if (type == IFMATCH) \
324 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
325 else rn += NEXT_OFF(rn); \
330 static void restore_pos(pTHX_ void *arg);
332 #define REGCP_PAREN_ELEMS 4
333 #define REGCP_OTHER_ELEMS 5
334 #define REGCP_FRAME_ELEMS 1
335 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
336 * are needed for the regexp context stack bookkeeping. */
339 S_regcppush(pTHX_ I32 parenfloor)
342 const int retval = PL_savestack_ix;
343 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
344 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
345 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
347 GET_RE_DEBUG_FLAGS_DECL;
349 if (paren_elems_to_push < 0)
350 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
352 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
353 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
354 " out of range (%lu-%ld)",
355 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
357 SSGROW(total_elems + REGCP_FRAME_ELEMS);
359 for (p = PL_regsize; p > parenfloor; p--) {
360 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
361 SSPUSHINT(PL_regoffs[p].end);
362 SSPUSHINT(PL_regoffs[p].start);
363 SSPUSHPTR(PL_reg_start_tmp[p]);
365 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
366 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
367 (UV)p, (IV)PL_regoffs[p].start,
368 (IV)(PL_reg_start_tmp[p] - PL_bostr),
369 (IV)PL_regoffs[p].end
372 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
373 SSPUSHPTR(PL_regoffs);
374 SSPUSHINT(PL_regsize);
375 SSPUSHINT(*PL_reglastparen);
376 SSPUSHINT(*PL_reglastcloseparen);
377 SSPUSHPTR(PL_reginput);
378 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
383 /* These are needed since we do not localize EVAL nodes: */
384 #define REGCP_SET(cp) \
386 PerlIO_printf(Perl_debug_log, \
387 " Setting an EVAL scope, savestack=%"IVdf"\n", \
388 (IV)PL_savestack_ix)); \
391 #define REGCP_UNWIND(cp) \
393 if (cp != PL_savestack_ix) \
394 PerlIO_printf(Perl_debug_log, \
395 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
396 (IV)(cp), (IV)PL_savestack_ix)); \
400 S_regcppop(pTHX_ const regexp *rex)
405 GET_RE_DEBUG_FLAGS_DECL;
407 PERL_ARGS_ASSERT_REGCPPOP;
409 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
411 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
412 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
413 input = (char *) SSPOPPTR;
414 *PL_reglastcloseparen = SSPOPINT;
415 *PL_reglastparen = SSPOPINT;
416 PL_regsize = SSPOPINT;
417 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
419 i -= REGCP_OTHER_ELEMS;
420 /* Now restore the parentheses context. */
421 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
423 U32 paren = (U32)SSPOPINT;
424 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
425 PL_regoffs[paren].start = SSPOPINT;
427 if (paren <= *PL_reglastparen)
428 PL_regoffs[paren].end = tmps;
430 PerlIO_printf(Perl_debug_log,
431 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
432 (UV)paren, (IV)PL_regoffs[paren].start,
433 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
434 (IV)PL_regoffs[paren].end,
435 (paren > *PL_reglastparen ? "(no)" : ""));
439 if (*PL_reglastparen + 1 <= rex->nparens) {
440 PerlIO_printf(Perl_debug_log,
441 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
442 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
446 /* It would seem that the similar code in regtry()
447 * already takes care of this, and in fact it is in
448 * a better location to since this code can #if 0-ed out
449 * but the code in regtry() is needed or otherwise tests
450 * requiring null fields (pat.t#187 and split.t#{13,14}
451 * (as of patchlevel 7877) will fail. Then again,
452 * this code seems to be necessary or otherwise
453 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
454 * --jhi updated by dapm */
455 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
457 PL_regoffs[i].start = -1;
458 PL_regoffs[i].end = -1;
464 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
467 * pregexec and friends
470 #ifndef PERL_IN_XSUB_RE
472 - pregexec - match a regexp against a string
475 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
476 char *strbeg, I32 minend, SV *screamer, U32 nosave)
477 /* strend: pointer to null at end of string */
478 /* strbeg: real beginning of string */
479 /* minend: end of match must be >=minend after stringarg. */
480 /* nosave: For optimizations. */
482 PERL_ARGS_ASSERT_PREGEXEC;
485 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
486 nosave ? 0 : REXEC_COPY_STR);
491 * Need to implement the following flags for reg_anch:
493 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
495 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
496 * INTUIT_AUTORITATIVE_ML
497 * INTUIT_ONCE_NOML - Intuit can match in one location only.
500 * Another flag for this function: SECOND_TIME (so that float substrs
501 * with giant delta may be not rechecked).
504 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
506 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
507 Otherwise, only SvCUR(sv) is used to get strbeg. */
509 /* XXXX We assume that strpos is strbeg unless sv. */
511 /* XXXX Some places assume that there is a fixed substring.
512 An update may be needed if optimizer marks as "INTUITable"
513 RExen without fixed substrings. Similarly, it is assumed that
514 lengths of all the strings are no more than minlen, thus they
515 cannot come from lookahead.
516 (Or minlen should take into account lookahead.)
517 NOTE: Some of this comment is not correct. minlen does now take account
518 of lookahead/behind. Further research is required. -- demerphq
522 /* A failure to find a constant substring means that there is no need to make
523 an expensive call to REx engine, thus we celebrate a failure. Similarly,
524 finding a substring too deep into the string means that less calls to
525 regtry() should be needed.
527 REx compiler's optimizer found 4 possible hints:
528 a) Anchored substring;
530 c) Whether we are anchored (beginning-of-line or \G);
531 d) First node (of those at offset 0) which may distinguish positions;
532 We use a)b)d) and multiline-part of c), and try to find a position in the
533 string which does not contradict any of them.
536 /* Most of decisions we do here should have been done at compile time.
537 The nodes of the REx which we used for the search should have been
538 deleted from the finite automaton. */
541 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
542 char *strend, const U32 flags, re_scream_pos_data *data)
545 struct regexp *const prog = (struct regexp *)SvANY(rx);
546 register I32 start_shift = 0;
547 /* Should be nonnegative! */
548 register I32 end_shift = 0;
553 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
555 register char *other_last = NULL; /* other substr checked before this */
556 char *check_at = NULL; /* check substr found at this pos */
557 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
558 RXi_GET_DECL(prog,progi);
560 const char * const i_strpos = strpos;
562 GET_RE_DEBUG_FLAGS_DECL;
564 PERL_ARGS_ASSERT_RE_INTUIT_START;
566 RX_MATCH_UTF8_set(rx,utf8_target);
569 PL_reg_flags |= RF_utf8;
572 debug_start_match(rx, utf8_target, strpos, strend,
573 sv ? "Guessing start of match in sv for"
574 : "Guessing start of match in string for");
577 /* CHR_DIST() would be more correct here but it makes things slow. */
578 if (prog->minlen > strend - strpos) {
579 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
580 "String too short... [re_intuit_start]\n"));
584 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
587 if (!prog->check_utf8 && prog->check_substr)
588 to_utf8_substr(prog);
589 check = prog->check_utf8;
591 if (!prog->check_substr && prog->check_utf8)
592 to_byte_substr(prog);
593 check = prog->check_substr;
595 if (check == &PL_sv_undef) {
596 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
597 "Non-utf8 string cannot match utf8 check string\n"));
600 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
601 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
602 || ( (prog->extflags & RXf_ANCH_BOL)
603 && !multiline ) ); /* Check after \n? */
606 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
607 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
608 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
610 && (strpos != strbeg)) {
611 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
614 if (prog->check_offset_min == prog->check_offset_max &&
615 !(prog->extflags & RXf_CANY_SEEN)) {
616 /* Substring at constant offset from beg-of-str... */
619 s = HOP3c(strpos, prog->check_offset_min, strend);
622 slen = SvCUR(check); /* >= 1 */
624 if ( strend - s > slen || strend - s < slen - 1
625 || (strend - s == slen && strend[-1] != '\n')) {
626 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
629 /* Now should match s[0..slen-2] */
631 if (slen && (*SvPVX_const(check) != *s
633 && memNE(SvPVX_const(check), s, slen)))) {
635 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
639 else if (*SvPVX_const(check) != *s
640 || ((slen = SvCUR(check)) > 1
641 && memNE(SvPVX_const(check), s, slen)))
644 goto success_at_start;
647 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
649 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
650 end_shift = prog->check_end_shift;
653 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
654 - (SvTAIL(check) != 0);
655 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
657 if (end_shift < eshift)
661 else { /* Can match at random position */
664 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
665 end_shift = prog->check_end_shift;
667 /* end shift should be non negative here */
670 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
672 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
673 (IV)end_shift, RX_PRECOMP(prog));
677 /* Find a possible match in the region s..strend by looking for
678 the "check" substring in the region corrected by start/end_shift. */
681 I32 srch_start_shift = start_shift;
682 I32 srch_end_shift = end_shift;
683 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
684 srch_end_shift -= ((strbeg - s) - srch_start_shift);
685 srch_start_shift = strbeg - s;
687 DEBUG_OPTIMISE_MORE_r({
688 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
689 (IV)prog->check_offset_min,
690 (IV)srch_start_shift,
692 (IV)prog->check_end_shift);
695 if (flags & REXEC_SCREAM) {
696 I32 p = -1; /* Internal iterator of scream. */
697 I32 * const pp = data ? data->scream_pos : &p;
699 if (PL_screamfirst[BmRARE(check)] >= 0
700 || ( BmRARE(check) == '\n'
701 && (BmPREVIOUS(check) == SvCUR(check) - 1)
703 s = screaminstr(sv, check,
704 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
707 /* we may be pointing at the wrong string */
708 if (s && RXp_MATCH_COPIED(prog))
709 s = strbeg + (s - SvPVX_const(sv));
711 *data->scream_olds = s;
716 if (prog->extflags & RXf_CANY_SEEN) {
717 start_point= (U8*)(s + srch_start_shift);
718 end_point= (U8*)(strend - srch_end_shift);
720 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
721 end_point= HOP3(strend, -srch_end_shift, strbeg);
723 DEBUG_OPTIMISE_MORE_r({
724 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
725 (int)(end_point - start_point),
726 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
730 s = fbm_instr( start_point, end_point,
731 check, multiline ? FBMrf_MULTILINE : 0);
734 /* Update the count-of-usability, remove useless subpatterns,
738 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
739 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
740 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
741 (s ? "Found" : "Did not find"),
742 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
743 ? "anchored" : "floating"),
746 (s ? " at offset " : "...\n") );
751 /* Finish the diagnostic message */
752 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
754 /* XXX dmq: first branch is for positive lookbehind...
755 Our check string is offset from the beginning of the pattern.
756 So we need to do any stclass tests offset forward from that
765 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
766 Start with the other substr.
767 XXXX no SCREAM optimization yet - and a very coarse implementation
768 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
769 *always* match. Probably should be marked during compile...
770 Probably it is right to do no SCREAM here...
773 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
774 : (prog->float_substr && prog->anchored_substr))
776 /* Take into account the "other" substring. */
777 /* XXXX May be hopelessly wrong for UTF... */
780 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
783 char * const last = HOP3c(s, -start_shift, strbeg);
785 char * const saved_s = s;
788 t = s - prog->check_offset_max;
789 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
791 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
796 t = HOP3c(t, prog->anchored_offset, strend);
797 if (t < other_last) /* These positions already checked */
799 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
802 /* XXXX It is not documented what units *_offsets are in.
803 We assume bytes, but this is clearly wrong.
804 Meaning this code needs to be carefully reviewed for errors.
808 /* On end-of-str: see comment below. */
809 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
810 if (must == &PL_sv_undef) {
812 DEBUG_r(must = prog->anchored_utf8); /* for debug */
817 HOP3(HOP3(last1, prog->anchored_offset, strend)
818 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
820 multiline ? FBMrf_MULTILINE : 0
823 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
824 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
825 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
826 (s ? "Found" : "Contradicts"),
827 quoted, RE_SV_TAIL(must));
832 if (last1 >= last2) {
833 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
834 ", giving up...\n"));
837 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
838 ", trying floating at offset %ld...\n",
839 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
840 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
841 s = HOP3c(last, 1, strend);
845 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
846 (long)(s - i_strpos)));
847 t = HOP3c(s, -prog->anchored_offset, strbeg);
848 other_last = HOP3c(s, 1, strend);
856 else { /* Take into account the floating substring. */
858 char * const saved_s = s;
861 t = HOP3c(s, -start_shift, strbeg);
863 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
864 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
865 last = HOP3c(t, prog->float_max_offset, strend);
866 s = HOP3c(t, prog->float_min_offset, strend);
869 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
870 must = utf8_target ? prog->float_utf8 : prog->float_substr;
871 /* fbm_instr() takes into account exact value of end-of-str
872 if the check is SvTAIL(ed). Since false positives are OK,
873 and end-of-str is not later than strend we are OK. */
874 if (must == &PL_sv_undef) {
876 DEBUG_r(must = prog->float_utf8); /* for debug message */
879 s = fbm_instr((unsigned char*)s,
880 (unsigned char*)last + SvCUR(must)
882 must, multiline ? FBMrf_MULTILINE : 0);
884 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
885 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
886 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
887 (s ? "Found" : "Contradicts"),
888 quoted, RE_SV_TAIL(must));
892 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
893 ", giving up...\n"));
896 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
897 ", trying anchored starting at offset %ld...\n",
898 (long)(saved_s + 1 - i_strpos)));
900 s = HOP3c(t, 1, strend);
904 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
905 (long)(s - i_strpos)));
906 other_last = s; /* Fix this later. --Hugo */
916 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
918 DEBUG_OPTIMISE_MORE_r(
919 PerlIO_printf(Perl_debug_log,
920 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
921 (IV)prog->check_offset_min,
922 (IV)prog->check_offset_max,
930 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
932 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
935 /* Fixed substring is found far enough so that the match
936 cannot start at strpos. */
938 if (ml_anch && t[-1] != '\n') {
939 /* Eventually fbm_*() should handle this, but often
940 anchored_offset is not 0, so this check will not be wasted. */
941 /* XXXX In the code below we prefer to look for "^" even in
942 presence of anchored substrings. And we search even
943 beyond the found float position. These pessimizations
944 are historical artefacts only. */
946 while (t < strend - prog->minlen) {
948 if (t < check_at - prog->check_offset_min) {
949 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
950 /* Since we moved from the found position,
951 we definitely contradict the found anchored
952 substr. Due to the above check we do not
953 contradict "check" substr.
954 Thus we can arrive here only if check substr
955 is float. Redo checking for "other"=="fixed".
958 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
959 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
960 goto do_other_anchored;
962 /* We don't contradict the found floating substring. */
963 /* XXXX Why not check for STCLASS? */
965 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
966 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
969 /* Position contradicts check-string */
970 /* XXXX probably better to look for check-string
971 than for "\n", so one should lower the limit for t? */
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
973 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
974 other_last = strpos = s = t + 1;
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
980 PL_colors[0], PL_colors[1]));
984 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
985 PL_colors[0], PL_colors[1]));
989 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
992 /* The found string does not prohibit matching at strpos,
993 - no optimization of calling REx engine can be performed,
994 unless it was an MBOL and we are not after MBOL,
995 or a future STCLASS check will fail this. */
997 /* Even in this situation we may use MBOL flag if strpos is offset
998 wrt the start of the string. */
999 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1000 && (strpos != strbeg) && strpos[-1] != '\n'
1001 /* May be due to an implicit anchor of m{.*foo} */
1002 && !(prog->intflags & PREGf_IMPLICIT))
1007 DEBUG_EXECUTE_r( if (ml_anch)
1008 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1009 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1012 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1014 prog->check_utf8 /* Could be deleted already */
1015 && --BmUSEFUL(prog->check_utf8) < 0
1016 && (prog->check_utf8 == prog->float_utf8)
1018 prog->check_substr /* Could be deleted already */
1019 && --BmUSEFUL(prog->check_substr) < 0
1020 && (prog->check_substr == prog->float_substr)
1023 /* If flags & SOMETHING - do not do it many times on the same match */
1024 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1025 /* XXX Does the destruction order has to change with utf8_target? */
1026 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1027 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1028 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1029 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1030 check = NULL; /* abort */
1032 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1033 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1034 if (prog->intflags & PREGf_IMPLICIT)
1035 prog->extflags &= ~RXf_ANCH_MBOL;
1036 /* XXXX This is a remnant of the old implementation. It
1037 looks wasteful, since now INTUIT can use many
1038 other heuristics. */
1039 prog->extflags &= ~RXf_USE_INTUIT;
1040 /* XXXX What other flags might need to be cleared in this branch? */
1046 /* Last resort... */
1047 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1048 /* trie stclasses are too expensive to use here, we are better off to
1049 leave it to regmatch itself */
1050 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1051 /* minlen == 0 is possible if regstclass is \b or \B,
1052 and the fixed substr is ''$.
1053 Since minlen is already taken into account, s+1 is before strend;
1054 accidentally, minlen >= 1 guaranties no false positives at s + 1
1055 even for \b or \B. But (minlen? 1 : 0) below assumes that
1056 regstclass does not come from lookahead... */
1057 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1058 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1059 const U8* const str = (U8*)STRING(progi->regstclass);
1060 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1061 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1064 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1065 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1066 else if (prog->float_substr || prog->float_utf8)
1067 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1071 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1072 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1075 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1078 const char *what = NULL;
1080 if (endpos == strend) {
1081 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1082 "Could not match STCLASS...\n") );
1085 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1086 "This position contradicts STCLASS...\n") );
1087 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1089 /* Contradict one of substrings */
1090 if (prog->anchored_substr || prog->anchored_utf8) {
1091 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1092 DEBUG_EXECUTE_r( what = "anchored" );
1094 s = HOP3c(t, 1, strend);
1095 if (s + start_shift + end_shift > strend) {
1096 /* XXXX Should be taken into account earlier? */
1097 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1098 "Could not match STCLASS...\n") );
1103 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1104 "Looking for %s substr starting at offset %ld...\n",
1105 what, (long)(s + start_shift - i_strpos)) );
1108 /* Have both, check_string is floating */
1109 if (t + start_shift >= check_at) /* Contradicts floating=check */
1110 goto retry_floating_check;
1111 /* Recheck anchored substring, but not floating... */
1115 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1116 "Looking for anchored substr starting at offset %ld...\n",
1117 (long)(other_last - i_strpos)) );
1118 goto do_other_anchored;
1120 /* Another way we could have checked stclass at the
1121 current position only: */
1126 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1127 "Looking for /%s^%s/m starting at offset %ld...\n",
1128 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1131 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1133 /* Check is floating substring. */
1134 retry_floating_check:
1135 t = check_at - start_shift;
1136 DEBUG_EXECUTE_r( what = "floating" );
1137 goto hop_and_restart;
1140 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1141 "By STCLASS: moving %ld --> %ld\n",
1142 (long)(t - i_strpos), (long)(s - i_strpos))
1146 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1147 "Does not contradict STCLASS...\n");
1152 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1153 PL_colors[4], (check ? "Guessed" : "Giving up"),
1154 PL_colors[5], (long)(s - i_strpos)) );
1157 fail_finish: /* Substring not found */
1158 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1159 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1161 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1162 PL_colors[4], PL_colors[5]));
1166 #define DECL_TRIE_TYPE(scan) \
1167 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1168 trie_type = (scan->flags != EXACT) \
1169 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1170 : (utf8_target ? trie_utf8 : trie_plain)
1172 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1173 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1174 switch (trie_type) { \
1175 case trie_utf8_fold: \
1176 if ( foldlen>0 ) { \
1177 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1182 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1183 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1184 foldlen -= UNISKIP( uvc ); \
1185 uscan = foldbuf + UNISKIP( uvc ); \
1188 case trie_latin_utf8_fold: \
1189 if ( foldlen>0 ) { \
1190 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1196 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1197 foldlen -= UNISKIP( uvc ); \
1198 uscan = foldbuf + UNISKIP( uvc ); \
1202 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1209 charid = trie->charmap[ uvc ]; \
1213 if (widecharmap) { \
1214 SV** const svpp = hv_fetch(widecharmap, \
1215 (char*)&uvc, sizeof(UV), 0); \
1217 charid = (U16)SvIV(*svpp); \
1222 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1226 && (ln == 1 || folder(s, pat_string, ln)) \
1227 && (!reginfo || regtry(reginfo, &s)) ) \
1233 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1235 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1241 #define REXEC_FBC_SCAN(CoDe) \
1243 while (s < strend) { \
1249 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1250 REXEC_FBC_UTF8_SCAN( \
1252 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1261 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1264 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1273 #define REXEC_FBC_TRYIT \
1274 if ((!reginfo || regtry(reginfo, &s))) \
1277 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1278 if (utf8_target) { \
1279 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1282 REXEC_FBC_CLASS_SCAN(CoNd); \
1285 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1286 if (utf8_target) { \
1288 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1291 REXEC_FBC_CLASS_SCAN(CoNd); \
1294 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1295 PL_reg_flags |= RF_tainted; \
1296 if (utf8_target) { \
1297 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1300 REXEC_FBC_CLASS_SCAN(CoNd); \
1303 #define DUMP_EXEC_POS(li,s,doutf8) \
1304 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1307 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1308 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1309 tmp = TEST_NON_UTF8(tmp); \
1310 REXEC_FBC_UTF8_SCAN( \
1311 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1320 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1321 if (s == PL_bostr) { \
1325 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1326 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1329 LOAD_UTF8_CHARCLASS_ALNUM(); \
1330 REXEC_FBC_UTF8_SCAN( \
1331 if (tmp == ! (TeSt2_UtF8)) { \
1340 /* The only difference between the BOUND and NBOUND cases is that
1341 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1342 * NBOUND. This is accomplished by passing it in either the if or else clause,
1343 * with the other one being empty */
1344 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1345 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1347 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1348 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1350 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1351 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1353 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1354 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1357 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1358 * be passed in completely with the variable name being tested, which isn't
1359 * such a clean interface, but this is easier to read than it was before. We
1360 * are looking for the boundary (or non-boundary between a word and non-word
1361 * character. The utf8 and non-utf8 cases have the same logic, but the details
1362 * must be different. Find the "wordness" of the character just prior to this
1363 * one, and compare it with the wordness of this one. If they differ, we have
1364 * a boundary. At the beginning of the string, pretend that the previous
1365 * character was a new-line */
1366 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1367 if (utf8_target) { \
1370 else { /* Not utf8 */ \
1371 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1372 tmp = TEST_NON_UTF8(tmp); \
1374 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1383 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1386 /* We know what class REx starts with. Try to find this position... */
1387 /* if reginfo is NULL, its a dryrun */
1388 /* annoyingly all the vars in this routine have different names from their counterparts
1389 in regmatch. /grrr */
1392 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1393 const char *strend, regmatch_info *reginfo)
1396 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1397 char *pat_string; /* The pattern's exactish string */
1398 char *pat_end; /* ptr to end char of pat_string */
1399 re_fold_t folder; /* Function for computing non-utf8 folds */
1400 const U8 *fold_array; /* array for folding ords < 256 */
1403 register STRLEN uskip;
1407 register I32 tmp = 1; /* Scratch variable? */
1408 register const bool utf8_target = PL_reg_match_utf8;
1409 UV utf8_fold_flags = 0;
1410 RXi_GET_DECL(prog,progi);
1412 PERL_ARGS_ASSERT_FIND_BYCLASS;
1414 /* We know what class it must start with. */
1418 if (utf8_target || OP(c) == ANYOFV) {
1419 STRLEN inclasslen = strend - s;
1420 REXEC_FBC_UTF8_CLASS_SCAN(
1421 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1424 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1429 if (tmp && (!reginfo || regtry(reginfo, &s)))
1437 if (UTF_PATTERN || utf8_target) {
1438 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1439 goto do_exactf_utf8;
1441 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1442 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1443 goto do_exactf_non_utf8; /* isn't dealt with by these */
1446 if (UTF_PATTERN || utf8_target) {
1447 utf8_fold_flags = 0;
1448 goto do_exactf_utf8;
1451 /* Any 'ss' in the pattern should have been replaced by regcomp,
1452 * so we don't have to worry here about this single special case
1453 * in the Latin1 range */
1454 fold_array = PL_fold_latin1;
1455 folder = foldEQ_latin1;
1456 goto do_exactf_non_utf8;
1459 if (UTF_PATTERN || utf8_target) {
1460 utf8_fold_flags = 0;
1461 goto do_exactf_utf8;
1463 fold_array = PL_fold;
1465 goto do_exactf_non_utf8;
1468 if (UTF_PATTERN || utf8_target) {
1469 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1470 goto do_exactf_utf8;
1472 fold_array = PL_fold_locale;
1473 folder = foldEQ_locale;
1477 do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */
1479 /* The idea in the non-utf8 EXACTF* cases is to first find the
1480 * first character of the EXACTF* node and then, if necessary,
1481 * case-insensitively compare the full text of the node. c1 is the
1482 * first character. c2 is its fold. This logic will not work for
1483 * Unicode semantics and the german sharp ss, which hence should
1484 * not be compiled into a node that gets here. */
1485 pat_string = STRING(c);
1486 ln = STR_LEN(c); /* length to match in octets/bytes */
1488 e = HOP3c(strend, -((I32)ln), s);
1490 if (!reginfo && e < s) {
1491 e = s; /* Due to minlen logic of intuit() */
1495 c2 = fold_array[c1];
1496 if (c1 == c2) { /* If char and fold are the same */
1497 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1500 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1506 /* If one of the operands is in utf8, we can't use the simpler
1507 * folding above, due to the fact that many different characters
1508 * can have the same fold, or portion of a fold, or different-
1510 pat_string = STRING(c);
1511 ln = STR_LEN(c); /* length to match in octets/bytes */
1512 pat_end = pat_string + ln;
1513 lnc = (UTF_PATTERN) /* length to match in characters */
1514 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1517 e = HOP3c(strend, -((I32)lnc), s);
1519 if (!reginfo && e < s) {
1520 e = s; /* Due to minlen logic of intuit() */
1524 char *my_strend= (char *)strend;
1525 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1526 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1527 && (!reginfo || regtry(reginfo, &s)) )
1535 PL_reg_flags |= RF_tainted;
1536 FBC_BOUND(isALNUM_LC,
1537 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1538 isALNUM_LC_utf8((U8*)s));
1541 PL_reg_flags |= RF_tainted;
1542 FBC_NBOUND(isALNUM_LC,
1543 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1544 isALNUM_LC_utf8((U8*)s));
1547 FBC_BOUND(isWORDCHAR,
1549 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1552 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1554 isWORDCHAR_A((U8*)s));
1557 FBC_NBOUND(isWORDCHAR,
1559 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1562 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1564 isWORDCHAR_A((U8*)s));
1567 FBC_BOUND(isWORDCHAR_L1,
1569 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1572 FBC_NBOUND(isWORDCHAR_L1,
1574 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1577 REXEC_FBC_CSCAN_TAINT(
1578 isALNUM_LC_utf8((U8*)s),
1583 REXEC_FBC_CSCAN_PRELOAD(
1584 LOAD_UTF8_CHARCLASS_ALNUM(),
1585 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1586 isWORDCHAR_L1((U8) *s)
1590 REXEC_FBC_CSCAN_PRELOAD(
1591 LOAD_UTF8_CHARCLASS_ALNUM(),
1592 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1597 /* Don't need to worry about utf8, as it can match only a single
1598 * byte invariant character */
1599 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1602 REXEC_FBC_CSCAN_PRELOAD(
1603 LOAD_UTF8_CHARCLASS_ALNUM(),
1604 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1605 ! isWORDCHAR_L1((U8) *s)
1609 REXEC_FBC_CSCAN_PRELOAD(
1610 LOAD_UTF8_CHARCLASS_ALNUM(),
1611 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1622 REXEC_FBC_CSCAN_TAINT(
1623 !isALNUM_LC_utf8((U8*)s),
1628 REXEC_FBC_CSCAN_PRELOAD(
1629 LOAD_UTF8_CHARCLASS_SPACE(),
1630 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1635 REXEC_FBC_CSCAN_PRELOAD(
1636 LOAD_UTF8_CHARCLASS_SPACE(),
1637 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1642 /* Don't need to worry about utf8, as it can match only a single
1643 * byte invariant character */
1644 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1647 REXEC_FBC_CSCAN_TAINT(
1648 isSPACE_LC_utf8((U8*)s),
1653 REXEC_FBC_CSCAN_PRELOAD(
1654 LOAD_UTF8_CHARCLASS_SPACE(),
1655 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1656 ! isSPACE_L1((U8) *s)
1660 REXEC_FBC_CSCAN_PRELOAD(
1661 LOAD_UTF8_CHARCLASS_SPACE(),
1662 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1673 REXEC_FBC_CSCAN_TAINT(
1674 !isSPACE_LC_utf8((U8*)s),
1679 REXEC_FBC_CSCAN_PRELOAD(
1680 LOAD_UTF8_CHARCLASS_DIGIT(),
1681 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1686 /* Don't need to worry about utf8, as it can match only a single
1687 * byte invariant character */
1688 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1691 REXEC_FBC_CSCAN_TAINT(
1692 isDIGIT_LC_utf8((U8*)s),
1697 REXEC_FBC_CSCAN_PRELOAD(
1698 LOAD_UTF8_CHARCLASS_DIGIT(),
1699 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1710 REXEC_FBC_CSCAN_TAINT(
1711 !isDIGIT_LC_utf8((U8*)s),
1718 is_LNBREAK_latin1(s)
1730 !is_VERTWS_latin1(s)
1736 is_HORIZWS_latin1(s)
1741 !is_HORIZWS_utf8(s),
1742 !is_HORIZWS_latin1(s)
1749 /* what trie are we using right now */
1751 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1753 = (reg_trie_data*)progi->data->data[ aho->trie ];
1754 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1756 const char *last_start = strend - trie->minlen;
1758 const char *real_start = s;
1760 STRLEN maxlen = trie->maxlen;
1762 U8 **points; /* map of where we were in the input string
1763 when reading a given char. For ASCII this
1764 is unnecessary overhead as the relationship
1765 is always 1:1, but for Unicode, especially
1766 case folded Unicode this is not true. */
1767 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1771 GET_RE_DEBUG_FLAGS_DECL;
1773 /* We can't just allocate points here. We need to wrap it in
1774 * an SV so it gets freed properly if there is a croak while
1775 * running the match */
1778 sv_points=newSV(maxlen * sizeof(U8 *));
1779 SvCUR_set(sv_points,
1780 maxlen * sizeof(U8 *));
1781 SvPOK_on(sv_points);
1782 sv_2mortal(sv_points);
1783 points=(U8**)SvPV_nolen(sv_points );
1784 if ( trie_type != trie_utf8_fold
1785 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1788 bitmap=(U8*)trie->bitmap;
1790 bitmap=(U8*)ANYOF_BITMAP(c);
1792 /* this is the Aho-Corasick algorithm modified a touch
1793 to include special handling for long "unknown char"
1794 sequences. The basic idea being that we use AC as long
1795 as we are dealing with a possible matching char, when
1796 we encounter an unknown char (and we have not encountered
1797 an accepting state) we scan forward until we find a legal
1799 AC matching is basically that of trie matching, except
1800 that when we encounter a failing transition, we fall back
1801 to the current states "fail state", and try the current char
1802 again, a process we repeat until we reach the root state,
1803 state 1, or a legal transition. If we fail on the root state
1804 then we can either terminate if we have reached an accepting
1805 state previously, or restart the entire process from the beginning
1809 while (s <= last_start) {
1810 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1818 U8 *uscan = (U8*)NULL;
1819 U8 *leftmost = NULL;
1821 U32 accepted_word= 0;
1825 while ( state && uc <= (U8*)strend ) {
1827 U32 word = aho->states[ state ].wordnum;
1831 DEBUG_TRIE_EXECUTE_r(
1832 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1833 dump_exec_pos( (char *)uc, c, strend, real_start,
1834 (char *)uc, utf8_target );
1835 PerlIO_printf( Perl_debug_log,
1836 " Scanning for legal start char...\n");
1840 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1844 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1850 if (uc >(U8*)last_start) break;
1854 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1855 if (!leftmost || lpos < leftmost) {
1856 DEBUG_r(accepted_word=word);
1862 points[pointpos++ % maxlen]= uc;
1863 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1864 uscan, len, uvc, charid, foldlen,
1866 DEBUG_TRIE_EXECUTE_r({
1867 dump_exec_pos( (char *)uc, c, strend, real_start,
1869 PerlIO_printf(Perl_debug_log,
1870 " Charid:%3u CP:%4"UVxf" ",
1876 word = aho->states[ state ].wordnum;
1878 base = aho->states[ state ].trans.base;
1880 DEBUG_TRIE_EXECUTE_r({
1882 dump_exec_pos( (char *)uc, c, strend, real_start,
1884 PerlIO_printf( Perl_debug_log,
1885 "%sState: %4"UVxf", word=%"UVxf,
1886 failed ? " Fail transition to " : "",
1887 (UV)state, (UV)word);
1893 ( ((offset = base + charid
1894 - 1 - trie->uniquecharcount)) >= 0)
1895 && ((U32)offset < trie->lasttrans)
1896 && trie->trans[offset].check == state
1897 && (tmp=trie->trans[offset].next))
1899 DEBUG_TRIE_EXECUTE_r(
1900 PerlIO_printf( Perl_debug_log," - legal\n"));
1905 DEBUG_TRIE_EXECUTE_r(
1906 PerlIO_printf( Perl_debug_log," - fail\n"));
1908 state = aho->fail[state];
1912 /* we must be accepting here */
1913 DEBUG_TRIE_EXECUTE_r(
1914 PerlIO_printf( Perl_debug_log," - accepting\n"));
1923 if (!state) state = 1;
1926 if ( aho->states[ state ].wordnum ) {
1927 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1928 if (!leftmost || lpos < leftmost) {
1929 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1934 s = (char*)leftmost;
1935 DEBUG_TRIE_EXECUTE_r({
1937 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1938 (UV)accepted_word, (IV)(s - real_start)
1941 if (!reginfo || regtry(reginfo, &s)) {
1947 DEBUG_TRIE_EXECUTE_r({
1948 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1951 DEBUG_TRIE_EXECUTE_r(
1952 PerlIO_printf( Perl_debug_log,"No match.\n"));
1961 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1971 - regexec_flags - match a regexp against a string
1974 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1975 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1976 /* strend: pointer to null at end of string */
1977 /* strbeg: real beginning of string */
1978 /* minend: end of match must be >=minend after stringarg. */
1979 /* data: May be used for some additional optimizations.
1980 Currently its only used, with a U32 cast, for transmitting
1981 the ganch offset when doing a /g match. This will change */
1982 /* nosave: For optimizations. */
1985 struct regexp *const prog = (struct regexp *)SvANY(rx);
1986 /*register*/ char *s;
1987 register regnode *c;
1988 /*register*/ char *startpos = stringarg;
1989 I32 minlen; /* must match at least this many chars */
1990 I32 dontbother = 0; /* how many characters not to try at end */
1991 I32 end_shift = 0; /* Same for the end. */ /* CC */
1992 I32 scream_pos = -1; /* Internal iterator of scream. */
1993 char *scream_olds = NULL;
1994 const bool utf8_target = cBOOL(DO_UTF8(sv));
1996 RXi_GET_DECL(prog,progi);
1997 regmatch_info reginfo; /* create some info to pass to regtry etc */
1998 regexp_paren_pair *swap = NULL;
1999 GET_RE_DEBUG_FLAGS_DECL;
2001 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2002 PERL_UNUSED_ARG(data);
2004 /* Be paranoid... */
2005 if (prog == NULL || startpos == NULL) {
2006 Perl_croak(aTHX_ "NULL regexp parameter");
2010 multiline = prog->extflags & RXf_PMf_MULTILINE;
2011 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2013 RX_MATCH_UTF8_set(rx, utf8_target);
2015 debug_start_match(rx, utf8_target, startpos, strend,
2019 minlen = prog->minlen;
2021 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2022 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2023 "String too short [regexec_flags]...\n"));
2028 /* Check validity of program. */
2029 if (UCHARAT(progi->program) != REG_MAGIC) {
2030 Perl_croak(aTHX_ "corrupted regexp program");
2034 PL_reg_eval_set = 0;
2038 PL_reg_flags |= RF_utf8;
2040 /* Mark beginning of line for ^ and lookbehind. */
2041 reginfo.bol = startpos; /* XXX not used ??? */
2045 /* Mark end of line for $ (and such) */
2048 /* see how far we have to get to not match where we matched before */
2049 reginfo.till = startpos+minend;
2051 /* If there is a "must appear" string, look for it. */
2054 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2056 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2057 reginfo.ganch = startpos + prog->gofs;
2058 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2059 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2060 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2062 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2063 && mg->mg_len >= 0) {
2064 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2065 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2066 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2068 if (prog->extflags & RXf_ANCH_GPOS) {
2069 if (s > reginfo.ganch)
2071 s = reginfo.ganch - prog->gofs;
2072 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2073 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2079 reginfo.ganch = strbeg + PTR2UV(data);
2080 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2081 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2083 } else { /* pos() not defined */
2084 reginfo.ganch = strbeg;
2085 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2086 "GPOS: reginfo.ganch = strbeg\n"));
2089 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2090 /* We have to be careful. If the previous successful match
2091 was from this regex we don't want a subsequent partially
2092 successful match to clobber the old results.
2093 So when we detect this possibility we add a swap buffer
2094 to the re, and switch the buffer each match. If we fail
2095 we switch it back, otherwise we leave it swapped.
2098 /* do we need a save destructor here for eval dies? */
2099 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2101 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2102 re_scream_pos_data d;
2104 d.scream_olds = &scream_olds;
2105 d.scream_pos = &scream_pos;
2106 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2108 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2109 goto phooey; /* not present */
2115 /* Simplest case: anchored match need be tried only once. */
2116 /* [unless only anchor is BOL and multiline is set] */
2117 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2118 if (s == startpos && regtry(®info, &startpos))
2120 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2121 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2126 dontbother = minlen - 1;
2127 end = HOP3c(strend, -dontbother, strbeg) - 1;
2128 /* for multiline we only have to try after newlines */
2129 if (prog->check_substr || prog->check_utf8) {
2130 /* because of the goto we can not easily reuse the macros for bifurcating the
2131 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2134 goto after_try_utf8;
2136 if (regtry(®info, &s)) {
2143 if (prog->extflags & RXf_USE_INTUIT) {
2144 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2153 } /* end search for check string in unicode */
2155 if (s == startpos) {
2156 goto after_try_latin;
2159 if (regtry(®info, &s)) {
2166 if (prog->extflags & RXf_USE_INTUIT) {
2167 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2176 } /* end search for check string in latin*/
2177 } /* end search for check string */
2178 else { /* search for newline */
2180 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2183 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2185 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2186 if (regtry(®info, &s))
2190 } /* end search for newline */
2191 } /* end anchored/multiline check string search */
2193 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2195 /* the warning about reginfo.ganch being used without initialization
2196 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2197 and we only enter this block when the same bit is set. */
2198 char *tmp_s = reginfo.ganch - prog->gofs;
2200 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2205 /* Messy cases: unanchored match. */
2206 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2207 /* we have /x+whatever/ */
2208 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2213 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2214 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2215 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2220 DEBUG_EXECUTE_r( did_match = 1 );
2221 if (regtry(®info, &s)) goto got_it;
2223 while (s < strend && *s == ch)
2231 DEBUG_EXECUTE_r( did_match = 1 );
2232 if (regtry(®info, &s)) goto got_it;
2234 while (s < strend && *s == ch)
2239 DEBUG_EXECUTE_r(if (!did_match)
2240 PerlIO_printf(Perl_debug_log,
2241 "Did not find anchored character...\n")
2244 else if (prog->anchored_substr != NULL
2245 || prog->anchored_utf8 != NULL
2246 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2247 && prog->float_max_offset < strend - s)) {
2252 char *last1; /* Last position checked before */
2256 if (prog->anchored_substr || prog->anchored_utf8) {
2257 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2258 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2259 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2260 back_max = back_min = prog->anchored_offset;
2262 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2263 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2264 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2265 back_max = prog->float_max_offset;
2266 back_min = prog->float_min_offset;
2270 if (must == &PL_sv_undef)
2271 /* could not downgrade utf8 check substring, so must fail */
2277 last = HOP3c(strend, /* Cannot start after this */
2278 -(I32)(CHR_SVLEN(must)
2279 - (SvTAIL(must) != 0) + back_min), strbeg);
2282 last1 = HOPc(s, -1);
2284 last1 = s - 1; /* bogus */
2286 /* XXXX check_substr already used to find "s", can optimize if
2287 check_substr==must. */
2289 dontbother = end_shift;
2290 strend = HOPc(strend, -dontbother);
2291 while ( (s <= last) &&
2292 ((flags & REXEC_SCREAM)
2293 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2294 end_shift, &scream_pos, 0))
2295 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2296 (unsigned char*)strend, must,
2297 multiline ? FBMrf_MULTILINE : 0))) ) {
2298 /* we may be pointing at the wrong string */
2299 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2300 s = strbeg + (s - SvPVX_const(sv));
2301 DEBUG_EXECUTE_r( did_match = 1 );
2302 if (HOPc(s, -back_max) > last1) {
2303 last1 = HOPc(s, -back_min);
2304 s = HOPc(s, -back_max);
2307 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2309 last1 = HOPc(s, -back_min);
2313 while (s <= last1) {
2314 if (regtry(®info, &s))
2320 while (s <= last1) {
2321 if (regtry(®info, &s))
2327 DEBUG_EXECUTE_r(if (!did_match) {
2328 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2329 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2330 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2331 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2332 ? "anchored" : "floating"),
2333 quoted, RE_SV_TAIL(must));
2337 else if ( (c = progi->regstclass) ) {
2339 const OPCODE op = OP(progi->regstclass);
2340 /* don't bother with what can't match */
2341 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2342 strend = HOPc(strend, -(minlen - 1));
2345 SV * const prop = sv_newmortal();
2346 regprop(prog, prop, c);
2348 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2350 PerlIO_printf(Perl_debug_log,
2351 "Matching stclass %.*s against %s (%d bytes)\n",
2352 (int)SvCUR(prop), SvPVX_const(prop),
2353 quoted, (int)(strend - s));
2356 if (find_byclass(prog, c, s, strend, ®info))
2358 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2362 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2367 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2368 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2369 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2371 if (flags & REXEC_SCREAM) {
2372 last = screaminstr(sv, float_real, s - strbeg,
2373 end_shift, &scream_pos, 1); /* last one */
2375 last = scream_olds; /* Only one occurrence. */
2376 /* we may be pointing at the wrong string */
2377 else if (RXp_MATCH_COPIED(prog))
2378 s = strbeg + (s - SvPVX_const(sv));
2382 const char * const little = SvPV_const(float_real, len);
2384 if (SvTAIL(float_real)) {
2385 if (memEQ(strend - len + 1, little, len - 1))
2386 last = strend - len + 1;
2387 else if (!multiline)
2388 last = memEQ(strend - len, little, len)
2389 ? strend - len : NULL;
2395 last = rninstr(s, strend, little, little + len);
2397 last = strend; /* matching "$" */
2402 PerlIO_printf(Perl_debug_log,
2403 "%sCan't trim the tail, match fails (should not happen)%s\n",
2404 PL_colors[4], PL_colors[5]));
2405 goto phooey; /* Should not happen! */
2407 dontbother = strend - last + prog->float_min_offset;
2409 if (minlen && (dontbother < minlen))
2410 dontbother = minlen - 1;
2411 strend -= dontbother; /* this one's always in bytes! */
2412 /* We don't know much -- general case. */
2415 if (regtry(®info, &s))
2424 if (regtry(®info, &s))
2426 } while (s++ < strend);
2435 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2437 if (PL_reg_eval_set)
2438 restore_pos(aTHX_ prog);
2439 if (RXp_PAREN_NAMES(prog))
2440 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2442 /* make sure $`, $&, $', and $digit will work later */
2443 if ( !(flags & REXEC_NOT_FIRST) ) {
2444 RX_MATCH_COPY_FREE(rx);
2445 if (flags & REXEC_COPY_STR) {
2446 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2447 #ifdef PERL_OLD_COPY_ON_WRITE
2449 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2451 PerlIO_printf(Perl_debug_log,
2452 "Copy on write: regexp capture, type %d\n",
2455 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2456 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2457 assert (SvPOKp(prog->saved_copy));
2461 RX_MATCH_COPIED_on(rx);
2462 s = savepvn(strbeg, i);
2468 prog->subbeg = strbeg;
2469 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2476 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2477 PL_colors[4], PL_colors[5]));
2478 if (PL_reg_eval_set)
2479 restore_pos(aTHX_ prog);
2481 /* we failed :-( roll it back */
2482 Safefree(prog->offs);
2491 - regtry - try match at specific point
2493 STATIC I32 /* 0 failure, 1 success */
2494 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2498 REGEXP *const rx = reginfo->prog;
2499 regexp *const prog = (struct regexp *)SvANY(rx);
2500 RXi_GET_DECL(prog,progi);
2501 GET_RE_DEBUG_FLAGS_DECL;
2503 PERL_ARGS_ASSERT_REGTRY;
2505 reginfo->cutpoint=NULL;
2507 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2510 PL_reg_eval_set = RS_init;
2511 DEBUG_EXECUTE_r(DEBUG_s(
2512 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2513 (IV)(PL_stack_sp - PL_stack_base));
2516 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2517 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2519 /* Apparently this is not needed, judging by wantarray. */
2520 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2521 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2524 /* Make $_ available to executed code. */
2525 if (reginfo->sv != DEFSV) {
2527 DEFSV_set(reginfo->sv);
2530 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2531 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2532 /* prepare for quick setting of pos */
2533 #ifdef PERL_OLD_COPY_ON_WRITE
2534 if (SvIsCOW(reginfo->sv))
2535 sv_force_normal_flags(reginfo->sv, 0);
2537 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2538 &PL_vtbl_mglob, NULL, 0);
2542 PL_reg_oldpos = mg->mg_len;
2543 SAVEDESTRUCTOR_X(restore_pos, prog);
2545 if (!PL_reg_curpm) {
2546 Newxz(PL_reg_curpm, 1, PMOP);
2549 SV* const repointer = &PL_sv_undef;
2550 /* this regexp is also owned by the new PL_reg_curpm, which
2551 will try to free it. */
2552 av_push(PL_regex_padav, repointer);
2553 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2554 PL_regex_pad = AvARRAY(PL_regex_padav);
2559 /* It seems that non-ithreads works both with and without this code.
2560 So for efficiency reasons it seems best not to have the code
2561 compiled when it is not needed. */
2562 /* This is safe against NULLs: */
2563 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2564 /* PM_reg_curpm owns a reference to this regexp. */
2565 (void)ReREFCNT_inc(rx);
2567 PM_SETRE(PL_reg_curpm, rx);
2568 PL_reg_oldcurpm = PL_curpm;
2569 PL_curpm = PL_reg_curpm;
2570 if (RXp_MATCH_COPIED(prog)) {
2571 /* Here is a serious problem: we cannot rewrite subbeg,
2572 since it may be needed if this match fails. Thus
2573 $` inside (?{}) could fail... */
2574 PL_reg_oldsaved = prog->subbeg;
2575 PL_reg_oldsavedlen = prog->sublen;
2576 #ifdef PERL_OLD_COPY_ON_WRITE
2577 PL_nrs = prog->saved_copy;
2579 RXp_MATCH_COPIED_off(prog);
2582 PL_reg_oldsaved = NULL;
2583 prog->subbeg = PL_bostr;
2584 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2586 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2587 prog->offs[0].start = *startpos - PL_bostr;
2588 PL_reginput = *startpos;
2589 PL_reglastparen = &prog->lastparen;
2590 PL_reglastcloseparen = &prog->lastcloseparen;
2591 prog->lastparen = 0;
2592 prog->lastcloseparen = 0;
2594 PL_regoffs = prog->offs;
2595 if (PL_reg_start_tmpl <= prog->nparens) {
2596 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2597 if(PL_reg_start_tmp)
2598 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2600 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2603 /* XXXX What this code is doing here?!!! There should be no need
2604 to do this again and again, PL_reglastparen should take care of
2607 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2608 * Actually, the code in regcppop() (which Ilya may be meaning by
2609 * PL_reglastparen), is not needed at all by the test suite
2610 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2611 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2612 * Meanwhile, this code *is* needed for the
2613 * above-mentioned test suite tests to succeed. The common theme
2614 * on those tests seems to be returning null fields from matches.
2615 * --jhi updated by dapm */
2617 if (prog->nparens) {
2618 regexp_paren_pair *pp = PL_regoffs;
2620 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2628 if (regmatch(reginfo, progi->program + 1)) {
2629 PL_regoffs[0].end = PL_reginput - PL_bostr;
2632 if (reginfo->cutpoint)
2633 *startpos= reginfo->cutpoint;
2634 REGCP_UNWIND(lastcp);
2639 #define sayYES goto yes
2640 #define sayNO goto no
2641 #define sayNO_SILENT goto no_silent
2643 /* we dont use STMT_START/END here because it leads to
2644 "unreachable code" warnings, which are bogus, but distracting. */
2645 #define CACHEsayNO \
2646 if (ST.cache_mask) \
2647 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2650 /* this is used to determine how far from the left messages like
2651 'failed...' are printed. It should be set such that messages
2652 are inline with the regop output that created them.
2654 #define REPORT_CODE_OFF 32
2657 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2658 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2660 #define SLAB_FIRST(s) (&(s)->states[0])
2661 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2663 /* grab a new slab and return the first slot in it */
2665 STATIC regmatch_state *
2668 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2671 regmatch_slab *s = PL_regmatch_slab->next;
2673 Newx(s, 1, regmatch_slab);
2674 s->prev = PL_regmatch_slab;
2676 PL_regmatch_slab->next = s;
2678 PL_regmatch_slab = s;
2679 return SLAB_FIRST(s);
2683 /* push a new state then goto it */
2685 #define PUSH_STATE_GOTO(state, node) \
2687 st->resume_state = state; \
2690 /* push a new state with success backtracking, then goto it */
2692 #define PUSH_YES_STATE_GOTO(state, node) \
2694 st->resume_state = state; \
2695 goto push_yes_state;
2701 regmatch() - main matching routine
2703 This is basically one big switch statement in a loop. We execute an op,
2704 set 'next' to point the next op, and continue. If we come to a point which
2705 we may need to backtrack to on failure such as (A|B|C), we push a
2706 backtrack state onto the backtrack stack. On failure, we pop the top
2707 state, and re-enter the loop at the state indicated. If there are no more
2708 states to pop, we return failure.
2710 Sometimes we also need to backtrack on success; for example /A+/, where
2711 after successfully matching one A, we need to go back and try to
2712 match another one; similarly for lookahead assertions: if the assertion
2713 completes successfully, we backtrack to the state just before the assertion
2714 and then carry on. In these cases, the pushed state is marked as
2715 'backtrack on success too'. This marking is in fact done by a chain of
2716 pointers, each pointing to the previous 'yes' state. On success, we pop to
2717 the nearest yes state, discarding any intermediate failure-only states.
2718 Sometimes a yes state is pushed just to force some cleanup code to be
2719 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2720 it to free the inner regex.
2722 Note that failure backtracking rewinds the cursor position, while
2723 success backtracking leaves it alone.
2725 A pattern is complete when the END op is executed, while a subpattern
2726 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2727 ops trigger the "pop to last yes state if any, otherwise return true"
2730 A common convention in this function is to use A and B to refer to the two
2731 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2732 the subpattern to be matched possibly multiple times, while B is the entire
2733 rest of the pattern. Variable and state names reflect this convention.
2735 The states in the main switch are the union of ops and failure/success of
2736 substates associated with with that op. For example, IFMATCH is the op
2737 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2738 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2739 successfully matched A and IFMATCH_A_fail is a state saying that we have
2740 just failed to match A. Resume states always come in pairs. The backtrack
2741 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2742 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2743 on success or failure.
2745 The struct that holds a backtracking state is actually a big union, with
2746 one variant for each major type of op. The variable st points to the
2747 top-most backtrack struct. To make the code clearer, within each
2748 block of code we #define ST to alias the relevant union.
2750 Here's a concrete example of a (vastly oversimplified) IFMATCH
2756 #define ST st->u.ifmatch
2758 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2759 ST.foo = ...; // some state we wish to save
2761 // push a yes backtrack state with a resume value of
2762 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2764 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2767 case IFMATCH_A: // we have successfully executed A; now continue with B
2769 bar = ST.foo; // do something with the preserved value
2772 case IFMATCH_A_fail: // A failed, so the assertion failed
2773 ...; // do some housekeeping, then ...
2774 sayNO; // propagate the failure
2781 For any old-timers reading this who are familiar with the old recursive
2782 approach, the code above is equivalent to:
2784 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2793 ...; // do some housekeeping, then ...
2794 sayNO; // propagate the failure
2797 The topmost backtrack state, pointed to by st, is usually free. If you
2798 want to claim it, populate any ST.foo fields in it with values you wish to
2799 save, then do one of
2801 PUSH_STATE_GOTO(resume_state, node);
2802 PUSH_YES_STATE_GOTO(resume_state, node);
2804 which sets that backtrack state's resume value to 'resume_state', pushes a
2805 new free entry to the top of the backtrack stack, then goes to 'node'.
2806 On backtracking, the free slot is popped, and the saved state becomes the
2807 new free state. An ST.foo field in this new top state can be temporarily
2808 accessed to retrieve values, but once the main loop is re-entered, it
2809 becomes available for reuse.
2811 Note that the depth of the backtrack stack constantly increases during the
2812 left-to-right execution of the pattern, rather than going up and down with
2813 the pattern nesting. For example the stack is at its maximum at Z at the
2814 end of the pattern, rather than at X in the following:
2816 /(((X)+)+)+....(Y)+....Z/
2818 The only exceptions to this are lookahead/behind assertions and the cut,
2819 (?>A), which pop all the backtrack states associated with A before
2822 Backtrack state structs are allocated in slabs of about 4K in size.
2823 PL_regmatch_state and st always point to the currently active state,
2824 and PL_regmatch_slab points to the slab currently containing
2825 PL_regmatch_state. The first time regmatch() is called, the first slab is
2826 allocated, and is never freed until interpreter destruction. When the slab
2827 is full, a new one is allocated and chained to the end. At exit from
2828 regmatch(), slabs allocated since entry are freed.
2833 #define DEBUG_STATE_pp(pp) \
2835 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2836 PerlIO_printf(Perl_debug_log, \
2837 " %*s"pp" %s%s%s%s%s\n", \
2839 PL_reg_name[st->resume_state], \
2840 ((st==yes_state||st==mark_state) ? "[" : ""), \
2841 ((st==yes_state) ? "Y" : ""), \
2842 ((st==mark_state) ? "M" : ""), \
2843 ((st==yes_state||st==mark_state) ? "]" : "") \
2848 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2853 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2854 const char *start, const char *end, const char *blurb)
2856 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2858 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2863 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2864 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2866 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2867 start, end - start, 60);
2869 PerlIO_printf(Perl_debug_log,
2870 "%s%s REx%s %s against %s\n",
2871 PL_colors[4], blurb, PL_colors[5], s0, s1);
2873 if (utf8_target||utf8_pat)
2874 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2875 utf8_pat ? "pattern" : "",
2876 utf8_pat && utf8_target ? " and " : "",
2877 utf8_target ? "string" : ""
2883 S_dump_exec_pos(pTHX_ const char *locinput,
2884 const regnode *scan,
2885 const char *loc_regeol,
2886 const char *loc_bostr,
2887 const char *loc_reg_starttry,
2888 const bool utf8_target)
2890 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2891 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2892 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2893 /* The part of the string before starttry has one color
2894 (pref0_len chars), between starttry and current
2895 position another one (pref_len - pref0_len chars),
2896 after the current position the third one.
2897 We assume that pref0_len <= pref_len, otherwise we
2898 decrease pref0_len. */
2899 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2900 ? (5 + taill) - l : locinput - loc_bostr;
2903 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2905 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2907 pref0_len = pref_len - (locinput - loc_reg_starttry);
2908 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2909 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2910 ? (5 + taill) - pref_len : loc_regeol - locinput);
2911 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2915 if (pref0_len > pref_len)
2916 pref0_len = pref_len;
2918 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2920 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2921 (locinput - pref_len),pref0_len, 60, 4, 5);
2923 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2924 (locinput - pref_len + pref0_len),
2925 pref_len - pref0_len, 60, 2, 3);
2927 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2928 locinput, loc_regeol - locinput, 10, 0, 1);
2930 const STRLEN tlen=len0+len1+len2;
2931 PerlIO_printf(Perl_debug_log,
2932 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2933 (IV)(locinput - loc_bostr),
2936 (docolor ? "" : "> <"),
2938 (int)(tlen > 19 ? 0 : 19 - tlen),
2945 /* reg_check_named_buff_matched()
2946 * Checks to see if a named buffer has matched. The data array of
2947 * buffer numbers corresponding to the buffer is expected to reside
2948 * in the regexp->data->data array in the slot stored in the ARG() of
2949 * node involved. Note that this routine doesn't actually care about the
2950 * name, that information is not preserved from compilation to execution.
2951 * Returns the index of the leftmost defined buffer with the given name
2952 * or 0 if non of the buffers matched.
2955 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2958 RXi_GET_DECL(rex,rexi);
2959 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2960 I32 *nums=(I32*)SvPVX(sv_dat);
2962 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2964 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2965 if ((I32)*PL_reglastparen >= nums[n] &&
2966 PL_regoffs[nums[n]].end != -1)
2975 /* free all slabs above current one - called during LEAVE_SCOPE */
2978 S_clear_backtrack_stack(pTHX_ void *p)
2980 regmatch_slab *s = PL_regmatch_slab->next;
2985 PL_regmatch_slab->next = NULL;
2987 regmatch_slab * const osl = s;
2994 #define SETREX(Re1,Re2) \
2995 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2998 STATIC I32 /* 0 failure, 1 success */
2999 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3001 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3005 register const bool utf8_target = PL_reg_match_utf8;
3006 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3007 REGEXP *rex_sv = reginfo->prog;
3008 regexp *rex = (struct regexp *)SvANY(rex_sv);
3009 RXi_GET_DECL(rex,rexi);
3011 /* the current state. This is a cached copy of PL_regmatch_state */
3012 register regmatch_state *st;
3013 /* cache heavy used fields of st in registers */
3014 register regnode *scan;
3015 register regnode *next;
3016 register U32 n = 0; /* general value; init to avoid compiler warning */
3017 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3018 register char *locinput = PL_reginput;
3019 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3021 bool result = 0; /* return value of S_regmatch */
3022 int depth = 0; /* depth of backtrack stack */
3023 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3024 const U32 max_nochange_depth =
3025 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3026 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3027 regmatch_state *yes_state = NULL; /* state to pop to on success of
3029 /* mark_state piggy backs on the yes_state logic so that when we unwind
3030 the stack on success we can update the mark_state as we go */
3031 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3032 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3033 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3035 bool no_final = 0; /* prevent failure from backtracking? */
3036 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3037 char *startpoint = PL_reginput;
3038 SV *popmark = NULL; /* are we looking for a mark? */
3039 SV *sv_commit = NULL; /* last mark name seen in failure */
3040 SV *sv_yes_mark = NULL; /* last mark name we have seen
3041 during a successful match */
3042 U32 lastopen = 0; /* last open we saw */
3043 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3044 SV* const oreplsv = GvSV(PL_replgv);
3045 /* these three flags are set by various ops to signal information to
3046 * the very next op. They have a useful lifetime of exactly one loop
3047 * iteration, and are not preserved or restored by state pushes/pops
3049 bool sw = 0; /* the condition value in (?(cond)a|b) */
3050 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3051 int logical = 0; /* the following EVAL is:
3055 or the following IFMATCH/UNLESSM is:
3056 false: plain (?=foo)
3057 true: used as a condition: (?(?=foo))
3060 GET_RE_DEBUG_FLAGS_DECL;
3063 PERL_ARGS_ASSERT_REGMATCH;
3065 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3066 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3068 /* on first ever call to regmatch, allocate first slab */
3069 if (!PL_regmatch_slab) {
3070 Newx(PL_regmatch_slab, 1, regmatch_slab);
3071 PL_regmatch_slab->prev = NULL;
3072 PL_regmatch_slab->next = NULL;
3073 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3076 oldsave = PL_savestack_ix;
3077 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3078 SAVEVPTR(PL_regmatch_slab);
3079 SAVEVPTR(PL_regmatch_state);
3081 /* grab next free state slot */
3082 st = ++PL_regmatch_state;
3083 if (st > SLAB_LAST(PL_regmatch_slab))
3084 st = PL_regmatch_state = S_push_slab(aTHX);
3086 /* Note that nextchr is a byte even in UTF */
3087 nextchr = UCHARAT(locinput);
3089 while (scan != NULL) {
3092 SV * const prop = sv_newmortal();
3093 regnode *rnext=regnext(scan);
3094 DUMP_EXEC_POS( locinput, scan, utf8_target );
3095 regprop(rex, prop, scan);
3097 PerlIO_printf(Perl_debug_log,
3098 "%3"IVdf":%*s%s(%"IVdf")\n",
3099 (IV)(scan - rexi->program), depth*2, "",
3101 (PL_regkind[OP(scan)] == END || !rnext) ?
3102 0 : (IV)(rnext - rexi->program));
3105 next = scan + NEXT_OFF(scan);
3108 state_num = OP(scan);
3112 assert(PL_reglastparen == &rex->lastparen);
3113 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3114 assert(PL_regoffs == rex->offs);
3116 switch (state_num) {
3118 if (locinput == PL_bostr)
3120 /* reginfo->till = reginfo->bol; */
3125 if (locinput == PL_bostr ||
3126 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3132 if (locinput == PL_bostr)
3136 if (locinput == reginfo->ganch)
3141 /* update the startpoint */
3142 st->u.keeper.val = PL_regoffs[0].start;
3143 PL_reginput = locinput;
3144 PL_regoffs[0].start = locinput - PL_bostr;
3145 PUSH_STATE_GOTO(KEEPS_next, next);
3147 case KEEPS_next_fail:
3148 /* rollback the start point change */
3149 PL_regoffs[0].start = st->u.keeper.val;
3155 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3160 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3162 if (PL_regeol - locinput > 1)
3166 if (PL_regeol != locinput)
3170 if (!nextchr && locinput >= PL_regeol)
3173 locinput += PL_utf8skip[nextchr];
3174 if (locinput > PL_regeol)
3176 nextchr = UCHARAT(locinput);
3179 nextchr = UCHARAT(++locinput);
3182 if (!nextchr && locinput >= PL_regeol)
3184 nextchr = UCHARAT(++locinput);
3187 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3190 locinput += PL_utf8skip[nextchr];
3191 if (locinput > PL_regeol)
3193 nextchr = UCHARAT(locinput);
3196 nextchr = UCHARAT(++locinput);
3200 #define ST st->u.trie
3202 /* In this case the charclass data is available inline so
3203 we can fail fast without a lot of extra overhead.
3205 if (scan->flags == EXACT || !utf8_target) {
3206 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3208 PerlIO_printf(Perl_debug_log,
3209 "%*s %sfailed to match trie start class...%s\n",
3210 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3218 /* the basic plan of execution of the trie is:
3219 * At the beginning, run though all the states, and
3220 * find the longest-matching word. Also remember the position
3221 * of the shortest matching word. For example, this pattern:
3224 * when matched against the string "abcde", will generate
3225 * accept states for all words except 3, with the longest
3226 * matching word being 4, and the shortest being 1 (with
3227 * the position being after char 1 of the string).
3229 * Then for each matching word, in word order (i.e. 1,2,4,5),
3230 * we run the remainder of the pattern; on each try setting
3231 * the current position to the character following the word,
3232 * returning to try the next word on failure.
3234 * We avoid having to build a list of words at runtime by
3235 * using a compile-time structure, wordinfo[].prev, which
3236 * gives, for each word, the previous accepting word (if any).
3237 * In the case above it would contain the mappings 1->2, 2->0,
3238 * 3->0, 4->5, 5->1. We can use this table to generate, from
3239 * the longest word (4 above), a list of all words, by
3240 * following the list of prev pointers; this gives us the
3241 * unordered list 4,5,1,2. Then given the current word we have
3242 * just tried, we can go through the list and find the
3243 * next-biggest word to try (so if we just failed on word 2,
3244 * the next in the list is 4).
3246 * Since at runtime we don't record the matching position in
3247 * the string for each word, we have to work that out for
3248 * each word we're about to process. The wordinfo table holds
3249 * the character length of each word; given that we recorded
3250 * at the start: the position of the shortest word and its
3251 * length in chars, we just need to move the pointer the
3252 * difference between the two char lengths. Depending on
3253 * Unicode status and folding, that's cheap or expensive.
3255 * This algorithm is optimised for the case where are only a
3256 * small number of accept states, i.e. 0,1, or maybe 2.
3257 * With lots of accepts states, and having to try all of them,
3258 * it becomes quadratic on number of accept states to find all
3263 /* what type of TRIE am I? (utf8 makes this contextual) */
3264 DECL_TRIE_TYPE(scan);
3266 /* what trie are we using right now */
3267 reg_trie_data * const trie
3268 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3269 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3270 U32 state = trie->startstate;
3272 if (trie->bitmap && trie_type != trie_utf8_fold &&
3273 !TRIE_BITMAP_TEST(trie,*locinput)
3275 if (trie->states[ state ].wordnum) {
3277 PerlIO_printf(Perl_debug_log,
3278 "%*s %smatched empty string...%s\n",
3279 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3285 PerlIO_printf(Perl_debug_log,
3286 "%*s %sfailed to match trie start class...%s\n",
3287 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3294 U8 *uc = ( U8* )locinput;
3298 U8 *uscan = (U8*)NULL;
3299 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3300 U32 charcount = 0; /* how many input chars we have matched */
3301 U32 accepted = 0; /* have we seen any accepting states? */
3304 ST.jump = trie->jump;
3307 ST.longfold = FALSE; /* char longer if folded => it's harder */
3310 /* fully traverse the TRIE; note the position of the
3311 shortest accept state and the wordnum of the longest
3314 while ( state && uc <= (U8*)PL_regeol ) {
3315 U32 base = trie->states[ state ].trans.base;
3319 wordnum = trie->states[ state ].wordnum;
3321 if (wordnum) { /* it's an accept state */
3324 /* record first match position */
3326 ST.firstpos = (U8*)locinput;
3331 ST.firstchars = charcount;
3334 if (!ST.nextword || wordnum < ST.nextword)
3335 ST.nextword = wordnum;
3336 ST.topword = wordnum;
3339 DEBUG_TRIE_EXECUTE_r({
3340 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3341 PerlIO_printf( Perl_debug_log,
3342 "%*s %sState: %4"UVxf" Accepted: %c ",
3343 2+depth * 2, "", PL_colors[4],
3344 (UV)state, (accepted ? 'Y' : 'N'));
3347 /* read a char and goto next state */
3350 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3351 uscan, len, uvc, charid, foldlen,
3358 base + charid - 1 - trie->uniquecharcount)) >= 0)
3360 && ((U32)offset < trie->lasttrans)
3361 && trie->trans[offset].check == state)
3363 state = trie->trans[offset].next;
3374 DEBUG_TRIE_EXECUTE_r(
3375 PerlIO_printf( Perl_debug_log,
3376 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3377 charid, uvc, (UV)state, PL_colors[5] );
3383 /* calculate total number of accept states */
3388 w = trie->wordinfo[w].prev;
3391 ST.accepted = accepted;
3395 PerlIO_printf( Perl_debug_log,
3396 "%*s %sgot %"IVdf" possible matches%s\n",
3397 REPORT_CODE_OFF + depth * 2, "",
3398 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3400 goto trie_first_try; /* jump into the fail handler */
3404 case TRIE_next_fail: /* we failed - try next alternative */
3406 REGCP_UNWIND(ST.cp);
3407 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3408 PL_regoffs[n].end = -1;
3409 *PL_reglastparen = n;
3411 if (!--ST.accepted) {
3413 PerlIO_printf( Perl_debug_log,
3414 "%*s %sTRIE failed...%s\n",
3415 REPORT_CODE_OFF+depth*2, "",
3422 /* Find next-highest word to process. Note that this code
3423 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3424 register U16 min = 0;
3426 register U16 const nextword = ST.nextword;
3427 register reg_trie_wordinfo * const wordinfo
3428 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3429 for (word=ST.topword; word; word=wordinfo[word].prev) {
3430 if (word > nextword && (!min || word < min))
3443 ST.lastparen = *PL_reglastparen;
3447 /* find start char of end of current word */
3449 U32 chars; /* how many chars to skip */
3450 U8 *uc = ST.firstpos;
3451 reg_trie_data * const trie
3452 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3454 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3456 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3460 /* the hard option - fold each char in turn and find
3461 * its folded length (which may be different */
3462 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3470 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3478 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3483 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3497 PL_reginput = (char *)uc;
3500 scan = (ST.jump && ST.jump[ST.nextword])
3501 ? ST.me + ST.jump[ST.nextword]
3505 PerlIO_printf( Perl_debug_log,
3506 "%*s %sTRIE matched word #%d, continuing%s\n",
3507 REPORT_CODE_OFF+depth*2, "",
3514 if (ST.accepted > 1 || has_cutgroup) {
3515 PUSH_STATE_GOTO(TRIE_next, scan);
3518 /* only one choice left - just continue */
3520 AV *const trie_words
3521 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3522 SV ** const tmp = av_fetch( trie_words,
3524 SV *sv= tmp ? sv_newmortal() : NULL;
3526 PerlIO_printf( Perl_debug_log,
3527 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3528 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3530 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3531 PL_colors[0], PL_colors[1],
3532 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3534 : "not compiled under -Dr",
3538 locinput = PL_reginput;
3539 nextchr = UCHARAT(locinput);
3540 continue; /* execute rest of RE */
3545 char *s = STRING(scan);
3547 if (utf8_target != UTF_PATTERN) {
3548 /* The target and the pattern have differing utf8ness. */
3550 const char * const e = s + ln;
3553 /* The target is utf8, the pattern is not utf8. */
3558 if (NATIVE_TO_UNI(*(U8*)s) !=
3559 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3567 /* The target is not utf8, the pattern is utf8. */
3572 if (NATIVE_TO_UNI(*((U8*)l)) !=
3573 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3581 nextchr = UCHARAT(locinput);
3584 /* The target and the pattern have the same utf8ness. */
3585 /* Inline the first character, for speed. */
3586 if (UCHARAT(s) != nextchr)
3588 if (PL_regeol - locinput < ln)
3590 if (ln > 1 && memNE(s, locinput, ln))
3593 nextchr = UCHARAT(locinput);
3598 const U8 * fold_array;
3600 U32 fold_utf8_flags;
3602 PL_reg_flags |= RF_tainted;
3603 folder = foldEQ_locale;
3604 fold_array = PL_fold_locale;
3605 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3609 folder = foldEQ_latin1;
3610 fold_array = PL_fold_latin1;
3611 fold_utf8_flags = 0;
3615 folder = foldEQ_latin1;
3616 fold_array = PL_fold_latin1;
3617 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3622 fold_array = PL_fold;
3623 fold_utf8_flags = 0;
3629 if (utf8_target || UTF_PATTERN) {
3630 /* Either target or the pattern are utf8. */
3631 const char * const l = locinput;
3632 char *e = PL_regeol;
3634 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3635 l, &e, 0, utf8_target, fold_utf8_flags))
3640 nextchr = UCHARAT(locinput);
3644 /* Neither the target nor the pattern are utf8 */
3645 if (UCHARAT(s) != nextchr &&
3646 UCHARAT(s) != fold_array[nextchr])
3650 if (PL_regeol - locinput < ln)
3652 if (ln > 1 && ! folder(s, locinput, ln))
3655 nextchr = UCHARAT(locinput);
3659 /* XXX Could improve efficiency by separating these all out using a
3660 * macro or in-line function. At that point regcomp.c would no longer
3661 * have to set the FLAGS fields of these */
3664 PL_reg_flags |= RF_tainted;
3672 /* was last char in word? */
3673 if (utf8_target && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET) {
3674 if (locinput == PL_bostr)
3677 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3679 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3681 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3682 ln = isALNUM_uni(ln);
3683 LOAD_UTF8_CHARCLASS_ALNUM();
3684 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3687 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3688 n = isALNUM_LC_utf8((U8*)locinput);
3693 /* Here the string isn't utf8, or is utf8 and only ascii
3694 * characters are to match \w. In the latter case looking at
3695 * the byte just prior to the current one may be just the final
3696 * byte of a multi-byte character. This is ok. There are two
3698 * 1) it is a single byte character, and then the test is doing
3699 * just what it's supposed to.
3700 * 2) it is a multi-byte character, in which case the final
3701 * byte is never mistakable for ASCII, and so the test
3702 * will say it is not a word character, which is the
3703 * correct answer. */
3704 ln = (locinput != PL_bostr) ?
3705 UCHARAT(locinput - 1) : '\n';
3706 switch (FLAGS(scan)) {
3707 case REGEX_UNICODE_CHARSET:
3708 ln = isWORDCHAR_L1(ln);
3709 n = isWORDCHAR_L1(nextchr);
3711 case REGEX_LOCALE_CHARSET:
3712 ln = isALNUM_LC(ln);
3713 n = isALNUM_LC(nextchr);
3715 case REGEX_DEPENDS_CHARSET:
3717 n = isALNUM(nextchr);
3719 case REGEX_ASCII_RESTRICTED_CHARSET:
3720 ln = isWORDCHAR_A(ln);
3721 n = isWORDCHAR_A(nextchr);
3724 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3728 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3730 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3735 if (utf8_target || state_num == ANYOFV) {
3736 STRLEN inclasslen = PL_regeol - locinput;
3737 if (locinput >= PL_regeol)
3740 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3742 locinput += inclasslen;
3743 nextchr = UCHARAT(locinput);
3748 nextchr = UCHARAT(locinput);
3749 if (!nextchr && locinput >= PL_regeol)
3751 if (!REGINCLASS(rex, scan, (U8*)locinput))
3753 nextchr = UCHARAT(++locinput);
3757 /* Special char classes - The defines start on line 129 or so */
3758 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3759 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3760 ALNUMU, NALNUMU, isWORDCHAR_L1,
3761 ALNUMA, NALNUMA, isWORDCHAR_A,
3764 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3765 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3766 SPACEU, NSPACEU, isSPACE_L1,
3767 SPACEA, NSPACEA, isSPACE_A,
3770 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3771 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3772 DIGITA, NDIGITA, isDIGIT_A,
3775 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3776 a Unicode extended Grapheme Cluster */
3777 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3778 extended Grapheme Cluster is:
3781 | Prepend* Begin Extend*
3784 Begin is (Hangul-syllable | ! Control)
3785 Extend is (Grapheme_Extend | Spacing_Mark)
3786 Control is [ GCB_Control CR LF ]
3788 The discussion below shows how the code for CLUMP is derived
3789 from this regex. Note that most of these concepts are from
3790 property values of the Grapheme Cluster Boundary (GCB) property.
3791 No code point can have multiple property values for a given
3792 property. Thus a code point in Prepend can't be in Control, but
3793 it must be in !Control. This is why Control above includes
3794 GCB_Control plus CR plus LF. The latter two are used in the GCB
3795 property separately, and so can't be in GCB_Control, even though
3796 they logically are controls. Control is not the same as gc=cc,
3797 but includes format and other characters as well.
3799 The Unicode definition of Hangul-syllable is:
3801 | (L* ( ( V | LV ) V* | LVT ) T*)
3804 Each of these is a value for the GCB property, and hence must be
3805 disjoint, so the order they are tested is immaterial, so the
3806 above can safely be changed to
3809 | (L* ( LVT | ( V | LV ) V*) T*)
3811 The last two terms can be combined like this:
3813 | (( LVT | ( V | LV ) V*) T*))
3815 And refactored into this:
3816 L* (L | LVT T* | V V* T* | LV V* T*)
3818 That means that if we have seen any L's at all we can quit
3819 there, but if the next character is a LVT, a V or and LV we
3822 There is a subtlety with Prepend* which showed up in testing.
3823 Note that the Begin, and only the Begin is required in:
3824 | Prepend* Begin Extend*
3825 Also, Begin contains '! Control'. A Prepend must be a '!
3826 Control', which means it must be a Begin. What it comes down to
3827 is that if we match Prepend* and then find no suitable Begin
3828 afterwards, that if we backtrack the last Prepend, that one will
3829 be a suitable Begin.
3832 if (locinput >= PL_regeol)
3834 if (! utf8_target) {
3836 /* Match either CR LF or '.', as all the other possibilities
3838 locinput++; /* Match the . or CR */
3840 && locinput < PL_regeol
3841 && UCHARAT(locinput) == '\n') locinput++;
3845 /* Utf8: See if is ( CR LF ); already know that locinput <
3846 * PL_regeol, so locinput+1 is in bounds */
3847 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3851 /* In case have to backtrack to beginning, then match '.' */
3852 char *starting = locinput;
3854 /* In case have to backtrack the last prepend */
3855 char *previous_prepend = 0;
3857 LOAD_UTF8_CHARCLASS_GCB();
3859 /* Match (prepend)* */
3860 while (locinput < PL_regeol
3861 && swash_fetch(PL_utf8_X_prepend,
3862 (U8*)locinput, utf8_target))
3864 previous_prepend = locinput;
3865 locinput += UTF8SKIP(locinput);
3868 /* As noted above, if we matched a prepend character, but
3869 * the next thing won't match, back off the last prepend we
3870 * matched, as it is guaranteed to match the begin */
3871 if (previous_prepend
3872 && (locinput >= PL_regeol
3873 || ! swash_fetch(PL_utf8_X_begin,
3874 (U8*)locinput, utf8_target)))
3876 locinput = previous_prepend;
3879 /* Note that here we know PL_regeol > locinput, as we
3880 * tested that upon input to this switch case, and if we
3881 * moved locinput forward, we tested the result just above
3882 * and it either passed, or we backed off so that it will
3884 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3886 /* Here did not match the required 'Begin' in the
3887 * second term. So just match the very first
3888 * character, the '.' of the final term of the regex */
3889 locinput = starting + UTF8SKIP(starting);
3892 /* Here is the beginning of a character that can have
3893 * an extender. It is either a hangul syllable, or a
3895 if (swash_fetch(PL_utf8_X_non_hangul,
3896 (U8*)locinput, utf8_target))
3899 /* Here not a Hangul syllable, must be a
3900 * ('! * Control') */
3901 locinput += UTF8SKIP(locinput);
3904 /* Here is a Hangul syllable. It can be composed
3905 * of several individual characters. One
3906 * possibility is T+ */
3907 if (swash_fetch(PL_utf8_X_T,
3908 (U8*)locinput, utf8_target))
3910 while (locinput < PL_regeol
3911 && swash_fetch(PL_utf8_X_T,
3912 (U8*)locinput, utf8_target))
3914 locinput += UTF8SKIP(locinput);
3918 /* Here, not T+, but is a Hangul. That means
3919 * it is one of the others: L, LV, LVT or V,
3921 * L* (L | LVT T* | V V* T* | LV V* T*) */
3924 while (locinput < PL_regeol
3925 && swash_fetch(PL_utf8_X_L,
3926 (U8*)locinput, utf8_target))
3928 locinput += UTF8SKIP(locinput);
3931 /* Here, have exhausted L*. If the next
3932 * character is not an LV, LVT nor V, it means
3933 * we had to have at least one L, so matches L+
3934 * in the original equation, we have a complete
3935 * hangul syllable. Are done. */
3937 if (locinput < PL_regeol
3938 && swash_fetch(PL_utf8_X_LV_LVT_V,
3939 (U8*)locinput, utf8_target))
3942 /* Otherwise keep going. Must be LV, LVT
3943 * or V. See if LVT */
3944 if (swash_fetch(PL_utf8_X_LVT,
3945 (U8*)locinput, utf8_target))
3947 locinput += UTF8SKIP(locinput);
3950 /* Must be V or LV. Take it, then
3952 locinput += UTF8SKIP(locinput);
3953 while (locinput < PL_regeol
3954 && swash_fetch(PL_utf8_X_V,
3955 (U8*)locinput, utf8_target))
3957 locinput += UTF8SKIP(locinput);
3961 /* And any of LV, LVT, or V can be followed
3963 while (locinput < PL_regeol
3964 && swash_fetch(PL_utf8_X_T,
3968 locinput += UTF8SKIP(locinput);
3974 /* Match any extender */
3975 while (locinput < PL_regeol
3976 && swash_fetch(PL_utf8_X_extend,
3977 (U8*)locinput, utf8_target))
3979 locinput += UTF8SKIP(locinput);
3983 if (locinput > PL_regeol) sayNO;
3985 nextchr = UCHARAT(locinput);
3989 { /* The capture buffer cases. The ones beginning with N for the
3990 named buffers just convert to the equivalent numbered and
3991 pretend they were called as the corresponding numbered buffer
3993 /* don't initialize these in the declaration, it makes C++
3998 const U8 *fold_array;
4001 PL_reg_flags |= RF_tainted;
4002 folder = foldEQ_locale;
4003 fold_array = PL_fold_locale;
4005 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4009 folder = foldEQ_latin1;
4010 fold_array = PL_fold_latin1;
4012 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4016 folder = foldEQ_latin1;
4017 fold_array = PL_fold_latin1;
4019 utf8_fold_flags = 0;
4024 fold_array = PL_fold;
4026 utf8_fold_flags = 0;
4033 utf8_fold_flags = 0;
4036 /* For the named back references, find the corresponding buffer
4038 n = reg_check_named_buff_matched(rex,scan);
4043 goto do_nref_ref_common;
4046 PL_reg_flags |= RF_tainted;
4047 folder = foldEQ_locale;
4048 fold_array = PL_fold_locale;
4049 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4053 folder = foldEQ_latin1;
4054 fold_array = PL_fold_latin1;
4055 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4059 folder = foldEQ_latin1;
4060 fold_array = PL_fold_latin1;
4061 utf8_fold_flags = 0;
4066 fold_array = PL_fold;
4067 utf8_fold_flags = 0;
4073 utf8_fold_flags = 0;
4077 n = ARG(scan); /* which paren pair */
4080 ln = PL_regoffs[n].start;
4081 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4082 if (*PL_reglastparen < n || ln == -1)
4083 sayNO; /* Do not match unless seen CLOSEn. */
4084 if (ln == PL_regoffs[n].end)
4088 if (type != REF /* REF can do byte comparison */
4089 && (utf8_target || type == REFFU))
4090 { /* XXX handle REFFL better */
4091 char * limit = PL_regeol;
4093 /* This call case insensitively compares the entire buffer
4094 * at s, with the current input starting at locinput, but
4095 * not going off the end given by PL_regeol, and returns in
4096 * limit upon success, how much of the current input was
4098 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4099 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4104 nextchr = UCHARAT(locinput);
4108 /* Not utf8: Inline the first character, for speed. */
4109 if (UCHARAT(s) != nextchr &&
4111 UCHARAT(s) != fold_array[nextchr]))
4113 ln = PL_regoffs[n].end - ln;
4114 if (locinput + ln > PL_regeol)
4116 if (ln > 1 && (type == REF
4117 ? memNE(s, locinput, ln)
4118 : ! folder(s, locinput, ln)))
4121 nextchr = UCHARAT(locinput);
4131 #define ST st->u.eval
4136 regexp_internal *rei;
4137 regnode *startpoint;
4140 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4141 if (cur_eval && cur_eval->locinput==locinput) {
4142 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4143 Perl_croak(aTHX_ "Infinite recursion in regex");
4144 if ( ++nochange_depth > max_nochange_depth )
4146 "Pattern subroutine nesting without pos change"
4147 " exceeded limit in regex");
4154 (void)ReREFCNT_inc(rex_sv);
4155 if (OP(scan)==GOSUB) {
4156 startpoint = scan + ARG2L(scan);
4157 ST.close_paren = ARG(scan);
4159 startpoint = rei->program+1;
4162 goto eval_recurse_doit;
4164 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4165 if (cur_eval && cur_eval->locinput==locinput) {
4166 if ( ++nochange_depth > max_nochange_depth )
4167 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4172 /* execute the code in the {...} */
4174 SV ** const before = SP;
4175 OP_4tree * const oop = PL_op;
4176 COP * const ocurcop = PL_curcop;
4178 char *saved_regeol = PL_regeol;
4179 struct re_save_state saved_state;
4181 /* To not corrupt the existing regex state while executing the
4182 * eval we would normally put it on the save stack, like with
4183 * save_re_context. However, re-evals have a weird scoping so we
4184 * can't just add ENTER/LEAVE here. With that, things like
4186 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4188 * would break, as they expect the localisation to be unwound
4189 * only when the re-engine backtracks through the bit that
4192 * What we do instead is just saving the state in a local c
4195 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4198 PL_op = (OP_4tree*)rexi->data->data[n];
4199 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4200 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4201 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4202 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4205 SV *sv_mrk = get_sv("REGMARK", 1);
4206 sv_setsv(sv_mrk, sv_yes_mark);
4209 CALLRUNOPS(aTHX); /* Scalar context. */
4212 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4218 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4221 PAD_RESTORE_LOCAL(old_comppad);
4222 PL_curcop = ocurcop;
4223 PL_regeol = saved_regeol;
4226 sv_setsv(save_scalar(PL_replgv), ret);
4230 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4233 /* extract RE object from returned value; compiling if
4239 SV *const sv = SvRV(ret);
4241 if (SvTYPE(sv) == SVt_REGEXP) {
4243 } else if (SvSMAGICAL(sv)) {
4244 mg = mg_find(sv, PERL_MAGIC_qr);
4247 } else if (SvTYPE(ret) == SVt_REGEXP) {
4249 } else if (SvSMAGICAL(ret)) {
4250 if (SvGMAGICAL(ret)) {
4251 /* I don't believe that there is ever qr magic
4253 assert(!mg_find(ret, PERL_MAGIC_qr));
4254 sv_unmagic(ret, PERL_MAGIC_qr);
4257 mg = mg_find(ret, PERL_MAGIC_qr);
4258 /* testing suggests mg only ends up non-NULL for
4259 scalars who were upgraded and compiled in the
4260 else block below. In turn, this is only
4261 triggered in the "postponed utf8 string" tests
4267 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4271 rx = reg_temp_copy(NULL, rx);
4275 const I32 osize = PL_regsize;
4278 assert (SvUTF8(ret));
4279 } else if (SvUTF8(ret)) {
4280 /* Not doing UTF-8, despite what the SV says. Is
4281 this only if we're trapped in use 'bytes'? */
4282 /* Make a copy of the octet sequence, but without
4283 the flag on, as the compiler now honours the
4284 SvUTF8 flag on ret. */
4286 const char *const p = SvPV(ret, len);
4287 ret = newSVpvn_flags(p, len, SVs_TEMP);
4289 rx = CALLREGCOMP(ret, pm_flags);
4291 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4293 /* This isn't a first class regexp. Instead, it's
4294 caching a regexp onto an existing, Perl visible
4296 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4301 re = (struct regexp *)SvANY(rx);
4303 RXp_MATCH_COPIED_off(re);
4304 re->subbeg = rex->subbeg;
4305 re->sublen = rex->sublen;
4308 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4309 "Matching embedded");
4311 startpoint = rei->program + 1;
4312 ST.close_paren = 0; /* only used for GOSUB */
4313 /* borrowed from regtry */
4314 if (PL_reg_start_tmpl <= re->nparens) {
4315 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4316 if(PL_reg_start_tmp)
4317 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4319 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4322 eval_recurse_doit: /* Share code with GOSUB below this line */
4323 /* run the pattern returned from (??{...}) */
4324 ST.cp = regcppush(0); /* Save *all* the positions. */
4325 REGCP_SET(ST.lastcp);
4327 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4329 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4330 PL_reglastparen = &re->lastparen;
4331 PL_reglastcloseparen = &re->lastcloseparen;
4333 re->lastcloseparen = 0;
4335 PL_reginput = locinput;
4338 /* XXXX This is too dramatic a measure... */
4341 ST.toggle_reg_flags = PL_reg_flags;
4343 PL_reg_flags |= RF_utf8;
4345 PL_reg_flags &= ~RF_utf8;
4346 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4348 ST.prev_rex = rex_sv;
4349 ST.prev_curlyx = cur_curlyx;
4350 SETREX(rex_sv,re_sv);
4355 ST.prev_eval = cur_eval;
4357 /* now continue from first node in postoned RE */
4358 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4361 /* logical is 1, /(?(?{...})X|Y)/ */
4362 sw = cBOOL(SvTRUE(ret));
4367 case EVAL_AB: /* cleanup after a successful (??{A})B */
4368 /* note: this is called twice; first after popping B, then A */
4369 PL_reg_flags ^= ST.toggle_reg_flags;
4370 ReREFCNT_dec(rex_sv);
4371 SETREX(rex_sv,ST.prev_rex);
4372 rex = (struct regexp *)SvANY(rex_sv);
4373 rexi = RXi_GET(rex);
4375 cur_eval = ST.prev_eval;
4376 cur_curlyx = ST.prev_curlyx;
4378 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4379 PL_reglastparen = &rex->lastparen;
4380 PL_reglastcloseparen = &rex->lastcloseparen;
4381 /* also update PL_regoffs */
4382 PL_regoffs = rex->offs;
4384 /* XXXX This is too dramatic a measure... */
4386 if ( nochange_depth )
4391 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4392 /* note: this is called twice; first after popping B, then A */
4393 PL_reg_flags ^= ST.toggle_reg_flags;
4394 ReREFCNT_dec(rex_sv);
4395 SETREX(rex_sv,ST.prev_rex);
4396 rex = (struct regexp *)SvANY(rex_sv);
4397 rexi = RXi_GET(rex);
4398 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4399 PL_reglastparen = &rex->lastparen;
4400 PL_reglastcloseparen = &rex->lastcloseparen;
4402 PL_reginput = locinput;
4403 REGCP_UNWIND(ST.lastcp);
4405 cur_eval = ST.prev_eval;
4406 cur_curlyx = ST.prev_curlyx;
4407 /* XXXX This is too dramatic a measure... */
4409 if ( nochange_depth )
4415 n = ARG(scan); /* which paren pair */
4416 PL_reg_start_tmp[n] = locinput;
4422 n = ARG(scan); /* which paren pair */
4423 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4424 PL_regoffs[n].end = locinput - PL_bostr;
4425 /*if (n > PL_regsize)
4427 if (n > *PL_reglastparen)
4428 *PL_reglastparen = n;
4429 *PL_reglastcloseparen = n;
4430 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4438 cursor && OP(cursor)!=END;
4439 cursor=regnext(cursor))
4441 if ( OP(cursor)==CLOSE ){
4443 if ( n <= lastopen ) {
4445 = PL_reg_start_tmp[n] - PL_bostr;
4446 PL_regoffs[n].end = locinput - PL_bostr;
4447 /*if (n > PL_regsize)
4449 if (n > *PL_reglastparen)
4450 *PL_reglastparen = n;
4451 *PL_reglastcloseparen = n;
4452 if ( n == ARG(scan) || (cur_eval &&
4453 cur_eval->u.eval.close_paren == n))
4462 n = ARG(scan); /* which paren pair */
4463 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4466 /* reg_check_named_buff_matched returns 0 for no match */
4467 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4471 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4477 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4479 next = NEXTOPER(NEXTOPER(scan));
4481 next = scan + ARG(scan);
4482 if (OP(next) == IFTHEN) /* Fake one. */
4483 next = NEXTOPER(NEXTOPER(next));
4487 logical = scan->flags;
4490 /*******************************************************************
4492 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4493 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4494 STAR/PLUS/CURLY/CURLYN are used instead.)
4496 A*B is compiled as <CURLYX><A><WHILEM><B>
4498 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4499 state, which contains the current count, initialised to -1. It also sets
4500 cur_curlyx to point to this state, with any previous value saved in the
4503 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4504 since the pattern may possibly match zero times (i.e. it's a while {} loop
4505 rather than a do {} while loop).
4507 Each entry to WHILEM represents a successful match of A. The count in the
4508 CURLYX block is incremented, another WHILEM state is pushed, and execution
4509 passes to A or B depending on greediness and the current count.
4511 For example, if matching against the string a1a2a3b (where the aN are
4512 substrings that match /A/), then the match progresses as follows: (the
4513 pushed states are interspersed with the bits of strings matched so far):
4516 <CURLYX cnt=0><WHILEM>
4517 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4518 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4519 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4520 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4522 (Contrast this with something like CURLYM, which maintains only a single
4526 a1 <CURLYM cnt=1> a2
4527 a1 a2 <CURLYM cnt=2> a3
4528 a1 a2 a3 <CURLYM cnt=3> b
4531 Each WHILEM state block marks a point to backtrack to upon partial failure
4532 of A or B, and also contains some minor state data related to that
4533 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4534 overall state, such as the count, and pointers to the A and B ops.
4536 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4537 must always point to the *current* CURLYX block, the rules are:
4539 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4540 and set cur_curlyx to point the new block.
4542 When popping the CURLYX block after a successful or unsuccessful match,
4543 restore the previous cur_curlyx.
4545 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4546 to the outer one saved in the CURLYX block.
4548 When popping the WHILEM block after a successful or unsuccessful B match,
4549 restore the previous cur_curlyx.
4551 Here's an example for the pattern (AI* BI)*BO
4552 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4555 curlyx backtrack stack
4556 ------ ---------------
4558 CO <CO prev=NULL> <WO>
4559 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4560 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4561 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4563 At this point the pattern succeeds, and we work back down the stack to
4564 clean up, restoring as we go:
4566 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4567 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4568 CO <CO prev=NULL> <WO>
4571 *******************************************************************/
4573 #define ST st->u.curlyx
4575 case CURLYX: /* start of /A*B/ (for complex A) */
4577 /* No need to save/restore up to this paren */
4578 I32 parenfloor = scan->flags;
4580 assert(next); /* keep Coverity happy */
4581 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4584 /* XXXX Probably it is better to teach regpush to support
4585 parenfloor > PL_regsize... */
4586 if (parenfloor > (I32)*PL_reglastparen)
4587 parenfloor = *PL_reglastparen; /* Pessimization... */
4589 ST.prev_curlyx= cur_curlyx;
4591 ST.cp = PL_savestack_ix;
4593 /* these fields contain the state of the current curly.
4594 * they are accessed by subsequent WHILEMs */
4595 ST.parenfloor = parenfloor;
4600 ST.count = -1; /* this will be updated by WHILEM */
4601 ST.lastloc = NULL; /* this will be updated by WHILEM */
4603 PL_reginput = locinput;
4604 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4608 case CURLYX_end: /* just finished matching all of A*B */
4609 cur_curlyx = ST.prev_curlyx;
4613 case CURLYX_end_fail: /* just failed to match all of A*B */
4615 cur_curlyx = ST.prev_curlyx;
4621 #define ST st->u.whilem
4623 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4625 /* see the discussion above about CURLYX/WHILEM */
4627 int min = ARG1(cur_curlyx->u.curlyx.me);
4628 int max = ARG2(cur_curlyx->u.curlyx.me);
4629 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4631 assert(cur_curlyx); /* keep Coverity happy */
4632 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4633 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4634 ST.cache_offset = 0;
4637 PL_reginput = locinput;
4639 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4640 "%*s whilem: matched %ld out of %d..%d\n",
4641 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4644 /* First just match a string of min A's. */
4647 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4648 cur_curlyx->u.curlyx.lastloc = locinput;
4649 REGCP_SET(ST.lastcp);
4651 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4655 /* If degenerate A matches "", assume A done. */
4657 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4658 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4659 "%*s whilem: empty match detected, trying continuation...\n",
4660 REPORT_CODE_OFF+depth*2, "")
4662 goto do_whilem_B_max;
4665 /* super-linear cache processing */
4669 if (!PL_reg_maxiter) {
4670 /* start the countdown: Postpone detection until we
4671 * know the match is not *that* much linear. */
4672 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4673 /* possible overflow for long strings and many CURLYX's */
4674 if (PL_reg_maxiter < 0)
4675 PL_reg_maxiter = I32_MAX;
4676 PL_reg_leftiter = PL_reg_maxiter;
4679 if (PL_reg_leftiter-- == 0) {
4680 /* initialise cache */
4681 const I32 size = (PL_reg_maxiter + 7)/8;
4682 if (PL_reg_poscache) {
4683 if ((I32)PL_reg_poscache_size < size) {
4684 Renew(PL_reg_poscache, size, char);
4685 PL_reg_poscache_size = size;
4687 Zero(PL_reg_poscache, size, char);
4690 PL_reg_poscache_size = size;
4691 Newxz(PL_reg_poscache, size, char);
4693 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4694 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4695 PL_colors[4], PL_colors[5])
4699 if (PL_reg_leftiter < 0) {
4700 /* have we already failed at this position? */
4702 offset = (scan->flags & 0xf) - 1
4703 + (locinput - PL_bostr) * (scan->flags>>4);
4704 mask = 1 << (offset % 8);
4706 if (PL_reg_poscache[offset] & mask) {
4707 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4708 "%*s whilem: (cache) already tried at this position...\n",
4709 REPORT_CODE_OFF+depth*2, "")
4711 sayNO; /* cache records failure */
4713 ST.cache_offset = offset;
4714 ST.cache_mask = mask;
4718 /* Prefer B over A for minimal matching. */
4720 if (cur_curlyx->u.curlyx.minmod) {
4721 ST.save_curlyx = cur_curlyx;
4722 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4723 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4724 REGCP_SET(ST.lastcp);
4725 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4729 /* Prefer A over B for maximal matching. */
4731 if (n < max) { /* More greed allowed? */
4732 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4733 cur_curlyx->u.curlyx.lastloc = locinput;
4734 REGCP_SET(ST.lastcp);
4735 PUSH_STATE_GOTO(WHILEM_A_max, A);
4738 goto do_whilem_B_max;
4742 case WHILEM_B_min: /* just matched B in a minimal match */
4743 case WHILEM_B_max: /* just matched B in a maximal match */
4744 cur_curlyx = ST.save_curlyx;
4748 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4749 cur_curlyx = ST.save_curlyx;
4750 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4751 cur_curlyx->u.curlyx.count--;
4755 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4757 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4758 REGCP_UNWIND(ST.lastcp);
4760 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4761 cur_curlyx->u.curlyx.count--;
4765 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4766 REGCP_UNWIND(ST.lastcp);
4767 regcppop(rex); /* Restore some previous $<digit>s? */
4768 PL_reginput = locinput;
4769 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4770 "%*s whilem: failed, trying continuation...\n",
4771 REPORT_CODE_OFF+depth*2, "")
4774 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4775 && ckWARN(WARN_REGEXP)
4776 && !(PL_reg_flags & RF_warned))
4778 PL_reg_flags |= RF_warned;
4779 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4780 "Complex regular subexpression recursion",
4785 ST.save_curlyx = cur_curlyx;
4786 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4787 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4790 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4791 cur_curlyx = ST.save_curlyx;
4792 REGCP_UNWIND(ST.lastcp);
4795 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4796 /* Maximum greed exceeded */
4797 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4798 && ckWARN(WARN_REGEXP)
4799 && !(PL_reg_flags & RF_warned))
4801 PL_reg_flags |= RF_warned;
4802 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4803 "%s limit (%d) exceeded",
4804 "Complex regular subexpression recursion",
4807 cur_curlyx->u.curlyx.count--;
4811 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4812 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4814 /* Try grabbing another A and see if it helps. */
4815 PL_reginput = locinput;
4816 cur_curlyx->u.curlyx.lastloc = locinput;
4817 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4818 REGCP_SET(ST.lastcp);
4819 PUSH_STATE_GOTO(WHILEM_A_min,
4820 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4824 #define ST st->u.branch
4826 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4827 next = scan + ARG(scan);
4830 scan = NEXTOPER(scan);
4833 case BRANCH: /* /(...|A|...)/ */
4834 scan = NEXTOPER(scan); /* scan now points to inner node */
4835 ST.lastparen = *PL_reglastparen;
4836 ST.next_branch = next;
4838 PL_reginput = locinput;
4840 /* Now go into the branch */
4842 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4844 PUSH_STATE_GOTO(BRANCH_next, scan);
4848 PL_reginput = locinput;
4849 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4850 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4851 PUSH_STATE_GOTO(CUTGROUP_next,next);
4853 case CUTGROUP_next_fail:
4856 if (st->u.mark.mark_name)
4857 sv_commit = st->u.mark.mark_name;
4863 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4868 REGCP_UNWIND(ST.cp);
4869 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4870 PL_regoffs[n].end = -1;
4871 *PL_reglastparen = n;
4872 /*dmq: *PL_reglastcloseparen = n; */
4873 scan = ST.next_branch;
4874 /* no more branches? */
4875 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4877 PerlIO_printf( Perl_debug_log,
4878 "%*s %sBRANCH failed...%s\n",
4879 REPORT_CODE_OFF+depth*2, "",
4885 continue; /* execute next BRANCH[J] op */
4893 #define ST st->u.curlym
4895 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4897 /* This is an optimisation of CURLYX that enables us to push
4898 * only a single backtracking state, no matter how many matches
4899 * there are in {m,n}. It relies on the pattern being constant
4900 * length, with no parens to influence future backrefs
4904 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4906 /* if paren positive, emulate an OPEN/CLOSE around A */
4908 U32 paren = ST.me->flags;
4909 if (paren > PL_regsize)
4911 if (paren > *PL_reglastparen)
4912 *PL_reglastparen = paren;
4913 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4921 ST.c1 = CHRTEST_UNINIT;
4924 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4927 curlym_do_A: /* execute the A in /A{m,n}B/ */
4928 PL_reginput = locinput;
4929 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4932 case CURLYM_A: /* we've just matched an A */
4933 locinput = st->locinput;
4934 nextchr = UCHARAT(locinput);
4937 /* after first match, determine A's length: u.curlym.alen */
4938 if (ST.count == 1) {
4939 if (PL_reg_match_utf8) {
4941 while (s < PL_reginput) {
4947 ST.alen = PL_reginput - locinput;
4950 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4953 PerlIO_printf(Perl_debug_log,
4954 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4955 (int)(REPORT_CODE_OFF+(depth*2)), "",
4956 (IV) ST.count, (IV)ST.alen)
4959 locinput = PL_reginput;
4961 if (cur_eval && cur_eval->u.eval.close_paren &&
4962 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4966 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4967 if ( max == REG_INFTY || ST.count < max )
4968 goto curlym_do_A; /* try to match another A */
4970 goto curlym_do_B; /* try to match B */
4972 case CURLYM_A_fail: /* just failed to match an A */
4973 REGCP_UNWIND(ST.cp);
4975 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4976 || (cur_eval && cur_eval->u.eval.close_paren &&
4977 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4980 curlym_do_B: /* execute the B in /A{m,n}B/ */
4981 PL_reginput = locinput;
4982 if (ST.c1 == CHRTEST_UNINIT) {
4983 /* calculate c1 and c2 for possible match of 1st char
4984 * following curly */
4985 ST.c1 = ST.c2 = CHRTEST_VOID;
4986 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4987 regnode *text_node = ST.B;
4988 if (! HAS_TEXT(text_node))
4989 FIND_NEXT_IMPT(text_node);
4992 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4994 But the former is redundant in light of the latter.
4996 if this changes back then the macro for
4997 IS_TEXT and friends need to change.
4999 if (PL_regkind[OP(text_node)] == EXACT)
5002 ST.c1 = (U8)*STRING(text_node);
5003 switch (OP(text_node)) {
5004 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5006 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5007 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5008 default: ST.c2 = ST.c1;
5015 PerlIO_printf(Perl_debug_log,
5016 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5017 (int)(REPORT_CODE_OFF+(depth*2)),
5020 if (ST.c1 != CHRTEST_VOID
5021 && UCHARAT(PL_reginput) != ST.c1
5022 && UCHARAT(PL_reginput) != ST.c2)
5024 /* simulate B failing */
5026 PerlIO_printf(Perl_debug_log,
5027 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5028 (int)(REPORT_CODE_OFF+(depth*2)),"",
5031 state_num = CURLYM_B_fail;
5032 goto reenter_switch;
5036 /* mark current A as captured */
5037 I32 paren = ST.me->flags;
5039 PL_regoffs[paren].start
5040 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5041 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5042 /*dmq: *PL_reglastcloseparen = paren; */
5045 PL_regoffs[paren].end = -1;
5046 if (cur_eval && cur_eval->u.eval.close_paren &&
5047 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5056 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5059 case CURLYM_B_fail: /* just failed to match a B */
5060 REGCP_UNWIND(ST.cp);
5062 I32 max = ARG2(ST.me);
5063 if (max != REG_INFTY && ST.count == max)
5065 goto curlym_do_A; /* try to match a further A */
5067 /* backtrack one A */
5068 if (ST.count == ARG1(ST.me) /* min */)
5071 locinput = HOPc(locinput, -ST.alen);
5072 goto curlym_do_B; /* try to match B */
5075 #define ST st->u.curly
5077 #define CURLY_SETPAREN(paren, success) \
5080 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5081 PL_regoffs[paren].end = locinput - PL_bostr; \
5082 *PL_reglastcloseparen = paren; \
5085 PL_regoffs[paren].end = -1; \
5088 case STAR: /* /A*B/ where A is width 1 */
5092 scan = NEXTOPER(scan);
5094 case PLUS: /* /A+B/ where A is width 1 */
5098 scan = NEXTOPER(scan);
5100 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5101 ST.paren = scan->flags; /* Which paren to set */
5102 if (ST.paren > PL_regsize)
5103 PL_regsize = ST.paren;
5104 if (ST.paren > *PL_reglastparen)
5105 *PL_reglastparen = ST.paren;
5106 ST.min = ARG1(scan); /* min to match */
5107 ST.max = ARG2(scan); /* max to match */
5108 if (cur_eval && cur_eval->u.eval.close_paren &&
5109 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5113 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5115 case CURLY: /* /A{m,n}B/ where A is width 1 */
5117 ST.min = ARG1(scan); /* min to match */
5118 ST.max = ARG2(scan); /* max to match */
5119 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5122 * Lookahead to avoid useless match attempts
5123 * when we know what character comes next.
5125 * Used to only do .*x and .*?x, but now it allows
5126 * for )'s, ('s and (?{ ... })'s to be in the way
5127 * of the quantifier and the EXACT-like node. -- japhy
5130 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5132 if (HAS_TEXT(next) || JUMPABLE(next)) {
5134 regnode *text_node = next;
5136 if (! HAS_TEXT(text_node))
5137 FIND_NEXT_IMPT(text_node);
5139 if (! HAS_TEXT(text_node))
5140 ST.c1 = ST.c2 = CHRTEST_VOID;
5142 if ( PL_regkind[OP(text_node)] != EXACT ) {
5143 ST.c1 = ST.c2 = CHRTEST_VOID;
5144 goto assume_ok_easy;
5147 s = (U8*)STRING(text_node);
5149 /* Currently we only get here when
5151 PL_rekind[OP(text_node)] == EXACT
5153 if this changes back then the macro for IS_TEXT and
5154 friends need to change. */
5157 switch (OP(text_node)) {
5158 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5160 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5161 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5162 default: ST.c2 = ST.c1; break;
5165 else { /* UTF_PATTERN */
5166 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5167 STRLEN ulen1, ulen2;
5168 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5169 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5171 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5172 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5174 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5176 0 : UTF8_ALLOW_ANY);
5177 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5179 0 : UTF8_ALLOW_ANY);
5181 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5183 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5188 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5195 ST.c1 = ST.c2 = CHRTEST_VOID;
5200 PL_reginput = locinput;
5203 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5206 locinput = PL_reginput;
5208 if (ST.c1 == CHRTEST_VOID)
5209 goto curly_try_B_min;
5211 ST.oldloc = locinput;
5213 /* set ST.maxpos to the furthest point along the
5214 * string that could possibly match */
5215 if (ST.max == REG_INFTY) {
5216 ST.maxpos = PL_regeol - 1;
5218 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5221 else if (utf8_target) {
5222 int m = ST.max - ST.min;
5223 for (ST.maxpos = locinput;
5224 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5225 ST.maxpos += UTF8SKIP(ST.maxpos);
5228 ST.maxpos = locinput + ST.max - ST.min;
5229 if (ST.maxpos >= PL_regeol)
5230 ST.maxpos = PL_regeol - 1;
5232 goto curly_try_B_min_known;
5236 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5237 locinput = PL_reginput;
5238 if (ST.count < ST.min)
5240 if ((ST.count > ST.min)
5241 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5243 /* A{m,n} must come at the end of the string, there's
5244 * no point in backing off ... */
5246 /* ...except that $ and \Z can match before *and* after
5247 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5248 We may back off by one in this case. */
5249 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5253 goto curly_try_B_max;
5258 case CURLY_B_min_known_fail:
5259 /* failed to find B in a non-greedy match where c1,c2 valid */
5260 if (ST.paren && ST.count)
5261 PL_regoffs[ST.paren].end = -1;
5263 PL_reginput = locinput; /* Could be reset... */
5264 REGCP_UNWIND(ST.cp);
5265 /* Couldn't or didn't -- move forward. */
5266 ST.oldloc = locinput;
5268 locinput += UTF8SKIP(locinput);
5272 curly_try_B_min_known:
5273 /* find the next place where 'B' could work, then call B */
5277 n = (ST.oldloc == locinput) ? 0 : 1;
5278 if (ST.c1 == ST.c2) {
5280 /* set n to utf8_distance(oldloc, locinput) */
5281 while (locinput <= ST.maxpos &&
5282 utf8n_to_uvchr((U8*)locinput,
5283 UTF8_MAXBYTES, &len,
5284 uniflags) != (UV)ST.c1) {
5290 /* set n to utf8_distance(oldloc, locinput) */
5291 while (locinput <= ST.maxpos) {
5293 const UV c = utf8n_to_uvchr((U8*)locinput,
5294 UTF8_MAXBYTES, &len,
5296 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5304 if (ST.c1 == ST.c2) {
5305 while (locinput <= ST.maxpos &&
5306 UCHARAT(locinput) != ST.c1)
5310 while (locinput <= ST.maxpos
5311 && UCHARAT(locinput) != ST.c1
5312 && UCHARAT(locinput) != ST.c2)
5315 n = locinput - ST.oldloc;
5317 if (locinput > ST.maxpos)
5319 /* PL_reginput == oldloc now */
5322 if (regrepeat(rex, ST.A, n, depth) < n)
5325 PL_reginput = locinput;
5326 CURLY_SETPAREN(ST.paren, ST.count);
5327 if (cur_eval && cur_eval->u.eval.close_paren &&
5328 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5331 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5336 case CURLY_B_min_fail:
5337 /* failed to find B in a non-greedy match where c1,c2 invalid */
5338 if (ST.paren && ST.count)
5339 PL_regoffs[ST.paren].end = -1;
5341 REGCP_UNWIND(ST.cp);
5342 /* failed -- move forward one */
5343 PL_reginput = locinput;
5344 if (regrepeat(rex, ST.A, 1, depth)) {
5346 locinput = PL_reginput;
5347 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5348 ST.count > 0)) /* count overflow ? */
5351 CURLY_SETPAREN(ST.paren, ST.count);
5352 if (cur_eval && cur_eval->u.eval.close_paren &&
5353 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5356 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5364 /* a successful greedy match: now try to match B */
5365 if (cur_eval && cur_eval->u.eval.close_paren &&
5366 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5371 if (ST.c1 != CHRTEST_VOID)
5372 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5373 UTF8_MAXBYTES, 0, uniflags)
5374 : (UV) UCHARAT(PL_reginput);
5375 /* If it could work, try it. */
5376 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5377 CURLY_SETPAREN(ST.paren, ST.count);
5378 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5383 case CURLY_B_max_fail:
5384 /* failed to find B in a greedy match */
5385 if (ST.paren && ST.count)
5386 PL_regoffs[ST.paren].end = -1;
5388 REGCP_UNWIND(ST.cp);
5390 if (--ST.count < ST.min)
5392 PL_reginput = locinput = HOPc(locinput, -1);
5393 goto curly_try_B_max;
5400 /* we've just finished A in /(??{A})B/; now continue with B */
5402 st->u.eval.toggle_reg_flags
5403 = cur_eval->u.eval.toggle_reg_flags;
5404 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5406 st->u.eval.prev_rex = rex_sv; /* inner */
5407 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5408 rex = (struct regexp *)SvANY(rex_sv);
5409 rexi = RXi_GET(rex);
5410 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5411 (void)ReREFCNT_inc(rex_sv);
5412 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5414 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5415 PL_reglastparen = &rex->lastparen;
5416 PL_reglastcloseparen = &rex->lastcloseparen;
5418 REGCP_SET(st->u.eval.lastcp);
5419 PL_reginput = locinput;
5421 /* Restore parens of the outer rex without popping the
5423 tmpix = PL_savestack_ix;
5424 PL_savestack_ix = cur_eval->u.eval.lastcp;
5426 PL_savestack_ix = tmpix;
5428 st->u.eval.prev_eval = cur_eval;
5429 cur_eval = cur_eval->u.eval.prev_eval;
5431 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5432 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5433 if ( nochange_depth )
5436 PUSH_YES_STATE_GOTO(EVAL_AB,
5437 st->u.eval.prev_eval->u.eval.B); /* match B */
5440 if (locinput < reginfo->till) {
5441 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5442 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5444 (long)(locinput - PL_reg_starttry),
5445 (long)(reginfo->till - PL_reg_starttry),
5448 sayNO_SILENT; /* Cannot match: too short. */
5450 PL_reginput = locinput; /* put where regtry can find it */
5451 sayYES; /* Success! */
5453 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5455 PerlIO_printf(Perl_debug_log,
5456 "%*s %ssubpattern success...%s\n",
5457 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5458 PL_reginput = locinput; /* put where regtry can find it */
5459 sayYES; /* Success! */
5462 #define ST st->u.ifmatch
5464 case SUSPEND: /* (?>A) */
5466 PL_reginput = locinput;
5469 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5471 goto ifmatch_trivial_fail_test;
5473 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5475 ifmatch_trivial_fail_test:
5477 char * const s = HOPBACKc(locinput, scan->flags);
5482 sw = 1 - cBOOL(ST.wanted);
5486 next = scan + ARG(scan);
5494 PL_reginput = locinput;
5498 ST.logical = logical;
5499 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5501 /* execute body of (?...A) */
5502 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5505 case IFMATCH_A_fail: /* body of (?...A) failed */
5506 ST.wanted = !ST.wanted;
5509 case IFMATCH_A: /* body of (?...A) succeeded */
5511 sw = cBOOL(ST.wanted);
5513 else if (!ST.wanted)
5516 if (OP(ST.me) == SUSPEND)
5517 locinput = PL_reginput;
5519 locinput = PL_reginput = st->locinput;
5520 nextchr = UCHARAT(locinput);
5522 scan = ST.me + ARG(ST.me);
5525 continue; /* execute B */
5530 next = scan + ARG(scan);
5535 reginfo->cutpoint = PL_regeol;
5538 PL_reginput = locinput;
5540 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5541 PUSH_STATE_GOTO(COMMIT_next,next);
5543 case COMMIT_next_fail:
5550 #define ST st->u.mark
5552 ST.prev_mark = mark_state;
5553 ST.mark_name = sv_commit = sv_yes_mark
5554 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5556 ST.mark_loc = PL_reginput = locinput;
5557 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5559 case MARKPOINT_next:
5560 mark_state = ST.prev_mark;
5563 case MARKPOINT_next_fail:
5564 if (popmark && sv_eq(ST.mark_name,popmark))
5566 if (ST.mark_loc > startpoint)
5567 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5568 popmark = NULL; /* we found our mark */
5569 sv_commit = ST.mark_name;
5572 PerlIO_printf(Perl_debug_log,
5573 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5574 REPORT_CODE_OFF+depth*2, "",
5575 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5578 mark_state = ST.prev_mark;
5579 sv_yes_mark = mark_state ?
5580 mark_state->u.mark.mark_name : NULL;
5584 PL_reginput = locinput;
5586 /* (*SKIP) : if we fail we cut here*/
5587 ST.mark_name = NULL;
5588 ST.mark_loc = locinput;
5589 PUSH_STATE_GOTO(SKIP_next,next);
5591 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5592 otherwise do nothing. Meaning we need to scan
5594 regmatch_state *cur = mark_state;
5595 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5598 if ( sv_eq( cur->u.mark.mark_name,
5601 ST.mark_name = find;
5602 PUSH_STATE_GOTO( SKIP_next, next );
5604 cur = cur->u.mark.prev_mark;
5607 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5609 case SKIP_next_fail:
5611 /* (*CUT:NAME) - Set up to search for the name as we
5612 collapse the stack*/
5613 popmark = ST.mark_name;
5615 /* (*CUT) - No name, we cut here.*/
5616 if (ST.mark_loc > startpoint)
5617 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5618 /* but we set sv_commit to latest mark_name if there
5619 is one so they can test to see how things lead to this
5622 sv_commit=mark_state->u.mark.mark_name;
5630 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5632 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5635 U8 folded[UTF8_MAXBYTES_CASE+1];
5637 const char * const l = locinput;
5638 char *e = PL_regeol;
5639 to_uni_fold(n, folded, &foldlen);
5641 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5642 l, &e, 0, utf8_target)) {
5647 nextchr = UCHARAT(locinput);
5650 if ((n=is_LNBREAK(locinput,utf8_target))) {
5652 nextchr = UCHARAT(locinput);
5657 #define CASE_CLASS(nAmE) \
5659 if ((n=is_##nAmE(locinput,utf8_target))) { \
5661 nextchr = UCHARAT(locinput); \
5666 if ((n=is_##nAmE(locinput,utf8_target))) { \
5669 locinput += UTF8SKIP(locinput); \
5670 nextchr = UCHARAT(locinput); \
5675 CASE_CLASS(HORIZWS);
5679 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5680 PTR2UV(scan), OP(scan));
5681 Perl_croak(aTHX_ "regexp memory corruption");
5685 /* switch break jumps here */
5686 scan = next; /* prepare to execute the next op and ... */
5687 continue; /* ... jump back to the top, reusing st */
5691 /* push a state that backtracks on success */
5692 st->u.yes.prev_yes_state = yes_state;
5696 /* push a new regex state, then continue at scan */
5698 regmatch_state *newst;
5701 regmatch_state *cur = st;
5702 regmatch_state *curyes = yes_state;
5704 regmatch_slab *slab = PL_regmatch_slab;
5705 for (;curd > -1;cur--,curd--) {
5706 if (cur < SLAB_FIRST(slab)) {
5708 cur = SLAB_LAST(slab);
5710 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5711 REPORT_CODE_OFF + 2 + depth * 2,"",
5712 curd, PL_reg_name[cur->resume_state],
5713 (curyes == cur) ? "yes" : ""
5716 curyes = cur->u.yes.prev_yes_state;
5719 DEBUG_STATE_pp("push")
5722 st->locinput = locinput;
5724 if (newst > SLAB_LAST(PL_regmatch_slab))
5725 newst = S_push_slab(aTHX);
5726 PL_regmatch_state = newst;
5728 locinput = PL_reginput;
5729 nextchr = UCHARAT(locinput);
5737 * We get here only if there's trouble -- normally "case END" is
5738 * the terminating point.
5740 Perl_croak(aTHX_ "corrupted regexp pointers");
5746 /* we have successfully completed a subexpression, but we must now
5747 * pop to the state marked by yes_state and continue from there */
5748 assert(st != yes_state);
5750 while (st != yes_state) {
5752 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5753 PL_regmatch_slab = PL_regmatch_slab->prev;
5754 st = SLAB_LAST(PL_regmatch_slab);
5758 DEBUG_STATE_pp("pop (no final)");
5760 DEBUG_STATE_pp("pop (yes)");
5766 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5767 || yes_state > SLAB_LAST(PL_regmatch_slab))
5769 /* not in this slab, pop slab */
5770 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5771 PL_regmatch_slab = PL_regmatch_slab->prev;
5772 st = SLAB_LAST(PL_regmatch_slab);
5774 depth -= (st - yes_state);
5777 yes_state = st->u.yes.prev_yes_state;
5778 PL_regmatch_state = st;
5781 locinput= st->locinput;
5782 nextchr = UCHARAT(locinput);
5784 state_num = st->resume_state + no_final;
5785 goto reenter_switch;
5788 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5789 PL_colors[4], PL_colors[5]));
5791 if (PL_reg_eval_set) {
5792 /* each successfully executed (?{...}) block does the equivalent of
5793 * local $^R = do {...}
5794 * When popping the save stack, all these locals would be undone;
5795 * bypass this by setting the outermost saved $^R to the latest
5797 if (oreplsv != GvSV(PL_replgv))
5798 sv_setsv(oreplsv, GvSV(PL_replgv));
5805 PerlIO_printf(Perl_debug_log,
5806 "%*s %sfailed...%s\n",
5807 REPORT_CODE_OFF+depth*2, "",
5808 PL_colors[4], PL_colors[5])
5820 /* there's a previous state to backtrack to */
5822 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5823 PL_regmatch_slab = PL_regmatch_slab->prev;
5824 st = SLAB_LAST(PL_regmatch_slab);
5826 PL_regmatch_state = st;
5827 locinput= st->locinput;
5828 nextchr = UCHARAT(locinput);
5830 DEBUG_STATE_pp("pop");
5832 if (yes_state == st)
5833 yes_state = st->u.yes.prev_yes_state;
5835 state_num = st->resume_state + 1; /* failure = success + 1 */
5836 goto reenter_switch;
5841 if (rex->intflags & PREGf_VERBARG_SEEN) {
5842 SV *sv_err = get_sv("REGERROR", 1);
5843 SV *sv_mrk = get_sv("REGMARK", 1);
5845 sv_commit = &PL_sv_no;
5847 sv_yes_mark = &PL_sv_yes;
5850 sv_commit = &PL_sv_yes;
5851 sv_yes_mark = &PL_sv_no;
5853 sv_setsv(sv_err, sv_commit);
5854 sv_setsv(sv_mrk, sv_yes_mark);
5857 /* clean up; in particular, free all slabs above current one */
5858 LEAVE_SCOPE(oldsave);
5864 - regrepeat - repeatedly match something simple, report how many
5867 * [This routine now assumes that it will only match on things of length 1.
5868 * That was true before, but now we assume scan - reginput is the count,
5869 * rather than incrementing count on every character. [Er, except utf8.]]
5872 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5875 register char *scan;
5877 register char *loceol = PL_regeol;
5878 register I32 hardcount = 0;
5879 register bool utf8_target = PL_reg_match_utf8;
5882 PERL_UNUSED_ARG(depth);
5885 PERL_ARGS_ASSERT_REGREPEAT;
5888 if (max == REG_INFTY)
5890 else if (max < loceol - scan)
5891 loceol = scan + max;
5896 while (scan < loceol && hardcount < max && *scan != '\n') {
5897 scan += UTF8SKIP(scan);
5901 while (scan < loceol && *scan != '\n')
5908 while (scan < loceol && hardcount < max) {
5909 scan += UTF8SKIP(scan);
5920 /* To get here, EXACTish nodes must have *byte* length == 1. That
5921 * means they match only characters in the string that can be expressed
5922 * as a single byte. For non-utf8 strings, that means a simple match.
5923 * For utf8 strings, the character matched must be an invariant, or
5924 * downgradable to a single byte. The pattern's utf8ness is
5925 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5926 * it is, it's an invariant */
5929 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5931 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5932 while (scan < loceol && UCHARAT(scan) == c) {
5938 /* Here, the string is utf8, and the pattern char is different
5939 * in utf8 than not, so can't compare them directly. Outside the
5940 * loop, find find the two utf8 bytes that represent c, and then
5941 * look for those in sequence in the utf8 string */
5942 U8 high = UTF8_TWO_BYTE_HI(c);
5943 U8 low = UTF8_TWO_BYTE_LO(c);
5946 while (hardcount < max
5947 && scan + 1 < loceol
5948 && UCHARAT(scan) == high
5949 && UCHARAT(scan + 1) == low)
5957 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5961 PL_reg_flags |= RF_tainted;
5962 utf8_flags = FOLDEQ_UTF8_LOCALE;
5969 /* The comments for the EXACT case above apply as well to these fold
5974 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5976 if (utf8_target) { /* Use full Unicode fold matching */
5977 char *tmpeol = loceol;
5978 while (hardcount < max
5979 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
5980 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
5987 /* XXX Note that the above handles properly the German sharp s in
5988 * the pattern matching ss in the string. But it doesn't handle
5989 * properly cases where the string contains say 'LIGATURE ff' and
5990 * the pattern is 'f+'. This would require, say, a new function or
5991 * revised interface to foldEQ_utf8(), in which the maximum number
5992 * of characters to match could be passed and it would return how
5993 * many actually did. This is just one of many cases where
5994 * multi-char folds don't work properly, and so the fix is being
6000 /* Here, the string isn't utf8 and c is a single byte; and either
6001 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6002 * doesn't affect c. Can just do simple comparisons for exact or
6005 case EXACTF: folded = PL_fold[c]; break;
6007 case EXACTFU: folded = PL_fold_latin1[c]; break;
6008 case EXACTFL: folded = PL_fold_locale[c]; break;
6009 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6011 while (scan < loceol &&
6012 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6020 if (utf8_target || OP(p) == ANYOFV) {
6023 inclasslen = loceol - scan;
6024 while (hardcount < max
6025 && ((inclasslen = loceol - scan) > 0)
6026 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6032 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6040 LOAD_UTF8_CHARCLASS_ALNUM();
6041 while (hardcount < max && scan < loceol &&
6042 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6044 scan += UTF8SKIP(scan);
6048 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6056 while (scan < loceol && isALNUM((U8) *scan)) {
6061 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6066 PL_reg_flags |= RF_tainted;
6069 while (hardcount < max && scan < loceol &&
6070 isALNUM_LC_utf8((U8*)scan)) {
6071 scan += UTF8SKIP(scan);
6075 while (scan < loceol && isALNUM_LC(*scan))
6085 LOAD_UTF8_CHARCLASS_ALNUM();
6086 while (hardcount < max && scan < loceol &&
6087 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6089 scan += UTF8SKIP(scan);
6093 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6100 goto utf8_Nwordchar;
6101 while (scan < loceol && ! isALNUM((U8) *scan)) {
6107 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6108 scan += UTF8SKIP(scan);
6112 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6118 PL_reg_flags |= RF_tainted;
6121 while (hardcount < max && scan < loceol &&
6122 !isALNUM_LC_utf8((U8*)scan)) {
6123 scan += UTF8SKIP(scan);
6127 while (scan < loceol && !isALNUM_LC(*scan))
6137 LOAD_UTF8_CHARCLASS_SPACE();
6138 while (hardcount < max && scan < loceol &&
6140 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6142 scan += UTF8SKIP(scan);
6148 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6157 while (scan < loceol && isSPACE((U8) *scan)) {
6162 while (scan < loceol && isSPACE_A((U8) *scan)) {
6167 PL_reg_flags |= RF_tainted;
6170 while (hardcount < max && scan < loceol &&
6171 isSPACE_LC_utf8((U8*)scan)) {
6172 scan += UTF8SKIP(scan);
6176 while (scan < loceol && isSPACE_LC(*scan))
6186 LOAD_UTF8_CHARCLASS_SPACE();
6187 while (hardcount < max && scan < loceol &&
6189 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6191 scan += UTF8SKIP(scan);
6197 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6206 while (scan < loceol && ! isSPACE((U8) *scan)) {
6212 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6213 scan += UTF8SKIP(scan);
6217 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6223 PL_reg_flags |= RF_tainted;
6226 while (hardcount < max && scan < loceol &&
6227 !isSPACE_LC_utf8((U8*)scan)) {
6228 scan += UTF8SKIP(scan);
6232 while (scan < loceol && !isSPACE_LC(*scan))
6239 LOAD_UTF8_CHARCLASS_DIGIT();
6240 while (hardcount < max && scan < loceol &&
6241 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6242 scan += UTF8SKIP(scan);
6246 while (scan < loceol && isDIGIT(*scan))
6251 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6256 PL_reg_flags |= RF_tainted;
6259 while (hardcount < max && scan < loceol &&
6260 isDIGIT_LC_utf8((U8*)scan)) {
6261 scan += UTF8SKIP(scan);
6265 while (scan < loceol && isDIGIT_LC(*scan))
6272 LOAD_UTF8_CHARCLASS_DIGIT();
6273 while (hardcount < max && scan < loceol &&
6274 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6275 scan += UTF8SKIP(scan);
6279 while (scan < loceol && !isDIGIT(*scan))
6285 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6286 scan += UTF8SKIP(scan);
6290 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6296 PL_reg_flags |= RF_tainted;
6299 while (hardcount < max && scan < loceol &&
6300 !isDIGIT_LC_utf8((U8*)scan)) {
6301 scan += UTF8SKIP(scan);
6305 while (scan < loceol && !isDIGIT_LC(*scan))
6312 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6318 LNBREAK can match two latin chars, which is ok,
6319 because we have a null terminated string, but we
6320 have to use hardcount in this situation
6322 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6331 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6336 while (scan < loceol && is_HORIZWS_latin1(scan))
6343 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6344 scan += UTF8SKIP(scan);
6348 while (scan < loceol && !is_HORIZWS_latin1(scan))
6356 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6361 while (scan < loceol && is_VERTWS_latin1(scan))
6369 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6370 scan += UTF8SKIP(scan);
6374 while (scan < loceol && !is_VERTWS_latin1(scan))
6380 default: /* Called on something of 0 width. */
6381 break; /* So match right here or not at all. */
6387 c = scan - PL_reginput;
6391 GET_RE_DEBUG_FLAGS_DECL;
6393 SV * const prop = sv_newmortal();
6394 regprop(prog, prop, p);
6395 PerlIO_printf(Perl_debug_log,
6396 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6397 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6405 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6407 - regclass_swash - prepare the utf8 swash
6411 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6417 RXi_GET_DECL(prog,progi);
6418 const struct reg_data * const data = prog ? progi->data : NULL;
6420 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6422 assert(ANYOF_NONBITMAP(node));
6424 if (data && data->count) {
6425 const U32 n = ARG(node);
6427 if (data->what[n] == 's') {
6428 SV * const rv = MUTABLE_SV(data->data[n]);
6429 AV * const av = MUTABLE_AV(SvRV(rv));
6430 SV **const ary = AvARRAY(av);
6433 /* See the end of regcomp.c:S_regclass() for
6434 * documentation of these array elements. */
6437 a = SvROK(ary[1]) ? &ary[1] : NULL;
6438 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6442 else if (si && doinit) {
6443 sw = swash_init("utf8", "", si, 1, 0);
6444 (void)av_store(av, 1, sw);
6461 - reginclass - determine if a character falls into a character class
6463 n is the ANYOF regnode
6464 p is the target string
6465 lenp is pointer to the maximum number of bytes of how far to go in p
6466 (This is assumed wthout checking to always be at least the current
6468 utf8_target tells whether p is in UTF-8.
6470 Returns true if matched; false otherwise. If lenp is not NULL, on return
6471 from a successful match, the value it points to will be updated to how many
6472 bytes in p were matched. If there was no match, the value is undefined,
6473 possibly changed from the input.
6475 Note that this can be a synthetic start class, a combination of various
6476 nodes, so things you think might be mutually exclusive, such as locale,
6477 aren't. It can match both locale and non-locale
6482 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6485 const char flags = ANYOF_FLAGS(n);
6491 PERL_ARGS_ASSERT_REGINCLASS;
6493 /* If c is not already the code point, get it */
6494 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6495 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6496 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6497 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6498 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6499 * UTF8_ALLOW_FFFF */
6500 if (c_len == (STRLEN)-1)
6501 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6507 /* Use passed in max length, or one character if none passed in or less
6508 * than one character. And assume will match just one character. This is
6509 * overwritten later if matched more. */
6511 maxlen = (*lenp > c_len) ? *lenp : c_len;
6519 /* If this character is potentially in the bitmap, check it */
6521 if (ANYOF_BITMAP_TEST(n, c))
6523 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6530 else if (flags & ANYOF_LOCALE) {
6531 PL_reg_flags |= RF_tainted;
6533 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6534 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6538 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6539 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6540 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6541 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6542 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6543 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6544 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6545 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6546 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6547 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6548 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6549 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6550 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6551 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6552 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6553 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6554 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6555 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6556 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6557 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6558 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6559 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6560 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6561 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6562 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6563 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6564 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6565 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6566 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6567 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6568 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6569 ) /* How's that for a conditional? */
6576 /* If the bitmap didn't (or couldn't) match, and something outside the
6577 * bitmap could match, try that. Locale nodes specifiy completely the
6578 * behavior of code points in the bit map (otherwise, a utf8 target would
6579 * cause them to be treated as Unicode and not locale), except in
6580 * the very unlikely event when this node is a synthetic start class, which
6581 * could be a combination of locale and non-locale nodes. So allow locale
6582 * to match for the synthetic start class, which will give a false
6583 * positive that will be resolved when the match is done again as not part
6584 * of the synthetic start class */
6586 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6587 match = TRUE; /* Everything above 255 matches */
6589 else if (ANYOF_NONBITMAP(n)
6590 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6593 || (! (flags & ANYOF_LOCALE))
6594 || (flags & ANYOF_IS_SYNTHETIC)))))
6597 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6605 /* Not utf8. Convert as much of the string as available up
6606 * to the limit of how far the (single) character in the
6607 * pattern can possibly match (no need to go further). If
6608 * the node is a straight ANYOF or not folding, it can't
6609 * match more than one. Otherwise, It can match up to how
6610 * far a single char can fold to. Since not utf8, each
6611 * character is a single byte, so the max it can be in
6612 * bytes is the same as the max it can be in characters */
6613 STRLEN len = (OP(n) == ANYOF
6614 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6616 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6618 : UTF8_MAX_FOLD_CHAR_EXPAND;
6619 utf8_p = bytes_to_utf8(p, &len);
6622 if (swash_fetch(sw, utf8_p, TRUE))
6624 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6626 /* Here, we need to test if the fold of the target string
6627 * matches. The non-multi char folds have all been moved to
6628 * the compilation phase, and the multi-char folds have
6629 * been stored by regcomp into 'av'; we linearly check to
6630 * see if any match the target string (folded). We know
6631 * that the originals were each one character, but we don't
6632 * currently know how many characters/bytes each folded to,
6633 * except we do know that there are small limits imposed by
6634 * Unicode. XXX A performance enhancement would be to have
6635 * regcomp.c store the max number of chars/bytes that are
6636 * in an av entry, as, say the 0th element. Even better
6637 * would be to have a hash of the few characters that can
6638 * start a multi-char fold to the max number of chars of
6641 * If there is a match, we will need to advance (if lenp is
6642 * specified) the match pointer in the target string. But
6643 * what we are comparing here isn't that string directly,
6644 * but its fold, whose length may differ from the original.
6645 * As we go along in constructing the fold, therefore, we
6646 * create a map so that we know how many bytes in the
6647 * source to advance given that we have matched a certain
6648 * number of bytes in the fold. This map is stored in
6649 * 'map_fold_len_back'. Let n mean the number of bytes in
6650 * the fold of the first character that we are folding.
6651 * Then map_fold_len_back[n] is set to the number of bytes
6652 * in that first character. Similarly let m be the
6653 * corresponding number for the second character to be
6654 * folded. Then map_fold_len_back[n+m] is set to the
6655 * number of bytes occupied by the first two source
6656 * characters. ... */
6657 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6658 U8 folded[UTF8_MAXBYTES_CASE+1];
6659 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6660 STRLEN total_foldlen = 0; /* num bytes in fold of all
6663 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6665 /* Here, only need to fold the first char of the target
6666 * string. It the source wasn't utf8, is 1 byte long */
6667 to_utf8_fold(utf8_p, folded, &foldlen);
6668 total_foldlen = foldlen;
6669 map_fold_len_back[foldlen] = (utf8_target)
6675 /* Here, need to fold more than the first char. Do so
6676 * up to the limits */
6677 U8* source_ptr = utf8_p; /* The source for the fold
6680 U8* folded_ptr = folded;
6681 U8* e = utf8_p + maxlen; /* Can't go beyond last
6682 available byte in the
6686 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6690 /* Fold the next character */
6691 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6692 STRLEN this_char_foldlen;
6693 to_utf8_fold(source_ptr,
6695 &this_char_foldlen);
6697 /* Bail if it would exceed the byte limit for
6698 * folding a single char. */
6699 if (this_char_foldlen + folded_ptr - folded >
6705 /* Add the fold of this character */
6706 Copy(this_char_folded,
6710 source_ptr += UTF8SKIP(source_ptr);
6711 folded_ptr += this_char_foldlen;
6712 total_foldlen = folded_ptr - folded;
6714 /* Create map from the number of bytes in the fold
6715 * back to the number of bytes in the source. If
6716 * the source isn't utf8, the byte count is just
6717 * the number of characters so far */
6718 map_fold_len_back[total_foldlen]
6720 ? source_ptr - utf8_p
6727 /* Do the linear search to see if the fold is in the list
6728 * of multi-char folds. */
6731 for (i = 0; i <= av_len(av); i++) {
6732 SV* const sv = *av_fetch(av, i, FALSE);
6734 const char * const s = SvPV_const(sv, len);
6736 if (len <= total_foldlen
6737 && memEQ(s, (char*)folded, len)
6739 /* If 0, means matched a partial char. See
6741 && map_fold_len_back[len])
6744 /* Advance the target string ptr to account for
6745 * this fold, but have to translate from the
6746 * folded length to the corresponding source
6749 *lenp = map_fold_len_back[len];
6758 /* If we allocated a string above, free it */
6759 if (! utf8_target) Safefree(utf8_p);
6764 return (flags & ANYOF_INVERT) ? !match : match;
6768 S_reghop3(U8 *s, I32 off, const U8* lim)
6772 PERL_ARGS_ASSERT_REGHOP3;
6775 while (off-- && s < lim) {
6776 /* XXX could check well-formedness here */
6781 while (off++ && s > lim) {
6783 if (UTF8_IS_CONTINUED(*s)) {
6784 while (s > lim && UTF8_IS_CONTINUATION(*s))
6787 /* XXX could check well-formedness here */
6794 /* there are a bunch of places where we use two reghop3's that should
6795 be replaced with this routine. but since thats not done yet
6796 we ifdef it out - dmq
6799 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6803 PERL_ARGS_ASSERT_REGHOP4;
6806 while (off-- && s < rlim) {
6807 /* XXX could check well-formedness here */
6812 while (off++ && s > llim) {
6814 if (UTF8_IS_CONTINUED(*s)) {
6815 while (s > llim && UTF8_IS_CONTINUATION(*s))
6818 /* XXX could check well-formedness here */
6826 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6830 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6833 while (off-- && s < lim) {
6834 /* XXX could check well-formedness here */
6841 while (off++ && s > lim) {
6843 if (UTF8_IS_CONTINUED(*s)) {
6844 while (s > lim && UTF8_IS_CONTINUATION(*s))
6847 /* XXX could check well-formedness here */
6856 restore_pos(pTHX_ void *arg)
6859 regexp * const rex = (regexp *)arg;
6860 if (PL_reg_eval_set) {
6861 if (PL_reg_oldsaved) {
6862 rex->subbeg = PL_reg_oldsaved;
6863 rex->sublen = PL_reg_oldsavedlen;
6864 #ifdef PERL_OLD_COPY_ON_WRITE
6865 rex->saved_copy = PL_nrs;
6867 RXp_MATCH_COPIED_on(rex);
6869 PL_reg_magic->mg_len = PL_reg_oldpos;
6870 PL_reg_eval_set = 0;
6871 PL_curpm = PL_reg_oldcurpm;
6876 S_to_utf8_substr(pTHX_ register regexp *prog)
6880 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6883 if (prog->substrs->data[i].substr
6884 && !prog->substrs->data[i].utf8_substr) {
6885 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6886 prog->substrs->data[i].utf8_substr = sv;
6887 sv_utf8_upgrade(sv);
6888 if (SvVALID(prog->substrs->data[i].substr)) {
6889 if (SvTAIL(prog->substrs->data[i].substr)) {
6890 /* Trim the trailing \n that fbm_compile added last
6892 SvCUR_set(sv, SvCUR(sv) - 1);
6893 /* Whilst this makes the SV technically "invalid" (as its
6894 buffer is no longer followed by "\0") when fbm_compile()
6895 adds the "\n" back, a "\0" is restored. */
6896 fbm_compile(sv, FBMcf_TAIL);
6900 if (prog->substrs->data[i].substr == prog->check_substr)
6901 prog->check_utf8 = sv;
6907 S_to_byte_substr(pTHX_ register regexp *prog)
6912 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6915 if (prog->substrs->data[i].utf8_substr
6916 && !prog->substrs->data[i].substr) {
6917 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6918 if (sv_utf8_downgrade(sv, TRUE)) {
6919 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6920 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
6921 /* Trim the trailing \n that fbm_compile added last
6923 SvCUR_set(sv, SvCUR(sv) - 1);
6924 fbm_compile(sv, FBMcf_TAIL);
6932 prog->substrs->data[i].substr = sv;
6933 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6934 prog->check_substr = sv;
6941 * c-indentation-style: bsd
6943 * indent-tabs-mode: t
6946 * ex: set ts=8 sts=4 sw=4 noet: