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
75 #undef PERL_IN_XSUB_RE
76 #define PERL_IN_XSUB_RE 1
79 #undef PERL_IN_XSUB_RE
81 #ifdef PERL_IN_XSUB_RE
87 #define RF_tainted 1 /* tainted information used? e.g. locale */
88 #define RF_warned 2 /* warned about big count? */
90 #define RF_utf8 8 /* Pattern contains multibyte chars? */
92 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
94 #define RS_init 1 /* eval environment created */
95 #define RS_set 2 /* replsv value is set */
101 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
102 * call if there are no complications: i.e., if everything matchable is
103 * straight forward in the bitmap */
104 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
105 : ANYOF_BITMAP_TEST(p,*(c)))
111 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
112 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
114 #define HOPc(pos,off) \
115 (char *)(PL_reg_match_utf8 \
116 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
118 #define HOPBACKc(pos, off) \
119 (char*)(PL_reg_match_utf8\
120 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
121 : (pos - off >= PL_bostr) \
125 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
126 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
128 /* these are unrolled below in the CCC_TRY_XXX defined */
129 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
130 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); 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)) { bool throw_away; ENTER; save_re_context(); throw_away = CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END
136 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
137 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
138 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
140 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
141 LOAD_UTF8_CHARCLASS(X_begin, " "); \
142 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
143 /* These are utf8 constants, and not utf-ebcdic constants, so the \
144 * assert should likely and hopefully fail on an EBCDIC machine */ \
145 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
147 /* No asserts are done for these, in case called on an early \
148 * Unicode version in which they map to nothing */ \
149 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
150 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
151 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
152 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
153 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
154 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
155 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
157 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
159 /* The actual code for CCC_TRY, which uses several variables from the routine
160 * it's callable from. It is designed to be the bulk of a case statement.
161 * FUNC is the macro or function to call on non-utf8 targets that indicate if
162 * nextchr matches the class.
163 * UTF8_TEST is the whole test string to use for utf8 targets
164 * LOAD is what to use to test, and if not present to load in the swash for the
166 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
168 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
169 * utf8 and a variant, load the swash if necessary and test using the utf8
170 * test. Advance to the next character if test is ok, otherwise fail; If not
171 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
172 * fails, or advance to the next character */
174 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
175 if (locinput >= PL_regeol) { \
178 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
179 LOAD_UTF8_CHARCLASS(CLASS, STR); \
180 if (POS_OR_NEG (UTF8_TEST)) { \
183 locinput += PL_utf8skip[nextchr]; \
184 nextchr = UCHARAT(locinput); \
187 if (POS_OR_NEG (FUNC(nextchr))) { \
190 nextchr = UCHARAT(++locinput); \
193 /* Handle the non-locale cases for a character class and its complement. It
194 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
195 * This is because that code fails when the test succeeds, so we want to have
196 * the test fail so that the code succeeds. The swash is stored in a
197 * predictable PL_ place */
198 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
201 _CCC_TRY_CODE( !, FUNC, \
202 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
203 (U8*)locinput, TRUE)), \
206 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
207 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
208 (U8*)locinput, TRUE)), \
211 /* Generate the case statements for both locale and non-locale character
212 * classes in regmatch for classes that don't have special unicode semantics.
213 * Locales don't use an immediate swash, but an intermediary special locale
214 * function that is called on the pointer to the current place in the input
215 * string. That function will resolve to needing the same swash. One might
216 * think that because we don't know what the locale will match, we shouldn't
217 * check with the swash loading function that it loaded properly; ie, that we
218 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
219 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
221 #define CCC_TRY(NAME, NNAME, FUNC, \
222 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
223 NAMEA, NNAMEA, FUNCA, \
226 PL_reg_flags |= RF_tainted; \
227 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
229 PL_reg_flags |= RF_tainted; \
230 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
233 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
236 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
237 nextchr = UCHARAT(++locinput); \
240 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
244 locinput += PL_utf8skip[nextchr]; \
245 nextchr = UCHARAT(locinput); \
248 nextchr = UCHARAT(++locinput); \
251 /* Generate the non-locale cases */ \
252 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
254 /* This is like CCC_TRY, but has an extra set of parameters for generating case
255 * statements to handle separate Unicode semantics nodes */
256 #define CCC_TRY_U(NAME, NNAME, FUNC, \
257 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
258 NAMEU, NNAMEU, FUNCU, \
259 NAMEA, NNAMEA, FUNCA, \
261 CCC_TRY(NAME, NNAME, FUNC, \
262 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
263 NAMEA, NNAMEA, FUNCA, \
265 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
267 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
269 /* for use after a quantifier and before an EXACT-like node -- japhy */
270 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
272 * NOTE that *nothing* that affects backtracking should be in here, specifically
273 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
274 * node that is in between two EXACT like nodes when ascertaining what the required
275 * "follow" character is. This should probably be moved to regex compile time
276 * although it may be done at run time beause of the REF possibility - more
277 * investigation required. -- demerphq
279 #define JUMPABLE(rn) ( \
281 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
283 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
284 OP(rn) == PLUS || OP(rn) == MINMOD || \
286 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
288 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
290 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
293 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
294 we don't need this definition. */
295 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
296 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
297 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
300 /* ... so we use this as its faster. */
301 #define IS_TEXT(rn) ( OP(rn)==EXACT )
302 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
303 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
304 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
309 Search for mandatory following text node; for lookahead, the text must
310 follow but for lookbehind (rn->flags != 0) we skip to the next step.
312 #define FIND_NEXT_IMPT(rn) STMT_START { \
313 while (JUMPABLE(rn)) { \
314 const OPCODE type = OP(rn); \
315 if (type == SUSPEND || PL_regkind[type] == CURLY) \
316 rn = NEXTOPER(NEXTOPER(rn)); \
317 else if (type == PLUS) \
319 else if (type == IFMATCH) \
320 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
321 else rn += NEXT_OFF(rn); \
326 static void restore_pos(pTHX_ void *arg);
328 #define REGCP_PAREN_ELEMS 4
329 #define REGCP_OTHER_ELEMS 5
330 #define REGCP_FRAME_ELEMS 1
331 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
332 * are needed for the regexp context stack bookkeeping. */
335 S_regcppush(pTHX_ I32 parenfloor)
338 const int retval = PL_savestack_ix;
339 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
340 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
341 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
343 GET_RE_DEBUG_FLAGS_DECL;
345 if (paren_elems_to_push < 0)
346 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
348 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
349 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
350 " out of range (%lu-%ld)",
351 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
353 SSGROW(total_elems + REGCP_FRAME_ELEMS);
355 for (p = PL_regsize; p > parenfloor; p--) {
356 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
357 SSPUSHINT(PL_regoffs[p].end);
358 SSPUSHINT(PL_regoffs[p].start);
359 SSPUSHPTR(PL_reg_start_tmp[p]);
361 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
362 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
363 (UV)p, (IV)PL_regoffs[p].start,
364 (IV)(PL_reg_start_tmp[p] - PL_bostr),
365 (IV)PL_regoffs[p].end
368 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
369 SSPUSHPTR(PL_regoffs);
370 SSPUSHINT(PL_regsize);
371 SSPUSHINT(*PL_reglastparen);
372 SSPUSHINT(*PL_reglastcloseparen);
373 SSPUSHPTR(PL_reginput);
374 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
379 /* These are needed since we do not localize EVAL nodes: */
380 #define REGCP_SET(cp) \
382 PerlIO_printf(Perl_debug_log, \
383 " Setting an EVAL scope, savestack=%"IVdf"\n", \
384 (IV)PL_savestack_ix)); \
387 #define REGCP_UNWIND(cp) \
389 if (cp != PL_savestack_ix) \
390 PerlIO_printf(Perl_debug_log, \
391 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
392 (IV)(cp), (IV)PL_savestack_ix)); \
396 S_regcppop(pTHX_ const regexp *rex)
401 GET_RE_DEBUG_FLAGS_DECL;
403 PERL_ARGS_ASSERT_REGCPPOP;
405 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
407 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
408 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
409 input = (char *) SSPOPPTR;
410 *PL_reglastcloseparen = SSPOPINT;
411 *PL_reglastparen = SSPOPINT;
412 PL_regsize = SSPOPINT;
413 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
415 i -= REGCP_OTHER_ELEMS;
416 /* Now restore the parentheses context. */
417 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
419 U32 paren = (U32)SSPOPINT;
420 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
421 PL_regoffs[paren].start = SSPOPINT;
423 if (paren <= *PL_reglastparen)
424 PL_regoffs[paren].end = tmps;
426 PerlIO_printf(Perl_debug_log,
427 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
428 (UV)paren, (IV)PL_regoffs[paren].start,
429 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
430 (IV)PL_regoffs[paren].end,
431 (paren > *PL_reglastparen ? "(no)" : ""));
435 if (*PL_reglastparen + 1 <= rex->nparens) {
436 PerlIO_printf(Perl_debug_log,
437 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
438 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
442 /* It would seem that the similar code in regtry()
443 * already takes care of this, and in fact it is in
444 * a better location to since this code can #if 0-ed out
445 * but the code in regtry() is needed or otherwise tests
446 * requiring null fields (pat.t#187 and split.t#{13,14}
447 * (as of patchlevel 7877) will fail. Then again,
448 * this code seems to be necessary or otherwise
449 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
450 * --jhi updated by dapm */
451 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
453 PL_regoffs[i].start = -1;
454 PL_regoffs[i].end = -1;
460 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
463 * pregexec and friends
466 #ifndef PERL_IN_XSUB_RE
468 - pregexec - match a regexp against a string
471 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
472 char *strbeg, I32 minend, SV *screamer, U32 nosave)
473 /* strend: pointer to null at end of string */
474 /* strbeg: real beginning of string */
475 /* minend: end of match must be >=minend after stringarg. */
476 /* nosave: For optimizations. */
478 PERL_ARGS_ASSERT_PREGEXEC;
481 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
482 nosave ? 0 : REXEC_COPY_STR);
487 * Need to implement the following flags for reg_anch:
489 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
491 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
492 * INTUIT_AUTORITATIVE_ML
493 * INTUIT_ONCE_NOML - Intuit can match in one location only.
496 * Another flag for this function: SECOND_TIME (so that float substrs
497 * with giant delta may be not rechecked).
500 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
502 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
503 Otherwise, only SvCUR(sv) is used to get strbeg. */
505 /* XXXX We assume that strpos is strbeg unless sv. */
507 /* XXXX Some places assume that there is a fixed substring.
508 An update may be needed if optimizer marks as "INTUITable"
509 RExen without fixed substrings. Similarly, it is assumed that
510 lengths of all the strings are no more than minlen, thus they
511 cannot come from lookahead.
512 (Or minlen should take into account lookahead.)
513 NOTE: Some of this comment is not correct. minlen does now take account
514 of lookahead/behind. Further research is required. -- demerphq
518 /* A failure to find a constant substring means that there is no need to make
519 an expensive call to REx engine, thus we celebrate a failure. Similarly,
520 finding a substring too deep into the string means that less calls to
521 regtry() should be needed.
523 REx compiler's optimizer found 4 possible hints:
524 a) Anchored substring;
526 c) Whether we are anchored (beginning-of-line or \G);
527 d) First node (of those at offset 0) which may distinguish positions;
528 We use a)b)d) and multiline-part of c), and try to find a position in the
529 string which does not contradict any of them.
532 /* Most of decisions we do here should have been done at compile time.
533 The nodes of the REx which we used for the search should have been
534 deleted from the finite automaton. */
537 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
538 char *strend, const U32 flags, re_scream_pos_data *data)
541 struct regexp *const prog = (struct regexp *)SvANY(rx);
542 register I32 start_shift = 0;
543 /* Should be nonnegative! */
544 register I32 end_shift = 0;
549 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
551 register char *other_last = NULL; /* other substr checked before this */
552 char *check_at = NULL; /* check substr found at this pos */
553 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
554 RXi_GET_DECL(prog,progi);
556 const char * const i_strpos = strpos;
558 GET_RE_DEBUG_FLAGS_DECL;
560 PERL_ARGS_ASSERT_RE_INTUIT_START;
562 RX_MATCH_UTF8_set(rx,utf8_target);
565 PL_reg_flags |= RF_utf8;
568 debug_start_match(rx, utf8_target, strpos, strend,
569 sv ? "Guessing start of match in sv for"
570 : "Guessing start of match in string for");
573 /* CHR_DIST() would be more correct here but it makes things slow. */
574 if (prog->minlen > strend - strpos) {
575 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
576 "String too short... [re_intuit_start]\n"));
580 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
583 if (!prog->check_utf8 && prog->check_substr)
584 to_utf8_substr(prog);
585 check = prog->check_utf8;
587 if (!prog->check_substr && prog->check_utf8)
588 to_byte_substr(prog);
589 check = prog->check_substr;
591 if (check == &PL_sv_undef) {
592 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
593 "Non-utf8 string cannot match utf8 check string\n"));
596 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
597 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
598 || ( (prog->extflags & RXf_ANCH_BOL)
599 && !multiline ) ); /* Check after \n? */
602 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
603 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
604 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
606 && (strpos != strbeg)) {
607 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
610 if (prog->check_offset_min == prog->check_offset_max &&
611 !(prog->extflags & RXf_CANY_SEEN)) {
612 /* Substring at constant offset from beg-of-str... */
615 s = HOP3c(strpos, prog->check_offset_min, strend);
618 slen = SvCUR(check); /* >= 1 */
620 if ( strend - s > slen || strend - s < slen - 1
621 || (strend - s == slen && strend[-1] != '\n')) {
622 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
625 /* Now should match s[0..slen-2] */
627 if (slen && (*SvPVX_const(check) != *s
629 && memNE(SvPVX_const(check), s, slen)))) {
631 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
635 else if (*SvPVX_const(check) != *s
636 || ((slen = SvCUR(check)) > 1
637 && memNE(SvPVX_const(check), s, slen)))
640 goto success_at_start;
643 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
645 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
646 end_shift = prog->check_end_shift;
649 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
650 - (SvTAIL(check) != 0);
651 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
653 if (end_shift < eshift)
657 else { /* Can match at random position */
660 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
661 end_shift = prog->check_end_shift;
663 /* end shift should be non negative here */
666 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
668 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
669 (IV)end_shift, RX_PRECOMP(prog));
673 /* Find a possible match in the region s..strend by looking for
674 the "check" substring in the region corrected by start/end_shift. */
677 I32 srch_start_shift = start_shift;
678 I32 srch_end_shift = end_shift;
679 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
680 srch_end_shift -= ((strbeg - s) - srch_start_shift);
681 srch_start_shift = strbeg - s;
683 DEBUG_OPTIMISE_MORE_r({
684 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
685 (IV)prog->check_offset_min,
686 (IV)srch_start_shift,
688 (IV)prog->check_end_shift);
691 if (flags & REXEC_SCREAM) {
692 I32 p = -1; /* Internal iterator of scream. */
693 I32 * const pp = data ? data->scream_pos : &p;
695 if (PL_screamfirst[BmRARE(check)] >= 0
696 || ( BmRARE(check) == '\n'
697 && (BmPREVIOUS(check) == SvCUR(check) - 1)
699 s = screaminstr(sv, check,
700 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
703 /* we may be pointing at the wrong string */
704 if (s && RXp_MATCH_COPIED(prog))
705 s = strbeg + (s - SvPVX_const(sv));
707 *data->scream_olds = s;
712 if (prog->extflags & RXf_CANY_SEEN) {
713 start_point= (U8*)(s + srch_start_shift);
714 end_point= (U8*)(strend - srch_end_shift);
716 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
717 end_point= HOP3(strend, -srch_end_shift, strbeg);
719 DEBUG_OPTIMISE_MORE_r({
720 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
721 (int)(end_point - start_point),
722 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
726 s = fbm_instr( start_point, end_point,
727 check, multiline ? FBMrf_MULTILINE : 0);
730 /* Update the count-of-usability, remove useless subpatterns,
734 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
735 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
736 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
737 (s ? "Found" : "Did not find"),
738 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
739 ? "anchored" : "floating"),
742 (s ? " at offset " : "...\n") );
747 /* Finish the diagnostic message */
748 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
750 /* XXX dmq: first branch is for positive lookbehind...
751 Our check string is offset from the beginning of the pattern.
752 So we need to do any stclass tests offset forward from that
761 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
762 Start with the other substr.
763 XXXX no SCREAM optimization yet - and a very coarse implementation
764 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
765 *always* match. Probably should be marked during compile...
766 Probably it is right to do no SCREAM here...
769 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
770 : (prog->float_substr && prog->anchored_substr))
772 /* Take into account the "other" substring. */
773 /* XXXX May be hopelessly wrong for UTF... */
776 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
779 char * const last = HOP3c(s, -start_shift, strbeg);
781 char * const saved_s = s;
784 t = s - prog->check_offset_max;
785 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
787 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
792 t = HOP3c(t, prog->anchored_offset, strend);
793 if (t < other_last) /* These positions already checked */
795 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
798 /* XXXX It is not documented what units *_offsets are in.
799 We assume bytes, but this is clearly wrong.
800 Meaning this code needs to be carefully reviewed for errors.
804 /* On end-of-str: see comment below. */
805 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
806 if (must == &PL_sv_undef) {
808 DEBUG_r(must = prog->anchored_utf8); /* for debug */
813 HOP3(HOP3(last1, prog->anchored_offset, strend)
814 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
816 multiline ? FBMrf_MULTILINE : 0
819 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
820 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
821 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
822 (s ? "Found" : "Contradicts"),
823 quoted, RE_SV_TAIL(must));
828 if (last1 >= last2) {
829 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
830 ", giving up...\n"));
833 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
834 ", trying floating at offset %ld...\n",
835 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
836 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
837 s = HOP3c(last, 1, strend);
841 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
842 (long)(s - i_strpos)));
843 t = HOP3c(s, -prog->anchored_offset, strbeg);
844 other_last = HOP3c(s, 1, strend);
852 else { /* Take into account the floating substring. */
854 char * const saved_s = s;
857 t = HOP3c(s, -start_shift, strbeg);
859 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
860 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
861 last = HOP3c(t, prog->float_max_offset, strend);
862 s = HOP3c(t, prog->float_min_offset, strend);
865 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
866 must = utf8_target ? prog->float_utf8 : prog->float_substr;
867 /* fbm_instr() takes into account exact value of end-of-str
868 if the check is SvTAIL(ed). Since false positives are OK,
869 and end-of-str is not later than strend we are OK. */
870 if (must == &PL_sv_undef) {
872 DEBUG_r(must = prog->float_utf8); /* for debug message */
875 s = fbm_instr((unsigned char*)s,
876 (unsigned char*)last + SvCUR(must)
878 must, multiline ? FBMrf_MULTILINE : 0);
880 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
881 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
882 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
883 (s ? "Found" : "Contradicts"),
884 quoted, RE_SV_TAIL(must));
888 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
889 ", giving up...\n"));
892 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
893 ", trying anchored starting at offset %ld...\n",
894 (long)(saved_s + 1 - i_strpos)));
896 s = HOP3c(t, 1, strend);
900 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
901 (long)(s - i_strpos)));
902 other_last = s; /* Fix this later. --Hugo */
912 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
914 DEBUG_OPTIMISE_MORE_r(
915 PerlIO_printf(Perl_debug_log,
916 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
917 (IV)prog->check_offset_min,
918 (IV)prog->check_offset_max,
926 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
928 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
931 /* Fixed substring is found far enough so that the match
932 cannot start at strpos. */
934 if (ml_anch && t[-1] != '\n') {
935 /* Eventually fbm_*() should handle this, but often
936 anchored_offset is not 0, so this check will not be wasted. */
937 /* XXXX In the code below we prefer to look for "^" even in
938 presence of anchored substrings. And we search even
939 beyond the found float position. These pessimizations
940 are historical artefacts only. */
942 while (t < strend - prog->minlen) {
944 if (t < check_at - prog->check_offset_min) {
945 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
946 /* Since we moved from the found position,
947 we definitely contradict the found anchored
948 substr. Due to the above check we do not
949 contradict "check" substr.
950 Thus we can arrive here only if check substr
951 is float. Redo checking for "other"=="fixed".
954 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
955 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
956 goto do_other_anchored;
958 /* We don't contradict the found floating substring. */
959 /* XXXX Why not check for STCLASS? */
961 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
962 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
965 /* Position contradicts check-string */
966 /* XXXX probably better to look for check-string
967 than for "\n", so one should lower the limit for t? */
968 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
969 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
970 other_last = strpos = s = t + 1;
975 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
976 PL_colors[0], PL_colors[1]));
980 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
981 PL_colors[0], PL_colors[1]));
985 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
988 /* The found string does not prohibit matching at strpos,
989 - no optimization of calling REx engine can be performed,
990 unless it was an MBOL and we are not after MBOL,
991 or a future STCLASS check will fail this. */
993 /* Even in this situation we may use MBOL flag if strpos is offset
994 wrt the start of the string. */
995 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
996 && (strpos != strbeg) && strpos[-1] != '\n'
997 /* May be due to an implicit anchor of m{.*foo} */
998 && !(prog->intflags & PREGf_IMPLICIT))
1003 DEBUG_EXECUTE_r( if (ml_anch)
1004 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1005 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1008 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1010 prog->check_utf8 /* Could be deleted already */
1011 && --BmUSEFUL(prog->check_utf8) < 0
1012 && (prog->check_utf8 == prog->float_utf8)
1014 prog->check_substr /* Could be deleted already */
1015 && --BmUSEFUL(prog->check_substr) < 0
1016 && (prog->check_substr == prog->float_substr)
1019 /* If flags & SOMETHING - do not do it many times on the same match */
1020 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1021 /* XXX Does the destruction order has to change with utf8_target? */
1022 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1023 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1024 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1025 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1026 check = NULL; /* abort */
1028 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1029 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1030 if (prog->intflags & PREGf_IMPLICIT)
1031 prog->extflags &= ~RXf_ANCH_MBOL;
1032 /* XXXX This is a remnant of the old implementation. It
1033 looks wasteful, since now INTUIT can use many
1034 other heuristics. */
1035 prog->extflags &= ~RXf_USE_INTUIT;
1036 /* XXXX What other flags might need to be cleared in this branch? */
1042 /* Last resort... */
1043 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1044 /* trie stclasses are too expensive to use here, we are better off to
1045 leave it to regmatch itself */
1046 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1047 /* minlen == 0 is possible if regstclass is \b or \B,
1048 and the fixed substr is ''$.
1049 Since minlen is already taken into account, s+1 is before strend;
1050 accidentally, minlen >= 1 guaranties no false positives at s + 1
1051 even for \b or \B. But (minlen? 1 : 0) below assumes that
1052 regstclass does not come from lookahead... */
1053 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1054 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1055 const U8* const str = (U8*)STRING(progi->regstclass);
1056 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1057 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1060 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1061 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1062 else if (prog->float_substr || prog->float_utf8)
1063 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1067 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1068 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1071 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1074 const char *what = NULL;
1076 if (endpos == strend) {
1077 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1078 "Could not match STCLASS...\n") );
1081 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1082 "This position contradicts STCLASS...\n") );
1083 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1085 /* Contradict one of substrings */
1086 if (prog->anchored_substr || prog->anchored_utf8) {
1087 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1088 DEBUG_EXECUTE_r( what = "anchored" );
1090 s = HOP3c(t, 1, strend);
1091 if (s + start_shift + end_shift > strend) {
1092 /* XXXX Should be taken into account earlier? */
1093 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1094 "Could not match STCLASS...\n") );
1099 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1100 "Looking for %s substr starting at offset %ld...\n",
1101 what, (long)(s + start_shift - i_strpos)) );
1104 /* Have both, check_string is floating */
1105 if (t + start_shift >= check_at) /* Contradicts floating=check */
1106 goto retry_floating_check;
1107 /* Recheck anchored substring, but not floating... */
1111 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1112 "Looking for anchored substr starting at offset %ld...\n",
1113 (long)(other_last - i_strpos)) );
1114 goto do_other_anchored;
1116 /* Another way we could have checked stclass at the
1117 current position only: */
1122 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1123 "Looking for /%s^%s/m starting at offset %ld...\n",
1124 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1127 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1129 /* Check is floating substring. */
1130 retry_floating_check:
1131 t = check_at - start_shift;
1132 DEBUG_EXECUTE_r( what = "floating" );
1133 goto hop_and_restart;
1136 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1137 "By STCLASS: moving %ld --> %ld\n",
1138 (long)(t - i_strpos), (long)(s - i_strpos))
1142 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1143 "Does not contradict STCLASS...\n");
1148 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1149 PL_colors[4], (check ? "Guessed" : "Giving up"),
1150 PL_colors[5], (long)(s - i_strpos)) );
1153 fail_finish: /* Substring not found */
1154 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1155 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1157 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1158 PL_colors[4], PL_colors[5]));
1162 #define DECL_TRIE_TYPE(scan) \
1163 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1164 trie_type = (scan->flags != EXACT) \
1165 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1166 : (utf8_target ? trie_utf8 : trie_plain)
1168 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1169 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1170 switch (trie_type) { \
1171 case trie_utf8_fold: \
1172 if ( foldlen>0 ) { \
1173 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1178 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1179 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1180 foldlen -= UNISKIP( uvc ); \
1181 uscan = foldbuf + UNISKIP( uvc ); \
1184 case trie_latin_utf8_fold: \
1185 if ( foldlen>0 ) { \
1186 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1192 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1193 foldlen -= UNISKIP( uvc ); \
1194 uscan = foldbuf + UNISKIP( uvc ); \
1198 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1205 charid = trie->charmap[ uvc ]; \
1209 if (widecharmap) { \
1210 SV** const svpp = hv_fetch(widecharmap, \
1211 (char*)&uvc, sizeof(UV), 0); \
1213 charid = (U16)SvIV(*svpp); \
1218 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1222 && (ln == 1 || folder(s, pat_string, ln)) \
1223 && (!reginfo || regtry(reginfo, &s)) ) \
1229 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1231 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1237 #define REXEC_FBC_SCAN(CoDe) \
1239 while (s < strend) { \
1245 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1246 REXEC_FBC_UTF8_SCAN( \
1248 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1257 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1260 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1269 #define REXEC_FBC_TRYIT \
1270 if ((!reginfo || regtry(reginfo, &s))) \
1273 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1274 if (utf8_target) { \
1275 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1278 REXEC_FBC_CLASS_SCAN(CoNd); \
1281 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1282 if (utf8_target) { \
1284 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1287 REXEC_FBC_CLASS_SCAN(CoNd); \
1290 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1291 PL_reg_flags |= RF_tainted; \
1292 if (utf8_target) { \
1293 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1296 REXEC_FBC_CLASS_SCAN(CoNd); \
1299 #define DUMP_EXEC_POS(li,s,doutf8) \
1300 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1303 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1304 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1305 tmp = TEST_NON_UTF8(tmp); \
1306 REXEC_FBC_UTF8_SCAN( \
1307 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1316 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1317 if (s == PL_bostr) { \
1321 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1322 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1325 LOAD_UTF8_CHARCLASS_ALNUM(); \
1326 REXEC_FBC_UTF8_SCAN( \
1327 if (tmp == ! (TeSt2_UtF8)) { \
1336 /* The only difference between the BOUND and NBOUND cases is that
1337 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1338 * NBOUND. This is accomplished by passing it in either the if or else clause,
1339 * with the other one being empty */
1340 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1341 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1343 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1344 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1346 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1347 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1349 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1350 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1353 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1354 * be passed in completely with the variable name being tested, which isn't
1355 * such a clean interface, but this is easier to read than it was before. We
1356 * are looking for the boundary (or non-boundary between a word and non-word
1357 * character. The utf8 and non-utf8 cases have the same logic, but the details
1358 * must be different. Find the "wordness" of the character just prior to this
1359 * one, and compare it with the wordness of this one. If they differ, we have
1360 * a boundary. At the beginning of the string, pretend that the previous
1361 * character was a new-line */
1362 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1363 if (utf8_target) { \
1366 else { /* Not utf8 */ \
1367 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1368 tmp = TEST_NON_UTF8(tmp); \
1370 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1379 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1382 /* We know what class REx starts with. Try to find this position... */
1383 /* if reginfo is NULL, its a dryrun */
1384 /* annoyingly all the vars in this routine have different names from their counterparts
1385 in regmatch. /grrr */
1388 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1389 const char *strend, regmatch_info *reginfo)
1392 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1393 char *pat_string; /* The pattern's exactish string */
1394 char *pat_end; /* ptr to end char of pat_string */
1395 re_fold_t folder; /* Function for computing non-utf8 folds */
1396 const U8 *fold_array; /* array for folding ords < 256 */
1399 register STRLEN uskip;
1403 register I32 tmp = 1; /* Scratch variable? */
1404 register const bool utf8_target = PL_reg_match_utf8;
1405 UV utf8_fold_flags = 0;
1406 RXi_GET_DECL(prog,progi);
1408 PERL_ARGS_ASSERT_FIND_BYCLASS;
1410 /* We know what class it must start with. */
1414 if (utf8_target || OP(c) == ANYOFV) {
1415 STRLEN inclasslen = strend - s;
1416 REXEC_FBC_UTF8_CLASS_SCAN(
1417 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1420 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1425 if (tmp && (!reginfo || regtry(reginfo, &s)))
1433 if (UTF_PATTERN || utf8_target) {
1434 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1435 goto do_exactf_utf8;
1437 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1438 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1439 goto do_exactf_non_utf8; /* isn't dealt with by these */
1442 if (UTF_PATTERN || utf8_target) {
1443 utf8_fold_flags = 0;
1444 goto do_exactf_utf8;
1446 fold_array = PL_fold_latin1;
1447 folder = foldEQ_latin1;
1448 /* XXX This uses the full utf8 fold because if the pattern contains
1449 * 'ss' it could match LATIN_SMALL_LETTER SHARP_S in the string.
1450 * There could be a new node type, say EXACTFU_SS, which is
1451 * generated by regcomp only if there is an 'ss', and then every
1452 * other case could goto do_exactf_non_utf8;*/
1453 goto do_exactf_utf8;
1456 if (UTF_PATTERN || utf8_target) {
1457 utf8_fold_flags = 0;
1458 goto do_exactf_utf8;
1460 fold_array = PL_fold;
1462 goto do_exactf_non_utf8;
1465 if (UTF_PATTERN || utf8_target) {
1466 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1467 goto do_exactf_utf8;
1469 fold_array = PL_fold_locale;
1470 folder = foldEQ_locale;
1474 do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */
1476 /* The idea in the non-utf8 EXACTF* cases is to first find the
1477 * first character of the EXACTF* node and then, if necessary,
1478 * case-insensitively compare the full text of the node. c1 is the
1479 * first character. c2 is its fold. This logic will not work for
1480 * Unicode semantics and the german sharp ss, which hence should
1481 * not be compiled into a node that gets here. */
1482 pat_string = STRING(c);
1483 ln = STR_LEN(c); /* length to match in octets/bytes */
1485 e = HOP3c(strend, -((I32)ln), s);
1487 if (!reginfo && e < s) {
1488 e = s; /* Due to minlen logic of intuit() */
1492 c2 = fold_array[c1];
1493 if (c1 == c2) { /* If char and fold are the same */
1494 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1497 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1503 /* If one of the operands is in utf8, we can't use the simpler
1504 * folding above, due to the fact that many different characters
1505 * can have the same fold, or portion of a fold, or different-
1507 pat_string = STRING(c);
1508 ln = STR_LEN(c); /* length to match in octets/bytes */
1509 pat_end = pat_string + ln;
1510 lnc = (UTF_PATTERN) /* length to match in characters */
1511 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1514 e = HOP3c(strend, -((I32)lnc), s);
1516 if (!reginfo && e < s) {
1517 e = s; /* Due to minlen logic of intuit() */
1521 char *my_strend= (char *)strend;
1522 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1523 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1524 && (!reginfo || regtry(reginfo, &s)) )
1532 PL_reg_flags |= RF_tainted;
1533 FBC_BOUND(isALNUM_LC,
1534 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1535 isALNUM_LC_utf8((U8*)s));
1538 PL_reg_flags |= RF_tainted;
1539 FBC_NBOUND(isALNUM_LC,
1540 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1541 isALNUM_LC_utf8((U8*)s));
1544 FBC_BOUND(isWORDCHAR,
1546 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1549 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1551 isWORDCHAR_A((U8*)s));
1554 FBC_NBOUND(isWORDCHAR,
1556 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1559 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1561 isWORDCHAR_A((U8*)s));
1564 FBC_BOUND(isWORDCHAR_L1,
1566 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1569 FBC_NBOUND(isWORDCHAR_L1,
1571 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1574 REXEC_FBC_CSCAN_TAINT(
1575 isALNUM_LC_utf8((U8*)s),
1580 REXEC_FBC_CSCAN_PRELOAD(
1581 LOAD_UTF8_CHARCLASS_ALNUM(),
1582 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1583 isWORDCHAR_L1((U8) *s)
1587 REXEC_FBC_CSCAN_PRELOAD(
1588 LOAD_UTF8_CHARCLASS_ALNUM(),
1589 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1594 /* Don't need to worry about utf8, as it can match only a single
1595 * byte invariant character */
1596 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1599 REXEC_FBC_CSCAN_PRELOAD(
1600 LOAD_UTF8_CHARCLASS_ALNUM(),
1601 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1602 ! isWORDCHAR_L1((U8) *s)
1606 REXEC_FBC_CSCAN_PRELOAD(
1607 LOAD_UTF8_CHARCLASS_ALNUM(),
1608 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1619 REXEC_FBC_CSCAN_TAINT(
1620 !isALNUM_LC_utf8((U8*)s),
1625 REXEC_FBC_CSCAN_PRELOAD(
1626 LOAD_UTF8_CHARCLASS_SPACE(),
1627 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1632 REXEC_FBC_CSCAN_PRELOAD(
1633 LOAD_UTF8_CHARCLASS_SPACE(),
1634 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1639 /* Don't need to worry about utf8, as it can match only a single
1640 * byte invariant character */
1641 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1644 REXEC_FBC_CSCAN_TAINT(
1645 isSPACE_LC_utf8((U8*)s),
1650 REXEC_FBC_CSCAN_PRELOAD(
1651 LOAD_UTF8_CHARCLASS_SPACE(),
1652 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1653 ! isSPACE_L1((U8) *s)
1657 REXEC_FBC_CSCAN_PRELOAD(
1658 LOAD_UTF8_CHARCLASS_SPACE(),
1659 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1670 REXEC_FBC_CSCAN_TAINT(
1671 !isSPACE_LC_utf8((U8*)s),
1676 REXEC_FBC_CSCAN_PRELOAD(
1677 LOAD_UTF8_CHARCLASS_DIGIT(),
1678 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1683 /* Don't need to worry about utf8, as it can match only a single
1684 * byte invariant character */
1685 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1688 REXEC_FBC_CSCAN_TAINT(
1689 isDIGIT_LC_utf8((U8*)s),
1694 REXEC_FBC_CSCAN_PRELOAD(
1695 LOAD_UTF8_CHARCLASS_DIGIT(),
1696 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1707 REXEC_FBC_CSCAN_TAINT(
1708 !isDIGIT_LC_utf8((U8*)s),
1715 is_LNBREAK_latin1(s)
1727 !is_VERTWS_latin1(s)
1733 is_HORIZWS_latin1(s)
1738 !is_HORIZWS_utf8(s),
1739 !is_HORIZWS_latin1(s)
1746 /* what trie are we using right now */
1748 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1750 = (reg_trie_data*)progi->data->data[ aho->trie ];
1751 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1753 const char *last_start = strend - trie->minlen;
1755 const char *real_start = s;
1757 STRLEN maxlen = trie->maxlen;
1759 U8 **points; /* map of where we were in the input string
1760 when reading a given char. For ASCII this
1761 is unnecessary overhead as the relationship
1762 is always 1:1, but for Unicode, especially
1763 case folded Unicode this is not true. */
1764 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1768 GET_RE_DEBUG_FLAGS_DECL;
1770 /* We can't just allocate points here. We need to wrap it in
1771 * an SV so it gets freed properly if there is a croak while
1772 * running the match */
1775 sv_points=newSV(maxlen * sizeof(U8 *));
1776 SvCUR_set(sv_points,
1777 maxlen * sizeof(U8 *));
1778 SvPOK_on(sv_points);
1779 sv_2mortal(sv_points);
1780 points=(U8**)SvPV_nolen(sv_points );
1781 if ( trie_type != trie_utf8_fold
1782 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1785 bitmap=(U8*)trie->bitmap;
1787 bitmap=(U8*)ANYOF_BITMAP(c);
1789 /* this is the Aho-Corasick algorithm modified a touch
1790 to include special handling for long "unknown char"
1791 sequences. The basic idea being that we use AC as long
1792 as we are dealing with a possible matching char, when
1793 we encounter an unknown char (and we have not encountered
1794 an accepting state) we scan forward until we find a legal
1796 AC matching is basically that of trie matching, except
1797 that when we encounter a failing transition, we fall back
1798 to the current states "fail state", and try the current char
1799 again, a process we repeat until we reach the root state,
1800 state 1, or a legal transition. If we fail on the root state
1801 then we can either terminate if we have reached an accepting
1802 state previously, or restart the entire process from the beginning
1806 while (s <= last_start) {
1807 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1815 U8 *uscan = (U8*)NULL;
1816 U8 *leftmost = NULL;
1818 U32 accepted_word= 0;
1822 while ( state && uc <= (U8*)strend ) {
1824 U32 word = aho->states[ state ].wordnum;
1828 DEBUG_TRIE_EXECUTE_r(
1829 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1830 dump_exec_pos( (char *)uc, c, strend, real_start,
1831 (char *)uc, utf8_target );
1832 PerlIO_printf( Perl_debug_log,
1833 " Scanning for legal start char...\n");
1837 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1841 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1847 if (uc >(U8*)last_start) break;
1851 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1852 if (!leftmost || lpos < leftmost) {
1853 DEBUG_r(accepted_word=word);
1859 points[pointpos++ % maxlen]= uc;
1860 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1861 uscan, len, uvc, charid, foldlen,
1863 DEBUG_TRIE_EXECUTE_r({
1864 dump_exec_pos( (char *)uc, c, strend, real_start,
1866 PerlIO_printf(Perl_debug_log,
1867 " Charid:%3u CP:%4"UVxf" ",
1873 word = aho->states[ state ].wordnum;
1875 base = aho->states[ state ].trans.base;
1877 DEBUG_TRIE_EXECUTE_r({
1879 dump_exec_pos( (char *)uc, c, strend, real_start,
1881 PerlIO_printf( Perl_debug_log,
1882 "%sState: %4"UVxf", word=%"UVxf,
1883 failed ? " Fail transition to " : "",
1884 (UV)state, (UV)word);
1890 ( ((offset = base + charid
1891 - 1 - trie->uniquecharcount)) >= 0)
1892 && ((U32)offset < trie->lasttrans)
1893 && trie->trans[offset].check == state
1894 && (tmp=trie->trans[offset].next))
1896 DEBUG_TRIE_EXECUTE_r(
1897 PerlIO_printf( Perl_debug_log," - legal\n"));
1902 DEBUG_TRIE_EXECUTE_r(
1903 PerlIO_printf( Perl_debug_log," - fail\n"));
1905 state = aho->fail[state];
1909 /* we must be accepting here */
1910 DEBUG_TRIE_EXECUTE_r(
1911 PerlIO_printf( Perl_debug_log," - accepting\n"));
1920 if (!state) state = 1;
1923 if ( aho->states[ state ].wordnum ) {
1924 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1925 if (!leftmost || lpos < leftmost) {
1926 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1931 s = (char*)leftmost;
1932 DEBUG_TRIE_EXECUTE_r({
1934 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1935 (UV)accepted_word, (IV)(s - real_start)
1938 if (!reginfo || regtry(reginfo, &s)) {
1944 DEBUG_TRIE_EXECUTE_r({
1945 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1948 DEBUG_TRIE_EXECUTE_r(
1949 PerlIO_printf( Perl_debug_log,"No match.\n"));
1958 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1968 - regexec_flags - match a regexp against a string
1971 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1972 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1973 /* strend: pointer to null at end of string */
1974 /* strbeg: real beginning of string */
1975 /* minend: end of match must be >=minend after stringarg. */
1976 /* data: May be used for some additional optimizations.
1977 Currently its only used, with a U32 cast, for transmitting
1978 the ganch offset when doing a /g match. This will change */
1979 /* nosave: For optimizations. */
1982 struct regexp *const prog = (struct regexp *)SvANY(rx);
1983 /*register*/ char *s;
1984 register regnode *c;
1985 /*register*/ char *startpos = stringarg;
1986 I32 minlen; /* must match at least this many chars */
1987 I32 dontbother = 0; /* how many characters not to try at end */
1988 I32 end_shift = 0; /* Same for the end. */ /* CC */
1989 I32 scream_pos = -1; /* Internal iterator of scream. */
1990 char *scream_olds = NULL;
1991 const bool utf8_target = cBOOL(DO_UTF8(sv));
1993 RXi_GET_DECL(prog,progi);
1994 regmatch_info reginfo; /* create some info to pass to regtry etc */
1995 regexp_paren_pair *swap = NULL;
1996 GET_RE_DEBUG_FLAGS_DECL;
1998 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1999 PERL_UNUSED_ARG(data);
2001 /* Be paranoid... */
2002 if (prog == NULL || startpos == NULL) {
2003 Perl_croak(aTHX_ "NULL regexp parameter");
2007 multiline = prog->extflags & RXf_PMf_MULTILINE;
2008 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2010 RX_MATCH_UTF8_set(rx, utf8_target);
2012 debug_start_match(rx, utf8_target, startpos, strend,
2016 minlen = prog->minlen;
2018 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2019 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2020 "String too short [regexec_flags]...\n"));
2025 /* Check validity of program. */
2026 if (UCHARAT(progi->program) != REG_MAGIC) {
2027 Perl_croak(aTHX_ "corrupted regexp program");
2031 PL_reg_eval_set = 0;
2035 PL_reg_flags |= RF_utf8;
2037 /* Mark beginning of line for ^ and lookbehind. */
2038 reginfo.bol = startpos; /* XXX not used ??? */
2042 /* Mark end of line for $ (and such) */
2045 /* see how far we have to get to not match where we matched before */
2046 reginfo.till = startpos+minend;
2048 /* If there is a "must appear" string, look for it. */
2051 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2053 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2054 reginfo.ganch = startpos + prog->gofs;
2055 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2056 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2057 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2059 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2060 && mg->mg_len >= 0) {
2061 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2062 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2063 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2065 if (prog->extflags & RXf_ANCH_GPOS) {
2066 if (s > reginfo.ganch)
2068 s = reginfo.ganch - prog->gofs;
2069 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2070 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2076 reginfo.ganch = strbeg + PTR2UV(data);
2077 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2078 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2080 } else { /* pos() not defined */
2081 reginfo.ganch = strbeg;
2082 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2083 "GPOS: reginfo.ganch = strbeg\n"));
2086 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2087 /* We have to be careful. If the previous successful match
2088 was from this regex we don't want a subsequent partially
2089 successful match to clobber the old results.
2090 So when we detect this possibility we add a swap buffer
2091 to the re, and switch the buffer each match. If we fail
2092 we switch it back, otherwise we leave it swapped.
2095 /* do we need a save destructor here for eval dies? */
2096 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2098 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2099 re_scream_pos_data d;
2101 d.scream_olds = &scream_olds;
2102 d.scream_pos = &scream_pos;
2103 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2105 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2106 goto phooey; /* not present */
2112 /* Simplest case: anchored match need be tried only once. */
2113 /* [unless only anchor is BOL and multiline is set] */
2114 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2115 if (s == startpos && regtry(®info, &startpos))
2117 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2118 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2123 dontbother = minlen - 1;
2124 end = HOP3c(strend, -dontbother, strbeg) - 1;
2125 /* for multiline we only have to try after newlines */
2126 if (prog->check_substr || prog->check_utf8) {
2127 /* because of the goto we can not easily reuse the macros for bifurcating the
2128 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2131 goto after_try_utf8;
2133 if (regtry(®info, &s)) {
2140 if (prog->extflags & RXf_USE_INTUIT) {
2141 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2150 } /* end search for check string in unicode */
2152 if (s == startpos) {
2153 goto after_try_latin;
2156 if (regtry(®info, &s)) {
2163 if (prog->extflags & RXf_USE_INTUIT) {
2164 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2173 } /* end search for check string in latin*/
2174 } /* end search for check string */
2175 else { /* search for newline */
2177 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2180 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2182 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2183 if (regtry(®info, &s))
2187 } /* end search for newline */
2188 } /* end anchored/multiline check string search */
2190 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2192 /* the warning about reginfo.ganch being used without initialization
2193 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2194 and we only enter this block when the same bit is set. */
2195 char *tmp_s = reginfo.ganch - prog->gofs;
2197 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2202 /* Messy cases: unanchored match. */
2203 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2204 /* we have /x+whatever/ */
2205 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2210 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2211 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2212 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2217 DEBUG_EXECUTE_r( did_match = 1 );
2218 if (regtry(®info, &s)) goto got_it;
2220 while (s < strend && *s == ch)
2228 DEBUG_EXECUTE_r( did_match = 1 );
2229 if (regtry(®info, &s)) goto got_it;
2231 while (s < strend && *s == ch)
2236 DEBUG_EXECUTE_r(if (!did_match)
2237 PerlIO_printf(Perl_debug_log,
2238 "Did not find anchored character...\n")
2241 else if (prog->anchored_substr != NULL
2242 || prog->anchored_utf8 != NULL
2243 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2244 && prog->float_max_offset < strend - s)) {
2249 char *last1; /* Last position checked before */
2253 if (prog->anchored_substr || prog->anchored_utf8) {
2254 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2255 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2256 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2257 back_max = back_min = prog->anchored_offset;
2259 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2260 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2261 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2262 back_max = prog->float_max_offset;
2263 back_min = prog->float_min_offset;
2267 if (must == &PL_sv_undef)
2268 /* could not downgrade utf8 check substring, so must fail */
2274 last = HOP3c(strend, /* Cannot start after this */
2275 -(I32)(CHR_SVLEN(must)
2276 - (SvTAIL(must) != 0) + back_min), strbeg);
2279 last1 = HOPc(s, -1);
2281 last1 = s - 1; /* bogus */
2283 /* XXXX check_substr already used to find "s", can optimize if
2284 check_substr==must. */
2286 dontbother = end_shift;
2287 strend = HOPc(strend, -dontbother);
2288 while ( (s <= last) &&
2289 ((flags & REXEC_SCREAM)
2290 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2291 end_shift, &scream_pos, 0))
2292 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2293 (unsigned char*)strend, must,
2294 multiline ? FBMrf_MULTILINE : 0))) ) {
2295 /* we may be pointing at the wrong string */
2296 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2297 s = strbeg + (s - SvPVX_const(sv));
2298 DEBUG_EXECUTE_r( did_match = 1 );
2299 if (HOPc(s, -back_max) > last1) {
2300 last1 = HOPc(s, -back_min);
2301 s = HOPc(s, -back_max);
2304 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2306 last1 = HOPc(s, -back_min);
2310 while (s <= last1) {
2311 if (regtry(®info, &s))
2317 while (s <= last1) {
2318 if (regtry(®info, &s))
2324 DEBUG_EXECUTE_r(if (!did_match) {
2325 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2326 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2327 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2328 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2329 ? "anchored" : "floating"),
2330 quoted, RE_SV_TAIL(must));
2334 else if ( (c = progi->regstclass) ) {
2336 const OPCODE op = OP(progi->regstclass);
2337 /* don't bother with what can't match */
2338 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2339 strend = HOPc(strend, -(minlen - 1));
2342 SV * const prop = sv_newmortal();
2343 regprop(prog, prop, c);
2345 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2347 PerlIO_printf(Perl_debug_log,
2348 "Matching stclass %.*s against %s (%d bytes)\n",
2349 (int)SvCUR(prop), SvPVX_const(prop),
2350 quoted, (int)(strend - s));
2353 if (find_byclass(prog, c, s, strend, ®info))
2355 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2359 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2364 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2365 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2366 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2368 if (flags & REXEC_SCREAM) {
2369 last = screaminstr(sv, float_real, s - strbeg,
2370 end_shift, &scream_pos, 1); /* last one */
2372 last = scream_olds; /* Only one occurrence. */
2373 /* we may be pointing at the wrong string */
2374 else if (RXp_MATCH_COPIED(prog))
2375 s = strbeg + (s - SvPVX_const(sv));
2379 const char * const little = SvPV_const(float_real, len);
2381 if (SvTAIL(float_real)) {
2382 if (memEQ(strend - len + 1, little, len - 1))
2383 last = strend - len + 1;
2384 else if (!multiline)
2385 last = memEQ(strend - len, little, len)
2386 ? strend - len : NULL;
2392 last = rninstr(s, strend, little, little + len);
2394 last = strend; /* matching "$" */
2399 PerlIO_printf(Perl_debug_log,
2400 "%sCan't trim the tail, match fails (should not happen)%s\n",
2401 PL_colors[4], PL_colors[5]));
2402 goto phooey; /* Should not happen! */
2404 dontbother = strend - last + prog->float_min_offset;
2406 if (minlen && (dontbother < minlen))
2407 dontbother = minlen - 1;
2408 strend -= dontbother; /* this one's always in bytes! */
2409 /* We don't know much -- general case. */
2412 if (regtry(®info, &s))
2421 if (regtry(®info, &s))
2423 } while (s++ < strend);
2432 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2434 if (PL_reg_eval_set)
2435 restore_pos(aTHX_ prog);
2436 if (RXp_PAREN_NAMES(prog))
2437 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2439 /* make sure $`, $&, $', and $digit will work later */
2440 if ( !(flags & REXEC_NOT_FIRST) ) {
2441 RX_MATCH_COPY_FREE(rx);
2442 if (flags & REXEC_COPY_STR) {
2443 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2444 #ifdef PERL_OLD_COPY_ON_WRITE
2446 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2448 PerlIO_printf(Perl_debug_log,
2449 "Copy on write: regexp capture, type %d\n",
2452 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2453 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2454 assert (SvPOKp(prog->saved_copy));
2458 RX_MATCH_COPIED_on(rx);
2459 s = savepvn(strbeg, i);
2465 prog->subbeg = strbeg;
2466 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2473 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2474 PL_colors[4], PL_colors[5]));
2475 if (PL_reg_eval_set)
2476 restore_pos(aTHX_ prog);
2478 /* we failed :-( roll it back */
2479 Safefree(prog->offs);
2488 - regtry - try match at specific point
2490 STATIC I32 /* 0 failure, 1 success */
2491 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2495 REGEXP *const rx = reginfo->prog;
2496 regexp *const prog = (struct regexp *)SvANY(rx);
2497 RXi_GET_DECL(prog,progi);
2498 GET_RE_DEBUG_FLAGS_DECL;
2500 PERL_ARGS_ASSERT_REGTRY;
2502 reginfo->cutpoint=NULL;
2504 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2507 PL_reg_eval_set = RS_init;
2508 DEBUG_EXECUTE_r(DEBUG_s(
2509 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2510 (IV)(PL_stack_sp - PL_stack_base));
2513 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2514 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2516 /* Apparently this is not needed, judging by wantarray. */
2517 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2518 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2521 /* Make $_ available to executed code. */
2522 if (reginfo->sv != DEFSV) {
2524 DEFSV_set(reginfo->sv);
2527 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2528 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2529 /* prepare for quick setting of pos */
2530 #ifdef PERL_OLD_COPY_ON_WRITE
2531 if (SvIsCOW(reginfo->sv))
2532 sv_force_normal_flags(reginfo->sv, 0);
2534 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2535 &PL_vtbl_mglob, NULL, 0);
2539 PL_reg_oldpos = mg->mg_len;
2540 SAVEDESTRUCTOR_X(restore_pos, prog);
2542 if (!PL_reg_curpm) {
2543 Newxz(PL_reg_curpm, 1, PMOP);
2546 SV* const repointer = &PL_sv_undef;
2547 /* this regexp is also owned by the new PL_reg_curpm, which
2548 will try to free it. */
2549 av_push(PL_regex_padav, repointer);
2550 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2551 PL_regex_pad = AvARRAY(PL_regex_padav);
2556 /* It seems that non-ithreads works both with and without this code.
2557 So for efficiency reasons it seems best not to have the code
2558 compiled when it is not needed. */
2559 /* This is safe against NULLs: */
2560 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2561 /* PM_reg_curpm owns a reference to this regexp. */
2562 (void)ReREFCNT_inc(rx);
2564 PM_SETRE(PL_reg_curpm, rx);
2565 PL_reg_oldcurpm = PL_curpm;
2566 PL_curpm = PL_reg_curpm;
2567 if (RXp_MATCH_COPIED(prog)) {
2568 /* Here is a serious problem: we cannot rewrite subbeg,
2569 since it may be needed if this match fails. Thus
2570 $` inside (?{}) could fail... */
2571 PL_reg_oldsaved = prog->subbeg;
2572 PL_reg_oldsavedlen = prog->sublen;
2573 #ifdef PERL_OLD_COPY_ON_WRITE
2574 PL_nrs = prog->saved_copy;
2576 RXp_MATCH_COPIED_off(prog);
2579 PL_reg_oldsaved = NULL;
2580 prog->subbeg = PL_bostr;
2581 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2583 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2584 prog->offs[0].start = *startpos - PL_bostr;
2585 PL_reginput = *startpos;
2586 PL_reglastparen = &prog->lastparen;
2587 PL_reglastcloseparen = &prog->lastcloseparen;
2588 prog->lastparen = 0;
2589 prog->lastcloseparen = 0;
2591 PL_regoffs = prog->offs;
2592 if (PL_reg_start_tmpl <= prog->nparens) {
2593 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2594 if(PL_reg_start_tmp)
2595 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2597 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2600 /* XXXX What this code is doing here?!!! There should be no need
2601 to do this again and again, PL_reglastparen should take care of
2604 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2605 * Actually, the code in regcppop() (which Ilya may be meaning by
2606 * PL_reglastparen), is not needed at all by the test suite
2607 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2608 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2609 * Meanwhile, this code *is* needed for the
2610 * above-mentioned test suite tests to succeed. The common theme
2611 * on those tests seems to be returning null fields from matches.
2612 * --jhi updated by dapm */
2614 if (prog->nparens) {
2615 regexp_paren_pair *pp = PL_regoffs;
2617 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2625 if (regmatch(reginfo, progi->program + 1)) {
2626 PL_regoffs[0].end = PL_reginput - PL_bostr;
2629 if (reginfo->cutpoint)
2630 *startpos= reginfo->cutpoint;
2631 REGCP_UNWIND(lastcp);
2636 #define sayYES goto yes
2637 #define sayNO goto no
2638 #define sayNO_SILENT goto no_silent
2640 /* we dont use STMT_START/END here because it leads to
2641 "unreachable code" warnings, which are bogus, but distracting. */
2642 #define CACHEsayNO \
2643 if (ST.cache_mask) \
2644 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2647 /* this is used to determine how far from the left messages like
2648 'failed...' are printed. It should be set such that messages
2649 are inline with the regop output that created them.
2651 #define REPORT_CODE_OFF 32
2654 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2655 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2657 #define SLAB_FIRST(s) (&(s)->states[0])
2658 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2660 /* grab a new slab and return the first slot in it */
2662 STATIC regmatch_state *
2665 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2668 regmatch_slab *s = PL_regmatch_slab->next;
2670 Newx(s, 1, regmatch_slab);
2671 s->prev = PL_regmatch_slab;
2673 PL_regmatch_slab->next = s;
2675 PL_regmatch_slab = s;
2676 return SLAB_FIRST(s);
2680 /* push a new state then goto it */
2682 #define PUSH_STATE_GOTO(state, node) \
2684 st->resume_state = state; \
2687 /* push a new state with success backtracking, then goto it */
2689 #define PUSH_YES_STATE_GOTO(state, node) \
2691 st->resume_state = state; \
2692 goto push_yes_state;
2698 regmatch() - main matching routine
2700 This is basically one big switch statement in a loop. We execute an op,
2701 set 'next' to point the next op, and continue. If we come to a point which
2702 we may need to backtrack to on failure such as (A|B|C), we push a
2703 backtrack state onto the backtrack stack. On failure, we pop the top
2704 state, and re-enter the loop at the state indicated. If there are no more
2705 states to pop, we return failure.
2707 Sometimes we also need to backtrack on success; for example /A+/, where
2708 after successfully matching one A, we need to go back and try to
2709 match another one; similarly for lookahead assertions: if the assertion
2710 completes successfully, we backtrack to the state just before the assertion
2711 and then carry on. In these cases, the pushed state is marked as
2712 'backtrack on success too'. This marking is in fact done by a chain of
2713 pointers, each pointing to the previous 'yes' state. On success, we pop to
2714 the nearest yes state, discarding any intermediate failure-only states.
2715 Sometimes a yes state is pushed just to force some cleanup code to be
2716 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2717 it to free the inner regex.
2719 Note that failure backtracking rewinds the cursor position, while
2720 success backtracking leaves it alone.
2722 A pattern is complete when the END op is executed, while a subpattern
2723 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2724 ops trigger the "pop to last yes state if any, otherwise return true"
2727 A common convention in this function is to use A and B to refer to the two
2728 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2729 the subpattern to be matched possibly multiple times, while B is the entire
2730 rest of the pattern. Variable and state names reflect this convention.
2732 The states in the main switch are the union of ops and failure/success of
2733 substates associated with with that op. For example, IFMATCH is the op
2734 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2735 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2736 successfully matched A and IFMATCH_A_fail is a state saying that we have
2737 just failed to match A. Resume states always come in pairs. The backtrack
2738 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2739 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2740 on success or failure.
2742 The struct that holds a backtracking state is actually a big union, with
2743 one variant for each major type of op. The variable st points to the
2744 top-most backtrack struct. To make the code clearer, within each
2745 block of code we #define ST to alias the relevant union.
2747 Here's a concrete example of a (vastly oversimplified) IFMATCH
2753 #define ST st->u.ifmatch
2755 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2756 ST.foo = ...; // some state we wish to save
2758 // push a yes backtrack state with a resume value of
2759 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2761 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2764 case IFMATCH_A: // we have successfully executed A; now continue with B
2766 bar = ST.foo; // do something with the preserved value
2769 case IFMATCH_A_fail: // A failed, so the assertion failed
2770 ...; // do some housekeeping, then ...
2771 sayNO; // propagate the failure
2778 For any old-timers reading this who are familiar with the old recursive
2779 approach, the code above is equivalent to:
2781 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2790 ...; // do some housekeeping, then ...
2791 sayNO; // propagate the failure
2794 The topmost backtrack state, pointed to by st, is usually free. If you
2795 want to claim it, populate any ST.foo fields in it with values you wish to
2796 save, then do one of
2798 PUSH_STATE_GOTO(resume_state, node);
2799 PUSH_YES_STATE_GOTO(resume_state, node);
2801 which sets that backtrack state's resume value to 'resume_state', pushes a
2802 new free entry to the top of the backtrack stack, then goes to 'node'.
2803 On backtracking, the free slot is popped, and the saved state becomes the
2804 new free state. An ST.foo field in this new top state can be temporarily
2805 accessed to retrieve values, but once the main loop is re-entered, it
2806 becomes available for reuse.
2808 Note that the depth of the backtrack stack constantly increases during the
2809 left-to-right execution of the pattern, rather than going up and down with
2810 the pattern nesting. For example the stack is at its maximum at Z at the
2811 end of the pattern, rather than at X in the following:
2813 /(((X)+)+)+....(Y)+....Z/
2815 The only exceptions to this are lookahead/behind assertions and the cut,
2816 (?>A), which pop all the backtrack states associated with A before
2819 Backtrack state structs are allocated in slabs of about 4K in size.
2820 PL_regmatch_state and st always point to the currently active state,
2821 and PL_regmatch_slab points to the slab currently containing
2822 PL_regmatch_state. The first time regmatch() is called, the first slab is
2823 allocated, and is never freed until interpreter destruction. When the slab
2824 is full, a new one is allocated and chained to the end. At exit from
2825 regmatch(), slabs allocated since entry are freed.
2830 #define DEBUG_STATE_pp(pp) \
2832 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2833 PerlIO_printf(Perl_debug_log, \
2834 " %*s"pp" %s%s%s%s%s\n", \
2836 PL_reg_name[st->resume_state], \
2837 ((st==yes_state||st==mark_state) ? "[" : ""), \
2838 ((st==yes_state) ? "Y" : ""), \
2839 ((st==mark_state) ? "M" : ""), \
2840 ((st==yes_state||st==mark_state) ? "]" : "") \
2845 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2850 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2851 const char *start, const char *end, const char *blurb)
2853 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2855 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2860 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2861 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2863 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2864 start, end - start, 60);
2866 PerlIO_printf(Perl_debug_log,
2867 "%s%s REx%s %s against %s\n",
2868 PL_colors[4], blurb, PL_colors[5], s0, s1);
2870 if (utf8_target||utf8_pat)
2871 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2872 utf8_pat ? "pattern" : "",
2873 utf8_pat && utf8_target ? " and " : "",
2874 utf8_target ? "string" : ""
2880 S_dump_exec_pos(pTHX_ const char *locinput,
2881 const regnode *scan,
2882 const char *loc_regeol,
2883 const char *loc_bostr,
2884 const char *loc_reg_starttry,
2885 const bool utf8_target)
2887 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2888 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2889 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2890 /* The part of the string before starttry has one color
2891 (pref0_len chars), between starttry and current
2892 position another one (pref_len - pref0_len chars),
2893 after the current position the third one.
2894 We assume that pref0_len <= pref_len, otherwise we
2895 decrease pref0_len. */
2896 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2897 ? (5 + taill) - l : locinput - loc_bostr;
2900 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2902 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2904 pref0_len = pref_len - (locinput - loc_reg_starttry);
2905 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2906 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2907 ? (5 + taill) - pref_len : loc_regeol - locinput);
2908 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2912 if (pref0_len > pref_len)
2913 pref0_len = pref_len;
2915 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2917 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2918 (locinput - pref_len),pref0_len, 60, 4, 5);
2920 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2921 (locinput - pref_len + pref0_len),
2922 pref_len - pref0_len, 60, 2, 3);
2924 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2925 locinput, loc_regeol - locinput, 10, 0, 1);
2927 const STRLEN tlen=len0+len1+len2;
2928 PerlIO_printf(Perl_debug_log,
2929 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2930 (IV)(locinput - loc_bostr),
2933 (docolor ? "" : "> <"),
2935 (int)(tlen > 19 ? 0 : 19 - tlen),
2942 /* reg_check_named_buff_matched()
2943 * Checks to see if a named buffer has matched. The data array of
2944 * buffer numbers corresponding to the buffer is expected to reside
2945 * in the regexp->data->data array in the slot stored in the ARG() of
2946 * node involved. Note that this routine doesn't actually care about the
2947 * name, that information is not preserved from compilation to execution.
2948 * Returns the index of the leftmost defined buffer with the given name
2949 * or 0 if non of the buffers matched.
2952 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2955 RXi_GET_DECL(rex,rexi);
2956 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2957 I32 *nums=(I32*)SvPVX(sv_dat);
2959 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2961 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2962 if ((I32)*PL_reglastparen >= nums[n] &&
2963 PL_regoffs[nums[n]].end != -1)
2972 /* free all slabs above current one - called during LEAVE_SCOPE */
2975 S_clear_backtrack_stack(pTHX_ void *p)
2977 regmatch_slab *s = PL_regmatch_slab->next;
2982 PL_regmatch_slab->next = NULL;
2984 regmatch_slab * const osl = s;
2991 #define SETREX(Re1,Re2) \
2992 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2995 STATIC I32 /* 0 failure, 1 success */
2996 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2998 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3002 register const bool utf8_target = PL_reg_match_utf8;
3003 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3004 REGEXP *rex_sv = reginfo->prog;
3005 regexp *rex = (struct regexp *)SvANY(rex_sv);
3006 RXi_GET_DECL(rex,rexi);
3008 /* the current state. This is a cached copy of PL_regmatch_state */
3009 register regmatch_state *st;
3010 /* cache heavy used fields of st in registers */
3011 register regnode *scan;
3012 register regnode *next;
3013 register U32 n = 0; /* general value; init to avoid compiler warning */
3014 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3015 register char *locinput = PL_reginput;
3016 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3018 bool result = 0; /* return value of S_regmatch */
3019 int depth = 0; /* depth of backtrack stack */
3020 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3021 const U32 max_nochange_depth =
3022 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3023 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3024 regmatch_state *yes_state = NULL; /* state to pop to on success of
3026 /* mark_state piggy backs on the yes_state logic so that when we unwind
3027 the stack on success we can update the mark_state as we go */
3028 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3029 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3030 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3032 bool no_final = 0; /* prevent failure from backtracking? */
3033 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3034 char *startpoint = PL_reginput;
3035 SV *popmark = NULL; /* are we looking for a mark? */
3036 SV *sv_commit = NULL; /* last mark name seen in failure */
3037 SV *sv_yes_mark = NULL; /* last mark name we have seen
3038 during a successful match */
3039 U32 lastopen = 0; /* last open we saw */
3040 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3041 SV* const oreplsv = GvSV(PL_replgv);
3042 /* these three flags are set by various ops to signal information to
3043 * the very next op. They have a useful lifetime of exactly one loop
3044 * iteration, and are not preserved or restored by state pushes/pops
3046 bool sw = 0; /* the condition value in (?(cond)a|b) */
3047 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3048 int logical = 0; /* the following EVAL is:
3052 or the following IFMATCH/UNLESSM is:
3053 false: plain (?=foo)
3054 true: used as a condition: (?(?=foo))
3057 GET_RE_DEBUG_FLAGS_DECL;
3060 PERL_ARGS_ASSERT_REGMATCH;
3062 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3063 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3065 /* on first ever call to regmatch, allocate first slab */
3066 if (!PL_regmatch_slab) {
3067 Newx(PL_regmatch_slab, 1, regmatch_slab);
3068 PL_regmatch_slab->prev = NULL;
3069 PL_regmatch_slab->next = NULL;
3070 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3073 oldsave = PL_savestack_ix;
3074 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3075 SAVEVPTR(PL_regmatch_slab);
3076 SAVEVPTR(PL_regmatch_state);
3078 /* grab next free state slot */
3079 st = ++PL_regmatch_state;
3080 if (st > SLAB_LAST(PL_regmatch_slab))
3081 st = PL_regmatch_state = S_push_slab(aTHX);
3083 /* Note that nextchr is a byte even in UTF */
3084 nextchr = UCHARAT(locinput);
3086 while (scan != NULL) {
3089 SV * const prop = sv_newmortal();
3090 regnode *rnext=regnext(scan);
3091 DUMP_EXEC_POS( locinput, scan, utf8_target );
3092 regprop(rex, prop, scan);
3094 PerlIO_printf(Perl_debug_log,
3095 "%3"IVdf":%*s%s(%"IVdf")\n",
3096 (IV)(scan - rexi->program), depth*2, "",
3098 (PL_regkind[OP(scan)] == END || !rnext) ?
3099 0 : (IV)(rnext - rexi->program));
3102 next = scan + NEXT_OFF(scan);
3105 state_num = OP(scan);
3107 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3110 assert(PL_reglastparen == &rex->lastparen);
3111 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3112 assert(PL_regoffs == rex->offs);
3114 switch (state_num) {
3116 if (locinput == PL_bostr)
3118 /* reginfo->till = reginfo->bol; */
3123 if (locinput == PL_bostr ||
3124 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3130 if (locinput == PL_bostr)
3134 if (locinput == reginfo->ganch)
3139 /* update the startpoint */
3140 st->u.keeper.val = PL_regoffs[0].start;
3141 PL_reginput = locinput;
3142 PL_regoffs[0].start = locinput - PL_bostr;
3143 PUSH_STATE_GOTO(KEEPS_next, next);
3145 case KEEPS_next_fail:
3146 /* rollback the start point change */
3147 PL_regoffs[0].start = st->u.keeper.val;
3153 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3158 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3160 if (PL_regeol - locinput > 1)
3164 if (PL_regeol != locinput)
3168 if (!nextchr && locinput >= PL_regeol)
3171 locinput += PL_utf8skip[nextchr];
3172 if (locinput > PL_regeol)
3174 nextchr = UCHARAT(locinput);
3177 nextchr = UCHARAT(++locinput);
3180 if (!nextchr && locinput >= PL_regeol)
3182 nextchr = UCHARAT(++locinput);
3185 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3188 locinput += PL_utf8skip[nextchr];
3189 if (locinput > PL_regeol)
3191 nextchr = UCHARAT(locinput);
3194 nextchr = UCHARAT(++locinput);
3198 #define ST st->u.trie
3200 /* In this case the charclass data is available inline so
3201 we can fail fast without a lot of extra overhead.
3203 if (scan->flags == EXACT || !utf8_target) {
3204 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3206 PerlIO_printf(Perl_debug_log,
3207 "%*s %sfailed to match trie start class...%s\n",
3208 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3216 /* the basic plan of execution of the trie is:
3217 * At the beginning, run though all the states, and
3218 * find the longest-matching word. Also remember the position
3219 * of the shortest matching word. For example, this pattern:
3222 * when matched against the string "abcde", will generate
3223 * accept states for all words except 3, with the longest
3224 * matching word being 4, and the shortest being 1 (with
3225 * the position being after char 1 of the string).
3227 * Then for each matching word, in word order (i.e. 1,2,4,5),
3228 * we run the remainder of the pattern; on each try setting
3229 * the current position to the character following the word,
3230 * returning to try the next word on failure.
3232 * We avoid having to build a list of words at runtime by
3233 * using a compile-time structure, wordinfo[].prev, which
3234 * gives, for each word, the previous accepting word (if any).
3235 * In the case above it would contain the mappings 1->2, 2->0,
3236 * 3->0, 4->5, 5->1. We can use this table to generate, from
3237 * the longest word (4 above), a list of all words, by
3238 * following the list of prev pointers; this gives us the
3239 * unordered list 4,5,1,2. Then given the current word we have
3240 * just tried, we can go through the list and find the
3241 * next-biggest word to try (so if we just failed on word 2,
3242 * the next in the list is 4).
3244 * Since at runtime we don't record the matching position in
3245 * the string for each word, we have to work that out for
3246 * each word we're about to process. The wordinfo table holds
3247 * the character length of each word; given that we recorded
3248 * at the start: the position of the shortest word and its
3249 * length in chars, we just need to move the pointer the
3250 * difference between the two char lengths. Depending on
3251 * Unicode status and folding, that's cheap or expensive.
3253 * This algorithm is optimised for the case where are only a
3254 * small number of accept states, i.e. 0,1, or maybe 2.
3255 * With lots of accepts states, and having to try all of them,
3256 * it becomes quadratic on number of accept states to find all
3261 /* what type of TRIE am I? (utf8 makes this contextual) */
3262 DECL_TRIE_TYPE(scan);
3264 /* what trie are we using right now */
3265 reg_trie_data * const trie
3266 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3267 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3268 U32 state = trie->startstate;
3270 if (trie->bitmap && trie_type != trie_utf8_fold &&
3271 !TRIE_BITMAP_TEST(trie,*locinput)
3273 if (trie->states[ state ].wordnum) {
3275 PerlIO_printf(Perl_debug_log,
3276 "%*s %smatched empty string...%s\n",
3277 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3283 PerlIO_printf(Perl_debug_log,
3284 "%*s %sfailed to match trie start class...%s\n",
3285 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3292 U8 *uc = ( U8* )locinput;
3296 U8 *uscan = (U8*)NULL;
3297 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3298 U32 charcount = 0; /* how many input chars we have matched */
3299 U32 accepted = 0; /* have we seen any accepting states? */
3302 ST.jump = trie->jump;
3305 ST.longfold = FALSE; /* char longer if folded => it's harder */
3308 /* fully traverse the TRIE; note the position of the
3309 shortest accept state and the wordnum of the longest
3312 while ( state && uc <= (U8*)PL_regeol ) {
3313 U32 base = trie->states[ state ].trans.base;
3317 wordnum = trie->states[ state ].wordnum;
3319 if (wordnum) { /* it's an accept state */
3322 /* record first match position */
3324 ST.firstpos = (U8*)locinput;
3329 ST.firstchars = charcount;
3332 if (!ST.nextword || wordnum < ST.nextword)
3333 ST.nextword = wordnum;
3334 ST.topword = wordnum;
3337 DEBUG_TRIE_EXECUTE_r({
3338 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3339 PerlIO_printf( Perl_debug_log,
3340 "%*s %sState: %4"UVxf" Accepted: %c ",
3341 2+depth * 2, "", PL_colors[4],
3342 (UV)state, (accepted ? 'Y' : 'N'));
3345 /* read a char and goto next state */
3348 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3349 uscan, len, uvc, charid, foldlen,
3356 base + charid - 1 - trie->uniquecharcount)) >= 0)
3358 && ((U32)offset < trie->lasttrans)
3359 && trie->trans[offset].check == state)
3361 state = trie->trans[offset].next;
3372 DEBUG_TRIE_EXECUTE_r(
3373 PerlIO_printf( Perl_debug_log,
3374 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3375 charid, uvc, (UV)state, PL_colors[5] );
3381 /* calculate total number of accept states */
3386 w = trie->wordinfo[w].prev;
3389 ST.accepted = accepted;
3393 PerlIO_printf( Perl_debug_log,
3394 "%*s %sgot %"IVdf" possible matches%s\n",
3395 REPORT_CODE_OFF + depth * 2, "",
3396 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3398 goto trie_first_try; /* jump into the fail handler */
3402 case TRIE_next_fail: /* we failed - try next alternative */
3404 REGCP_UNWIND(ST.cp);
3405 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3406 PL_regoffs[n].end = -1;
3407 *PL_reglastparen = n;
3409 if (!--ST.accepted) {
3411 PerlIO_printf( Perl_debug_log,
3412 "%*s %sTRIE failed...%s\n",
3413 REPORT_CODE_OFF+depth*2, "",
3420 /* Find next-highest word to process. Note that this code
3421 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3422 register U16 min = 0;
3424 register U16 const nextword = ST.nextword;
3425 register reg_trie_wordinfo * const wordinfo
3426 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3427 for (word=ST.topword; word; word=wordinfo[word].prev) {
3428 if (word > nextword && (!min || word < min))
3441 ST.lastparen = *PL_reglastparen;
3445 /* find start char of end of current word */
3447 U32 chars; /* how many chars to skip */
3448 U8 *uc = ST.firstpos;
3449 reg_trie_data * const trie
3450 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3452 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3454 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3458 /* the hard option - fold each char in turn and find
3459 * its folded length (which may be different */
3460 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3468 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3476 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3481 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3495 PL_reginput = (char *)uc;
3498 scan = (ST.jump && ST.jump[ST.nextword])
3499 ? ST.me + ST.jump[ST.nextword]
3503 PerlIO_printf( Perl_debug_log,
3504 "%*s %sTRIE matched word #%d, continuing%s\n",
3505 REPORT_CODE_OFF+depth*2, "",
3512 if (ST.accepted > 1 || has_cutgroup) {
3513 PUSH_STATE_GOTO(TRIE_next, scan);
3516 /* only one choice left - just continue */
3518 AV *const trie_words
3519 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3520 SV ** const tmp = av_fetch( trie_words,
3522 SV *sv= tmp ? sv_newmortal() : NULL;
3524 PerlIO_printf( Perl_debug_log,
3525 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3526 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3528 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3529 PL_colors[0], PL_colors[1],
3530 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3532 : "not compiled under -Dr",
3536 locinput = PL_reginput;
3537 nextchr = UCHARAT(locinput);
3538 continue; /* execute rest of RE */
3543 char *s = STRING(scan);
3545 if (utf8_target != UTF_PATTERN) {
3546 /* The target and the pattern have differing utf8ness. */
3548 const char * const e = s + ln;
3551 /* The target is utf8, the pattern is not utf8. */
3556 if (NATIVE_TO_UNI(*(U8*)s) !=
3557 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3565 /* The target is not utf8, the pattern is utf8. */
3570 if (NATIVE_TO_UNI(*((U8*)l)) !=
3571 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3579 nextchr = UCHARAT(locinput);
3582 /* The target and the pattern have the same utf8ness. */
3583 /* Inline the first character, for speed. */
3584 if (UCHARAT(s) != nextchr)
3586 if (PL_regeol - locinput < ln)
3588 if (ln > 1 && memNE(s, locinput, ln))
3591 nextchr = UCHARAT(locinput);
3596 const U8 * fold_array;
3598 U32 fold_utf8_flags;
3600 PL_reg_flags |= RF_tainted;
3601 folder = foldEQ_locale;
3602 fold_array = PL_fold_locale;
3603 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3607 folder = foldEQ_latin1;
3608 fold_array = PL_fold_latin1;
3609 fold_utf8_flags = 0;
3613 folder = foldEQ_latin1;
3614 fold_array = PL_fold_latin1;
3615 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3620 fold_array = PL_fold;
3621 fold_utf8_flags = 0;
3627 if (utf8_target || UTF_PATTERN) {
3628 /* Either target or the pattern are utf8. */
3629 const char * const l = locinput;
3630 char *e = PL_regeol;
3632 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3633 l, &e, 0, utf8_target, fold_utf8_flags))
3638 nextchr = UCHARAT(locinput);
3642 /* Neither the target nor the pattern are utf8 */
3643 if (UCHARAT(s) != nextchr &&
3644 UCHARAT(s) != fold_array[nextchr])
3648 if (PL_regeol - locinput < ln)
3650 if (ln > 1 && ! folder(s, locinput, ln))
3653 nextchr = UCHARAT(locinput);
3657 /* XXX Could improve efficiency by separating these all out using a
3658 * macro or in-line function. At that point regcomp.c would no longer
3659 * have to set the FLAGS fields of these */
3662 PL_reg_flags |= RF_tainted;
3670 /* was last char in word? */
3671 if (utf8_target && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET) {
3672 if (locinput == PL_bostr)
3675 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3677 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3679 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3680 ln = isALNUM_uni(ln);
3681 LOAD_UTF8_CHARCLASS_ALNUM();
3682 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3685 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3686 n = isALNUM_LC_utf8((U8*)locinput);
3691 /* Here the string isn't utf8, or is utf8 and only ascii
3692 * characters are to match \w. In the latter case looking at
3693 * the byte just prior to the current one may be just the final
3694 * byte of a multi-byte character. This is ok. There are two
3696 * 1) it is a single byte character, and then the test is doing
3697 * just what it's supposed to.
3698 * 2) it is a multi-byte character, in which case the final
3699 * byte is never mistakable for ASCII, and so the test
3700 * will say it is not a word character, which is the
3701 * correct answer. */
3702 ln = (locinput != PL_bostr) ?
3703 UCHARAT(locinput - 1) : '\n';
3704 switch (FLAGS(scan)) {
3705 case REGEX_UNICODE_CHARSET:
3706 ln = isWORDCHAR_L1(ln);
3707 n = isWORDCHAR_L1(nextchr);
3709 case REGEX_LOCALE_CHARSET:
3710 ln = isALNUM_LC(ln);
3711 n = isALNUM_LC(nextchr);
3713 case REGEX_DEPENDS_CHARSET:
3715 n = isALNUM(nextchr);
3717 case REGEX_ASCII_RESTRICTED_CHARSET:
3718 ln = isWORDCHAR_A(ln);
3719 n = isWORDCHAR_A(nextchr);
3722 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3726 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3728 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3733 if (utf8_target || state_num == ANYOFV) {
3734 STRLEN inclasslen = PL_regeol - locinput;
3735 if (locinput >= PL_regeol)
3738 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3740 locinput += inclasslen;
3741 nextchr = UCHARAT(locinput);
3746 nextchr = UCHARAT(locinput);
3747 if (!nextchr && locinput >= PL_regeol)
3749 if (!REGINCLASS(rex, scan, (U8*)locinput))
3751 nextchr = UCHARAT(++locinput);
3755 /* Special char classes - The defines start on line 129 or so */
3756 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3757 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3758 ALNUMU, NALNUMU, isWORDCHAR_L1,
3759 ALNUMA, NALNUMA, isWORDCHAR_A,
3762 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3763 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3764 SPACEU, NSPACEU, isSPACE_L1,
3765 SPACEA, NSPACEA, isSPACE_A,
3768 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3769 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3770 DIGITA, NDIGITA, isDIGIT_A,
3773 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3774 a Unicode extended Grapheme Cluster */
3775 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3776 extended Grapheme Cluster is:
3779 | Prepend* Begin Extend*
3782 Begin is (Hangul-syllable | ! Control)
3783 Extend is (Grapheme_Extend | Spacing_Mark)
3784 Control is [ GCB_Control CR LF ]
3786 The discussion below shows how the code for CLUMP is derived
3787 from this regex. Note that most of these concepts are from
3788 property values of the Grapheme Cluster Boundary (GCB) property.
3789 No code point can have multiple property values for a given
3790 property. Thus a code point in Prepend can't be in Control, but
3791 it must be in !Control. This is why Control above includes
3792 GCB_Control plus CR plus LF. The latter two are used in the GCB
3793 property separately, and so can't be in GCB_Control, even though
3794 they logically are controls. Control is not the same as gc=cc,
3795 but includes format and other characters as well.
3797 The Unicode definition of Hangul-syllable is:
3799 | (L* ( ( V | LV ) V* | LVT ) T*)
3802 Each of these is a value for the GCB property, and hence must be
3803 disjoint, so the order they are tested is immaterial, so the
3804 above can safely be changed to
3807 | (L* ( LVT | ( V | LV ) V*) T*)
3809 The last two terms can be combined like this:
3811 | (( LVT | ( V | LV ) V*) T*))
3813 And refactored into this:
3814 L* (L | LVT T* | V V* T* | LV V* T*)
3816 That means that if we have seen any L's at all we can quit
3817 there, but if the next character is a LVT, a V or and LV we
3820 There is a subtlety with Prepend* which showed up in testing.
3821 Note that the Begin, and only the Begin is required in:
3822 | Prepend* Begin Extend*
3823 Also, Begin contains '! Control'. A Prepend must be a '!
3824 Control', which means it must be a Begin. What it comes down to
3825 is that if we match Prepend* and then find no suitable Begin
3826 afterwards, that if we backtrack the last Prepend, that one will
3827 be a suitable Begin.
3830 if (locinput >= PL_regeol)
3832 if (! utf8_target) {
3834 /* Match either CR LF or '.', as all the other possibilities
3836 locinput++; /* Match the . or CR */
3838 && locinput < PL_regeol
3839 && UCHARAT(locinput) == '\n') locinput++;
3843 /* Utf8: See if is ( CR LF ); already know that locinput <
3844 * PL_regeol, so locinput+1 is in bounds */
3845 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3849 /* In case have to backtrack to beginning, then match '.' */
3850 char *starting = locinput;
3852 /* In case have to backtrack the last prepend */
3853 char *previous_prepend = 0;
3855 LOAD_UTF8_CHARCLASS_GCB();
3857 /* Match (prepend)* */
3858 while (locinput < PL_regeol
3859 && swash_fetch(PL_utf8_X_prepend,
3860 (U8*)locinput, utf8_target))
3862 previous_prepend = locinput;
3863 locinput += UTF8SKIP(locinput);
3866 /* As noted above, if we matched a prepend character, but
3867 * the next thing won't match, back off the last prepend we
3868 * matched, as it is guaranteed to match the begin */
3869 if (previous_prepend
3870 && (locinput >= PL_regeol
3871 || ! swash_fetch(PL_utf8_X_begin,
3872 (U8*)locinput, utf8_target)))
3874 locinput = previous_prepend;
3877 /* Note that here we know PL_regeol > locinput, as we
3878 * tested that upon input to this switch case, and if we
3879 * moved locinput forward, we tested the result just above
3880 * and it either passed, or we backed off so that it will
3882 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3884 /* Here did not match the required 'Begin' in the
3885 * second term. So just match the very first
3886 * character, the '.' of the final term of the regex */
3887 locinput = starting + UTF8SKIP(starting);
3890 /* Here is the beginning of a character that can have
3891 * an extender. It is either a hangul syllable, or a
3893 if (swash_fetch(PL_utf8_X_non_hangul,
3894 (U8*)locinput, utf8_target))
3897 /* Here not a Hangul syllable, must be a
3898 * ('! * Control') */
3899 locinput += UTF8SKIP(locinput);
3902 /* Here is a Hangul syllable. It can be composed
3903 * of several individual characters. One
3904 * possibility is T+ */
3905 if (swash_fetch(PL_utf8_X_T,
3906 (U8*)locinput, utf8_target))
3908 while (locinput < PL_regeol
3909 && swash_fetch(PL_utf8_X_T,
3910 (U8*)locinput, utf8_target))
3912 locinput += UTF8SKIP(locinput);
3916 /* Here, not T+, but is a Hangul. That means
3917 * it is one of the others: L, LV, LVT or V,
3919 * L* (L | LVT T* | V V* T* | LV V* T*) */
3922 while (locinput < PL_regeol
3923 && swash_fetch(PL_utf8_X_L,
3924 (U8*)locinput, utf8_target))
3926 locinput += UTF8SKIP(locinput);
3929 /* Here, have exhausted L*. If the next
3930 * character is not an LV, LVT nor V, it means
3931 * we had to have at least one L, so matches L+
3932 * in the original equation, we have a complete
3933 * hangul syllable. Are done. */
3935 if (locinput < PL_regeol
3936 && swash_fetch(PL_utf8_X_LV_LVT_V,
3937 (U8*)locinput, utf8_target))
3940 /* Otherwise keep going. Must be LV, LVT
3941 * or V. See if LVT */
3942 if (swash_fetch(PL_utf8_X_LVT,
3943 (U8*)locinput, utf8_target))
3945 locinput += UTF8SKIP(locinput);
3948 /* Must be V or LV. Take it, then
3950 locinput += UTF8SKIP(locinput);
3951 while (locinput < PL_regeol
3952 && swash_fetch(PL_utf8_X_V,
3953 (U8*)locinput, utf8_target))
3955 locinput += UTF8SKIP(locinput);
3959 /* And any of LV, LVT, or V can be followed
3961 while (locinput < PL_regeol
3962 && swash_fetch(PL_utf8_X_T,
3966 locinput += UTF8SKIP(locinput);
3972 /* Match any extender */
3973 while (locinput < PL_regeol
3974 && swash_fetch(PL_utf8_X_extend,
3975 (U8*)locinput, utf8_target))
3977 locinput += UTF8SKIP(locinput);
3981 if (locinput > PL_regeol) sayNO;
3983 nextchr = UCHARAT(locinput);
3987 { /* The capture buffer cases. The ones beginning with N for the
3988 named buffers just convert to the equivalent numbered and
3989 pretend they were called as the corresponding numbered buffer
3991 /* don't initialize these in the declaration, it makes C++
3996 const U8 *fold_array;
3999 PL_reg_flags |= RF_tainted;
4000 folder = foldEQ_locale;
4001 fold_array = PL_fold_locale;
4003 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4007 folder = foldEQ_latin1;
4008 fold_array = PL_fold_latin1;
4010 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4014 folder = foldEQ_latin1;
4015 fold_array = PL_fold_latin1;
4017 utf8_fold_flags = 0;
4022 fold_array = PL_fold;
4024 utf8_fold_flags = 0;
4031 utf8_fold_flags = 0;
4034 /* For the named back references, find the corresponding buffer
4036 n = reg_check_named_buff_matched(rex,scan);
4041 goto do_nref_ref_common;
4044 PL_reg_flags |= RF_tainted;
4045 folder = foldEQ_locale;
4046 fold_array = PL_fold_locale;
4047 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4051 folder = foldEQ_latin1;
4052 fold_array = PL_fold_latin1;
4053 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4057 folder = foldEQ_latin1;
4058 fold_array = PL_fold_latin1;
4059 utf8_fold_flags = 0;
4064 fold_array = PL_fold;
4065 utf8_fold_flags = 0;
4071 utf8_fold_flags = 0;
4075 n = ARG(scan); /* which paren pair */
4078 ln = PL_regoffs[n].start;
4079 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4080 if (*PL_reglastparen < n || ln == -1)
4081 sayNO; /* Do not match unless seen CLOSEn. */
4082 if (ln == PL_regoffs[n].end)
4086 if (type != REF /* REF can do byte comparison */
4087 && (utf8_target || type == REFFU))
4088 { /* XXX handle REFFL better */
4089 char * limit = PL_regeol;
4091 /* This call case insensitively compares the entire buffer
4092 * at s, with the current input starting at locinput, but
4093 * not going off the end given by PL_regeol, and returns in
4094 * limit upon success, how much of the current input was
4096 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4097 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4102 nextchr = UCHARAT(locinput);
4106 /* Not utf8: Inline the first character, for speed. */
4107 if (UCHARAT(s) != nextchr &&
4109 UCHARAT(s) != fold_array[nextchr]))
4111 ln = PL_regoffs[n].end - ln;
4112 if (locinput + ln > PL_regeol)
4114 if (ln > 1 && (type == REF
4115 ? memNE(s, locinput, ln)
4116 : ! folder(s, locinput, ln)))
4119 nextchr = UCHARAT(locinput);
4129 #define ST st->u.eval
4134 regexp_internal *rei;
4135 regnode *startpoint;
4138 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4139 if (cur_eval && cur_eval->locinput==locinput) {
4140 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4141 Perl_croak(aTHX_ "Infinite recursion in regex");
4142 if ( ++nochange_depth > max_nochange_depth )
4144 "Pattern subroutine nesting without pos change"
4145 " exceeded limit in regex");
4152 (void)ReREFCNT_inc(rex_sv);
4153 if (OP(scan)==GOSUB) {
4154 startpoint = scan + ARG2L(scan);
4155 ST.close_paren = ARG(scan);
4157 startpoint = rei->program+1;
4160 goto eval_recurse_doit;
4162 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4163 if (cur_eval && cur_eval->locinput==locinput) {
4164 if ( ++nochange_depth > max_nochange_depth )
4165 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4170 /* execute the code in the {...} */
4172 SV ** const before = SP;
4173 OP_4tree * const oop = PL_op;
4174 COP * const ocurcop = PL_curcop;
4176 char *saved_regeol = PL_regeol;
4177 struct re_save_state saved_state;
4179 /* To not corrupt the existing regex state while executing the
4180 * eval we would normally put it on the save stack, like with
4181 * save_re_context. However, re-evals have a weird scoping so we
4182 * can't just add ENTER/LEAVE here. With that, things like
4184 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4186 * would break, as they expect the localisation to be unwound
4187 * only when the re-engine backtracks through the bit that
4190 * What we do instead is just saving the state in a local c
4193 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4196 PL_op = (OP_4tree*)rexi->data->data[n];
4197 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4198 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4199 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4200 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4203 SV *sv_mrk = get_sv("REGMARK", 1);
4204 sv_setsv(sv_mrk, sv_yes_mark);
4207 CALLRUNOPS(aTHX); /* Scalar context. */
4210 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4216 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4219 PAD_RESTORE_LOCAL(old_comppad);
4220 PL_curcop = ocurcop;
4221 PL_regeol = saved_regeol;
4224 sv_setsv(save_scalar(PL_replgv), ret);
4228 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4231 /* extract RE object from returned value; compiling if
4237 SV *const sv = SvRV(ret);
4239 if (SvTYPE(sv) == SVt_REGEXP) {
4241 } else if (SvSMAGICAL(sv)) {
4242 mg = mg_find(sv, PERL_MAGIC_qr);
4245 } else if (SvTYPE(ret) == SVt_REGEXP) {
4247 } else if (SvSMAGICAL(ret)) {
4248 if (SvGMAGICAL(ret)) {
4249 /* I don't believe that there is ever qr magic
4251 assert(!mg_find(ret, PERL_MAGIC_qr));
4252 sv_unmagic(ret, PERL_MAGIC_qr);
4255 mg = mg_find(ret, PERL_MAGIC_qr);
4256 /* testing suggests mg only ends up non-NULL for
4257 scalars who were upgraded and compiled in the
4258 else block below. In turn, this is only
4259 triggered in the "postponed utf8 string" tests
4265 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4269 rx = reg_temp_copy(NULL, rx);
4273 const I32 osize = PL_regsize;
4276 assert (SvUTF8(ret));
4277 } else if (SvUTF8(ret)) {
4278 /* Not doing UTF-8, despite what the SV says. Is
4279 this only if we're trapped in use 'bytes'? */
4280 /* Make a copy of the octet sequence, but without
4281 the flag on, as the compiler now honours the
4282 SvUTF8 flag on ret. */
4284 const char *const p = SvPV(ret, len);
4285 ret = newSVpvn_flags(p, len, SVs_TEMP);
4287 rx = CALLREGCOMP(ret, pm_flags);
4289 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4291 /* This isn't a first class regexp. Instead, it's
4292 caching a regexp onto an existing, Perl visible
4294 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4299 re = (struct regexp *)SvANY(rx);
4301 RXp_MATCH_COPIED_off(re);
4302 re->subbeg = rex->subbeg;
4303 re->sublen = rex->sublen;
4306 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4307 "Matching embedded");
4309 startpoint = rei->program + 1;
4310 ST.close_paren = 0; /* only used for GOSUB */
4311 /* borrowed from regtry */
4312 if (PL_reg_start_tmpl <= re->nparens) {
4313 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4314 if(PL_reg_start_tmp)
4315 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4317 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4320 eval_recurse_doit: /* Share code with GOSUB below this line */
4321 /* run the pattern returned from (??{...}) */
4322 ST.cp = regcppush(0); /* Save *all* the positions. */
4323 REGCP_SET(ST.lastcp);
4325 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4327 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4328 PL_reglastparen = &re->lastparen;
4329 PL_reglastcloseparen = &re->lastcloseparen;
4331 re->lastcloseparen = 0;
4333 PL_reginput = locinput;
4336 /* XXXX This is too dramatic a measure... */
4339 ST.toggle_reg_flags = PL_reg_flags;
4341 PL_reg_flags |= RF_utf8;
4343 PL_reg_flags &= ~RF_utf8;
4344 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4346 ST.prev_rex = rex_sv;
4347 ST.prev_curlyx = cur_curlyx;
4348 SETREX(rex_sv,re_sv);
4353 ST.prev_eval = cur_eval;
4355 /* now continue from first node in postoned RE */
4356 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4359 /* logical is 1, /(?(?{...})X|Y)/ */
4360 sw = cBOOL(SvTRUE(ret));
4365 case EVAL_AB: /* cleanup after a successful (??{A})B */
4366 /* note: this is called twice; first after popping B, then A */
4367 PL_reg_flags ^= ST.toggle_reg_flags;
4368 ReREFCNT_dec(rex_sv);
4369 SETREX(rex_sv,ST.prev_rex);
4370 rex = (struct regexp *)SvANY(rex_sv);
4371 rexi = RXi_GET(rex);
4373 cur_eval = ST.prev_eval;
4374 cur_curlyx = ST.prev_curlyx;
4376 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4377 PL_reglastparen = &rex->lastparen;
4378 PL_reglastcloseparen = &rex->lastcloseparen;
4379 /* also update PL_regoffs */
4380 PL_regoffs = rex->offs;
4382 /* XXXX This is too dramatic a measure... */
4384 if ( nochange_depth )
4389 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4390 /* note: this is called twice; first after popping B, then A */
4391 PL_reg_flags ^= ST.toggle_reg_flags;
4392 ReREFCNT_dec(rex_sv);
4393 SETREX(rex_sv,ST.prev_rex);
4394 rex = (struct regexp *)SvANY(rex_sv);
4395 rexi = RXi_GET(rex);
4396 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4397 PL_reglastparen = &rex->lastparen;
4398 PL_reglastcloseparen = &rex->lastcloseparen;
4400 PL_reginput = locinput;
4401 REGCP_UNWIND(ST.lastcp);
4403 cur_eval = ST.prev_eval;
4404 cur_curlyx = ST.prev_curlyx;
4405 /* XXXX This is too dramatic a measure... */
4407 if ( nochange_depth )
4413 n = ARG(scan); /* which paren pair */
4414 PL_reg_start_tmp[n] = locinput;
4420 n = ARG(scan); /* which paren pair */
4421 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4422 PL_regoffs[n].end = locinput - PL_bostr;
4423 /*if (n > PL_regsize)
4425 if (n > *PL_reglastparen)
4426 *PL_reglastparen = n;
4427 *PL_reglastcloseparen = n;
4428 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4436 cursor && OP(cursor)!=END;
4437 cursor=regnext(cursor))
4439 if ( OP(cursor)==CLOSE ){
4441 if ( n <= lastopen ) {
4443 = PL_reg_start_tmp[n] - PL_bostr;
4444 PL_regoffs[n].end = locinput - PL_bostr;
4445 /*if (n > PL_regsize)
4447 if (n > *PL_reglastparen)
4448 *PL_reglastparen = n;
4449 *PL_reglastcloseparen = n;
4450 if ( n == ARG(scan) || (cur_eval &&
4451 cur_eval->u.eval.close_paren == n))
4460 n = ARG(scan); /* which paren pair */
4461 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4464 /* reg_check_named_buff_matched returns 0 for no match */
4465 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4469 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4475 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4477 next = NEXTOPER(NEXTOPER(scan));
4479 next = scan + ARG(scan);
4480 if (OP(next) == IFTHEN) /* Fake one. */
4481 next = NEXTOPER(NEXTOPER(next));
4485 logical = scan->flags;
4488 /*******************************************************************
4490 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4491 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4492 STAR/PLUS/CURLY/CURLYN are used instead.)
4494 A*B is compiled as <CURLYX><A><WHILEM><B>
4496 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4497 state, which contains the current count, initialised to -1. It also sets
4498 cur_curlyx to point to this state, with any previous value saved in the
4501 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4502 since the pattern may possibly match zero times (i.e. it's a while {} loop
4503 rather than a do {} while loop).
4505 Each entry to WHILEM represents a successful match of A. The count in the
4506 CURLYX block is incremented, another WHILEM state is pushed, and execution
4507 passes to A or B depending on greediness and the current count.
4509 For example, if matching against the string a1a2a3b (where the aN are
4510 substrings that match /A/), then the match progresses as follows: (the
4511 pushed states are interspersed with the bits of strings matched so far):
4514 <CURLYX cnt=0><WHILEM>
4515 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4516 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4517 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4518 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4520 (Contrast this with something like CURLYM, which maintains only a single
4524 a1 <CURLYM cnt=1> a2
4525 a1 a2 <CURLYM cnt=2> a3
4526 a1 a2 a3 <CURLYM cnt=3> b
4529 Each WHILEM state block marks a point to backtrack to upon partial failure
4530 of A or B, and also contains some minor state data related to that
4531 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4532 overall state, such as the count, and pointers to the A and B ops.
4534 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4535 must always point to the *current* CURLYX block, the rules are:
4537 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4538 and set cur_curlyx to point the new block.
4540 When popping the CURLYX block after a successful or unsuccessful match,
4541 restore the previous cur_curlyx.
4543 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4544 to the outer one saved in the CURLYX block.
4546 When popping the WHILEM block after a successful or unsuccessful B match,
4547 restore the previous cur_curlyx.
4549 Here's an example for the pattern (AI* BI)*BO
4550 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4553 curlyx backtrack stack
4554 ------ ---------------
4556 CO <CO prev=NULL> <WO>
4557 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4558 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4559 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4561 At this point the pattern succeeds, and we work back down the stack to
4562 clean up, restoring as we go:
4564 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4565 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4566 CO <CO prev=NULL> <WO>
4569 *******************************************************************/
4571 #define ST st->u.curlyx
4573 case CURLYX: /* start of /A*B/ (for complex A) */
4575 /* No need to save/restore up to this paren */
4576 I32 parenfloor = scan->flags;
4578 assert(next); /* keep Coverity happy */
4579 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4582 /* XXXX Probably it is better to teach regpush to support
4583 parenfloor > PL_regsize... */
4584 if (parenfloor > (I32)*PL_reglastparen)
4585 parenfloor = *PL_reglastparen; /* Pessimization... */
4587 ST.prev_curlyx= cur_curlyx;
4589 ST.cp = PL_savestack_ix;
4591 /* these fields contain the state of the current curly.
4592 * they are accessed by subsequent WHILEMs */
4593 ST.parenfloor = parenfloor;
4598 ST.count = -1; /* this will be updated by WHILEM */
4599 ST.lastloc = NULL; /* this will be updated by WHILEM */
4601 PL_reginput = locinput;
4602 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4606 case CURLYX_end: /* just finished matching all of A*B */
4607 cur_curlyx = ST.prev_curlyx;
4611 case CURLYX_end_fail: /* just failed to match all of A*B */
4613 cur_curlyx = ST.prev_curlyx;
4619 #define ST st->u.whilem
4621 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4623 /* see the discussion above about CURLYX/WHILEM */
4625 int min = ARG1(cur_curlyx->u.curlyx.me);
4626 int max = ARG2(cur_curlyx->u.curlyx.me);
4627 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4629 assert(cur_curlyx); /* keep Coverity happy */
4630 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4631 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4632 ST.cache_offset = 0;
4635 PL_reginput = locinput;
4637 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4638 "%*s whilem: matched %ld out of %d..%d\n",
4639 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4642 /* First just match a string of min A's. */
4645 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4646 cur_curlyx->u.curlyx.lastloc = locinput;
4647 REGCP_SET(ST.lastcp);
4649 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4653 /* If degenerate A matches "", assume A done. */
4655 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4656 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4657 "%*s whilem: empty match detected, trying continuation...\n",
4658 REPORT_CODE_OFF+depth*2, "")
4660 goto do_whilem_B_max;
4663 /* super-linear cache processing */
4667 if (!PL_reg_maxiter) {
4668 /* start the countdown: Postpone detection until we
4669 * know the match is not *that* much linear. */
4670 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4671 /* possible overflow for long strings and many CURLYX's */
4672 if (PL_reg_maxiter < 0)
4673 PL_reg_maxiter = I32_MAX;
4674 PL_reg_leftiter = PL_reg_maxiter;
4677 if (PL_reg_leftiter-- == 0) {
4678 /* initialise cache */
4679 const I32 size = (PL_reg_maxiter + 7)/8;
4680 if (PL_reg_poscache) {
4681 if ((I32)PL_reg_poscache_size < size) {
4682 Renew(PL_reg_poscache, size, char);
4683 PL_reg_poscache_size = size;
4685 Zero(PL_reg_poscache, size, char);
4688 PL_reg_poscache_size = size;
4689 Newxz(PL_reg_poscache, size, char);
4691 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4692 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4693 PL_colors[4], PL_colors[5])
4697 if (PL_reg_leftiter < 0) {
4698 /* have we already failed at this position? */
4700 offset = (scan->flags & 0xf) - 1
4701 + (locinput - PL_bostr) * (scan->flags>>4);
4702 mask = 1 << (offset % 8);
4704 if (PL_reg_poscache[offset] & mask) {
4705 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4706 "%*s whilem: (cache) already tried at this position...\n",
4707 REPORT_CODE_OFF+depth*2, "")
4709 sayNO; /* cache records failure */
4711 ST.cache_offset = offset;
4712 ST.cache_mask = mask;
4716 /* Prefer B over A for minimal matching. */
4718 if (cur_curlyx->u.curlyx.minmod) {
4719 ST.save_curlyx = cur_curlyx;
4720 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4721 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4722 REGCP_SET(ST.lastcp);
4723 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4727 /* Prefer A over B for maximal matching. */
4729 if (n < max) { /* More greed allowed? */
4730 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4731 cur_curlyx->u.curlyx.lastloc = locinput;
4732 REGCP_SET(ST.lastcp);
4733 PUSH_STATE_GOTO(WHILEM_A_max, A);
4736 goto do_whilem_B_max;
4740 case WHILEM_B_min: /* just matched B in a minimal match */
4741 case WHILEM_B_max: /* just matched B in a maximal match */
4742 cur_curlyx = ST.save_curlyx;
4746 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4747 cur_curlyx = ST.save_curlyx;
4748 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4749 cur_curlyx->u.curlyx.count--;
4753 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4755 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4756 REGCP_UNWIND(ST.lastcp);
4758 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4759 cur_curlyx->u.curlyx.count--;
4763 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4764 REGCP_UNWIND(ST.lastcp);
4765 regcppop(rex); /* Restore some previous $<digit>s? */
4766 PL_reginput = locinput;
4767 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4768 "%*s whilem: failed, trying continuation...\n",
4769 REPORT_CODE_OFF+depth*2, "")
4772 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4773 && ckWARN(WARN_REGEXP)
4774 && !(PL_reg_flags & RF_warned))
4776 PL_reg_flags |= RF_warned;
4777 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4778 "Complex regular subexpression recursion",
4783 ST.save_curlyx = cur_curlyx;
4784 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4785 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4788 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4789 cur_curlyx = ST.save_curlyx;
4790 REGCP_UNWIND(ST.lastcp);
4793 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4794 /* Maximum greed exceeded */
4795 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4796 && ckWARN(WARN_REGEXP)
4797 && !(PL_reg_flags & RF_warned))
4799 PL_reg_flags |= RF_warned;
4800 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4801 "%s limit (%d) exceeded",
4802 "Complex regular subexpression recursion",
4805 cur_curlyx->u.curlyx.count--;
4809 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4810 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4812 /* Try grabbing another A and see if it helps. */
4813 PL_reginput = locinput;
4814 cur_curlyx->u.curlyx.lastloc = locinput;
4815 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4816 REGCP_SET(ST.lastcp);
4817 PUSH_STATE_GOTO(WHILEM_A_min,
4818 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4822 #define ST st->u.branch
4824 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4825 next = scan + ARG(scan);
4828 scan = NEXTOPER(scan);
4831 case BRANCH: /* /(...|A|...)/ */
4832 scan = NEXTOPER(scan); /* scan now points to inner node */
4833 ST.lastparen = *PL_reglastparen;
4834 ST.next_branch = next;
4836 PL_reginput = locinput;
4838 /* Now go into the branch */
4840 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4842 PUSH_STATE_GOTO(BRANCH_next, scan);
4846 PL_reginput = locinput;
4847 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4848 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4849 PUSH_STATE_GOTO(CUTGROUP_next,next);
4851 case CUTGROUP_next_fail:
4854 if (st->u.mark.mark_name)
4855 sv_commit = st->u.mark.mark_name;
4861 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4866 REGCP_UNWIND(ST.cp);
4867 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4868 PL_regoffs[n].end = -1;
4869 *PL_reglastparen = n;
4870 /*dmq: *PL_reglastcloseparen = n; */
4871 scan = ST.next_branch;
4872 /* no more branches? */
4873 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4875 PerlIO_printf( Perl_debug_log,
4876 "%*s %sBRANCH failed...%s\n",
4877 REPORT_CODE_OFF+depth*2, "",
4883 continue; /* execute next BRANCH[J] op */
4891 #define ST st->u.curlym
4893 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4895 /* This is an optimisation of CURLYX that enables us to push
4896 * only a single backtracking state, no matter how many matches
4897 * there are in {m,n}. It relies on the pattern being constant
4898 * length, with no parens to influence future backrefs
4902 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4904 /* if paren positive, emulate an OPEN/CLOSE around A */
4906 U32 paren = ST.me->flags;
4907 if (paren > PL_regsize)
4909 if (paren > *PL_reglastparen)
4910 *PL_reglastparen = paren;
4911 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4919 ST.c1 = CHRTEST_UNINIT;
4922 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4925 curlym_do_A: /* execute the A in /A{m,n}B/ */
4926 PL_reginput = locinput;
4927 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4930 case CURLYM_A: /* we've just matched an A */
4931 locinput = st->locinput;
4932 nextchr = UCHARAT(locinput);
4935 /* after first match, determine A's length: u.curlym.alen */
4936 if (ST.count == 1) {
4937 if (PL_reg_match_utf8) {
4939 while (s < PL_reginput) {
4945 ST.alen = PL_reginput - locinput;
4948 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4951 PerlIO_printf(Perl_debug_log,
4952 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4953 (int)(REPORT_CODE_OFF+(depth*2)), "",
4954 (IV) ST.count, (IV)ST.alen)
4957 locinput = PL_reginput;
4959 if (cur_eval && cur_eval->u.eval.close_paren &&
4960 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4964 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4965 if ( max == REG_INFTY || ST.count < max )
4966 goto curlym_do_A; /* try to match another A */
4968 goto curlym_do_B; /* try to match B */
4970 case CURLYM_A_fail: /* just failed to match an A */
4971 REGCP_UNWIND(ST.cp);
4973 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4974 || (cur_eval && cur_eval->u.eval.close_paren &&
4975 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4978 curlym_do_B: /* execute the B in /A{m,n}B/ */
4979 PL_reginput = locinput;
4980 if (ST.c1 == CHRTEST_UNINIT) {
4981 /* calculate c1 and c2 for possible match of 1st char
4982 * following curly */
4983 ST.c1 = ST.c2 = CHRTEST_VOID;
4984 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4985 regnode *text_node = ST.B;
4986 if (! HAS_TEXT(text_node))
4987 FIND_NEXT_IMPT(text_node);
4990 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4992 But the former is redundant in light of the latter.
4994 if this changes back then the macro for
4995 IS_TEXT and friends need to change.
4997 if (PL_regkind[OP(text_node)] == EXACT)
5000 ST.c1 = (U8)*STRING(text_node);
5001 switch (OP(text_node)) {
5002 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5004 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5005 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5006 default: ST.c2 = ST.c1;
5013 PerlIO_printf(Perl_debug_log,
5014 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5015 (int)(REPORT_CODE_OFF+(depth*2)),
5018 if (ST.c1 != CHRTEST_VOID
5019 && UCHARAT(PL_reginput) != ST.c1
5020 && UCHARAT(PL_reginput) != ST.c2)
5022 /* simulate B failing */
5024 PerlIO_printf(Perl_debug_log,
5025 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5026 (int)(REPORT_CODE_OFF+(depth*2)),"",
5029 state_num = CURLYM_B_fail;
5030 goto reenter_switch;
5034 /* mark current A as captured */
5035 I32 paren = ST.me->flags;
5037 PL_regoffs[paren].start
5038 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5039 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5040 /*dmq: *PL_reglastcloseparen = paren; */
5043 PL_regoffs[paren].end = -1;
5044 if (cur_eval && cur_eval->u.eval.close_paren &&
5045 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5054 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5057 case CURLYM_B_fail: /* just failed to match a B */
5058 REGCP_UNWIND(ST.cp);
5060 I32 max = ARG2(ST.me);
5061 if (max != REG_INFTY && ST.count == max)
5063 goto curlym_do_A; /* try to match a further A */
5065 /* backtrack one A */
5066 if (ST.count == ARG1(ST.me) /* min */)
5069 locinput = HOPc(locinput, -ST.alen);
5070 goto curlym_do_B; /* try to match B */
5073 #define ST st->u.curly
5075 #define CURLY_SETPAREN(paren, success) \
5078 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5079 PL_regoffs[paren].end = locinput - PL_bostr; \
5080 *PL_reglastcloseparen = paren; \
5083 PL_regoffs[paren].end = -1; \
5086 case STAR: /* /A*B/ where A is width 1 */
5090 scan = NEXTOPER(scan);
5092 case PLUS: /* /A+B/ where A is width 1 */
5096 scan = NEXTOPER(scan);
5098 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5099 ST.paren = scan->flags; /* Which paren to set */
5100 if (ST.paren > PL_regsize)
5101 PL_regsize = ST.paren;
5102 if (ST.paren > *PL_reglastparen)
5103 *PL_reglastparen = ST.paren;
5104 ST.min = ARG1(scan); /* min to match */
5105 ST.max = ARG2(scan); /* max to match */
5106 if (cur_eval && cur_eval->u.eval.close_paren &&
5107 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5111 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5113 case CURLY: /* /A{m,n}B/ where A is width 1 */
5115 ST.min = ARG1(scan); /* min to match */
5116 ST.max = ARG2(scan); /* max to match */
5117 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5120 * Lookahead to avoid useless match attempts
5121 * when we know what character comes next.
5123 * Used to only do .*x and .*?x, but now it allows
5124 * for )'s, ('s and (?{ ... })'s to be in the way
5125 * of the quantifier and the EXACT-like node. -- japhy
5128 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5130 if (HAS_TEXT(next) || JUMPABLE(next)) {
5132 regnode *text_node = next;
5134 if (! HAS_TEXT(text_node))
5135 FIND_NEXT_IMPT(text_node);
5137 if (! HAS_TEXT(text_node))
5138 ST.c1 = ST.c2 = CHRTEST_VOID;
5140 if ( PL_regkind[OP(text_node)] != EXACT ) {
5141 ST.c1 = ST.c2 = CHRTEST_VOID;
5142 goto assume_ok_easy;
5145 s = (U8*)STRING(text_node);
5147 /* Currently we only get here when
5149 PL_rekind[OP(text_node)] == EXACT
5151 if this changes back then the macro for IS_TEXT and
5152 friends need to change. */
5155 switch (OP(text_node)) {
5156 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5158 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5159 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5160 default: ST.c2 = ST.c1; break;
5163 else { /* UTF_PATTERN */
5164 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5165 STRLEN ulen1, ulen2;
5166 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5167 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5169 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5170 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5172 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5174 0 : UTF8_ALLOW_ANY);
5175 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5177 0 : UTF8_ALLOW_ANY);
5179 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5181 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5186 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5193 ST.c1 = ST.c2 = CHRTEST_VOID;
5198 PL_reginput = locinput;
5201 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5204 locinput = PL_reginput;
5206 if (ST.c1 == CHRTEST_VOID)
5207 goto curly_try_B_min;
5209 ST.oldloc = locinput;
5211 /* set ST.maxpos to the furthest point along the
5212 * string that could possibly match */
5213 if (ST.max == REG_INFTY) {
5214 ST.maxpos = PL_regeol - 1;
5216 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5219 else if (utf8_target) {
5220 int m = ST.max - ST.min;
5221 for (ST.maxpos = locinput;
5222 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5223 ST.maxpos += UTF8SKIP(ST.maxpos);
5226 ST.maxpos = locinput + ST.max - ST.min;
5227 if (ST.maxpos >= PL_regeol)
5228 ST.maxpos = PL_regeol - 1;
5230 goto curly_try_B_min_known;
5234 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5235 locinput = PL_reginput;
5236 if (ST.count < ST.min)
5238 if ((ST.count > ST.min)
5239 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5241 /* A{m,n} must come at the end of the string, there's
5242 * no point in backing off ... */
5244 /* ...except that $ and \Z can match before *and* after
5245 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5246 We may back off by one in this case. */
5247 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5251 goto curly_try_B_max;
5256 case CURLY_B_min_known_fail:
5257 /* failed to find B in a non-greedy match where c1,c2 valid */
5258 if (ST.paren && ST.count)
5259 PL_regoffs[ST.paren].end = -1;
5261 PL_reginput = locinput; /* Could be reset... */
5262 REGCP_UNWIND(ST.cp);
5263 /* Couldn't or didn't -- move forward. */
5264 ST.oldloc = locinput;
5266 locinput += UTF8SKIP(locinput);
5270 curly_try_B_min_known:
5271 /* find the next place where 'B' could work, then call B */
5275 n = (ST.oldloc == locinput) ? 0 : 1;
5276 if (ST.c1 == ST.c2) {
5278 /* set n to utf8_distance(oldloc, locinput) */
5279 while (locinput <= ST.maxpos &&
5280 utf8n_to_uvchr((U8*)locinput,
5281 UTF8_MAXBYTES, &len,
5282 uniflags) != (UV)ST.c1) {
5288 /* set n to utf8_distance(oldloc, locinput) */
5289 while (locinput <= ST.maxpos) {
5291 const UV c = utf8n_to_uvchr((U8*)locinput,
5292 UTF8_MAXBYTES, &len,
5294 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5302 if (ST.c1 == ST.c2) {
5303 while (locinput <= ST.maxpos &&
5304 UCHARAT(locinput) != ST.c1)
5308 while (locinput <= ST.maxpos
5309 && UCHARAT(locinput) != ST.c1
5310 && UCHARAT(locinput) != ST.c2)
5313 n = locinput - ST.oldloc;
5315 if (locinput > ST.maxpos)
5317 /* PL_reginput == oldloc now */
5320 if (regrepeat(rex, ST.A, n, depth) < n)
5323 PL_reginput = locinput;
5324 CURLY_SETPAREN(ST.paren, ST.count);
5325 if (cur_eval && cur_eval->u.eval.close_paren &&
5326 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5329 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5334 case CURLY_B_min_fail:
5335 /* failed to find B in a non-greedy match where c1,c2 invalid */
5336 if (ST.paren && ST.count)
5337 PL_regoffs[ST.paren].end = -1;
5339 REGCP_UNWIND(ST.cp);
5340 /* failed -- move forward one */
5341 PL_reginput = locinput;
5342 if (regrepeat(rex, ST.A, 1, depth)) {
5344 locinput = PL_reginput;
5345 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5346 ST.count > 0)) /* count overflow ? */
5349 CURLY_SETPAREN(ST.paren, ST.count);
5350 if (cur_eval && cur_eval->u.eval.close_paren &&
5351 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5354 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5362 /* a successful greedy match: now try to match B */
5363 if (cur_eval && cur_eval->u.eval.close_paren &&
5364 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5369 if (ST.c1 != CHRTEST_VOID)
5370 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5371 UTF8_MAXBYTES, 0, uniflags)
5372 : (UV) UCHARAT(PL_reginput);
5373 /* If it could work, try it. */
5374 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5375 CURLY_SETPAREN(ST.paren, ST.count);
5376 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5381 case CURLY_B_max_fail:
5382 /* failed to find B in a greedy match */
5383 if (ST.paren && ST.count)
5384 PL_regoffs[ST.paren].end = -1;
5386 REGCP_UNWIND(ST.cp);
5388 if (--ST.count < ST.min)
5390 PL_reginput = locinput = HOPc(locinput, -1);
5391 goto curly_try_B_max;
5398 /* we've just finished A in /(??{A})B/; now continue with B */
5400 st->u.eval.toggle_reg_flags
5401 = cur_eval->u.eval.toggle_reg_flags;
5402 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5404 st->u.eval.prev_rex = rex_sv; /* inner */
5405 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5406 rex = (struct regexp *)SvANY(rex_sv);
5407 rexi = RXi_GET(rex);
5408 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5409 (void)ReREFCNT_inc(rex_sv);
5410 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5412 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5413 PL_reglastparen = &rex->lastparen;
5414 PL_reglastcloseparen = &rex->lastcloseparen;
5416 REGCP_SET(st->u.eval.lastcp);
5417 PL_reginput = locinput;
5419 /* Restore parens of the outer rex without popping the
5421 tmpix = PL_savestack_ix;
5422 PL_savestack_ix = cur_eval->u.eval.lastcp;
5424 PL_savestack_ix = tmpix;
5426 st->u.eval.prev_eval = cur_eval;
5427 cur_eval = cur_eval->u.eval.prev_eval;
5429 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5430 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5431 if ( nochange_depth )
5434 PUSH_YES_STATE_GOTO(EVAL_AB,
5435 st->u.eval.prev_eval->u.eval.B); /* match B */
5438 if (locinput < reginfo->till) {
5439 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5440 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5442 (long)(locinput - PL_reg_starttry),
5443 (long)(reginfo->till - PL_reg_starttry),
5446 sayNO_SILENT; /* Cannot match: too short. */
5448 PL_reginput = locinput; /* put where regtry can find it */
5449 sayYES; /* Success! */
5451 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5453 PerlIO_printf(Perl_debug_log,
5454 "%*s %ssubpattern success...%s\n",
5455 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5456 PL_reginput = locinput; /* put where regtry can find it */
5457 sayYES; /* Success! */
5460 #define ST st->u.ifmatch
5462 case SUSPEND: /* (?>A) */
5464 PL_reginput = locinput;
5467 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5469 goto ifmatch_trivial_fail_test;
5471 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5473 ifmatch_trivial_fail_test:
5475 char * const s = HOPBACKc(locinput, scan->flags);
5480 sw = 1 - cBOOL(ST.wanted);
5484 next = scan + ARG(scan);
5492 PL_reginput = locinput;
5496 ST.logical = logical;
5497 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5499 /* execute body of (?...A) */
5500 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5503 case IFMATCH_A_fail: /* body of (?...A) failed */
5504 ST.wanted = !ST.wanted;
5507 case IFMATCH_A: /* body of (?...A) succeeded */
5509 sw = cBOOL(ST.wanted);
5511 else if (!ST.wanted)
5514 if (OP(ST.me) == SUSPEND)
5515 locinput = PL_reginput;
5517 locinput = PL_reginput = st->locinput;
5518 nextchr = UCHARAT(locinput);
5520 scan = ST.me + ARG(ST.me);
5523 continue; /* execute B */
5528 next = scan + ARG(scan);
5533 reginfo->cutpoint = PL_regeol;
5536 PL_reginput = locinput;
5538 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5539 PUSH_STATE_GOTO(COMMIT_next,next);
5541 case COMMIT_next_fail:
5548 #define ST st->u.mark
5550 ST.prev_mark = mark_state;
5551 ST.mark_name = sv_commit = sv_yes_mark
5552 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5554 ST.mark_loc = PL_reginput = locinput;
5555 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5557 case MARKPOINT_next:
5558 mark_state = ST.prev_mark;
5561 case MARKPOINT_next_fail:
5562 if (popmark && sv_eq(ST.mark_name,popmark))
5564 if (ST.mark_loc > startpoint)
5565 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5566 popmark = NULL; /* we found our mark */
5567 sv_commit = ST.mark_name;
5570 PerlIO_printf(Perl_debug_log,
5571 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5572 REPORT_CODE_OFF+depth*2, "",
5573 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5576 mark_state = ST.prev_mark;
5577 sv_yes_mark = mark_state ?
5578 mark_state->u.mark.mark_name : NULL;
5582 PL_reginput = locinput;
5584 /* (*SKIP) : if we fail we cut here*/
5585 ST.mark_name = NULL;
5586 ST.mark_loc = locinput;
5587 PUSH_STATE_GOTO(SKIP_next,next);
5589 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5590 otherwise do nothing. Meaning we need to scan
5592 regmatch_state *cur = mark_state;
5593 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5596 if ( sv_eq( cur->u.mark.mark_name,
5599 ST.mark_name = find;
5600 PUSH_STATE_GOTO( SKIP_next, next );
5602 cur = cur->u.mark.prev_mark;
5605 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5607 case SKIP_next_fail:
5609 /* (*CUT:NAME) - Set up to search for the name as we
5610 collapse the stack*/
5611 popmark = ST.mark_name;
5613 /* (*CUT) - No name, we cut here.*/
5614 if (ST.mark_loc > startpoint)
5615 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5616 /* but we set sv_commit to latest mark_name if there
5617 is one so they can test to see how things lead to this
5620 sv_commit=mark_state->u.mark.mark_name;
5628 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5630 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5633 U8 folded[UTF8_MAXBYTES_CASE+1];
5635 const char * const l = locinput;
5636 char *e = PL_regeol;
5637 to_uni_fold(n, folded, &foldlen);
5639 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5640 l, &e, 0, utf8_target)) {
5645 nextchr = UCHARAT(locinput);
5648 if ((n=is_LNBREAK(locinput,utf8_target))) {
5650 nextchr = UCHARAT(locinput);
5655 #define CASE_CLASS(nAmE) \
5657 if ((n=is_##nAmE(locinput,utf8_target))) { \
5659 nextchr = UCHARAT(locinput); \
5664 if ((n=is_##nAmE(locinput,utf8_target))) { \
5667 locinput += UTF8SKIP(locinput); \
5668 nextchr = UCHARAT(locinput); \
5673 CASE_CLASS(HORIZWS);
5677 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5678 PTR2UV(scan), OP(scan));
5679 Perl_croak(aTHX_ "regexp memory corruption");
5683 /* switch break jumps here */
5684 scan = next; /* prepare to execute the next op and ... */
5685 continue; /* ... jump back to the top, reusing st */
5689 /* push a state that backtracks on success */
5690 st->u.yes.prev_yes_state = yes_state;
5694 /* push a new regex state, then continue at scan */
5696 regmatch_state *newst;
5699 regmatch_state *cur = st;
5700 regmatch_state *curyes = yes_state;
5702 regmatch_slab *slab = PL_regmatch_slab;
5703 for (;curd > -1;cur--,curd--) {
5704 if (cur < SLAB_FIRST(slab)) {
5706 cur = SLAB_LAST(slab);
5708 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5709 REPORT_CODE_OFF + 2 + depth * 2,"",
5710 curd, PL_reg_name[cur->resume_state],
5711 (curyes == cur) ? "yes" : ""
5714 curyes = cur->u.yes.prev_yes_state;
5717 DEBUG_STATE_pp("push")
5720 st->locinput = locinput;
5722 if (newst > SLAB_LAST(PL_regmatch_slab))
5723 newst = S_push_slab(aTHX);
5724 PL_regmatch_state = newst;
5726 locinput = PL_reginput;
5727 nextchr = UCHARAT(locinput);
5735 * We get here only if there's trouble -- normally "case END" is
5736 * the terminating point.
5738 Perl_croak(aTHX_ "corrupted regexp pointers");
5744 /* we have successfully completed a subexpression, but we must now
5745 * pop to the state marked by yes_state and continue from there */
5746 assert(st != yes_state);
5748 while (st != yes_state) {
5750 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5751 PL_regmatch_slab = PL_regmatch_slab->prev;
5752 st = SLAB_LAST(PL_regmatch_slab);
5756 DEBUG_STATE_pp("pop (no final)");
5758 DEBUG_STATE_pp("pop (yes)");
5764 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5765 || yes_state > SLAB_LAST(PL_regmatch_slab))
5767 /* not in this slab, pop slab */
5768 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5769 PL_regmatch_slab = PL_regmatch_slab->prev;
5770 st = SLAB_LAST(PL_regmatch_slab);
5772 depth -= (st - yes_state);
5775 yes_state = st->u.yes.prev_yes_state;
5776 PL_regmatch_state = st;
5779 locinput= st->locinput;
5780 nextchr = UCHARAT(locinput);
5782 state_num = st->resume_state + no_final;
5783 goto reenter_switch;
5786 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5787 PL_colors[4], PL_colors[5]));
5789 if (PL_reg_eval_set) {
5790 /* each successfully executed (?{...}) block does the equivalent of
5791 * local $^R = do {...}
5792 * When popping the save stack, all these locals would be undone;
5793 * bypass this by setting the outermost saved $^R to the latest
5795 if (oreplsv != GvSV(PL_replgv))
5796 sv_setsv(oreplsv, GvSV(PL_replgv));
5803 PerlIO_printf(Perl_debug_log,
5804 "%*s %sfailed...%s\n",
5805 REPORT_CODE_OFF+depth*2, "",
5806 PL_colors[4], PL_colors[5])
5818 /* there's a previous state to backtrack to */
5820 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5821 PL_regmatch_slab = PL_regmatch_slab->prev;
5822 st = SLAB_LAST(PL_regmatch_slab);
5824 PL_regmatch_state = st;
5825 locinput= st->locinput;
5826 nextchr = UCHARAT(locinput);
5828 DEBUG_STATE_pp("pop");
5830 if (yes_state == st)
5831 yes_state = st->u.yes.prev_yes_state;
5833 state_num = st->resume_state + 1; /* failure = success + 1 */
5834 goto reenter_switch;
5839 if (rex->intflags & PREGf_VERBARG_SEEN) {
5840 SV *sv_err = get_sv("REGERROR", 1);
5841 SV *sv_mrk = get_sv("REGMARK", 1);
5843 sv_commit = &PL_sv_no;
5845 sv_yes_mark = &PL_sv_yes;
5848 sv_commit = &PL_sv_yes;
5849 sv_yes_mark = &PL_sv_no;
5851 sv_setsv(sv_err, sv_commit);
5852 sv_setsv(sv_mrk, sv_yes_mark);
5855 /* clean up; in particular, free all slabs above current one */
5856 LEAVE_SCOPE(oldsave);
5862 - regrepeat - repeatedly match something simple, report how many
5865 * [This routine now assumes that it will only match on things of length 1.
5866 * That was true before, but now we assume scan - reginput is the count,
5867 * rather than incrementing count on every character. [Er, except utf8.]]
5870 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5873 register char *scan;
5875 register char *loceol = PL_regeol;
5876 register I32 hardcount = 0;
5877 register bool utf8_target = PL_reg_match_utf8;
5880 PERL_UNUSED_ARG(depth);
5883 PERL_ARGS_ASSERT_REGREPEAT;
5886 if (max == REG_INFTY)
5888 else if (max < loceol - scan)
5889 loceol = scan + max;
5894 while (scan < loceol && hardcount < max && *scan != '\n') {
5895 scan += UTF8SKIP(scan);
5899 while (scan < loceol && *scan != '\n')
5906 while (scan < loceol && hardcount < max) {
5907 scan += UTF8SKIP(scan);
5918 /* To get here, EXACTish nodes must have *byte* length == 1. That
5919 * means they match only characters in the string that can be expressed
5920 * as a single byte. For non-utf8 strings, that means a simple match.
5921 * For utf8 strings, the character matched must be an invariant, or
5922 * downgradable to a single byte. The pattern's utf8ness is
5923 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5924 * it is, it's an invariant */
5927 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5929 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5930 while (scan < loceol && UCHARAT(scan) == c) {
5936 /* Here, the string is utf8, and the pattern char is different
5937 * in utf8 than not, so can't compare them directly. Outside the
5938 * loop, find find the two utf8 bytes that represent c, and then
5939 * look for those in sequence in the utf8 string */
5940 U8 high = UTF8_TWO_BYTE_HI(c);
5941 U8 low = UTF8_TWO_BYTE_LO(c);
5944 while (hardcount < max
5945 && scan + 1 < loceol
5946 && UCHARAT(scan) == high
5947 && UCHARAT(scan + 1) == low)
5955 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5959 PL_reg_flags |= RF_tainted;
5960 utf8_flags = FOLDEQ_UTF8_LOCALE;
5967 /* The comments for the EXACT case above apply as well to these fold
5972 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5974 if (utf8_target) { /* Use full Unicode fold matching */
5975 char *tmpeol = loceol;
5976 while (hardcount < max
5977 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
5978 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
5985 /* XXX Note that the above handles properly the German sharp s in
5986 * the pattern matching ss in the string. But it doesn't handle
5987 * properly cases where the string contains say 'LIGATURE ff' and
5988 * the pattern is 'f+'. This would require, say, a new function or
5989 * revised interface to foldEQ_utf8(), in which the maximum number
5990 * of characters to match could be passed and it would return how
5991 * many actually did. This is just one of many cases where
5992 * multi-char folds don't work properly, and so the fix is being
5998 /* Here, the string isn't utf8 and c is a single byte; and either
5999 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6000 * doesn't affect c. Can just do simple comparisons for exact or
6003 case EXACTF: folded = PL_fold[c]; break;
6005 case EXACTFU: folded = PL_fold_latin1[c]; break;
6006 case EXACTFL: folded = PL_fold_locale[c]; break;
6007 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6009 while (scan < loceol &&
6010 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6018 if (utf8_target || OP(p) == ANYOFV) {
6021 inclasslen = loceol - scan;
6022 while (hardcount < max
6023 && ((inclasslen = loceol - scan) > 0)
6024 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6030 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6038 LOAD_UTF8_CHARCLASS_ALNUM();
6039 while (hardcount < max && scan < loceol &&
6040 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6042 scan += UTF8SKIP(scan);
6046 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6054 while (scan < loceol && isALNUM((U8) *scan)) {
6059 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6064 PL_reg_flags |= RF_tainted;
6067 while (hardcount < max && scan < loceol &&
6068 isALNUM_LC_utf8((U8*)scan)) {
6069 scan += UTF8SKIP(scan);
6073 while (scan < loceol && isALNUM_LC(*scan))
6083 LOAD_UTF8_CHARCLASS_ALNUM();
6084 while (hardcount < max && scan < loceol &&
6085 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6087 scan += UTF8SKIP(scan);
6091 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6098 goto utf8_Nwordchar;
6099 while (scan < loceol && ! isALNUM((U8) *scan)) {
6105 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6106 scan += UTF8SKIP(scan);
6110 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6116 PL_reg_flags |= RF_tainted;
6119 while (hardcount < max && scan < loceol &&
6120 !isALNUM_LC_utf8((U8*)scan)) {
6121 scan += UTF8SKIP(scan);
6125 while (scan < loceol && !isALNUM_LC(*scan))
6135 LOAD_UTF8_CHARCLASS_SPACE();
6136 while (hardcount < max && scan < loceol &&
6138 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6140 scan += UTF8SKIP(scan);
6146 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6155 while (scan < loceol && isSPACE((U8) *scan)) {
6160 while (scan < loceol && isSPACE_A((U8) *scan)) {
6165 PL_reg_flags |= RF_tainted;
6168 while (hardcount < max && scan < loceol &&
6169 isSPACE_LC_utf8((U8*)scan)) {
6170 scan += UTF8SKIP(scan);
6174 while (scan < loceol && isSPACE_LC(*scan))
6184 LOAD_UTF8_CHARCLASS_SPACE();
6185 while (hardcount < max && scan < loceol &&
6187 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6189 scan += UTF8SKIP(scan);
6195 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6204 while (scan < loceol && ! isSPACE((U8) *scan)) {
6210 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6211 scan += UTF8SKIP(scan);
6215 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6221 PL_reg_flags |= RF_tainted;
6224 while (hardcount < max && scan < loceol &&
6225 !isSPACE_LC_utf8((U8*)scan)) {
6226 scan += UTF8SKIP(scan);
6230 while (scan < loceol && !isSPACE_LC(*scan))
6237 LOAD_UTF8_CHARCLASS_DIGIT();
6238 while (hardcount < max && scan < loceol &&
6239 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6240 scan += UTF8SKIP(scan);
6244 while (scan < loceol && isDIGIT(*scan))
6249 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6254 PL_reg_flags |= RF_tainted;
6257 while (hardcount < max && scan < loceol &&
6258 isDIGIT_LC_utf8((U8*)scan)) {
6259 scan += UTF8SKIP(scan);
6263 while (scan < loceol && isDIGIT_LC(*scan))
6270 LOAD_UTF8_CHARCLASS_DIGIT();
6271 while (hardcount < max && scan < loceol &&
6272 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6273 scan += UTF8SKIP(scan);
6277 while (scan < loceol && !isDIGIT(*scan))
6283 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6284 scan += UTF8SKIP(scan);
6288 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6294 PL_reg_flags |= RF_tainted;
6297 while (hardcount < max && scan < loceol &&
6298 !isDIGIT_LC_utf8((U8*)scan)) {
6299 scan += UTF8SKIP(scan);
6303 while (scan < loceol && !isDIGIT_LC(*scan))
6310 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6316 LNBREAK can match two latin chars, which is ok,
6317 because we have a null terminated string, but we
6318 have to use hardcount in this situation
6320 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6329 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6334 while (scan < loceol && is_HORIZWS_latin1(scan))
6341 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6342 scan += UTF8SKIP(scan);
6346 while (scan < loceol && !is_HORIZWS_latin1(scan))
6354 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6359 while (scan < loceol && is_VERTWS_latin1(scan))
6367 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6368 scan += UTF8SKIP(scan);
6372 while (scan < loceol && !is_VERTWS_latin1(scan))
6378 default: /* Called on something of 0 width. */
6379 break; /* So match right here or not at all. */
6385 c = scan - PL_reginput;
6389 GET_RE_DEBUG_FLAGS_DECL;
6391 SV * const prop = sv_newmortal();
6392 regprop(prog, prop, p);
6393 PerlIO_printf(Perl_debug_log,
6394 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6395 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6403 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6405 - regclass_swash - prepare the utf8 swash
6409 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6415 RXi_GET_DECL(prog,progi);
6416 const struct reg_data * const data = prog ? progi->data : NULL;
6418 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6420 assert(ANYOF_NONBITMAP(node));
6422 if (data && data->count) {
6423 const U32 n = ARG(node);
6425 if (data->what[n] == 's') {
6426 SV * const rv = MUTABLE_SV(data->data[n]);
6427 AV * const av = MUTABLE_AV(SvRV(rv));
6428 SV **const ary = AvARRAY(av);
6431 /* See the end of regcomp.c:S_regclass() for
6432 * documentation of these array elements. */
6435 a = SvROK(ary[1]) ? &ary[1] : NULL;
6436 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6440 else if (si && doinit) {
6441 sw = swash_init("utf8", "", si, 1, 0);
6442 (void)av_store(av, 1, sw);
6459 - reginclass - determine if a character falls into a character class
6461 n is the ANYOF regnode
6462 p is the target string
6463 lenp is pointer to the maximum number of bytes of how far to go in p
6464 (This is assumed wthout checking to always be at least the current
6466 utf8_target tells whether p is in UTF-8.
6468 Returns true if matched; false otherwise. If lenp is not NULL, on return
6469 from a successful match, the value it points to will be updated to how many
6470 bytes in p were matched. If there was no match, the value is undefined,
6471 possibly changed from the input.
6473 Note that this can be a synthetic start class, a combination of various
6474 nodes, so things you think might be mutually exclusive, such as locale,
6475 aren't. It can match both locale and non-locale
6480 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6483 const char flags = ANYOF_FLAGS(n);
6489 PERL_ARGS_ASSERT_REGINCLASS;
6491 /* If c is not already the code point, get it */
6492 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6493 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6494 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6495 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6496 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6497 * UTF8_ALLOW_FFFF */
6498 if (c_len == (STRLEN)-1)
6499 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6505 /* Use passed in max length, or one character if none passed in or less
6506 * than one character. And assume will match just one character. This is
6507 * overwritten later if matched more. */
6509 maxlen = (*lenp > c_len) ? *lenp : c_len;
6517 /* If this character is potentially in the bitmap, check it */
6519 if (ANYOF_BITMAP_TEST(n, c))
6521 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6528 else if (flags & ANYOF_LOCALE) {
6529 PL_reg_flags |= RF_tainted;
6531 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6532 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6536 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6537 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6538 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6539 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6540 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6541 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6542 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6543 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6544 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6545 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6546 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6547 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6548 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6549 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6550 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6551 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6552 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6553 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6554 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6555 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6556 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6557 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6558 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6559 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6560 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6561 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6562 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6563 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6564 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6565 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6566 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6567 ) /* How's that for a conditional? */
6574 /* If the bitmap didn't (or couldn't) match, and something outside the
6575 * bitmap could match, try that. Locale nodes specifiy completely the
6576 * behavior of code points in the bit map (otherwise, a utf8 target would
6577 * cause them to be treated as Unicode and not locale), except in
6578 * the very unlikely event when this node is a synthetic start class, which
6579 * could be a combination of locale and non-locale nodes. So allow locale
6580 * to match for the synthetic start class, which will give a false
6581 * positive that will be resolved when the match is done again as not part
6582 * of the synthetic start class */
6584 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6585 match = TRUE; /* Everything above 255 matches */
6587 else if (ANYOF_NONBITMAP(n)
6588 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6591 || (! (flags & ANYOF_LOCALE))
6592 || (flags & ANYOF_IS_SYNTHETIC)))))
6595 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6603 /* Not utf8. Convert as much of the string as available up
6604 * to the limit of how far the (single) character in the
6605 * pattern can possibly match (no need to go further). If
6606 * the node is a straight ANYOF or not folding, it can't
6607 * match more than one. Otherwise, It can match up to how
6608 * far a single char can fold to. Since not utf8, each
6609 * character is a single byte, so the max it can be in
6610 * bytes is the same as the max it can be in characters */
6611 STRLEN len = (OP(n) == ANYOF
6612 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6614 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6616 : UTF8_MAX_FOLD_CHAR_EXPAND;
6617 utf8_p = bytes_to_utf8(p, &len);
6620 if (swash_fetch(sw, utf8_p, TRUE))
6622 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6624 /* Here, we need to test if the fold of the target string
6625 * matches. The non-multi char folds have all been moved to
6626 * the compilation phase, and the multi-char folds have
6627 * been stored by regcomp into 'av'; we linearly check to
6628 * see if any match the target string (folded). We know
6629 * that the originals were each one character, but we don't
6630 * currently know how many characters/bytes each folded to,
6631 * except we do know that there are small limits imposed by
6632 * Unicode. XXX A performance enhancement would be to have
6633 * regcomp.c store the max number of chars/bytes that are
6634 * in an av entry, as, say the 0th element. Even better
6635 * would be to have a hash of the few characters that can
6636 * start a multi-char fold to the max number of chars of
6639 * If there is a match, we will need to advance (if lenp is
6640 * specified) the match pointer in the target string. But
6641 * what we are comparing here isn't that string directly,
6642 * but its fold, whose length may differ from the original.
6643 * As we go along in constructing the fold, therefore, we
6644 * create a map so that we know how many bytes in the
6645 * source to advance given that we have matched a certain
6646 * number of bytes in the fold. This map is stored in
6647 * 'map_fold_len_back'. Let n mean the number of bytes in
6648 * the fold of the first character that we are folding.
6649 * Then map_fold_len_back[n] is set to the number of bytes
6650 * in that first character. Similarly let m be the
6651 * corresponding number for the second character to be
6652 * folded. Then map_fold_len_back[n+m] is set to the
6653 * number of bytes occupied by the first two source
6654 * characters. ... */
6655 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6656 U8 folded[UTF8_MAXBYTES_CASE+1];
6657 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6658 STRLEN total_foldlen = 0; /* num bytes in fold of all
6661 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6663 /* Here, only need to fold the first char of the target
6664 * string. It the source wasn't utf8, is 1 byte long */
6665 to_utf8_fold(utf8_p, folded, &foldlen);
6666 total_foldlen = foldlen;
6667 map_fold_len_back[foldlen] = (utf8_target)
6673 /* Here, need to fold more than the first char. Do so
6674 * up to the limits */
6675 U8* source_ptr = utf8_p; /* The source for the fold
6678 U8* folded_ptr = folded;
6679 U8* e = utf8_p + maxlen; /* Can't go beyond last
6680 available byte in the
6684 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6688 /* Fold the next character */
6689 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6690 STRLEN this_char_foldlen;
6691 to_utf8_fold(source_ptr,
6693 &this_char_foldlen);
6695 /* Bail if it would exceed the byte limit for
6696 * folding a single char. */
6697 if (this_char_foldlen + folded_ptr - folded >
6703 /* Add the fold of this character */
6704 Copy(this_char_folded,
6708 source_ptr += UTF8SKIP(source_ptr);
6709 folded_ptr += this_char_foldlen;
6710 total_foldlen = folded_ptr - folded;
6712 /* Create map from the number of bytes in the fold
6713 * back to the number of bytes in the source. If
6714 * the source isn't utf8, the byte count is just
6715 * the number of characters so far */
6716 map_fold_len_back[total_foldlen]
6718 ? source_ptr - utf8_p
6725 /* Do the linear search to see if the fold is in the list
6726 * of multi-char folds. */
6729 for (i = 0; i <= av_len(av); i++) {
6730 SV* const sv = *av_fetch(av, i, FALSE);
6732 const char * const s = SvPV_const(sv, len);
6734 if (len <= total_foldlen && memEQ(s,
6739 /* Advance the target string ptr to account for
6740 * this fold, but have to translate from the
6741 * folded length to the corresponding source
6744 *lenp = map_fold_len_back[len];
6745 assert(*lenp != 0); /* Otherwise will loop */
6754 /* If we allocated a string above, free it */
6755 if (! utf8_target) Safefree(utf8_p);
6760 return (flags & ANYOF_INVERT) ? !match : match;
6764 S_reghop3(U8 *s, I32 off, const U8* lim)
6768 PERL_ARGS_ASSERT_REGHOP3;
6771 while (off-- && s < lim) {
6772 /* XXX could check well-formedness here */
6777 while (off++ && s > lim) {
6779 if (UTF8_IS_CONTINUED(*s)) {
6780 while (s > lim && UTF8_IS_CONTINUATION(*s))
6783 /* XXX could check well-formedness here */
6790 /* there are a bunch of places where we use two reghop3's that should
6791 be replaced with this routine. but since thats not done yet
6792 we ifdef it out - dmq
6795 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6799 PERL_ARGS_ASSERT_REGHOP4;
6802 while (off-- && s < rlim) {
6803 /* XXX could check well-formedness here */
6808 while (off++ && s > llim) {
6810 if (UTF8_IS_CONTINUED(*s)) {
6811 while (s > llim && UTF8_IS_CONTINUATION(*s))
6814 /* XXX could check well-formedness here */
6822 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6826 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6829 while (off-- && s < lim) {
6830 /* XXX could check well-formedness here */
6837 while (off++ && s > lim) {
6839 if (UTF8_IS_CONTINUED(*s)) {
6840 while (s > lim && UTF8_IS_CONTINUATION(*s))
6843 /* XXX could check well-formedness here */
6852 restore_pos(pTHX_ void *arg)
6855 regexp * const rex = (regexp *)arg;
6856 if (PL_reg_eval_set) {
6857 if (PL_reg_oldsaved) {
6858 rex->subbeg = PL_reg_oldsaved;
6859 rex->sublen = PL_reg_oldsavedlen;
6860 #ifdef PERL_OLD_COPY_ON_WRITE
6861 rex->saved_copy = PL_nrs;
6863 RXp_MATCH_COPIED_on(rex);
6865 PL_reg_magic->mg_len = PL_reg_oldpos;
6866 PL_reg_eval_set = 0;
6867 PL_curpm = PL_reg_oldcurpm;
6872 S_to_utf8_substr(pTHX_ register regexp *prog)
6876 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6879 if (prog->substrs->data[i].substr
6880 && !prog->substrs->data[i].utf8_substr) {
6881 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6882 prog->substrs->data[i].utf8_substr = sv;
6883 sv_utf8_upgrade(sv);
6884 if (SvVALID(prog->substrs->data[i].substr)) {
6885 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6886 if (flags & FBMcf_TAIL) {
6887 /* Trim the trailing \n that fbm_compile added last
6889 SvCUR_set(sv, SvCUR(sv) - 1);
6890 /* Whilst this makes the SV technically "invalid" (as its
6891 buffer is no longer followed by "\0") when fbm_compile()
6892 adds the "\n" back, a "\0" is restored. */
6894 fbm_compile(sv, flags);
6896 if (prog->substrs->data[i].substr == prog->check_substr)
6897 prog->check_utf8 = sv;
6903 S_to_byte_substr(pTHX_ register regexp *prog)
6908 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6911 if (prog->substrs->data[i].utf8_substr
6912 && !prog->substrs->data[i].substr) {
6913 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6914 if (sv_utf8_downgrade(sv, TRUE)) {
6915 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6917 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6918 if (flags & FBMcf_TAIL) {
6919 /* Trim the trailing \n that fbm_compile added last
6921 SvCUR_set(sv, SvCUR(sv) - 1);
6923 fbm_compile(sv, flags);
6929 prog->substrs->data[i].substr = sv;
6930 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6931 prog->check_substr = sv;
6938 * c-indentation-style: bsd
6940 * indent-tabs-mode: t
6943 * ex: set ts=8 sts=4 sw=4 noet: