5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #define RF_tainted 1 /* tainted information used? e.g. locale */
84 #define RF_warned 2 /* warned about big count? */
86 #define RF_utf8 8 /* Pattern contains multibyte chars? */
88 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
90 #define RS_init 1 /* eval environment created */
91 #define RS_set 2 /* replsv value is set */
97 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
126 /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
127 * skip the check on EBCDIC platforms */
128 # define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
130 # define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
131 if (!CAT2(PL_utf8_,class)) { \
133 ENTER; save_re_context(); \
134 ok=CAT2(is_utf8_,class)((const U8*)str); \
135 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
138 /* Doesn't do an assert to verify that is correct */
139 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
140 if (!CAT2(PL_utf8_,class)) { \
141 bool throw_away __attribute__unused__; \
142 ENTER; save_re_context(); \
143 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
146 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
147 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
148 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
150 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
151 LOAD_UTF8_CHARCLASS(X_begin, " "); \
152 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
153 /* These are utf8 constants, and not utf-ebcdic constants, so the \
154 * assert should likely and hopefully fail on an EBCDIC machine */ \
155 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
157 /* No asserts are done for these, in case called on an early \
158 * Unicode version in which they map to nothing */ \
159 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
160 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
161 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
162 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
163 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
164 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
165 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
167 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
169 /* The actual code for CCC_TRY, which uses several variables from the routine
170 * it's callable from. It is designed to be the bulk of a case statement.
171 * FUNC is the macro or function to call on non-utf8 targets that indicate if
172 * nextchr matches the class.
173 * UTF8_TEST is the whole test string to use for utf8 targets
174 * LOAD is what to use to test, and if not present to load in the swash for the
176 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
178 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
179 * utf8 and a variant, load the swash if necessary and test using the utf8
180 * test. Advance to the next character if test is ok, otherwise fail; If not
181 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
182 * fails, or advance to the next character */
184 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
185 if (locinput >= PL_regeol) { \
188 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
189 LOAD_UTF8_CHARCLASS(CLASS, STR); \
190 if (POS_OR_NEG (UTF8_TEST)) { \
193 locinput += PL_utf8skip[nextchr]; \
194 nextchr = UCHARAT(locinput); \
197 if (POS_OR_NEG (FUNC(nextchr))) { \
200 nextchr = UCHARAT(++locinput); \
203 /* Handle the non-locale cases for a character class and its complement. It
204 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
205 * This is because that code fails when the test succeeds, so we want to have
206 * the test fail so that the code succeeds. The swash is stored in a
207 * predictable PL_ place */
208 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
211 _CCC_TRY_CODE( !, FUNC, \
212 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
213 (U8*)locinput, TRUE)), \
216 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
217 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
218 (U8*)locinput, TRUE)), \
221 /* Generate the case statements for both locale and non-locale character
222 * classes in regmatch for classes that don't have special unicode semantics.
223 * Locales don't use an immediate swash, but an intermediary special locale
224 * function that is called on the pointer to the current place in the input
225 * string. That function will resolve to needing the same swash. One might
226 * think that because we don't know what the locale will match, we shouldn't
227 * check with the swash loading function that it loaded properly; ie, that we
228 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
229 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
231 #define CCC_TRY(NAME, NNAME, FUNC, \
232 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
233 NAMEA, NNAMEA, FUNCA, \
236 PL_reg_flags |= RF_tainted; \
237 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
239 PL_reg_flags |= RF_tainted; \
240 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
243 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
246 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
247 nextchr = UCHARAT(++locinput); \
250 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
254 locinput += PL_utf8skip[nextchr]; \
255 nextchr = UCHARAT(locinput); \
258 nextchr = UCHARAT(++locinput); \
261 /* Generate the non-locale cases */ \
262 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
264 /* This is like CCC_TRY, but has an extra set of parameters for generating case
265 * statements to handle separate Unicode semantics nodes */
266 #define CCC_TRY_U(NAME, NNAME, FUNC, \
267 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
268 NAMEU, NNAMEU, FUNCU, \
269 NAMEA, NNAMEA, FUNCA, \
271 CCC_TRY(NAME, NNAME, FUNC, \
272 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
273 NAMEA, NNAMEA, FUNCA, \
275 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
277 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
279 /* for use after a quantifier and before an EXACT-like node -- japhy */
280 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
282 * NOTE that *nothing* that affects backtracking should be in here, specifically
283 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
284 * node that is in between two EXACT like nodes when ascertaining what the required
285 * "follow" character is. This should probably be moved to regex compile time
286 * although it may be done at run time beause of the REF possibility - more
287 * investigation required. -- demerphq
289 #define JUMPABLE(rn) ( \
291 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
293 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
294 OP(rn) == PLUS || OP(rn) == MINMOD || \
296 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
298 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
300 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
303 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
304 we don't need this definition. */
305 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
306 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
307 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
310 /* ... so we use this as its faster. */
311 #define IS_TEXT(rn) ( OP(rn)==EXACT )
312 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
313 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
314 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
319 Search for mandatory following text node; for lookahead, the text must
320 follow but for lookbehind (rn->flags != 0) we skip to the next step.
322 #define FIND_NEXT_IMPT(rn) STMT_START { \
323 while (JUMPABLE(rn)) { \
324 const OPCODE type = OP(rn); \
325 if (type == SUSPEND || PL_regkind[type] == CURLY) \
326 rn = NEXTOPER(NEXTOPER(rn)); \
327 else if (type == PLUS) \
329 else if (type == IFMATCH) \
330 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
331 else rn += NEXT_OFF(rn); \
336 static void restore_pos(pTHX_ void *arg);
338 #define REGCP_PAREN_ELEMS 4
339 #define REGCP_OTHER_ELEMS 5
340 #define REGCP_FRAME_ELEMS 1
341 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
342 * are needed for the regexp context stack bookkeeping. */
345 S_regcppush(pTHX_ I32 parenfloor)
348 const int retval = PL_savestack_ix;
349 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
350 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
351 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
353 GET_RE_DEBUG_FLAGS_DECL;
355 if (paren_elems_to_push < 0)
356 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
358 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
359 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
360 " out of range (%lu-%ld)",
361 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
363 SSGROW(total_elems + REGCP_FRAME_ELEMS);
365 for (p = PL_regsize; p > parenfloor; p--) {
366 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
367 SSPUSHINT(PL_regoffs[p].end);
368 SSPUSHINT(PL_regoffs[p].start);
369 SSPUSHPTR(PL_reg_start_tmp[p]);
371 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
372 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
373 (UV)p, (IV)PL_regoffs[p].start,
374 (IV)(PL_reg_start_tmp[p] - PL_bostr),
375 (IV)PL_regoffs[p].end
378 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
379 SSPUSHPTR(PL_regoffs);
380 SSPUSHINT(PL_regsize);
381 SSPUSHINT(*PL_reglastparen);
382 SSPUSHINT(*PL_reglastcloseparen);
383 SSPUSHPTR(PL_reginput);
384 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
389 /* These are needed since we do not localize EVAL nodes: */
390 #define REGCP_SET(cp) \
392 PerlIO_printf(Perl_debug_log, \
393 " Setting an EVAL scope, savestack=%"IVdf"\n", \
394 (IV)PL_savestack_ix)); \
397 #define REGCP_UNWIND(cp) \
399 if (cp != PL_savestack_ix) \
400 PerlIO_printf(Perl_debug_log, \
401 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
402 (IV)(cp), (IV)PL_savestack_ix)); \
406 S_regcppop(pTHX_ const regexp *rex)
411 GET_RE_DEBUG_FLAGS_DECL;
413 PERL_ARGS_ASSERT_REGCPPOP;
415 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
417 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
418 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
419 input = (char *) SSPOPPTR;
420 *PL_reglastcloseparen = SSPOPINT;
421 *PL_reglastparen = SSPOPINT;
422 PL_regsize = SSPOPINT;
423 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
425 i -= REGCP_OTHER_ELEMS;
426 /* Now restore the parentheses context. */
427 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
429 U32 paren = (U32)SSPOPINT;
430 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
431 PL_regoffs[paren].start = SSPOPINT;
433 if (paren <= *PL_reglastparen)
434 PL_regoffs[paren].end = tmps;
436 PerlIO_printf(Perl_debug_log,
437 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
438 (UV)paren, (IV)PL_regoffs[paren].start,
439 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
440 (IV)PL_regoffs[paren].end,
441 (paren > *PL_reglastparen ? "(no)" : ""));
445 if (*PL_reglastparen + 1 <= rex->nparens) {
446 PerlIO_printf(Perl_debug_log,
447 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
448 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
452 /* It would seem that the similar code in regtry()
453 * already takes care of this, and in fact it is in
454 * a better location to since this code can #if 0-ed out
455 * but the code in regtry() is needed or otherwise tests
456 * requiring null fields (pat.t#187 and split.t#{13,14}
457 * (as of patchlevel 7877) will fail. Then again,
458 * this code seems to be necessary or otherwise
459 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
460 * --jhi updated by dapm */
461 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
463 PL_regoffs[i].start = -1;
464 PL_regoffs[i].end = -1;
470 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
473 * pregexec and friends
476 #ifndef PERL_IN_XSUB_RE
478 - pregexec - match a regexp against a string
481 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
482 char *strbeg, I32 minend, SV *screamer, U32 nosave)
483 /* strend: pointer to null at end of string */
484 /* strbeg: real beginning of string */
485 /* minend: end of match must be >=minend after stringarg. */
486 /* nosave: For optimizations. */
488 PERL_ARGS_ASSERT_PREGEXEC;
491 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
492 nosave ? 0 : REXEC_COPY_STR);
497 * Need to implement the following flags for reg_anch:
499 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
501 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
502 * INTUIT_AUTORITATIVE_ML
503 * INTUIT_ONCE_NOML - Intuit can match in one location only.
506 * Another flag for this function: SECOND_TIME (so that float substrs
507 * with giant delta may be not rechecked).
510 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
512 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
513 Otherwise, only SvCUR(sv) is used to get strbeg. */
515 /* XXXX We assume that strpos is strbeg unless sv. */
517 /* XXXX Some places assume that there is a fixed substring.
518 An update may be needed if optimizer marks as "INTUITable"
519 RExen without fixed substrings. Similarly, it is assumed that
520 lengths of all the strings are no more than minlen, thus they
521 cannot come from lookahead.
522 (Or minlen should take into account lookahead.)
523 NOTE: Some of this comment is not correct. minlen does now take account
524 of lookahead/behind. Further research is required. -- demerphq
528 /* A failure to find a constant substring means that there is no need to make
529 an expensive call to REx engine, thus we celebrate a failure. Similarly,
530 finding a substring too deep into the string means that less calls to
531 regtry() should be needed.
533 REx compiler's optimizer found 4 possible hints:
534 a) Anchored substring;
536 c) Whether we are anchored (beginning-of-line or \G);
537 d) First node (of those at offset 0) which may distinguish positions;
538 We use a)b)d) and multiline-part of c), and try to find a position in the
539 string which does not contradict any of them.
542 /* Most of decisions we do here should have been done at compile time.
543 The nodes of the REx which we used for the search should have been
544 deleted from the finite automaton. */
547 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
548 char *strend, const U32 flags, re_scream_pos_data *data)
551 struct regexp *const prog = (struct regexp *)SvANY(rx);
552 register I32 start_shift = 0;
553 /* Should be nonnegative! */
554 register I32 end_shift = 0;
559 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
561 register char *other_last = NULL; /* other substr checked before this */
562 char *check_at = NULL; /* check substr found at this pos */
563 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
564 RXi_GET_DECL(prog,progi);
566 const char * const i_strpos = strpos;
568 GET_RE_DEBUG_FLAGS_DECL;
570 PERL_ARGS_ASSERT_RE_INTUIT_START;
572 RX_MATCH_UTF8_set(rx,utf8_target);
575 PL_reg_flags |= RF_utf8;
578 debug_start_match(rx, utf8_target, strpos, strend,
579 sv ? "Guessing start of match in sv for"
580 : "Guessing start of match in string for");
583 /* CHR_DIST() would be more correct here but it makes things slow. */
584 if (prog->minlen > strend - strpos) {
585 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
586 "String too short... [re_intuit_start]\n"));
590 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
593 if (!prog->check_utf8 && prog->check_substr)
594 to_utf8_substr(prog);
595 check = prog->check_utf8;
597 if (!prog->check_substr && prog->check_utf8)
598 to_byte_substr(prog);
599 check = prog->check_substr;
601 if (check == &PL_sv_undef) {
602 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
603 "Non-utf8 string cannot match utf8 check string\n"));
606 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
607 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
608 || ( (prog->extflags & RXf_ANCH_BOL)
609 && !multiline ) ); /* Check after \n? */
612 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
613 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
614 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
616 && (strpos != strbeg)) {
617 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
620 if (prog->check_offset_min == prog->check_offset_max &&
621 !(prog->extflags & RXf_CANY_SEEN)) {
622 /* Substring at constant offset from beg-of-str... */
625 s = HOP3c(strpos, prog->check_offset_min, strend);
628 slen = SvCUR(check); /* >= 1 */
630 if ( strend - s > slen || strend - s < slen - 1
631 || (strend - s == slen && strend[-1] != '\n')) {
632 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
635 /* Now should match s[0..slen-2] */
637 if (slen && (*SvPVX_const(check) != *s
639 && memNE(SvPVX_const(check), s, slen)))) {
641 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
645 else if (*SvPVX_const(check) != *s
646 || ((slen = SvCUR(check)) > 1
647 && memNE(SvPVX_const(check), s, slen)))
650 goto success_at_start;
653 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
655 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
656 end_shift = prog->check_end_shift;
659 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
660 - (SvTAIL(check) != 0);
661 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
663 if (end_shift < eshift)
667 else { /* Can match at random position */
670 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
671 end_shift = prog->check_end_shift;
673 /* end shift should be non negative here */
676 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
678 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
679 (IV)end_shift, RX_PRECOMP(prog));
683 /* Find a possible match in the region s..strend by looking for
684 the "check" substring in the region corrected by start/end_shift. */
687 I32 srch_start_shift = start_shift;
688 I32 srch_end_shift = end_shift;
689 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
690 srch_end_shift -= ((strbeg - s) - srch_start_shift);
691 srch_start_shift = strbeg - s;
693 DEBUG_OPTIMISE_MORE_r({
694 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
695 (IV)prog->check_offset_min,
696 (IV)srch_start_shift,
698 (IV)prog->check_end_shift);
701 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
702 I32 p = -1; /* Internal iterator of scream. */
703 I32 * const pp = data ? data->scream_pos : &p;
707 assert(SvMAGICAL(sv));
708 mg = mg_find(sv, PERL_MAGIC_study);
711 if (mg->mg_private == 1) {
712 found = ((U8 *)mg->mg_ptr)[BmRARE(check)] != (U8)~0;
713 } else if (mg->mg_private == 2) {
714 found = ((U16 *)mg->mg_ptr)[BmRARE(check)] != (U16)~0;
716 assert (mg->mg_private == 4);
717 found = ((U32 *)mg->mg_ptr)[BmRARE(check)] != (U32)~0;
721 || ( BmRARE(check) == '\n'
722 && (BmPREVIOUS(check) == SvCUR(check) - 1)
724 s = screaminstr(sv, check,
725 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
728 /* we may be pointing at the wrong string */
729 if (s && RXp_MATCH_COPIED(prog))
730 s = strbeg + (s - SvPVX_const(sv));
732 *data->scream_olds = s;
737 if (prog->extflags & RXf_CANY_SEEN) {
738 start_point= (U8*)(s + srch_start_shift);
739 end_point= (U8*)(strend - srch_end_shift);
741 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
742 end_point= HOP3(strend, -srch_end_shift, strbeg);
744 DEBUG_OPTIMISE_MORE_r({
745 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
746 (int)(end_point - start_point),
747 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
751 s = fbm_instr( start_point, end_point,
752 check, multiline ? FBMrf_MULTILINE : 0);
755 /* Update the count-of-usability, remove useless subpatterns,
759 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
760 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
761 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
762 (s ? "Found" : "Did not find"),
763 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
764 ? "anchored" : "floating"),
767 (s ? " at offset " : "...\n") );
772 /* Finish the diagnostic message */
773 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
775 /* XXX dmq: first branch is for positive lookbehind...
776 Our check string is offset from the beginning of the pattern.
777 So we need to do any stclass tests offset forward from that
786 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
787 Start with the other substr.
788 XXXX no SCREAM optimization yet - and a very coarse implementation
789 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
790 *always* match. Probably should be marked during compile...
791 Probably it is right to do no SCREAM here...
794 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
795 : (prog->float_substr && prog->anchored_substr))
797 /* Take into account the "other" substring. */
798 /* XXXX May be hopelessly wrong for UTF... */
801 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
804 char * const last = HOP3c(s, -start_shift, strbeg);
806 char * const saved_s = s;
809 t = s - prog->check_offset_max;
810 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
812 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
817 t = HOP3c(t, prog->anchored_offset, strend);
818 if (t < other_last) /* These positions already checked */
820 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
823 /* XXXX It is not documented what units *_offsets are in.
824 We assume bytes, but this is clearly wrong.
825 Meaning this code needs to be carefully reviewed for errors.
829 /* On end-of-str: see comment below. */
830 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
831 if (must == &PL_sv_undef) {
833 DEBUG_r(must = prog->anchored_utf8); /* for debug */
838 HOP3(HOP3(last1, prog->anchored_offset, strend)
839 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
841 multiline ? FBMrf_MULTILINE : 0
844 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
845 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
846 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
847 (s ? "Found" : "Contradicts"),
848 quoted, RE_SV_TAIL(must));
853 if (last1 >= last2) {
854 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
855 ", giving up...\n"));
858 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
859 ", trying floating at offset %ld...\n",
860 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
861 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
862 s = HOP3c(last, 1, strend);
866 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
867 (long)(s - i_strpos)));
868 t = HOP3c(s, -prog->anchored_offset, strbeg);
869 other_last = HOP3c(s, 1, strend);
877 else { /* Take into account the floating substring. */
879 char * const saved_s = s;
882 t = HOP3c(s, -start_shift, strbeg);
884 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
885 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
886 last = HOP3c(t, prog->float_max_offset, strend);
887 s = HOP3c(t, prog->float_min_offset, strend);
890 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
891 must = utf8_target ? prog->float_utf8 : prog->float_substr;
892 /* fbm_instr() takes into account exact value of end-of-str
893 if the check is SvTAIL(ed). Since false positives are OK,
894 and end-of-str is not later than strend we are OK. */
895 if (must == &PL_sv_undef) {
897 DEBUG_r(must = prog->float_utf8); /* for debug message */
900 s = fbm_instr((unsigned char*)s,
901 (unsigned char*)last + SvCUR(must)
903 must, multiline ? FBMrf_MULTILINE : 0);
905 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
906 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
907 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
908 (s ? "Found" : "Contradicts"),
909 quoted, RE_SV_TAIL(must));
913 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
914 ", giving up...\n"));
917 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
918 ", trying anchored starting at offset %ld...\n",
919 (long)(saved_s + 1 - i_strpos)));
921 s = HOP3c(t, 1, strend);
925 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
926 (long)(s - i_strpos)));
927 other_last = s; /* Fix this later. --Hugo */
937 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
939 DEBUG_OPTIMISE_MORE_r(
940 PerlIO_printf(Perl_debug_log,
941 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
942 (IV)prog->check_offset_min,
943 (IV)prog->check_offset_max,
951 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
953 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
956 /* Fixed substring is found far enough so that the match
957 cannot start at strpos. */
959 if (ml_anch && t[-1] != '\n') {
960 /* Eventually fbm_*() should handle this, but often
961 anchored_offset is not 0, so this check will not be wasted. */
962 /* XXXX In the code below we prefer to look for "^" even in
963 presence of anchored substrings. And we search even
964 beyond the found float position. These pessimizations
965 are historical artefacts only. */
967 while (t < strend - prog->minlen) {
969 if (t < check_at - prog->check_offset_min) {
970 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
971 /* Since we moved from the found position,
972 we definitely contradict the found anchored
973 substr. Due to the above check we do not
974 contradict "check" substr.
975 Thus we can arrive here only if check substr
976 is float. Redo checking for "other"=="fixed".
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
980 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
981 goto do_other_anchored;
983 /* We don't contradict the found floating substring. */
984 /* XXXX Why not check for STCLASS? */
986 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
987 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
990 /* Position contradicts check-string */
991 /* XXXX probably better to look for check-string
992 than for "\n", so one should lower the limit for t? */
993 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
994 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
995 other_last = strpos = s = t + 1;
1000 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1001 PL_colors[0], PL_colors[1]));
1005 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1006 PL_colors[0], PL_colors[1]));
1010 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1013 /* The found string does not prohibit matching at strpos,
1014 - no optimization of calling REx engine can be performed,
1015 unless it was an MBOL and we are not after MBOL,
1016 or a future STCLASS check will fail this. */
1018 /* Even in this situation we may use MBOL flag if strpos is offset
1019 wrt the start of the string. */
1020 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1021 && (strpos != strbeg) && strpos[-1] != '\n'
1022 /* May be due to an implicit anchor of m{.*foo} */
1023 && !(prog->intflags & PREGf_IMPLICIT))
1028 DEBUG_EXECUTE_r( if (ml_anch)
1029 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1030 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1033 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1035 prog->check_utf8 /* Could be deleted already */
1036 && --BmUSEFUL(prog->check_utf8) < 0
1037 && (prog->check_utf8 == prog->float_utf8)
1039 prog->check_substr /* Could be deleted already */
1040 && --BmUSEFUL(prog->check_substr) < 0
1041 && (prog->check_substr == prog->float_substr)
1044 /* If flags & SOMETHING - do not do it many times on the same match */
1045 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1046 /* XXX Does the destruction order has to change with utf8_target? */
1047 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1048 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1049 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1050 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1051 check = NULL; /* abort */
1053 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1054 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1055 if (prog->intflags & PREGf_IMPLICIT)
1056 prog->extflags &= ~RXf_ANCH_MBOL;
1057 /* XXXX This is a remnant of the old implementation. It
1058 looks wasteful, since now INTUIT can use many
1059 other heuristics. */
1060 prog->extflags &= ~RXf_USE_INTUIT;
1061 /* XXXX What other flags might need to be cleared in this branch? */
1067 /* Last resort... */
1068 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1069 /* trie stclasses are too expensive to use here, we are better off to
1070 leave it to regmatch itself */
1071 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1072 /* minlen == 0 is possible if regstclass is \b or \B,
1073 and the fixed substr is ''$.
1074 Since minlen is already taken into account, s+1 is before strend;
1075 accidentally, minlen >= 1 guaranties no false positives at s + 1
1076 even for \b or \B. But (minlen? 1 : 0) below assumes that
1077 regstclass does not come from lookahead... */
1078 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1079 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1080 const U8* const str = (U8*)STRING(progi->regstclass);
1081 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1082 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1085 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1086 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1087 else if (prog->float_substr || prog->float_utf8)
1088 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1092 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1093 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1096 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1099 const char *what = NULL;
1101 if (endpos == strend) {
1102 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1103 "Could not match STCLASS...\n") );
1106 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1107 "This position contradicts STCLASS...\n") );
1108 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1110 /* Contradict one of substrings */
1111 if (prog->anchored_substr || prog->anchored_utf8) {
1112 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1113 DEBUG_EXECUTE_r( what = "anchored" );
1115 s = HOP3c(t, 1, strend);
1116 if (s + start_shift + end_shift > strend) {
1117 /* XXXX Should be taken into account earlier? */
1118 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1119 "Could not match STCLASS...\n") );
1124 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1125 "Looking for %s substr starting at offset %ld...\n",
1126 what, (long)(s + start_shift - i_strpos)) );
1129 /* Have both, check_string is floating */
1130 if (t + start_shift >= check_at) /* Contradicts floating=check */
1131 goto retry_floating_check;
1132 /* Recheck anchored substring, but not floating... */
1136 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1137 "Looking for anchored substr starting at offset %ld...\n",
1138 (long)(other_last - i_strpos)) );
1139 goto do_other_anchored;
1141 /* Another way we could have checked stclass at the
1142 current position only: */
1147 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1148 "Looking for /%s^%s/m starting at offset %ld...\n",
1149 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1152 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1154 /* Check is floating substring. */
1155 retry_floating_check:
1156 t = check_at - start_shift;
1157 DEBUG_EXECUTE_r( what = "floating" );
1158 goto hop_and_restart;
1161 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1162 "By STCLASS: moving %ld --> %ld\n",
1163 (long)(t - i_strpos), (long)(s - i_strpos))
1167 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1168 "Does not contradict STCLASS...\n");
1173 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1174 PL_colors[4], (check ? "Guessed" : "Giving up"),
1175 PL_colors[5], (long)(s - i_strpos)) );
1178 fail_finish: /* Substring not found */
1179 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1180 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1182 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1183 PL_colors[4], PL_colors[5]));
1187 #define DECL_TRIE_TYPE(scan) \
1188 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1189 trie_type = (scan->flags != EXACT) \
1190 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1191 : (utf8_target ? trie_utf8 : trie_plain)
1193 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1194 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1195 switch (trie_type) { \
1196 case trie_utf8_fold: \
1197 if ( foldlen>0 ) { \
1198 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1203 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1204 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1205 foldlen -= UNISKIP( uvc ); \
1206 uscan = foldbuf + UNISKIP( uvc ); \
1209 case trie_latin_utf8_fold: \
1210 if ( foldlen>0 ) { \
1211 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1217 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1218 foldlen -= UNISKIP( uvc ); \
1219 uscan = foldbuf + UNISKIP( uvc ); \
1223 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1230 charid = trie->charmap[ uvc ]; \
1234 if (widecharmap) { \
1235 SV** const svpp = hv_fetch(widecharmap, \
1236 (char*)&uvc, sizeof(UV), 0); \
1238 charid = (U16)SvIV(*svpp); \
1243 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1247 && (ln == 1 || folder(s, pat_string, ln)) \
1248 && (!reginfo || regtry(reginfo, &s)) ) \
1254 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1256 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1262 #define REXEC_FBC_SCAN(CoDe) \
1264 while (s < strend) { \
1270 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1271 REXEC_FBC_UTF8_SCAN( \
1273 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1282 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1285 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1294 #define REXEC_FBC_TRYIT \
1295 if ((!reginfo || regtry(reginfo, &s))) \
1298 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1299 if (utf8_target) { \
1300 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1303 REXEC_FBC_CLASS_SCAN(CoNd); \
1306 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1307 if (utf8_target) { \
1309 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1312 REXEC_FBC_CLASS_SCAN(CoNd); \
1315 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1316 PL_reg_flags |= RF_tainted; \
1317 if (utf8_target) { \
1318 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1321 REXEC_FBC_CLASS_SCAN(CoNd); \
1324 #define DUMP_EXEC_POS(li,s,doutf8) \
1325 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1328 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1329 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1330 tmp = TEST_NON_UTF8(tmp); \
1331 REXEC_FBC_UTF8_SCAN( \
1332 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1341 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1342 if (s == PL_bostr) { \
1346 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1347 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1350 LOAD_UTF8_CHARCLASS_ALNUM(); \
1351 REXEC_FBC_UTF8_SCAN( \
1352 if (tmp == ! (TeSt2_UtF8)) { \
1361 /* The only difference between the BOUND and NBOUND cases is that
1362 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1363 * NBOUND. This is accomplished by passing it in either the if or else clause,
1364 * with the other one being empty */
1365 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1366 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1368 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1369 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1371 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1372 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1374 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1375 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1378 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1379 * be passed in completely with the variable name being tested, which isn't
1380 * such a clean interface, but this is easier to read than it was before. We
1381 * are looking for the boundary (or non-boundary between a word and non-word
1382 * character. The utf8 and non-utf8 cases have the same logic, but the details
1383 * must be different. Find the "wordness" of the character just prior to this
1384 * one, and compare it with the wordness of this one. If they differ, we have
1385 * a boundary. At the beginning of the string, pretend that the previous
1386 * character was a new-line */
1387 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1388 if (utf8_target) { \
1391 else { /* Not utf8 */ \
1392 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1393 tmp = TEST_NON_UTF8(tmp); \
1395 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1404 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1407 /* We know what class REx starts with. Try to find this position... */
1408 /* if reginfo is NULL, its a dryrun */
1409 /* annoyingly all the vars in this routine have different names from their counterparts
1410 in regmatch. /grrr */
1413 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1414 const char *strend, regmatch_info *reginfo)
1417 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1418 char *pat_string; /* The pattern's exactish string */
1419 char *pat_end; /* ptr to end char of pat_string */
1420 re_fold_t folder; /* Function for computing non-utf8 folds */
1421 const U8 *fold_array; /* array for folding ords < 256 */
1424 register STRLEN uskip;
1428 register I32 tmp = 1; /* Scratch variable? */
1429 register const bool utf8_target = PL_reg_match_utf8;
1430 UV utf8_fold_flags = 0;
1431 RXi_GET_DECL(prog,progi);
1433 PERL_ARGS_ASSERT_FIND_BYCLASS;
1435 /* We know what class it must start with. */
1439 if (utf8_target || OP(c) == ANYOFV) {
1440 STRLEN inclasslen = strend - s;
1441 REXEC_FBC_UTF8_CLASS_SCAN(
1442 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1445 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1450 if (tmp && (!reginfo || regtry(reginfo, &s)))
1458 if (UTF_PATTERN || utf8_target) {
1459 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1460 goto do_exactf_utf8;
1462 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1463 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1464 goto do_exactf_non_utf8; /* isn't dealt with by these */
1467 if (UTF_PATTERN || utf8_target) {
1468 utf8_fold_flags = 0;
1469 goto do_exactf_utf8;
1472 /* Any 'ss' in the pattern should have been replaced by regcomp,
1473 * so we don't have to worry here about this single special case
1474 * in the Latin1 range */
1475 fold_array = PL_fold_latin1;
1476 folder = foldEQ_latin1;
1477 goto do_exactf_non_utf8;
1480 if (UTF_PATTERN || utf8_target) {
1481 utf8_fold_flags = 0;
1482 goto do_exactf_utf8;
1484 fold_array = PL_fold;
1486 goto do_exactf_non_utf8;
1489 if (UTF_PATTERN || utf8_target) {
1490 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1491 goto do_exactf_utf8;
1493 fold_array = PL_fold_locale;
1494 folder = foldEQ_locale;
1498 do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */
1500 /* The idea in the non-utf8 EXACTF* cases is to first find the
1501 * first character of the EXACTF* node and then, if necessary,
1502 * case-insensitively compare the full text of the node. c1 is the
1503 * first character. c2 is its fold. This logic will not work for
1504 * Unicode semantics and the german sharp ss, which hence should
1505 * not be compiled into a node that gets here. */
1506 pat_string = STRING(c);
1507 ln = STR_LEN(c); /* length to match in octets/bytes */
1509 e = HOP3c(strend, -((I32)ln), s);
1511 if (!reginfo && e < s) {
1512 e = s; /* Due to minlen logic of intuit() */
1516 c2 = fold_array[c1];
1517 if (c1 == c2) { /* If char and fold are the same */
1518 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1521 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1527 /* If one of the operands is in utf8, we can't use the simpler
1528 * folding above, due to the fact that many different characters
1529 * can have the same fold, or portion of a fold, or different-
1531 pat_string = STRING(c);
1532 ln = STR_LEN(c); /* length to match in octets/bytes */
1533 pat_end = pat_string + ln;
1534 lnc = (UTF_PATTERN) /* length to match in characters */
1535 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1538 e = HOP3c(strend, -((I32)lnc), s);
1540 if (!reginfo && e < s) {
1541 e = s; /* Due to minlen logic of intuit() */
1545 char *my_strend= (char *)strend;
1546 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1547 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1548 && (!reginfo || regtry(reginfo, &s)) )
1556 PL_reg_flags |= RF_tainted;
1557 FBC_BOUND(isALNUM_LC,
1558 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1559 isALNUM_LC_utf8((U8*)s));
1562 PL_reg_flags |= RF_tainted;
1563 FBC_NBOUND(isALNUM_LC,
1564 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1565 isALNUM_LC_utf8((U8*)s));
1568 FBC_BOUND(isWORDCHAR,
1570 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1573 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1575 isWORDCHAR_A((U8*)s));
1578 FBC_NBOUND(isWORDCHAR,
1580 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1583 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1585 isWORDCHAR_A((U8*)s));
1588 FBC_BOUND(isWORDCHAR_L1,
1590 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1593 FBC_NBOUND(isWORDCHAR_L1,
1595 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1598 REXEC_FBC_CSCAN_TAINT(
1599 isALNUM_LC_utf8((U8*)s),
1604 REXEC_FBC_CSCAN_PRELOAD(
1605 LOAD_UTF8_CHARCLASS_ALNUM(),
1606 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1607 isWORDCHAR_L1((U8) *s)
1611 REXEC_FBC_CSCAN_PRELOAD(
1612 LOAD_UTF8_CHARCLASS_ALNUM(),
1613 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1618 /* Don't need to worry about utf8, as it can match only a single
1619 * byte invariant character */
1620 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1623 REXEC_FBC_CSCAN_PRELOAD(
1624 LOAD_UTF8_CHARCLASS_ALNUM(),
1625 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1626 ! isWORDCHAR_L1((U8) *s)
1630 REXEC_FBC_CSCAN_PRELOAD(
1631 LOAD_UTF8_CHARCLASS_ALNUM(),
1632 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1643 REXEC_FBC_CSCAN_TAINT(
1644 !isALNUM_LC_utf8((U8*)s),
1649 REXEC_FBC_CSCAN_PRELOAD(
1650 LOAD_UTF8_CHARCLASS_SPACE(),
1651 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1656 REXEC_FBC_CSCAN_PRELOAD(
1657 LOAD_UTF8_CHARCLASS_SPACE(),
1658 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1663 /* Don't need to worry about utf8, as it can match only a single
1664 * byte invariant character */
1665 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1668 REXEC_FBC_CSCAN_TAINT(
1669 isSPACE_LC_utf8((U8*)s),
1674 REXEC_FBC_CSCAN_PRELOAD(
1675 LOAD_UTF8_CHARCLASS_SPACE(),
1676 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1677 ! isSPACE_L1((U8) *s)
1681 REXEC_FBC_CSCAN_PRELOAD(
1682 LOAD_UTF8_CHARCLASS_SPACE(),
1683 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1694 REXEC_FBC_CSCAN_TAINT(
1695 !isSPACE_LC_utf8((U8*)s),
1700 REXEC_FBC_CSCAN_PRELOAD(
1701 LOAD_UTF8_CHARCLASS_DIGIT(),
1702 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1707 /* Don't need to worry about utf8, as it can match only a single
1708 * byte invariant character */
1709 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1712 REXEC_FBC_CSCAN_TAINT(
1713 isDIGIT_LC_utf8((U8*)s),
1718 REXEC_FBC_CSCAN_PRELOAD(
1719 LOAD_UTF8_CHARCLASS_DIGIT(),
1720 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1731 REXEC_FBC_CSCAN_TAINT(
1732 !isDIGIT_LC_utf8((U8*)s),
1739 is_LNBREAK_latin1(s)
1751 !is_VERTWS_latin1(s)
1757 is_HORIZWS_latin1(s)
1762 !is_HORIZWS_utf8(s),
1763 !is_HORIZWS_latin1(s)
1770 /* what trie are we using right now */
1772 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1774 = (reg_trie_data*)progi->data->data[ aho->trie ];
1775 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1777 const char *last_start = strend - trie->minlen;
1779 const char *real_start = s;
1781 STRLEN maxlen = trie->maxlen;
1783 U8 **points; /* map of where we were in the input string
1784 when reading a given char. For ASCII this
1785 is unnecessary overhead as the relationship
1786 is always 1:1, but for Unicode, especially
1787 case folded Unicode this is not true. */
1788 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1792 GET_RE_DEBUG_FLAGS_DECL;
1794 /* We can't just allocate points here. We need to wrap it in
1795 * an SV so it gets freed properly if there is a croak while
1796 * running the match */
1799 sv_points=newSV(maxlen * sizeof(U8 *));
1800 SvCUR_set(sv_points,
1801 maxlen * sizeof(U8 *));
1802 SvPOK_on(sv_points);
1803 sv_2mortal(sv_points);
1804 points=(U8**)SvPV_nolen(sv_points );
1805 if ( trie_type != trie_utf8_fold
1806 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1809 bitmap=(U8*)trie->bitmap;
1811 bitmap=(U8*)ANYOF_BITMAP(c);
1813 /* this is the Aho-Corasick algorithm modified a touch
1814 to include special handling for long "unknown char"
1815 sequences. The basic idea being that we use AC as long
1816 as we are dealing with a possible matching char, when
1817 we encounter an unknown char (and we have not encountered
1818 an accepting state) we scan forward until we find a legal
1820 AC matching is basically that of trie matching, except
1821 that when we encounter a failing transition, we fall back
1822 to the current states "fail state", and try the current char
1823 again, a process we repeat until we reach the root state,
1824 state 1, or a legal transition. If we fail on the root state
1825 then we can either terminate if we have reached an accepting
1826 state previously, or restart the entire process from the beginning
1830 while (s <= last_start) {
1831 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1839 U8 *uscan = (U8*)NULL;
1840 U8 *leftmost = NULL;
1842 U32 accepted_word= 0;
1846 while ( state && uc <= (U8*)strend ) {
1848 U32 word = aho->states[ state ].wordnum;
1852 DEBUG_TRIE_EXECUTE_r(
1853 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1854 dump_exec_pos( (char *)uc, c, strend, real_start,
1855 (char *)uc, utf8_target );
1856 PerlIO_printf( Perl_debug_log,
1857 " Scanning for legal start char...\n");
1861 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1865 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1871 if (uc >(U8*)last_start) break;
1875 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1876 if (!leftmost || lpos < leftmost) {
1877 DEBUG_r(accepted_word=word);
1883 points[pointpos++ % maxlen]= uc;
1884 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1885 uscan, len, uvc, charid, foldlen,
1887 DEBUG_TRIE_EXECUTE_r({
1888 dump_exec_pos( (char *)uc, c, strend, real_start,
1890 PerlIO_printf(Perl_debug_log,
1891 " Charid:%3u CP:%4"UVxf" ",
1897 word = aho->states[ state ].wordnum;
1899 base = aho->states[ state ].trans.base;
1901 DEBUG_TRIE_EXECUTE_r({
1903 dump_exec_pos( (char *)uc, c, strend, real_start,
1905 PerlIO_printf( Perl_debug_log,
1906 "%sState: %4"UVxf", word=%"UVxf,
1907 failed ? " Fail transition to " : "",
1908 (UV)state, (UV)word);
1914 ( ((offset = base + charid
1915 - 1 - trie->uniquecharcount)) >= 0)
1916 && ((U32)offset < trie->lasttrans)
1917 && trie->trans[offset].check == state
1918 && (tmp=trie->trans[offset].next))
1920 DEBUG_TRIE_EXECUTE_r(
1921 PerlIO_printf( Perl_debug_log," - legal\n"));
1926 DEBUG_TRIE_EXECUTE_r(
1927 PerlIO_printf( Perl_debug_log," - fail\n"));
1929 state = aho->fail[state];
1933 /* we must be accepting here */
1934 DEBUG_TRIE_EXECUTE_r(
1935 PerlIO_printf( Perl_debug_log," - accepting\n"));
1944 if (!state) state = 1;
1947 if ( aho->states[ state ].wordnum ) {
1948 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1949 if (!leftmost || lpos < leftmost) {
1950 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1955 s = (char*)leftmost;
1956 DEBUG_TRIE_EXECUTE_r({
1958 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1959 (UV)accepted_word, (IV)(s - real_start)
1962 if (!reginfo || regtry(reginfo, &s)) {
1968 DEBUG_TRIE_EXECUTE_r({
1969 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1972 DEBUG_TRIE_EXECUTE_r(
1973 PerlIO_printf( Perl_debug_log,"No match.\n"));
1982 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1992 - regexec_flags - match a regexp against a string
1995 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1996 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1997 /* strend: pointer to null at end of string */
1998 /* strbeg: real beginning of string */
1999 /* minend: end of match must be >=minend after stringarg. */
2000 /* data: May be used for some additional optimizations.
2001 Currently its only used, with a U32 cast, for transmitting
2002 the ganch offset when doing a /g match. This will change */
2003 /* nosave: For optimizations. */
2006 struct regexp *const prog = (struct regexp *)SvANY(rx);
2007 /*register*/ char *s;
2008 register regnode *c;
2009 /*register*/ char *startpos = stringarg;
2010 I32 minlen; /* must match at least this many chars */
2011 I32 dontbother = 0; /* how many characters not to try at end */
2012 I32 end_shift = 0; /* Same for the end. */ /* CC */
2013 I32 scream_pos = -1; /* Internal iterator of scream. */
2014 char *scream_olds = NULL;
2015 const bool utf8_target = cBOOL(DO_UTF8(sv));
2017 RXi_GET_DECL(prog,progi);
2018 regmatch_info reginfo; /* create some info to pass to regtry etc */
2019 regexp_paren_pair *swap = NULL;
2020 GET_RE_DEBUG_FLAGS_DECL;
2022 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2023 PERL_UNUSED_ARG(data);
2025 /* Be paranoid... */
2026 if (prog == NULL || startpos == NULL) {
2027 Perl_croak(aTHX_ "NULL regexp parameter");
2031 multiline = prog->extflags & RXf_PMf_MULTILINE;
2032 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2034 RX_MATCH_UTF8_set(rx, utf8_target);
2036 debug_start_match(rx, utf8_target, startpos, strend,
2040 minlen = prog->minlen;
2042 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2043 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2044 "String too short [regexec_flags]...\n"));
2049 /* Check validity of program. */
2050 if (UCHARAT(progi->program) != REG_MAGIC) {
2051 Perl_croak(aTHX_ "corrupted regexp program");
2055 PL_reg_eval_set = 0;
2059 PL_reg_flags |= RF_utf8;
2061 /* Mark beginning of line for ^ and lookbehind. */
2062 reginfo.bol = startpos; /* XXX not used ??? */
2066 /* Mark end of line for $ (and such) */
2069 /* see how far we have to get to not match where we matched before */
2070 reginfo.till = startpos+minend;
2072 /* If there is a "must appear" string, look for it. */
2075 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2077 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2078 reginfo.ganch = startpos + prog->gofs;
2079 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2080 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2081 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2083 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2084 && mg->mg_len >= 0) {
2085 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2086 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2087 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2089 if (prog->extflags & RXf_ANCH_GPOS) {
2090 if (s > reginfo.ganch)
2092 s = reginfo.ganch - prog->gofs;
2093 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2094 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2100 reginfo.ganch = strbeg + PTR2UV(data);
2101 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2102 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2104 } else { /* pos() not defined */
2105 reginfo.ganch = strbeg;
2106 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2107 "GPOS: reginfo.ganch = strbeg\n"));
2110 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2111 /* We have to be careful. If the previous successful match
2112 was from this regex we don't want a subsequent partially
2113 successful match to clobber the old results.
2114 So when we detect this possibility we add a swap buffer
2115 to the re, and switch the buffer each match. If we fail
2116 we switch it back, otherwise we leave it swapped.
2119 /* do we need a save destructor here for eval dies? */
2120 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2122 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2123 re_scream_pos_data d;
2125 d.scream_olds = &scream_olds;
2126 d.scream_pos = &scream_pos;
2127 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2129 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2130 goto phooey; /* not present */
2136 /* Simplest case: anchored match need be tried only once. */
2137 /* [unless only anchor is BOL and multiline is set] */
2138 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2139 if (s == startpos && regtry(®info, &startpos))
2141 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2142 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2147 dontbother = minlen - 1;
2148 end = HOP3c(strend, -dontbother, strbeg) - 1;
2149 /* for multiline we only have to try after newlines */
2150 if (prog->check_substr || prog->check_utf8) {
2151 /* because of the goto we can not easily reuse the macros for bifurcating the
2152 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2155 goto after_try_utf8;
2157 if (regtry(®info, &s)) {
2164 if (prog->extflags & RXf_USE_INTUIT) {
2165 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2174 } /* end search for check string in unicode */
2176 if (s == startpos) {
2177 goto after_try_latin;
2180 if (regtry(®info, &s)) {
2187 if (prog->extflags & RXf_USE_INTUIT) {
2188 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2197 } /* end search for check string in latin*/
2198 } /* end search for check string */
2199 else { /* search for newline */
2201 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2204 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2206 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2207 if (regtry(®info, &s))
2211 } /* end search for newline */
2212 } /* end anchored/multiline check string search */
2214 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2216 /* the warning about reginfo.ganch being used without initialization
2217 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2218 and we only enter this block when the same bit is set. */
2219 char *tmp_s = reginfo.ganch - prog->gofs;
2221 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2226 /* Messy cases: unanchored match. */
2227 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2228 /* we have /x+whatever/ */
2229 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2234 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2235 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2236 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2241 DEBUG_EXECUTE_r( did_match = 1 );
2242 if (regtry(®info, &s)) goto got_it;
2244 while (s < strend && *s == ch)
2252 DEBUG_EXECUTE_r( did_match = 1 );
2253 if (regtry(®info, &s)) goto got_it;
2255 while (s < strend && *s == ch)
2260 DEBUG_EXECUTE_r(if (!did_match)
2261 PerlIO_printf(Perl_debug_log,
2262 "Did not find anchored character...\n")
2265 else if (prog->anchored_substr != NULL
2266 || prog->anchored_utf8 != NULL
2267 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2268 && prog->float_max_offset < strend - s)) {
2273 char *last1; /* Last position checked before */
2277 if (prog->anchored_substr || prog->anchored_utf8) {
2278 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2279 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2280 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2281 back_max = back_min = prog->anchored_offset;
2283 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2284 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2285 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2286 back_max = prog->float_max_offset;
2287 back_min = prog->float_min_offset;
2291 if (must == &PL_sv_undef)
2292 /* could not downgrade utf8 check substring, so must fail */
2298 last = HOP3c(strend, /* Cannot start after this */
2299 -(I32)(CHR_SVLEN(must)
2300 - (SvTAIL(must) != 0) + back_min), strbeg);
2303 last1 = HOPc(s, -1);
2305 last1 = s - 1; /* bogus */
2307 /* XXXX check_substr already used to find "s", can optimize if
2308 check_substr==must. */
2310 dontbother = end_shift;
2311 strend = HOPc(strend, -dontbother);
2312 while ( (s <= last) &&
2313 ((flags & REXEC_SCREAM) && SvSCREAM(sv)
2314 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2315 end_shift, &scream_pos, 0))
2316 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2317 (unsigned char*)strend, must,
2318 multiline ? FBMrf_MULTILINE : 0))) ) {
2319 /* we may be pointing at the wrong string */
2320 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2321 s = strbeg + (s - SvPVX_const(sv));
2322 DEBUG_EXECUTE_r( did_match = 1 );
2323 if (HOPc(s, -back_max) > last1) {
2324 last1 = HOPc(s, -back_min);
2325 s = HOPc(s, -back_max);
2328 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2330 last1 = HOPc(s, -back_min);
2334 while (s <= last1) {
2335 if (regtry(®info, &s))
2341 while (s <= last1) {
2342 if (regtry(®info, &s))
2348 DEBUG_EXECUTE_r(if (!did_match) {
2349 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2350 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2351 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2352 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2353 ? "anchored" : "floating"),
2354 quoted, RE_SV_TAIL(must));
2358 else if ( (c = progi->regstclass) ) {
2360 const OPCODE op = OP(progi->regstclass);
2361 /* don't bother with what can't match */
2362 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2363 strend = HOPc(strend, -(minlen - 1));
2366 SV * const prop = sv_newmortal();
2367 regprop(prog, prop, c);
2369 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2371 PerlIO_printf(Perl_debug_log,
2372 "Matching stclass %.*s against %s (%d bytes)\n",
2373 (int)SvCUR(prop), SvPVX_const(prop),
2374 quoted, (int)(strend - s));
2377 if (find_byclass(prog, c, s, strend, ®info))
2379 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2383 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2388 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2389 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2390 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2392 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
2393 last = screaminstr(sv, float_real, s - strbeg,
2394 end_shift, &scream_pos, 1); /* last one */
2396 last = scream_olds; /* Only one occurrence. */
2397 /* we may be pointing at the wrong string */
2398 else if (RXp_MATCH_COPIED(prog))
2399 s = strbeg + (s - SvPVX_const(sv));
2403 const char * const little = SvPV_const(float_real, len);
2405 if (SvTAIL(float_real)) {
2406 if (memEQ(strend - len + 1, little, len - 1))
2407 last = strend - len + 1;
2408 else if (!multiline)
2409 last = memEQ(strend - len, little, len)
2410 ? strend - len : NULL;
2416 last = rninstr(s, strend, little, little + len);
2418 last = strend; /* matching "$" */
2423 PerlIO_printf(Perl_debug_log,
2424 "%sCan't trim the tail, match fails (should not happen)%s\n",
2425 PL_colors[4], PL_colors[5]));
2426 goto phooey; /* Should not happen! */
2428 dontbother = strend - last + prog->float_min_offset;
2430 if (minlen && (dontbother < minlen))
2431 dontbother = minlen - 1;
2432 strend -= dontbother; /* this one's always in bytes! */
2433 /* We don't know much -- general case. */
2436 if (regtry(®info, &s))
2445 if (regtry(®info, &s))
2447 } while (s++ < strend);
2456 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2458 if (PL_reg_eval_set)
2459 restore_pos(aTHX_ prog);
2460 if (RXp_PAREN_NAMES(prog))
2461 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2463 /* make sure $`, $&, $', and $digit will work later */
2464 if ( !(flags & REXEC_NOT_FIRST) ) {
2465 RX_MATCH_COPY_FREE(rx);
2466 if (flags & REXEC_COPY_STR) {
2467 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2468 #ifdef PERL_OLD_COPY_ON_WRITE
2470 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2472 PerlIO_printf(Perl_debug_log,
2473 "Copy on write: regexp capture, type %d\n",
2476 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2477 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2478 assert (SvPOKp(prog->saved_copy));
2482 RX_MATCH_COPIED_on(rx);
2483 s = savepvn(strbeg, i);
2489 prog->subbeg = strbeg;
2490 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2497 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2498 PL_colors[4], PL_colors[5]));
2499 if (PL_reg_eval_set)
2500 restore_pos(aTHX_ prog);
2502 /* we failed :-( roll it back */
2503 Safefree(prog->offs);
2512 - regtry - try match at specific point
2514 STATIC I32 /* 0 failure, 1 success */
2515 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2519 REGEXP *const rx = reginfo->prog;
2520 regexp *const prog = (struct regexp *)SvANY(rx);
2521 RXi_GET_DECL(prog,progi);
2522 GET_RE_DEBUG_FLAGS_DECL;
2524 PERL_ARGS_ASSERT_REGTRY;
2526 reginfo->cutpoint=NULL;
2528 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2531 PL_reg_eval_set = RS_init;
2532 DEBUG_EXECUTE_r(DEBUG_s(
2533 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2534 (IV)(PL_stack_sp - PL_stack_base));
2537 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2538 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2540 /* Apparently this is not needed, judging by wantarray. */
2541 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2542 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2545 /* Make $_ available to executed code. */
2546 if (reginfo->sv != DEFSV) {
2548 DEFSV_set(reginfo->sv);
2551 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2552 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2553 /* prepare for quick setting of pos */
2554 #ifdef PERL_OLD_COPY_ON_WRITE
2555 if (SvIsCOW(reginfo->sv))
2556 sv_force_normal_flags(reginfo->sv, 0);
2558 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2559 &PL_vtbl_mglob, NULL, 0);
2563 PL_reg_oldpos = mg->mg_len;
2564 SAVEDESTRUCTOR_X(restore_pos, prog);
2566 if (!PL_reg_curpm) {
2567 Newxz(PL_reg_curpm, 1, PMOP);
2570 SV* const repointer = &PL_sv_undef;
2571 /* this regexp is also owned by the new PL_reg_curpm, which
2572 will try to free it. */
2573 av_push(PL_regex_padav, repointer);
2574 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2575 PL_regex_pad = AvARRAY(PL_regex_padav);
2580 /* It seems that non-ithreads works both with and without this code.
2581 So for efficiency reasons it seems best not to have the code
2582 compiled when it is not needed. */
2583 /* This is safe against NULLs: */
2584 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2585 /* PM_reg_curpm owns a reference to this regexp. */
2586 (void)ReREFCNT_inc(rx);
2588 PM_SETRE(PL_reg_curpm, rx);
2589 PL_reg_oldcurpm = PL_curpm;
2590 PL_curpm = PL_reg_curpm;
2591 if (RXp_MATCH_COPIED(prog)) {
2592 /* Here is a serious problem: we cannot rewrite subbeg,
2593 since it may be needed if this match fails. Thus
2594 $` inside (?{}) could fail... */
2595 PL_reg_oldsaved = prog->subbeg;
2596 PL_reg_oldsavedlen = prog->sublen;
2597 #ifdef PERL_OLD_COPY_ON_WRITE
2598 PL_nrs = prog->saved_copy;
2600 RXp_MATCH_COPIED_off(prog);
2603 PL_reg_oldsaved = NULL;
2604 prog->subbeg = PL_bostr;
2605 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2607 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2608 prog->offs[0].start = *startpos - PL_bostr;
2609 PL_reginput = *startpos;
2610 PL_reglastparen = &prog->lastparen;
2611 PL_reglastcloseparen = &prog->lastcloseparen;
2612 prog->lastparen = 0;
2613 prog->lastcloseparen = 0;
2615 PL_regoffs = prog->offs;
2616 if (PL_reg_start_tmpl <= prog->nparens) {
2617 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2618 if(PL_reg_start_tmp)
2619 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2621 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2624 /* XXXX What this code is doing here?!!! There should be no need
2625 to do this again and again, PL_reglastparen should take care of
2628 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2629 * Actually, the code in regcppop() (which Ilya may be meaning by
2630 * PL_reglastparen), is not needed at all by the test suite
2631 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2632 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2633 * Meanwhile, this code *is* needed for the
2634 * above-mentioned test suite tests to succeed. The common theme
2635 * on those tests seems to be returning null fields from matches.
2636 * --jhi updated by dapm */
2638 if (prog->nparens) {
2639 regexp_paren_pair *pp = PL_regoffs;
2641 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2649 if (regmatch(reginfo, progi->program + 1)) {
2650 PL_regoffs[0].end = PL_reginput - PL_bostr;
2653 if (reginfo->cutpoint)
2654 *startpos= reginfo->cutpoint;
2655 REGCP_UNWIND(lastcp);
2660 #define sayYES goto yes
2661 #define sayNO goto no
2662 #define sayNO_SILENT goto no_silent
2664 /* we dont use STMT_START/END here because it leads to
2665 "unreachable code" warnings, which are bogus, but distracting. */
2666 #define CACHEsayNO \
2667 if (ST.cache_mask) \
2668 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2671 /* this is used to determine how far from the left messages like
2672 'failed...' are printed. It should be set such that messages
2673 are inline with the regop output that created them.
2675 #define REPORT_CODE_OFF 32
2678 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2679 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2681 #define SLAB_FIRST(s) (&(s)->states[0])
2682 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2684 /* grab a new slab and return the first slot in it */
2686 STATIC regmatch_state *
2689 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2692 regmatch_slab *s = PL_regmatch_slab->next;
2694 Newx(s, 1, regmatch_slab);
2695 s->prev = PL_regmatch_slab;
2697 PL_regmatch_slab->next = s;
2699 PL_regmatch_slab = s;
2700 return SLAB_FIRST(s);
2704 /* push a new state then goto it */
2706 #define PUSH_STATE_GOTO(state, node) \
2708 st->resume_state = state; \
2711 /* push a new state with success backtracking, then goto it */
2713 #define PUSH_YES_STATE_GOTO(state, node) \
2715 st->resume_state = state; \
2716 goto push_yes_state;
2722 regmatch() - main matching routine
2724 This is basically one big switch statement in a loop. We execute an op,
2725 set 'next' to point the next op, and continue. If we come to a point which
2726 we may need to backtrack to on failure such as (A|B|C), we push a
2727 backtrack state onto the backtrack stack. On failure, we pop the top
2728 state, and re-enter the loop at the state indicated. If there are no more
2729 states to pop, we return failure.
2731 Sometimes we also need to backtrack on success; for example /A+/, where
2732 after successfully matching one A, we need to go back and try to
2733 match another one; similarly for lookahead assertions: if the assertion
2734 completes successfully, we backtrack to the state just before the assertion
2735 and then carry on. In these cases, the pushed state is marked as
2736 'backtrack on success too'. This marking is in fact done by a chain of
2737 pointers, each pointing to the previous 'yes' state. On success, we pop to
2738 the nearest yes state, discarding any intermediate failure-only states.
2739 Sometimes a yes state is pushed just to force some cleanup code to be
2740 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2741 it to free the inner regex.
2743 Note that failure backtracking rewinds the cursor position, while
2744 success backtracking leaves it alone.
2746 A pattern is complete when the END op is executed, while a subpattern
2747 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2748 ops trigger the "pop to last yes state if any, otherwise return true"
2751 A common convention in this function is to use A and B to refer to the two
2752 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2753 the subpattern to be matched possibly multiple times, while B is the entire
2754 rest of the pattern. Variable and state names reflect this convention.
2756 The states in the main switch are the union of ops and failure/success of
2757 substates associated with with that op. For example, IFMATCH is the op
2758 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2759 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2760 successfully matched A and IFMATCH_A_fail is a state saying that we have
2761 just failed to match A. Resume states always come in pairs. The backtrack
2762 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2763 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2764 on success or failure.
2766 The struct that holds a backtracking state is actually a big union, with
2767 one variant for each major type of op. The variable st points to the
2768 top-most backtrack struct. To make the code clearer, within each
2769 block of code we #define ST to alias the relevant union.
2771 Here's a concrete example of a (vastly oversimplified) IFMATCH
2777 #define ST st->u.ifmatch
2779 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2780 ST.foo = ...; // some state we wish to save
2782 // push a yes backtrack state with a resume value of
2783 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2785 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2788 case IFMATCH_A: // we have successfully executed A; now continue with B
2790 bar = ST.foo; // do something with the preserved value
2793 case IFMATCH_A_fail: // A failed, so the assertion failed
2794 ...; // do some housekeeping, then ...
2795 sayNO; // propagate the failure
2802 For any old-timers reading this who are familiar with the old recursive
2803 approach, the code above is equivalent to:
2805 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2814 ...; // do some housekeeping, then ...
2815 sayNO; // propagate the failure
2818 The topmost backtrack state, pointed to by st, is usually free. If you
2819 want to claim it, populate any ST.foo fields in it with values you wish to
2820 save, then do one of
2822 PUSH_STATE_GOTO(resume_state, node);
2823 PUSH_YES_STATE_GOTO(resume_state, node);
2825 which sets that backtrack state's resume value to 'resume_state', pushes a
2826 new free entry to the top of the backtrack stack, then goes to 'node'.
2827 On backtracking, the free slot is popped, and the saved state becomes the
2828 new free state. An ST.foo field in this new top state can be temporarily
2829 accessed to retrieve values, but once the main loop is re-entered, it
2830 becomes available for reuse.
2832 Note that the depth of the backtrack stack constantly increases during the
2833 left-to-right execution of the pattern, rather than going up and down with
2834 the pattern nesting. For example the stack is at its maximum at Z at the
2835 end of the pattern, rather than at X in the following:
2837 /(((X)+)+)+....(Y)+....Z/
2839 The only exceptions to this are lookahead/behind assertions and the cut,
2840 (?>A), which pop all the backtrack states associated with A before
2843 Backtrack state structs are allocated in slabs of about 4K in size.
2844 PL_regmatch_state and st always point to the currently active state,
2845 and PL_regmatch_slab points to the slab currently containing
2846 PL_regmatch_state. The first time regmatch() is called, the first slab is
2847 allocated, and is never freed until interpreter destruction. When the slab
2848 is full, a new one is allocated and chained to the end. At exit from
2849 regmatch(), slabs allocated since entry are freed.
2854 #define DEBUG_STATE_pp(pp) \
2856 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2857 PerlIO_printf(Perl_debug_log, \
2858 " %*s"pp" %s%s%s%s%s\n", \
2860 PL_reg_name[st->resume_state], \
2861 ((st==yes_state||st==mark_state) ? "[" : ""), \
2862 ((st==yes_state) ? "Y" : ""), \
2863 ((st==mark_state) ? "M" : ""), \
2864 ((st==yes_state||st==mark_state) ? "]" : "") \
2869 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2874 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2875 const char *start, const char *end, const char *blurb)
2877 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2879 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2884 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2885 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2887 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2888 start, end - start, 60);
2890 PerlIO_printf(Perl_debug_log,
2891 "%s%s REx%s %s against %s\n",
2892 PL_colors[4], blurb, PL_colors[5], s0, s1);
2894 if (utf8_target||utf8_pat)
2895 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2896 utf8_pat ? "pattern" : "",
2897 utf8_pat && utf8_target ? " and " : "",
2898 utf8_target ? "string" : ""
2904 S_dump_exec_pos(pTHX_ const char *locinput,
2905 const regnode *scan,
2906 const char *loc_regeol,
2907 const char *loc_bostr,
2908 const char *loc_reg_starttry,
2909 const bool utf8_target)
2911 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2912 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2913 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2914 /* The part of the string before starttry has one color
2915 (pref0_len chars), between starttry and current
2916 position another one (pref_len - pref0_len chars),
2917 after the current position the third one.
2918 We assume that pref0_len <= pref_len, otherwise we
2919 decrease pref0_len. */
2920 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2921 ? (5 + taill) - l : locinput - loc_bostr;
2924 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2926 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2928 pref0_len = pref_len - (locinput - loc_reg_starttry);
2929 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2930 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2931 ? (5 + taill) - pref_len : loc_regeol - locinput);
2932 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2936 if (pref0_len > pref_len)
2937 pref0_len = pref_len;
2939 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2941 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2942 (locinput - pref_len),pref0_len, 60, 4, 5);
2944 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2945 (locinput - pref_len + pref0_len),
2946 pref_len - pref0_len, 60, 2, 3);
2948 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2949 locinput, loc_regeol - locinput, 10, 0, 1);
2951 const STRLEN tlen=len0+len1+len2;
2952 PerlIO_printf(Perl_debug_log,
2953 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2954 (IV)(locinput - loc_bostr),
2957 (docolor ? "" : "> <"),
2959 (int)(tlen > 19 ? 0 : 19 - tlen),
2966 /* reg_check_named_buff_matched()
2967 * Checks to see if a named buffer has matched. The data array of
2968 * buffer numbers corresponding to the buffer is expected to reside
2969 * in the regexp->data->data array in the slot stored in the ARG() of
2970 * node involved. Note that this routine doesn't actually care about the
2971 * name, that information is not preserved from compilation to execution.
2972 * Returns the index of the leftmost defined buffer with the given name
2973 * or 0 if non of the buffers matched.
2976 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2979 RXi_GET_DECL(rex,rexi);
2980 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2981 I32 *nums=(I32*)SvPVX(sv_dat);
2983 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2985 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2986 if ((I32)*PL_reglastparen >= nums[n] &&
2987 PL_regoffs[nums[n]].end != -1)
2996 /* free all slabs above current one - called during LEAVE_SCOPE */
2999 S_clear_backtrack_stack(pTHX_ void *p)
3001 regmatch_slab *s = PL_regmatch_slab->next;
3006 PL_regmatch_slab->next = NULL;
3008 regmatch_slab * const osl = s;
3015 #define SETREX(Re1,Re2) \
3016 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3019 STATIC I32 /* 0 failure, 1 success */
3020 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3022 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3026 register const bool utf8_target = PL_reg_match_utf8;
3027 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3028 REGEXP *rex_sv = reginfo->prog;
3029 regexp *rex = (struct regexp *)SvANY(rex_sv);
3030 RXi_GET_DECL(rex,rexi);
3032 /* the current state. This is a cached copy of PL_regmatch_state */
3033 register regmatch_state *st;
3034 /* cache heavy used fields of st in registers */
3035 register regnode *scan;
3036 register regnode *next;
3037 register U32 n = 0; /* general value; init to avoid compiler warning */
3038 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3039 register char *locinput = PL_reginput;
3040 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3042 bool result = 0; /* return value of S_regmatch */
3043 int depth = 0; /* depth of backtrack stack */
3044 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3045 const U32 max_nochange_depth =
3046 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3047 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3048 regmatch_state *yes_state = NULL; /* state to pop to on success of
3050 /* mark_state piggy backs on the yes_state logic so that when we unwind
3051 the stack on success we can update the mark_state as we go */
3052 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3053 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3054 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3056 bool no_final = 0; /* prevent failure from backtracking? */
3057 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3058 char *startpoint = PL_reginput;
3059 SV *popmark = NULL; /* are we looking for a mark? */
3060 SV *sv_commit = NULL; /* last mark name seen in failure */
3061 SV *sv_yes_mark = NULL; /* last mark name we have seen
3062 during a successful match */
3063 U32 lastopen = 0; /* last open we saw */
3064 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3065 SV* const oreplsv = GvSV(PL_replgv);
3066 /* these three flags are set by various ops to signal information to
3067 * the very next op. They have a useful lifetime of exactly one loop
3068 * iteration, and are not preserved or restored by state pushes/pops
3070 bool sw = 0; /* the condition value in (?(cond)a|b) */
3071 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3072 int logical = 0; /* the following EVAL is:
3076 or the following IFMATCH/UNLESSM is:
3077 false: plain (?=foo)
3078 true: used as a condition: (?(?=foo))
3081 GET_RE_DEBUG_FLAGS_DECL;
3084 PERL_ARGS_ASSERT_REGMATCH;
3086 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3087 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3089 /* on first ever call to regmatch, allocate first slab */
3090 if (!PL_regmatch_slab) {
3091 Newx(PL_regmatch_slab, 1, regmatch_slab);
3092 PL_regmatch_slab->prev = NULL;
3093 PL_regmatch_slab->next = NULL;
3094 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3097 oldsave = PL_savestack_ix;
3098 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3099 SAVEVPTR(PL_regmatch_slab);
3100 SAVEVPTR(PL_regmatch_state);
3102 /* grab next free state slot */
3103 st = ++PL_regmatch_state;
3104 if (st > SLAB_LAST(PL_regmatch_slab))
3105 st = PL_regmatch_state = S_push_slab(aTHX);
3107 /* Note that nextchr is a byte even in UTF */
3108 nextchr = UCHARAT(locinput);
3110 while (scan != NULL) {
3113 SV * const prop = sv_newmortal();
3114 regnode *rnext=regnext(scan);
3115 DUMP_EXEC_POS( locinput, scan, utf8_target );
3116 regprop(rex, prop, scan);
3118 PerlIO_printf(Perl_debug_log,
3119 "%3"IVdf":%*s%s(%"IVdf")\n",
3120 (IV)(scan - rexi->program), depth*2, "",
3122 (PL_regkind[OP(scan)] == END || !rnext) ?
3123 0 : (IV)(rnext - rexi->program));
3126 next = scan + NEXT_OFF(scan);
3129 state_num = OP(scan);
3133 assert(PL_reglastparen == &rex->lastparen);
3134 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3135 assert(PL_regoffs == rex->offs);
3137 switch (state_num) {
3139 if (locinput == PL_bostr)
3141 /* reginfo->till = reginfo->bol; */
3146 if (locinput == PL_bostr ||
3147 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3153 if (locinput == PL_bostr)
3157 if (locinput == reginfo->ganch)
3162 /* update the startpoint */
3163 st->u.keeper.val = PL_regoffs[0].start;
3164 PL_reginput = locinput;
3165 PL_regoffs[0].start = locinput - PL_bostr;
3166 PUSH_STATE_GOTO(KEEPS_next, next);
3168 case KEEPS_next_fail:
3169 /* rollback the start point change */
3170 PL_regoffs[0].start = st->u.keeper.val;
3176 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3181 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3183 if (PL_regeol - locinput > 1)
3187 if (PL_regeol != locinput)
3191 if (!nextchr && locinput >= PL_regeol)
3194 locinput += PL_utf8skip[nextchr];
3195 if (locinput > PL_regeol)
3197 nextchr = UCHARAT(locinput);
3200 nextchr = UCHARAT(++locinput);
3203 if (!nextchr && locinput >= PL_regeol)
3205 nextchr = UCHARAT(++locinput);
3208 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3211 locinput += PL_utf8skip[nextchr];
3212 if (locinput > PL_regeol)
3214 nextchr = UCHARAT(locinput);
3217 nextchr = UCHARAT(++locinput);
3221 #define ST st->u.trie
3223 /* In this case the charclass data is available inline so
3224 we can fail fast without a lot of extra overhead.
3226 if (scan->flags == EXACT || !utf8_target) {
3227 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3229 PerlIO_printf(Perl_debug_log,
3230 "%*s %sfailed to match trie start class...%s\n",
3231 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3239 /* the basic plan of execution of the trie is:
3240 * At the beginning, run though all the states, and
3241 * find the longest-matching word. Also remember the position
3242 * of the shortest matching word. For example, this pattern:
3245 * when matched against the string "abcde", will generate
3246 * accept states for all words except 3, with the longest
3247 * matching word being 4, and the shortest being 1 (with
3248 * the position being after char 1 of the string).
3250 * Then for each matching word, in word order (i.e. 1,2,4,5),
3251 * we run the remainder of the pattern; on each try setting
3252 * the current position to the character following the word,
3253 * returning to try the next word on failure.
3255 * We avoid having to build a list of words at runtime by
3256 * using a compile-time structure, wordinfo[].prev, which
3257 * gives, for each word, the previous accepting word (if any).
3258 * In the case above it would contain the mappings 1->2, 2->0,
3259 * 3->0, 4->5, 5->1. We can use this table to generate, from
3260 * the longest word (4 above), a list of all words, by
3261 * following the list of prev pointers; this gives us the
3262 * unordered list 4,5,1,2. Then given the current word we have
3263 * just tried, we can go through the list and find the
3264 * next-biggest word to try (so if we just failed on word 2,
3265 * the next in the list is 4).
3267 * Since at runtime we don't record the matching position in
3268 * the string for each word, we have to work that out for
3269 * each word we're about to process. The wordinfo table holds
3270 * the character length of each word; given that we recorded
3271 * at the start: the position of the shortest word and its
3272 * length in chars, we just need to move the pointer the
3273 * difference between the two char lengths. Depending on
3274 * Unicode status and folding, that's cheap or expensive.
3276 * This algorithm is optimised for the case where are only a
3277 * small number of accept states, i.e. 0,1, or maybe 2.
3278 * With lots of accepts states, and having to try all of them,
3279 * it becomes quadratic on number of accept states to find all
3284 /* what type of TRIE am I? (utf8 makes this contextual) */
3285 DECL_TRIE_TYPE(scan);
3287 /* what trie are we using right now */
3288 reg_trie_data * const trie
3289 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3290 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3291 U32 state = trie->startstate;
3293 if (trie->bitmap && trie_type != trie_utf8_fold &&
3294 !TRIE_BITMAP_TEST(trie,*locinput)
3296 if (trie->states[ state ].wordnum) {
3298 PerlIO_printf(Perl_debug_log,
3299 "%*s %smatched empty string...%s\n",
3300 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3306 PerlIO_printf(Perl_debug_log,
3307 "%*s %sfailed to match trie start class...%s\n",
3308 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3315 U8 *uc = ( U8* )locinput;
3319 U8 *uscan = (U8*)NULL;
3320 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3321 U32 charcount = 0; /* how many input chars we have matched */
3322 U32 accepted = 0; /* have we seen any accepting states? */
3325 ST.jump = trie->jump;
3328 ST.longfold = FALSE; /* char longer if folded => it's harder */
3331 /* fully traverse the TRIE; note the position of the
3332 shortest accept state and the wordnum of the longest
3335 while ( state && uc <= (U8*)PL_regeol ) {
3336 U32 base = trie->states[ state ].trans.base;
3340 wordnum = trie->states[ state ].wordnum;
3342 if (wordnum) { /* it's an accept state */
3345 /* record first match position */
3347 ST.firstpos = (U8*)locinput;
3352 ST.firstchars = charcount;
3355 if (!ST.nextword || wordnum < ST.nextword)
3356 ST.nextword = wordnum;
3357 ST.topword = wordnum;
3360 DEBUG_TRIE_EXECUTE_r({
3361 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3362 PerlIO_printf( Perl_debug_log,
3363 "%*s %sState: %4"UVxf" Accepted: %c ",
3364 2+depth * 2, "", PL_colors[4],
3365 (UV)state, (accepted ? 'Y' : 'N'));
3368 /* read a char and goto next state */
3371 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3372 uscan, len, uvc, charid, foldlen,
3379 base + charid - 1 - trie->uniquecharcount)) >= 0)
3381 && ((U32)offset < trie->lasttrans)
3382 && trie->trans[offset].check == state)
3384 state = trie->trans[offset].next;
3395 DEBUG_TRIE_EXECUTE_r(
3396 PerlIO_printf( Perl_debug_log,
3397 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3398 charid, uvc, (UV)state, PL_colors[5] );
3404 /* calculate total number of accept states */
3409 w = trie->wordinfo[w].prev;
3412 ST.accepted = accepted;
3416 PerlIO_printf( Perl_debug_log,
3417 "%*s %sgot %"IVdf" possible matches%s\n",
3418 REPORT_CODE_OFF + depth * 2, "",
3419 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3421 goto trie_first_try; /* jump into the fail handler */
3425 case TRIE_next_fail: /* we failed - try next alternative */
3427 REGCP_UNWIND(ST.cp);
3428 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3429 PL_regoffs[n].end = -1;
3430 *PL_reglastparen = n;
3432 if (!--ST.accepted) {
3434 PerlIO_printf( Perl_debug_log,
3435 "%*s %sTRIE failed...%s\n",
3436 REPORT_CODE_OFF+depth*2, "",
3443 /* Find next-highest word to process. Note that this code
3444 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3445 register U16 min = 0;
3447 register U16 const nextword = ST.nextword;
3448 register reg_trie_wordinfo * const wordinfo
3449 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3450 for (word=ST.topword; word; word=wordinfo[word].prev) {
3451 if (word > nextword && (!min || word < min))
3464 ST.lastparen = *PL_reglastparen;
3468 /* find start char of end of current word */
3470 U32 chars; /* how many chars to skip */
3471 U8 *uc = ST.firstpos;
3472 reg_trie_data * const trie
3473 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3475 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3477 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3481 /* the hard option - fold each char in turn and find
3482 * its folded length (which may be different */
3483 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3491 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3499 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3504 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3518 PL_reginput = (char *)uc;
3521 scan = (ST.jump && ST.jump[ST.nextword])
3522 ? ST.me + ST.jump[ST.nextword]
3526 PerlIO_printf( Perl_debug_log,
3527 "%*s %sTRIE matched word #%d, continuing%s\n",
3528 REPORT_CODE_OFF+depth*2, "",
3535 if (ST.accepted > 1 || has_cutgroup) {
3536 PUSH_STATE_GOTO(TRIE_next, scan);
3539 /* only one choice left - just continue */
3541 AV *const trie_words
3542 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3543 SV ** const tmp = av_fetch( trie_words,
3545 SV *sv= tmp ? sv_newmortal() : NULL;
3547 PerlIO_printf( Perl_debug_log,
3548 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3549 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3551 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3552 PL_colors[0], PL_colors[1],
3553 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3555 : "not compiled under -Dr",
3559 locinput = PL_reginput;
3560 nextchr = UCHARAT(locinput);
3561 continue; /* execute rest of RE */
3566 char *s = STRING(scan);
3568 if (utf8_target != UTF_PATTERN) {
3569 /* The target and the pattern have differing utf8ness. */
3571 const char * const e = s + ln;
3574 /* The target is utf8, the pattern is not utf8. */
3579 if (NATIVE_TO_UNI(*(U8*)s) !=
3580 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3588 /* The target is not utf8, the pattern is utf8. */
3593 if (NATIVE_TO_UNI(*((U8*)l)) !=
3594 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3602 nextchr = UCHARAT(locinput);
3605 /* The target and the pattern have the same utf8ness. */
3606 /* Inline the first character, for speed. */
3607 if (UCHARAT(s) != nextchr)
3609 if (PL_regeol - locinput < ln)
3611 if (ln > 1 && memNE(s, locinput, ln))
3614 nextchr = UCHARAT(locinput);
3619 const U8 * fold_array;
3621 U32 fold_utf8_flags;
3623 PL_reg_flags |= RF_tainted;
3624 folder = foldEQ_locale;
3625 fold_array = PL_fold_locale;
3626 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3630 folder = foldEQ_latin1;
3631 fold_array = PL_fold_latin1;
3632 fold_utf8_flags = 0;
3636 folder = foldEQ_latin1;
3637 fold_array = PL_fold_latin1;
3638 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3643 fold_array = PL_fold;
3644 fold_utf8_flags = 0;
3650 if (utf8_target || UTF_PATTERN) {
3651 /* Either target or the pattern are utf8. */
3652 const char * const l = locinput;
3653 char *e = PL_regeol;
3655 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3656 l, &e, 0, utf8_target, fold_utf8_flags))
3661 nextchr = UCHARAT(locinput);
3665 /* Neither the target nor the pattern are utf8 */
3666 if (UCHARAT(s) != nextchr &&
3667 UCHARAT(s) != fold_array[nextchr])
3671 if (PL_regeol - locinput < ln)
3673 if (ln > 1 && ! folder(s, locinput, ln))
3676 nextchr = UCHARAT(locinput);
3680 /* XXX Could improve efficiency by separating these all out using a
3681 * macro or in-line function. At that point regcomp.c would no longer
3682 * have to set the FLAGS fields of these */
3685 PL_reg_flags |= RF_tainted;
3693 /* was last char in word? */
3695 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3696 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3698 if (locinput == PL_bostr)
3701 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3703 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3705 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3706 ln = isALNUM_uni(ln);
3707 LOAD_UTF8_CHARCLASS_ALNUM();
3708 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3711 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3712 n = isALNUM_LC_utf8((U8*)locinput);
3717 /* Here the string isn't utf8, or is utf8 and only ascii
3718 * characters are to match \w. In the latter case looking at
3719 * the byte just prior to the current one may be just the final
3720 * byte of a multi-byte character. This is ok. There are two
3722 * 1) it is a single byte character, and then the test is doing
3723 * just what it's supposed to.
3724 * 2) it is a multi-byte character, in which case the final
3725 * byte is never mistakable for ASCII, and so the test
3726 * will say it is not a word character, which is the
3727 * correct answer. */
3728 ln = (locinput != PL_bostr) ?
3729 UCHARAT(locinput - 1) : '\n';
3730 switch (FLAGS(scan)) {
3731 case REGEX_UNICODE_CHARSET:
3732 ln = isWORDCHAR_L1(ln);
3733 n = isWORDCHAR_L1(nextchr);
3735 case REGEX_LOCALE_CHARSET:
3736 ln = isALNUM_LC(ln);
3737 n = isALNUM_LC(nextchr);
3739 case REGEX_DEPENDS_CHARSET:
3741 n = isALNUM(nextchr);
3743 case REGEX_ASCII_RESTRICTED_CHARSET:
3744 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3745 ln = isWORDCHAR_A(ln);
3746 n = isWORDCHAR_A(nextchr);
3749 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3753 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3755 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3760 if (utf8_target || state_num == ANYOFV) {
3761 STRLEN inclasslen = PL_regeol - locinput;
3762 if (locinput >= PL_regeol)
3765 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3767 locinput += inclasslen;
3768 nextchr = UCHARAT(locinput);
3773 nextchr = UCHARAT(locinput);
3774 if (!nextchr && locinput >= PL_regeol)
3776 if (!REGINCLASS(rex, scan, (U8*)locinput))
3778 nextchr = UCHARAT(++locinput);
3782 /* Special char classes - The defines start on line 129 or so */
3783 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3784 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3785 ALNUMU, NALNUMU, isWORDCHAR_L1,
3786 ALNUMA, NALNUMA, isWORDCHAR_A,
3789 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3790 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3791 SPACEU, NSPACEU, isSPACE_L1,
3792 SPACEA, NSPACEA, isSPACE_A,
3795 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3796 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3797 DIGITA, NDIGITA, isDIGIT_A,
3800 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3801 a Unicode extended Grapheme Cluster */
3802 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3803 extended Grapheme Cluster is:
3806 | Prepend* Begin Extend*
3809 Begin is (Hangul-syllable | ! Control)
3810 Extend is (Grapheme_Extend | Spacing_Mark)
3811 Control is [ GCB_Control CR LF ]
3813 The discussion below shows how the code for CLUMP is derived
3814 from this regex. Note that most of these concepts are from
3815 property values of the Grapheme Cluster Boundary (GCB) property.
3816 No code point can have multiple property values for a given
3817 property. Thus a code point in Prepend can't be in Control, but
3818 it must be in !Control. This is why Control above includes
3819 GCB_Control plus CR plus LF. The latter two are used in the GCB
3820 property separately, and so can't be in GCB_Control, even though
3821 they logically are controls. Control is not the same as gc=cc,
3822 but includes format and other characters as well.
3824 The Unicode definition of Hangul-syllable is:
3826 | (L* ( ( V | LV ) V* | LVT ) T*)
3829 Each of these is a value for the GCB property, and hence must be
3830 disjoint, so the order they are tested is immaterial, so the
3831 above can safely be changed to
3834 | (L* ( LVT | ( V | LV ) V*) T*)
3836 The last two terms can be combined like this:
3838 | (( LVT | ( V | LV ) V*) T*))
3840 And refactored into this:
3841 L* (L | LVT T* | V V* T* | LV V* T*)
3843 That means that if we have seen any L's at all we can quit
3844 there, but if the next character is an LVT, a V, or an LV we
3847 There is a subtlety with Prepend* which showed up in testing.
3848 Note that the Begin, and only the Begin is required in:
3849 | Prepend* Begin Extend*
3850 Also, Begin contains '! Control'. A Prepend must be a
3851 '! Control', which means it must also be a Begin. What it
3852 comes down to is that if we match Prepend* and then find no
3853 suitable Begin afterwards, that if we backtrack the last
3854 Prepend, that one will be a suitable Begin.
3857 if (locinput >= PL_regeol)
3859 if (! utf8_target) {
3861 /* Match either CR LF or '.', as all the other possibilities
3863 locinput++; /* Match the . or CR */
3864 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3866 && locinput < PL_regeol
3867 && UCHARAT(locinput) == '\n') locinput++;
3871 /* Utf8: See if is ( CR LF ); already know that locinput <
3872 * PL_regeol, so locinput+1 is in bounds */
3873 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3877 /* In case have to backtrack to beginning, then match '.' */
3878 char *starting = locinput;
3880 /* In case have to backtrack the last prepend */
3881 char *previous_prepend = 0;
3883 LOAD_UTF8_CHARCLASS_GCB();
3885 /* Match (prepend)* */
3886 while (locinput < PL_regeol
3887 && swash_fetch(PL_utf8_X_prepend,
3888 (U8*)locinput, utf8_target))
3890 previous_prepend = locinput;
3891 locinput += UTF8SKIP(locinput);
3894 /* As noted above, if we matched a prepend character, but
3895 * the next thing won't match, back off the last prepend we
3896 * matched, as it is guaranteed to match the begin */
3897 if (previous_prepend
3898 && (locinput >= PL_regeol
3899 || ! swash_fetch(PL_utf8_X_begin,
3900 (U8*)locinput, utf8_target)))
3902 locinput = previous_prepend;
3905 /* Note that here we know PL_regeol > locinput, as we
3906 * tested that upon input to this switch case, and if we
3907 * moved locinput forward, we tested the result just above
3908 * and it either passed, or we backed off so that it will
3910 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3912 /* Here did not match the required 'Begin' in the
3913 * second term. So just match the very first
3914 * character, the '.' of the final term of the regex */
3915 locinput = starting + UTF8SKIP(starting);
3918 /* Here is the beginning of a character that can have
3919 * an extender. It is either a hangul syllable, or a
3921 if (swash_fetch(PL_utf8_X_non_hangul,
3922 (U8*)locinput, utf8_target))
3925 /* Here not a Hangul syllable, must be a
3926 * ('! * Control') */
3927 locinput += UTF8SKIP(locinput);
3930 /* Here is a Hangul syllable. It can be composed
3931 * of several individual characters. One
3932 * possibility is T+ */
3933 if (swash_fetch(PL_utf8_X_T,
3934 (U8*)locinput, utf8_target))
3936 while (locinput < PL_regeol
3937 && swash_fetch(PL_utf8_X_T,
3938 (U8*)locinput, utf8_target))
3940 locinput += UTF8SKIP(locinput);
3944 /* Here, not T+, but is a Hangul. That means
3945 * it is one of the others: L, LV, LVT or V,
3947 * L* (L | LVT T* | V V* T* | LV V* T*) */
3950 while (locinput < PL_regeol
3951 && swash_fetch(PL_utf8_X_L,
3952 (U8*)locinput, utf8_target))
3954 locinput += UTF8SKIP(locinput);
3957 /* Here, have exhausted L*. If the next
3958 * character is not an LV, LVT nor V, it means
3959 * we had to have at least one L, so matches L+
3960 * in the original equation, we have a complete
3961 * hangul syllable. Are done. */
3963 if (locinput < PL_regeol
3964 && swash_fetch(PL_utf8_X_LV_LVT_V,
3965 (U8*)locinput, utf8_target))
3968 /* Otherwise keep going. Must be LV, LVT
3969 * or V. See if LVT */
3970 if (swash_fetch(PL_utf8_X_LVT,
3971 (U8*)locinput, utf8_target))
3973 locinput += UTF8SKIP(locinput);
3976 /* Must be V or LV. Take it, then
3978 locinput += UTF8SKIP(locinput);
3979 while (locinput < PL_regeol
3980 && swash_fetch(PL_utf8_X_V,
3981 (U8*)locinput, utf8_target))
3983 locinput += UTF8SKIP(locinput);
3987 /* And any of LV, LVT, or V can be followed
3989 while (locinput < PL_regeol
3990 && swash_fetch(PL_utf8_X_T,
3994 locinput += UTF8SKIP(locinput);
4000 /* Match any extender */
4001 while (locinput < PL_regeol
4002 && swash_fetch(PL_utf8_X_extend,
4003 (U8*)locinput, utf8_target))
4005 locinput += UTF8SKIP(locinput);
4009 if (locinput > PL_regeol) sayNO;
4011 nextchr = UCHARAT(locinput);
4015 { /* The capture buffer cases. The ones beginning with N for the
4016 named buffers just convert to the equivalent numbered and
4017 pretend they were called as the corresponding numbered buffer
4019 /* don't initialize these in the declaration, it makes C++
4024 const U8 *fold_array;
4027 PL_reg_flags |= RF_tainted;
4028 folder = foldEQ_locale;
4029 fold_array = PL_fold_locale;
4031 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4035 folder = foldEQ_latin1;
4036 fold_array = PL_fold_latin1;
4038 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4042 folder = foldEQ_latin1;
4043 fold_array = PL_fold_latin1;
4045 utf8_fold_flags = 0;
4050 fold_array = PL_fold;
4052 utf8_fold_flags = 0;
4059 utf8_fold_flags = 0;
4062 /* For the named back references, find the corresponding buffer
4064 n = reg_check_named_buff_matched(rex,scan);
4069 goto do_nref_ref_common;
4072 PL_reg_flags |= RF_tainted;
4073 folder = foldEQ_locale;
4074 fold_array = PL_fold_locale;
4075 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4079 folder = foldEQ_latin1;
4080 fold_array = PL_fold_latin1;
4081 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4085 folder = foldEQ_latin1;
4086 fold_array = PL_fold_latin1;
4087 utf8_fold_flags = 0;
4092 fold_array = PL_fold;
4093 utf8_fold_flags = 0;
4099 utf8_fold_flags = 0;
4103 n = ARG(scan); /* which paren pair */
4106 ln = PL_regoffs[n].start;
4107 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4108 if (*PL_reglastparen < n || ln == -1)
4109 sayNO; /* Do not match unless seen CLOSEn. */
4110 if (ln == PL_regoffs[n].end)
4114 if (type != REF /* REF can do byte comparison */
4115 && (utf8_target || type == REFFU))
4116 { /* XXX handle REFFL better */
4117 char * limit = PL_regeol;
4119 /* This call case insensitively compares the entire buffer
4120 * at s, with the current input starting at locinput, but
4121 * not going off the end given by PL_regeol, and returns in
4122 * limit upon success, how much of the current input was
4124 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4125 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4130 nextchr = UCHARAT(locinput);
4134 /* Not utf8: Inline the first character, for speed. */
4135 if (UCHARAT(s) != nextchr &&
4137 UCHARAT(s) != fold_array[nextchr]))
4139 ln = PL_regoffs[n].end - ln;
4140 if (locinput + ln > PL_regeol)
4142 if (ln > 1 && (type == REF
4143 ? memNE(s, locinput, ln)
4144 : ! folder(s, locinput, ln)))
4147 nextchr = UCHARAT(locinput);
4157 #define ST st->u.eval
4162 regexp_internal *rei;
4163 regnode *startpoint;
4166 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4167 if (cur_eval && cur_eval->locinput==locinput) {
4168 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4169 Perl_croak(aTHX_ "Infinite recursion in regex");
4170 if ( ++nochange_depth > max_nochange_depth )
4172 "Pattern subroutine nesting without pos change"
4173 " exceeded limit in regex");
4180 (void)ReREFCNT_inc(rex_sv);
4181 if (OP(scan)==GOSUB) {
4182 startpoint = scan + ARG2L(scan);
4183 ST.close_paren = ARG(scan);
4185 startpoint = rei->program+1;
4188 goto eval_recurse_doit;
4190 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4191 if (cur_eval && cur_eval->locinput==locinput) {
4192 if ( ++nochange_depth > max_nochange_depth )
4193 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4198 /* execute the code in the {...} */
4200 SV ** const before = SP;
4201 OP_4tree * const oop = PL_op;
4202 COP * const ocurcop = PL_curcop;
4204 char *saved_regeol = PL_regeol;
4205 struct re_save_state saved_state;
4207 /* To not corrupt the existing regex state while executing the
4208 * eval we would normally put it on the save stack, like with
4209 * save_re_context. However, re-evals have a weird scoping so we
4210 * can't just add ENTER/LEAVE here. With that, things like
4212 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4214 * would break, as they expect the localisation to be unwound
4215 * only when the re-engine backtracks through the bit that
4218 * What we do instead is just saving the state in a local c
4221 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4224 PL_op = (OP_4tree*)rexi->data->data[n];
4225 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4226 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4227 /* wrap the call in two SAVECOMPPADs. This ensures that
4228 * when the save stack is eventually unwound, all the
4229 * accumulated SAVEt_CLEARSV's will be processed with
4230 * interspersed SAVEt_COMPPAD's to ensure that lexicals
4231 * are cleared in the right pad */
4233 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4234 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4237 SV *sv_mrk = get_sv("REGMARK", 1);
4238 sv_setsv(sv_mrk, sv_yes_mark);
4241 CALLRUNOPS(aTHX); /* Scalar context. */
4244 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4250 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4254 PAD_RESTORE_LOCAL(old_comppad);
4255 PL_curcop = ocurcop;
4256 PL_regeol = saved_regeol;
4259 sv_setsv(save_scalar(PL_replgv), ret);
4263 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4266 /* extract RE object from returned value; compiling if
4272 SV *const sv = SvRV(ret);
4274 if (SvTYPE(sv) == SVt_REGEXP) {
4276 } else if (SvSMAGICAL(sv)) {
4277 mg = mg_find(sv, PERL_MAGIC_qr);
4280 } else if (SvTYPE(ret) == SVt_REGEXP) {
4282 } else if (SvSMAGICAL(ret)) {
4283 if (SvGMAGICAL(ret)) {
4284 /* I don't believe that there is ever qr magic
4286 assert(!mg_find(ret, PERL_MAGIC_qr));
4287 sv_unmagic(ret, PERL_MAGIC_qr);
4290 mg = mg_find(ret, PERL_MAGIC_qr);
4291 /* testing suggests mg only ends up non-NULL for
4292 scalars who were upgraded and compiled in the
4293 else block below. In turn, this is only
4294 triggered in the "postponed utf8 string" tests
4300 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4304 rx = reg_temp_copy(NULL, rx);
4308 const I32 osize = PL_regsize;
4311 assert (SvUTF8(ret));
4312 } else if (SvUTF8(ret)) {
4313 /* Not doing UTF-8, despite what the SV says. Is
4314 this only if we're trapped in use 'bytes'? */
4315 /* Make a copy of the octet sequence, but without
4316 the flag on, as the compiler now honours the
4317 SvUTF8 flag on ret. */
4319 const char *const p = SvPV(ret, len);
4320 ret = newSVpvn_flags(p, len, SVs_TEMP);
4322 rx = CALLREGCOMP(ret, pm_flags);
4324 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4326 /* This isn't a first class regexp. Instead, it's
4327 caching a regexp onto an existing, Perl visible
4329 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4334 re = (struct regexp *)SvANY(rx);
4336 RXp_MATCH_COPIED_off(re);
4337 re->subbeg = rex->subbeg;
4338 re->sublen = rex->sublen;
4341 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4342 "Matching embedded");
4344 startpoint = rei->program + 1;
4345 ST.close_paren = 0; /* only used for GOSUB */
4346 /* borrowed from regtry */
4347 if (PL_reg_start_tmpl <= re->nparens) {
4348 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4349 if(PL_reg_start_tmp)
4350 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4352 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4355 eval_recurse_doit: /* Share code with GOSUB below this line */
4356 /* run the pattern returned from (??{...}) */
4357 ST.cp = regcppush(0); /* Save *all* the positions. */
4358 REGCP_SET(ST.lastcp);
4360 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4362 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4363 PL_reglastparen = &re->lastparen;
4364 PL_reglastcloseparen = &re->lastcloseparen;
4366 re->lastcloseparen = 0;
4368 PL_reginput = locinput;
4371 /* XXXX This is too dramatic a measure... */
4374 ST.toggle_reg_flags = PL_reg_flags;
4376 PL_reg_flags |= RF_utf8;
4378 PL_reg_flags &= ~RF_utf8;
4379 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4381 ST.prev_rex = rex_sv;
4382 ST.prev_curlyx = cur_curlyx;
4383 SETREX(rex_sv,re_sv);
4388 ST.prev_eval = cur_eval;
4390 /* now continue from first node in postoned RE */
4391 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4394 /* logical is 1, /(?(?{...})X|Y)/ */
4395 sw = cBOOL(SvTRUE(ret));
4400 case EVAL_AB: /* cleanup after a successful (??{A})B */
4401 /* note: this is called twice; first after popping B, then A */
4402 PL_reg_flags ^= ST.toggle_reg_flags;
4403 ReREFCNT_dec(rex_sv);
4404 SETREX(rex_sv,ST.prev_rex);
4405 rex = (struct regexp *)SvANY(rex_sv);
4406 rexi = RXi_GET(rex);
4408 cur_eval = ST.prev_eval;
4409 cur_curlyx = ST.prev_curlyx;
4411 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4412 PL_reglastparen = &rex->lastparen;
4413 PL_reglastcloseparen = &rex->lastcloseparen;
4414 /* also update PL_regoffs */
4415 PL_regoffs = rex->offs;
4417 /* XXXX This is too dramatic a measure... */
4419 if ( nochange_depth )
4424 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4425 /* note: this is called twice; first after popping B, then A */
4426 PL_reg_flags ^= ST.toggle_reg_flags;
4427 ReREFCNT_dec(rex_sv);
4428 SETREX(rex_sv,ST.prev_rex);
4429 rex = (struct regexp *)SvANY(rex_sv);
4430 rexi = RXi_GET(rex);
4431 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4432 PL_reglastparen = &rex->lastparen;
4433 PL_reglastcloseparen = &rex->lastcloseparen;
4435 PL_reginput = locinput;
4436 REGCP_UNWIND(ST.lastcp);
4438 cur_eval = ST.prev_eval;
4439 cur_curlyx = ST.prev_curlyx;
4440 /* XXXX This is too dramatic a measure... */
4442 if ( nochange_depth )
4448 n = ARG(scan); /* which paren pair */
4449 PL_reg_start_tmp[n] = locinput;
4455 n = ARG(scan); /* which paren pair */
4456 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4457 PL_regoffs[n].end = locinput - PL_bostr;
4458 /*if (n > PL_regsize)
4460 if (n > *PL_reglastparen)
4461 *PL_reglastparen = n;
4462 *PL_reglastcloseparen = n;
4463 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4471 cursor && OP(cursor)!=END;
4472 cursor=regnext(cursor))
4474 if ( OP(cursor)==CLOSE ){
4476 if ( n <= lastopen ) {
4478 = PL_reg_start_tmp[n] - PL_bostr;
4479 PL_regoffs[n].end = locinput - PL_bostr;
4480 /*if (n > PL_regsize)
4482 if (n > *PL_reglastparen)
4483 *PL_reglastparen = n;
4484 *PL_reglastcloseparen = n;
4485 if ( n == ARG(scan) || (cur_eval &&
4486 cur_eval->u.eval.close_paren == n))
4495 n = ARG(scan); /* which paren pair */
4496 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4499 /* reg_check_named_buff_matched returns 0 for no match */
4500 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4504 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4510 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4512 next = NEXTOPER(NEXTOPER(scan));
4514 next = scan + ARG(scan);
4515 if (OP(next) == IFTHEN) /* Fake one. */
4516 next = NEXTOPER(NEXTOPER(next));
4520 logical = scan->flags;
4523 /*******************************************************************
4525 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4526 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4527 STAR/PLUS/CURLY/CURLYN are used instead.)
4529 A*B is compiled as <CURLYX><A><WHILEM><B>
4531 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4532 state, which contains the current count, initialised to -1. It also sets
4533 cur_curlyx to point to this state, with any previous value saved in the
4536 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4537 since the pattern may possibly match zero times (i.e. it's a while {} loop
4538 rather than a do {} while loop).
4540 Each entry to WHILEM represents a successful match of A. The count in the
4541 CURLYX block is incremented, another WHILEM state is pushed, and execution
4542 passes to A or B depending on greediness and the current count.
4544 For example, if matching against the string a1a2a3b (where the aN are
4545 substrings that match /A/), then the match progresses as follows: (the
4546 pushed states are interspersed with the bits of strings matched so far):
4549 <CURLYX cnt=0><WHILEM>
4550 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4551 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4552 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4553 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4555 (Contrast this with something like CURLYM, which maintains only a single
4559 a1 <CURLYM cnt=1> a2
4560 a1 a2 <CURLYM cnt=2> a3
4561 a1 a2 a3 <CURLYM cnt=3> b
4564 Each WHILEM state block marks a point to backtrack to upon partial failure
4565 of A or B, and also contains some minor state data related to that
4566 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4567 overall state, such as the count, and pointers to the A and B ops.
4569 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4570 must always point to the *current* CURLYX block, the rules are:
4572 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4573 and set cur_curlyx to point the new block.
4575 When popping the CURLYX block after a successful or unsuccessful match,
4576 restore the previous cur_curlyx.
4578 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4579 to the outer one saved in the CURLYX block.
4581 When popping the WHILEM block after a successful or unsuccessful B match,
4582 restore the previous cur_curlyx.
4584 Here's an example for the pattern (AI* BI)*BO
4585 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4588 curlyx backtrack stack
4589 ------ ---------------
4591 CO <CO prev=NULL> <WO>
4592 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4593 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4594 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4596 At this point the pattern succeeds, and we work back down the stack to
4597 clean up, restoring as we go:
4599 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4600 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4601 CO <CO prev=NULL> <WO>
4604 *******************************************************************/
4606 #define ST st->u.curlyx
4608 case CURLYX: /* start of /A*B/ (for complex A) */
4610 /* No need to save/restore up to this paren */
4611 I32 parenfloor = scan->flags;
4613 assert(next); /* keep Coverity happy */
4614 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4617 /* XXXX Probably it is better to teach regpush to support
4618 parenfloor > PL_regsize... */
4619 if (parenfloor > (I32)*PL_reglastparen)
4620 parenfloor = *PL_reglastparen; /* Pessimization... */
4622 ST.prev_curlyx= cur_curlyx;
4624 ST.cp = PL_savestack_ix;
4626 /* these fields contain the state of the current curly.
4627 * they are accessed by subsequent WHILEMs */
4628 ST.parenfloor = parenfloor;
4633 ST.count = -1; /* this will be updated by WHILEM */
4634 ST.lastloc = NULL; /* this will be updated by WHILEM */
4636 PL_reginput = locinput;
4637 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4641 case CURLYX_end: /* just finished matching all of A*B */
4642 cur_curlyx = ST.prev_curlyx;
4646 case CURLYX_end_fail: /* just failed to match all of A*B */
4648 cur_curlyx = ST.prev_curlyx;
4654 #define ST st->u.whilem
4656 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4658 /* see the discussion above about CURLYX/WHILEM */
4660 int min = ARG1(cur_curlyx->u.curlyx.me);
4661 int max = ARG2(cur_curlyx->u.curlyx.me);
4662 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4664 assert(cur_curlyx); /* keep Coverity happy */
4665 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4666 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4667 ST.cache_offset = 0;
4670 PL_reginput = locinput;
4672 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4673 "%*s whilem: matched %ld out of %d..%d\n",
4674 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4677 /* First just match a string of min A's. */
4680 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4681 cur_curlyx->u.curlyx.lastloc = locinput;
4682 REGCP_SET(ST.lastcp);
4684 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4688 /* If degenerate A matches "", assume A done. */
4690 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4691 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4692 "%*s whilem: empty match detected, trying continuation...\n",
4693 REPORT_CODE_OFF+depth*2, "")
4695 goto do_whilem_B_max;
4698 /* super-linear cache processing */
4702 if (!PL_reg_maxiter) {
4703 /* start the countdown: Postpone detection until we
4704 * know the match is not *that* much linear. */
4705 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4706 /* possible overflow for long strings and many CURLYX's */
4707 if (PL_reg_maxiter < 0)
4708 PL_reg_maxiter = I32_MAX;
4709 PL_reg_leftiter = PL_reg_maxiter;
4712 if (PL_reg_leftiter-- == 0) {
4713 /* initialise cache */
4714 const I32 size = (PL_reg_maxiter + 7)/8;
4715 if (PL_reg_poscache) {
4716 if ((I32)PL_reg_poscache_size < size) {
4717 Renew(PL_reg_poscache, size, char);
4718 PL_reg_poscache_size = size;
4720 Zero(PL_reg_poscache, size, char);
4723 PL_reg_poscache_size = size;
4724 Newxz(PL_reg_poscache, size, char);
4726 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4727 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4728 PL_colors[4], PL_colors[5])
4732 if (PL_reg_leftiter < 0) {
4733 /* have we already failed at this position? */
4735 offset = (scan->flags & 0xf) - 1
4736 + (locinput - PL_bostr) * (scan->flags>>4);
4737 mask = 1 << (offset % 8);
4739 if (PL_reg_poscache[offset] & mask) {
4740 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4741 "%*s whilem: (cache) already tried at this position...\n",
4742 REPORT_CODE_OFF+depth*2, "")
4744 sayNO; /* cache records failure */
4746 ST.cache_offset = offset;
4747 ST.cache_mask = mask;
4751 /* Prefer B over A for minimal matching. */
4753 if (cur_curlyx->u.curlyx.minmod) {
4754 ST.save_curlyx = cur_curlyx;
4755 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4756 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4757 REGCP_SET(ST.lastcp);
4758 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4762 /* Prefer A over B for maximal matching. */
4764 if (n < max) { /* More greed allowed? */
4765 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4766 cur_curlyx->u.curlyx.lastloc = locinput;
4767 REGCP_SET(ST.lastcp);
4768 PUSH_STATE_GOTO(WHILEM_A_max, A);
4771 goto do_whilem_B_max;
4775 case WHILEM_B_min: /* just matched B in a minimal match */
4776 case WHILEM_B_max: /* just matched B in a maximal match */
4777 cur_curlyx = ST.save_curlyx;
4781 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4782 cur_curlyx = ST.save_curlyx;
4783 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4784 cur_curlyx->u.curlyx.count--;
4788 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4790 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4791 REGCP_UNWIND(ST.lastcp);
4793 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4794 cur_curlyx->u.curlyx.count--;
4798 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4799 REGCP_UNWIND(ST.lastcp);
4800 regcppop(rex); /* Restore some previous $<digit>s? */
4801 PL_reginput = locinput;
4802 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4803 "%*s whilem: failed, trying continuation...\n",
4804 REPORT_CODE_OFF+depth*2, "")
4807 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4808 && ckWARN(WARN_REGEXP)
4809 && !(PL_reg_flags & RF_warned))
4811 PL_reg_flags |= RF_warned;
4812 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4813 "Complex regular subexpression recursion",
4818 ST.save_curlyx = cur_curlyx;
4819 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4820 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4823 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4824 cur_curlyx = ST.save_curlyx;
4825 REGCP_UNWIND(ST.lastcp);
4828 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4829 /* Maximum greed exceeded */
4830 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4831 && ckWARN(WARN_REGEXP)
4832 && !(PL_reg_flags & RF_warned))
4834 PL_reg_flags |= RF_warned;
4835 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4836 "%s limit (%d) exceeded",
4837 "Complex regular subexpression recursion",
4840 cur_curlyx->u.curlyx.count--;
4844 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4845 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4847 /* Try grabbing another A and see if it helps. */
4848 PL_reginput = locinput;
4849 cur_curlyx->u.curlyx.lastloc = locinput;
4850 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4851 REGCP_SET(ST.lastcp);
4852 PUSH_STATE_GOTO(WHILEM_A_min,
4853 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4857 #define ST st->u.branch
4859 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4860 next = scan + ARG(scan);
4863 scan = NEXTOPER(scan);
4866 case BRANCH: /* /(...|A|...)/ */
4867 scan = NEXTOPER(scan); /* scan now points to inner node */
4868 ST.lastparen = *PL_reglastparen;
4869 ST.next_branch = next;
4871 PL_reginput = locinput;
4873 /* Now go into the branch */
4875 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4877 PUSH_STATE_GOTO(BRANCH_next, scan);
4881 PL_reginput = locinput;
4882 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4883 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4884 PUSH_STATE_GOTO(CUTGROUP_next,next);
4886 case CUTGROUP_next_fail:
4889 if (st->u.mark.mark_name)
4890 sv_commit = st->u.mark.mark_name;
4896 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4901 REGCP_UNWIND(ST.cp);
4902 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4903 PL_regoffs[n].end = -1;
4904 *PL_reglastparen = n;
4905 /*dmq: *PL_reglastcloseparen = n; */
4906 scan = ST.next_branch;
4907 /* no more branches? */
4908 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4910 PerlIO_printf( Perl_debug_log,
4911 "%*s %sBRANCH failed...%s\n",
4912 REPORT_CODE_OFF+depth*2, "",
4918 continue; /* execute next BRANCH[J] op */
4926 #define ST st->u.curlym
4928 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4930 /* This is an optimisation of CURLYX that enables us to push
4931 * only a single backtracking state, no matter how many matches
4932 * there are in {m,n}. It relies on the pattern being constant
4933 * length, with no parens to influence future backrefs
4937 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4939 /* if paren positive, emulate an OPEN/CLOSE around A */
4941 U32 paren = ST.me->flags;
4942 if (paren > PL_regsize)
4944 if (paren > *PL_reglastparen)
4945 *PL_reglastparen = paren;
4946 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4954 ST.c1 = CHRTEST_UNINIT;
4957 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4960 curlym_do_A: /* execute the A in /A{m,n}B/ */
4961 PL_reginput = locinput;
4962 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4965 case CURLYM_A: /* we've just matched an A */
4966 locinput = st->locinput;
4967 nextchr = UCHARAT(locinput);
4970 /* after first match, determine A's length: u.curlym.alen */
4971 if (ST.count == 1) {
4972 if (PL_reg_match_utf8) {
4974 while (s < PL_reginput) {
4980 ST.alen = PL_reginput - locinput;
4983 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4986 PerlIO_printf(Perl_debug_log,
4987 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4988 (int)(REPORT_CODE_OFF+(depth*2)), "",
4989 (IV) ST.count, (IV)ST.alen)
4992 locinput = PL_reginput;
4994 if (cur_eval && cur_eval->u.eval.close_paren &&
4995 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4999 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5000 if ( max == REG_INFTY || ST.count < max )
5001 goto curlym_do_A; /* try to match another A */
5003 goto curlym_do_B; /* try to match B */
5005 case CURLYM_A_fail: /* just failed to match an A */
5006 REGCP_UNWIND(ST.cp);
5008 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5009 || (cur_eval && cur_eval->u.eval.close_paren &&
5010 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5013 curlym_do_B: /* execute the B in /A{m,n}B/ */
5014 PL_reginput = locinput;
5015 if (ST.c1 == CHRTEST_UNINIT) {
5016 /* calculate c1 and c2 for possible match of 1st char
5017 * following curly */
5018 ST.c1 = ST.c2 = CHRTEST_VOID;
5019 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5020 regnode *text_node = ST.B;
5021 if (! HAS_TEXT(text_node))
5022 FIND_NEXT_IMPT(text_node);
5025 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5027 But the former is redundant in light of the latter.
5029 if this changes back then the macro for
5030 IS_TEXT and friends need to change.
5032 if (PL_regkind[OP(text_node)] == EXACT)
5035 ST.c1 = (U8)*STRING(text_node);
5036 switch (OP(text_node)) {
5037 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5039 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5040 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5041 default: ST.c2 = ST.c1;
5048 PerlIO_printf(Perl_debug_log,
5049 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5050 (int)(REPORT_CODE_OFF+(depth*2)),
5053 if (ST.c1 != CHRTEST_VOID
5054 && UCHARAT(PL_reginput) != ST.c1
5055 && UCHARAT(PL_reginput) != ST.c2)
5057 /* simulate B failing */
5059 PerlIO_printf(Perl_debug_log,
5060 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5061 (int)(REPORT_CODE_OFF+(depth*2)),"",
5064 state_num = CURLYM_B_fail;
5065 goto reenter_switch;
5069 /* mark current A as captured */
5070 I32 paren = ST.me->flags;
5072 PL_regoffs[paren].start
5073 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5074 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5075 /*dmq: *PL_reglastcloseparen = paren; */
5078 PL_regoffs[paren].end = -1;
5079 if (cur_eval && cur_eval->u.eval.close_paren &&
5080 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5089 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5092 case CURLYM_B_fail: /* just failed to match a B */
5093 REGCP_UNWIND(ST.cp);
5095 I32 max = ARG2(ST.me);
5096 if (max != REG_INFTY && ST.count == max)
5098 goto curlym_do_A; /* try to match a further A */
5100 /* backtrack one A */
5101 if (ST.count == ARG1(ST.me) /* min */)
5104 locinput = HOPc(locinput, -ST.alen);
5105 goto curlym_do_B; /* try to match B */
5108 #define ST st->u.curly
5110 #define CURLY_SETPAREN(paren, success) \
5113 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5114 PL_regoffs[paren].end = locinput - PL_bostr; \
5115 *PL_reglastcloseparen = paren; \
5118 PL_regoffs[paren].end = -1; \
5121 case STAR: /* /A*B/ where A is width 1 */
5125 scan = NEXTOPER(scan);
5127 case PLUS: /* /A+B/ where A is width 1 */
5131 scan = NEXTOPER(scan);
5133 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5134 ST.paren = scan->flags; /* Which paren to set */
5135 if (ST.paren > PL_regsize)
5136 PL_regsize = ST.paren;
5137 if (ST.paren > *PL_reglastparen)
5138 *PL_reglastparen = ST.paren;
5139 ST.min = ARG1(scan); /* min to match */
5140 ST.max = ARG2(scan); /* max to match */
5141 if (cur_eval && cur_eval->u.eval.close_paren &&
5142 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5146 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5148 case CURLY: /* /A{m,n}B/ where A is width 1 */
5150 ST.min = ARG1(scan); /* min to match */
5151 ST.max = ARG2(scan); /* max to match */
5152 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5155 * Lookahead to avoid useless match attempts
5156 * when we know what character comes next.
5158 * Used to only do .*x and .*?x, but now it allows
5159 * for )'s, ('s and (?{ ... })'s to be in the way
5160 * of the quantifier and the EXACT-like node. -- japhy
5163 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5165 if (HAS_TEXT(next) || JUMPABLE(next)) {
5167 regnode *text_node = next;
5169 if (! HAS_TEXT(text_node))
5170 FIND_NEXT_IMPT(text_node);
5172 if (! HAS_TEXT(text_node))
5173 ST.c1 = ST.c2 = CHRTEST_VOID;
5175 if ( PL_regkind[OP(text_node)] != EXACT ) {
5176 ST.c1 = ST.c2 = CHRTEST_VOID;
5177 goto assume_ok_easy;
5180 s = (U8*)STRING(text_node);
5182 /* Currently we only get here when
5184 PL_rekind[OP(text_node)] == EXACT
5186 if this changes back then the macro for IS_TEXT and
5187 friends need to change. */
5190 switch (OP(text_node)) {
5191 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5193 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5194 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5195 default: ST.c2 = ST.c1; break;
5198 else { /* UTF_PATTERN */
5199 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5200 STRLEN ulen1, ulen2;
5201 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5202 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5204 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5205 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5207 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5209 0 : UTF8_ALLOW_ANY);
5210 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5212 0 : UTF8_ALLOW_ANY);
5214 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5216 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5221 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5228 ST.c1 = ST.c2 = CHRTEST_VOID;
5233 PL_reginput = locinput;
5236 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5239 locinput = PL_reginput;
5241 if (ST.c1 == CHRTEST_VOID)
5242 goto curly_try_B_min;
5244 ST.oldloc = locinput;
5246 /* set ST.maxpos to the furthest point along the
5247 * string that could possibly match */
5248 if (ST.max == REG_INFTY) {
5249 ST.maxpos = PL_regeol - 1;
5251 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5254 else if (utf8_target) {
5255 int m = ST.max - ST.min;
5256 for (ST.maxpos = locinput;
5257 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5258 ST.maxpos += UTF8SKIP(ST.maxpos);
5261 ST.maxpos = locinput + ST.max - ST.min;
5262 if (ST.maxpos >= PL_regeol)
5263 ST.maxpos = PL_regeol - 1;
5265 goto curly_try_B_min_known;
5269 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5270 locinput = PL_reginput;
5271 if (ST.count < ST.min)
5273 if ((ST.count > ST.min)
5274 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5276 /* A{m,n} must come at the end of the string, there's
5277 * no point in backing off ... */
5279 /* ...except that $ and \Z can match before *and* after
5280 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5281 We may back off by one in this case. */
5282 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5286 goto curly_try_B_max;
5291 case CURLY_B_min_known_fail:
5292 /* failed to find B in a non-greedy match where c1,c2 valid */
5293 if (ST.paren && ST.count)
5294 PL_regoffs[ST.paren].end = -1;
5296 PL_reginput = locinput; /* Could be reset... */
5297 REGCP_UNWIND(ST.cp);
5298 /* Couldn't or didn't -- move forward. */
5299 ST.oldloc = locinput;
5301 locinput += UTF8SKIP(locinput);
5305 curly_try_B_min_known:
5306 /* find the next place where 'B' could work, then call B */
5310 n = (ST.oldloc == locinput) ? 0 : 1;
5311 if (ST.c1 == ST.c2) {
5313 /* set n to utf8_distance(oldloc, locinput) */
5314 while (locinput <= ST.maxpos &&
5315 utf8n_to_uvchr((U8*)locinput,
5316 UTF8_MAXBYTES, &len,
5317 uniflags) != (UV)ST.c1) {
5323 /* set n to utf8_distance(oldloc, locinput) */
5324 while (locinput <= ST.maxpos) {
5326 const UV c = utf8n_to_uvchr((U8*)locinput,
5327 UTF8_MAXBYTES, &len,
5329 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5337 if (ST.c1 == ST.c2) {
5338 while (locinput <= ST.maxpos &&
5339 UCHARAT(locinput) != ST.c1)
5343 while (locinput <= ST.maxpos
5344 && UCHARAT(locinput) != ST.c1
5345 && UCHARAT(locinput) != ST.c2)
5348 n = locinput - ST.oldloc;
5350 if (locinput > ST.maxpos)
5352 /* PL_reginput == oldloc now */
5355 if (regrepeat(rex, ST.A, n, depth) < n)
5358 PL_reginput = locinput;
5359 CURLY_SETPAREN(ST.paren, ST.count);
5360 if (cur_eval && cur_eval->u.eval.close_paren &&
5361 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5364 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5369 case CURLY_B_min_fail:
5370 /* failed to find B in a non-greedy match where c1,c2 invalid */
5371 if (ST.paren && ST.count)
5372 PL_regoffs[ST.paren].end = -1;
5374 REGCP_UNWIND(ST.cp);
5375 /* failed -- move forward one */
5376 PL_reginput = locinput;
5377 if (regrepeat(rex, ST.A, 1, depth)) {
5379 locinput = PL_reginput;
5380 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5381 ST.count > 0)) /* count overflow ? */
5384 CURLY_SETPAREN(ST.paren, ST.count);
5385 if (cur_eval && cur_eval->u.eval.close_paren &&
5386 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5389 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5397 /* a successful greedy match: now try to match B */
5398 if (cur_eval && cur_eval->u.eval.close_paren &&
5399 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5404 if (ST.c1 != CHRTEST_VOID)
5405 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5406 UTF8_MAXBYTES, 0, uniflags)
5407 : (UV) UCHARAT(PL_reginput);
5408 /* If it could work, try it. */
5409 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5410 CURLY_SETPAREN(ST.paren, ST.count);
5411 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5416 case CURLY_B_max_fail:
5417 /* failed to find B in a greedy match */
5418 if (ST.paren && ST.count)
5419 PL_regoffs[ST.paren].end = -1;
5421 REGCP_UNWIND(ST.cp);
5423 if (--ST.count < ST.min)
5425 PL_reginput = locinput = HOPc(locinput, -1);
5426 goto curly_try_B_max;
5433 /* we've just finished A in /(??{A})B/; now continue with B */
5435 st->u.eval.toggle_reg_flags
5436 = cur_eval->u.eval.toggle_reg_flags;
5437 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5439 st->u.eval.prev_rex = rex_sv; /* inner */
5440 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5441 rex = (struct regexp *)SvANY(rex_sv);
5442 rexi = RXi_GET(rex);
5443 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5444 (void)ReREFCNT_inc(rex_sv);
5445 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5447 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5448 PL_reglastparen = &rex->lastparen;
5449 PL_reglastcloseparen = &rex->lastcloseparen;
5451 REGCP_SET(st->u.eval.lastcp);
5452 PL_reginput = locinput;
5454 /* Restore parens of the outer rex without popping the
5456 tmpix = PL_savestack_ix;
5457 PL_savestack_ix = cur_eval->u.eval.lastcp;
5459 PL_savestack_ix = tmpix;
5461 st->u.eval.prev_eval = cur_eval;
5462 cur_eval = cur_eval->u.eval.prev_eval;
5464 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5465 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5466 if ( nochange_depth )
5469 PUSH_YES_STATE_GOTO(EVAL_AB,
5470 st->u.eval.prev_eval->u.eval.B); /* match B */
5473 if (locinput < reginfo->till) {
5474 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5475 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5477 (long)(locinput - PL_reg_starttry),
5478 (long)(reginfo->till - PL_reg_starttry),
5481 sayNO_SILENT; /* Cannot match: too short. */
5483 PL_reginput = locinput; /* put where regtry can find it */
5484 sayYES; /* Success! */
5486 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5488 PerlIO_printf(Perl_debug_log,
5489 "%*s %ssubpattern success...%s\n",
5490 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5491 PL_reginput = locinput; /* put where regtry can find it */
5492 sayYES; /* Success! */
5495 #define ST st->u.ifmatch
5497 case SUSPEND: /* (?>A) */
5499 PL_reginput = locinput;
5502 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5504 goto ifmatch_trivial_fail_test;
5506 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5508 ifmatch_trivial_fail_test:
5510 char * const s = HOPBACKc(locinput, scan->flags);
5515 sw = 1 - cBOOL(ST.wanted);
5519 next = scan + ARG(scan);
5527 PL_reginput = locinput;
5531 ST.logical = logical;
5532 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5534 /* execute body of (?...A) */
5535 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5538 case IFMATCH_A_fail: /* body of (?...A) failed */
5539 ST.wanted = !ST.wanted;
5542 case IFMATCH_A: /* body of (?...A) succeeded */
5544 sw = cBOOL(ST.wanted);
5546 else if (!ST.wanted)
5549 if (OP(ST.me) == SUSPEND)
5550 locinput = PL_reginput;
5552 locinput = PL_reginput = st->locinput;
5553 nextchr = UCHARAT(locinput);
5555 scan = ST.me + ARG(ST.me);
5558 continue; /* execute B */
5563 next = scan + ARG(scan);
5568 reginfo->cutpoint = PL_regeol;
5571 PL_reginput = locinput;
5573 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5574 PUSH_STATE_GOTO(COMMIT_next,next);
5576 case COMMIT_next_fail:
5583 #define ST st->u.mark
5585 ST.prev_mark = mark_state;
5586 ST.mark_name = sv_commit = sv_yes_mark
5587 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5589 ST.mark_loc = PL_reginput = locinput;
5590 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5592 case MARKPOINT_next:
5593 mark_state = ST.prev_mark;
5596 case MARKPOINT_next_fail:
5597 if (popmark && sv_eq(ST.mark_name,popmark))
5599 if (ST.mark_loc > startpoint)
5600 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5601 popmark = NULL; /* we found our mark */
5602 sv_commit = ST.mark_name;
5605 PerlIO_printf(Perl_debug_log,
5606 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5607 REPORT_CODE_OFF+depth*2, "",
5608 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5611 mark_state = ST.prev_mark;
5612 sv_yes_mark = mark_state ?
5613 mark_state->u.mark.mark_name : NULL;
5617 PL_reginput = locinput;
5619 /* (*SKIP) : if we fail we cut here*/
5620 ST.mark_name = NULL;
5621 ST.mark_loc = locinput;
5622 PUSH_STATE_GOTO(SKIP_next,next);
5624 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5625 otherwise do nothing. Meaning we need to scan
5627 regmatch_state *cur = mark_state;
5628 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5631 if ( sv_eq( cur->u.mark.mark_name,
5634 ST.mark_name = find;
5635 PUSH_STATE_GOTO( SKIP_next, next );
5637 cur = cur->u.mark.prev_mark;
5640 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5642 case SKIP_next_fail:
5644 /* (*CUT:NAME) - Set up to search for the name as we
5645 collapse the stack*/
5646 popmark = ST.mark_name;
5648 /* (*CUT) - No name, we cut here.*/
5649 if (ST.mark_loc > startpoint)
5650 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5651 /* but we set sv_commit to latest mark_name if there
5652 is one so they can test to see how things lead to this
5655 sv_commit=mark_state->u.mark.mark_name;
5663 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5665 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5668 U8 folded[UTF8_MAXBYTES_CASE+1];
5670 const char * const l = locinput;
5671 char *e = PL_regeol;
5672 to_uni_fold(n, folded, &foldlen);
5674 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5675 l, &e, 0, utf8_target)) {
5680 nextchr = UCHARAT(locinput);
5683 if ((n=is_LNBREAK(locinput,utf8_target))) {
5685 nextchr = UCHARAT(locinput);
5690 #define CASE_CLASS(nAmE) \
5692 if (locinput >= PL_regeol) \
5694 if ((n=is_##nAmE(locinput,utf8_target))) { \
5696 nextchr = UCHARAT(locinput); \
5701 if (locinput >= PL_regeol) \
5703 if ((n=is_##nAmE(locinput,utf8_target))) { \
5706 locinput += UTF8SKIP(locinput); \
5707 nextchr = UCHARAT(locinput); \
5712 CASE_CLASS(HORIZWS);
5716 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5717 PTR2UV(scan), OP(scan));
5718 Perl_croak(aTHX_ "regexp memory corruption");
5722 /* switch break jumps here */
5723 scan = next; /* prepare to execute the next op and ... */
5724 continue; /* ... jump back to the top, reusing st */
5728 /* push a state that backtracks on success */
5729 st->u.yes.prev_yes_state = yes_state;
5733 /* push a new regex state, then continue at scan */
5735 regmatch_state *newst;
5738 regmatch_state *cur = st;
5739 regmatch_state *curyes = yes_state;
5741 regmatch_slab *slab = PL_regmatch_slab;
5742 for (;curd > -1;cur--,curd--) {
5743 if (cur < SLAB_FIRST(slab)) {
5745 cur = SLAB_LAST(slab);
5747 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5748 REPORT_CODE_OFF + 2 + depth * 2,"",
5749 curd, PL_reg_name[cur->resume_state],
5750 (curyes == cur) ? "yes" : ""
5753 curyes = cur->u.yes.prev_yes_state;
5756 DEBUG_STATE_pp("push")
5759 st->locinput = locinput;
5761 if (newst > SLAB_LAST(PL_regmatch_slab))
5762 newst = S_push_slab(aTHX);
5763 PL_regmatch_state = newst;
5765 locinput = PL_reginput;
5766 nextchr = UCHARAT(locinput);
5774 * We get here only if there's trouble -- normally "case END" is
5775 * the terminating point.
5777 Perl_croak(aTHX_ "corrupted regexp pointers");
5783 /* we have successfully completed a subexpression, but we must now
5784 * pop to the state marked by yes_state and continue from there */
5785 assert(st != yes_state);
5787 while (st != yes_state) {
5789 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5790 PL_regmatch_slab = PL_regmatch_slab->prev;
5791 st = SLAB_LAST(PL_regmatch_slab);
5795 DEBUG_STATE_pp("pop (no final)");
5797 DEBUG_STATE_pp("pop (yes)");
5803 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5804 || yes_state > SLAB_LAST(PL_regmatch_slab))
5806 /* not in this slab, pop slab */
5807 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5808 PL_regmatch_slab = PL_regmatch_slab->prev;
5809 st = SLAB_LAST(PL_regmatch_slab);
5811 depth -= (st - yes_state);
5814 yes_state = st->u.yes.prev_yes_state;
5815 PL_regmatch_state = st;
5818 locinput= st->locinput;
5819 nextchr = UCHARAT(locinput);
5821 state_num = st->resume_state + no_final;
5822 goto reenter_switch;
5825 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5826 PL_colors[4], PL_colors[5]));
5828 if (PL_reg_eval_set) {
5829 /* each successfully executed (?{...}) block does the equivalent of
5830 * local $^R = do {...}
5831 * When popping the save stack, all these locals would be undone;
5832 * bypass this by setting the outermost saved $^R to the latest
5834 if (oreplsv != GvSV(PL_replgv))
5835 sv_setsv(oreplsv, GvSV(PL_replgv));
5842 PerlIO_printf(Perl_debug_log,
5843 "%*s %sfailed...%s\n",
5844 REPORT_CODE_OFF+depth*2, "",
5845 PL_colors[4], PL_colors[5])
5857 /* there's a previous state to backtrack to */
5859 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5860 PL_regmatch_slab = PL_regmatch_slab->prev;
5861 st = SLAB_LAST(PL_regmatch_slab);
5863 PL_regmatch_state = st;
5864 locinput= st->locinput;
5865 nextchr = UCHARAT(locinput);
5867 DEBUG_STATE_pp("pop");
5869 if (yes_state == st)
5870 yes_state = st->u.yes.prev_yes_state;
5872 state_num = st->resume_state + 1; /* failure = success + 1 */
5873 goto reenter_switch;
5878 if (rex->intflags & PREGf_VERBARG_SEEN) {
5879 SV *sv_err = get_sv("REGERROR", 1);
5880 SV *sv_mrk = get_sv("REGMARK", 1);
5882 sv_commit = &PL_sv_no;
5884 sv_yes_mark = &PL_sv_yes;
5887 sv_commit = &PL_sv_yes;
5888 sv_yes_mark = &PL_sv_no;
5890 sv_setsv(sv_err, sv_commit);
5891 sv_setsv(sv_mrk, sv_yes_mark);
5894 /* clean up; in particular, free all slabs above current one */
5895 LEAVE_SCOPE(oldsave);
5901 - regrepeat - repeatedly match something simple, report how many
5904 * [This routine now assumes that it will only match on things of length 1.
5905 * That was true before, but now we assume scan - reginput is the count,
5906 * rather than incrementing count on every character. [Er, except utf8.]]
5909 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5912 register char *scan;
5914 register char *loceol = PL_regeol;
5915 register I32 hardcount = 0;
5916 register bool utf8_target = PL_reg_match_utf8;
5919 PERL_UNUSED_ARG(depth);
5922 PERL_ARGS_ASSERT_REGREPEAT;
5925 if (max == REG_INFTY)
5927 else if (max < loceol - scan)
5928 loceol = scan + max;
5933 while (scan < loceol && hardcount < max && *scan != '\n') {
5934 scan += UTF8SKIP(scan);
5938 while (scan < loceol && *scan != '\n')
5945 while (scan < loceol && hardcount < max) {
5946 scan += UTF8SKIP(scan);
5957 /* To get here, EXACTish nodes must have *byte* length == 1. That
5958 * means they match only characters in the string that can be expressed
5959 * as a single byte. For non-utf8 strings, that means a simple match.
5960 * For utf8 strings, the character matched must be an invariant, or
5961 * downgradable to a single byte. The pattern's utf8ness is
5962 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5963 * it is, it's an invariant */
5966 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5968 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5969 while (scan < loceol && UCHARAT(scan) == c) {
5975 /* Here, the string is utf8, and the pattern char is different
5976 * in utf8 than not, so can't compare them directly. Outside the
5977 * loop, find find the two utf8 bytes that represent c, and then
5978 * look for those in sequence in the utf8 string */
5979 U8 high = UTF8_TWO_BYTE_HI(c);
5980 U8 low = UTF8_TWO_BYTE_LO(c);
5983 while (hardcount < max
5984 && scan + 1 < loceol
5985 && UCHARAT(scan) == high
5986 && UCHARAT(scan + 1) == low)
5994 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5998 PL_reg_flags |= RF_tainted;
5999 utf8_flags = FOLDEQ_UTF8_LOCALE;
6006 /* The comments for the EXACT case above apply as well to these fold
6011 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6013 if (utf8_target) { /* Use full Unicode fold matching */
6014 char *tmpeol = loceol;
6015 while (hardcount < max
6016 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6017 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6024 /* XXX Note that the above handles properly the German sharp s in
6025 * the pattern matching ss in the string. But it doesn't handle
6026 * properly cases where the string contains say 'LIGATURE ff' and
6027 * the pattern is 'f+'. This would require, say, a new function or
6028 * revised interface to foldEQ_utf8(), in which the maximum number
6029 * of characters to match could be passed and it would return how
6030 * many actually did. This is just one of many cases where
6031 * multi-char folds don't work properly, and so the fix is being
6037 /* Here, the string isn't utf8 and c is a single byte; and either
6038 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6039 * doesn't affect c. Can just do simple comparisons for exact or
6042 case EXACTF: folded = PL_fold[c]; break;
6044 case EXACTFU: folded = PL_fold_latin1[c]; break;
6045 case EXACTFL: folded = PL_fold_locale[c]; break;
6046 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6048 while (scan < loceol &&
6049 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6057 if (utf8_target || OP(p) == ANYOFV) {
6060 inclasslen = loceol - scan;
6061 while (hardcount < max
6062 && ((inclasslen = loceol - scan) > 0)
6063 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6069 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6077 LOAD_UTF8_CHARCLASS_ALNUM();
6078 while (hardcount < max && scan < loceol &&
6079 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6081 scan += UTF8SKIP(scan);
6085 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6093 while (scan < loceol && isALNUM((U8) *scan)) {
6098 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6103 PL_reg_flags |= RF_tainted;
6106 while (hardcount < max && scan < loceol &&
6107 isALNUM_LC_utf8((U8*)scan)) {
6108 scan += UTF8SKIP(scan);
6112 while (scan < loceol && isALNUM_LC(*scan))
6122 LOAD_UTF8_CHARCLASS_ALNUM();
6123 while (hardcount < max && scan < loceol &&
6124 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6126 scan += UTF8SKIP(scan);
6130 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6137 goto utf8_Nwordchar;
6138 while (scan < loceol && ! isALNUM((U8) *scan)) {
6144 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6145 scan += UTF8SKIP(scan);
6149 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6155 PL_reg_flags |= RF_tainted;
6158 while (hardcount < max && scan < loceol &&
6159 !isALNUM_LC_utf8((U8*)scan)) {
6160 scan += UTF8SKIP(scan);
6164 while (scan < loceol && !isALNUM_LC(*scan))
6174 LOAD_UTF8_CHARCLASS_SPACE();
6175 while (hardcount < max && scan < loceol &&
6177 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6179 scan += UTF8SKIP(scan);
6185 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6194 while (scan < loceol && isSPACE((U8) *scan)) {
6199 while (scan < loceol && isSPACE_A((U8) *scan)) {
6204 PL_reg_flags |= RF_tainted;
6207 while (hardcount < max && scan < loceol &&
6208 isSPACE_LC_utf8((U8*)scan)) {
6209 scan += UTF8SKIP(scan);
6213 while (scan < loceol && isSPACE_LC(*scan))
6223 LOAD_UTF8_CHARCLASS_SPACE();
6224 while (hardcount < max && scan < loceol &&
6226 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6228 scan += UTF8SKIP(scan);
6234 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6243 while (scan < loceol && ! isSPACE((U8) *scan)) {
6249 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6250 scan += UTF8SKIP(scan);
6254 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6260 PL_reg_flags |= RF_tainted;
6263 while (hardcount < max && scan < loceol &&
6264 !isSPACE_LC_utf8((U8*)scan)) {
6265 scan += UTF8SKIP(scan);
6269 while (scan < loceol && !isSPACE_LC(*scan))
6276 LOAD_UTF8_CHARCLASS_DIGIT();
6277 while (hardcount < max && scan < loceol &&
6278 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6279 scan += UTF8SKIP(scan);
6283 while (scan < loceol && isDIGIT(*scan))
6288 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6293 PL_reg_flags |= RF_tainted;
6296 while (hardcount < max && scan < loceol &&
6297 isDIGIT_LC_utf8((U8*)scan)) {
6298 scan += UTF8SKIP(scan);
6302 while (scan < loceol && isDIGIT_LC(*scan))
6309 LOAD_UTF8_CHARCLASS_DIGIT();
6310 while (hardcount < max && scan < loceol &&
6311 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6312 scan += UTF8SKIP(scan);
6316 while (scan < loceol && !isDIGIT(*scan))
6322 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6323 scan += UTF8SKIP(scan);
6327 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6333 PL_reg_flags |= RF_tainted;
6336 while (hardcount < max && scan < loceol &&
6337 !isDIGIT_LC_utf8((U8*)scan)) {
6338 scan += UTF8SKIP(scan);
6342 while (scan < loceol && !isDIGIT_LC(*scan))
6349 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6355 LNBREAK can match two latin chars, which is ok,
6356 because we have a null terminated string, but we
6357 have to use hardcount in this situation
6359 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6368 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6373 while (scan < loceol && is_HORIZWS_latin1(scan))
6380 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6381 scan += UTF8SKIP(scan);
6385 while (scan < loceol && !is_HORIZWS_latin1(scan))
6393 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6398 while (scan < loceol && is_VERTWS_latin1(scan))
6406 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6407 scan += UTF8SKIP(scan);
6411 while (scan < loceol && !is_VERTWS_latin1(scan))
6417 default: /* Called on something of 0 width. */
6418 break; /* So match right here or not at all. */
6424 c = scan - PL_reginput;
6428 GET_RE_DEBUG_FLAGS_DECL;
6430 SV * const prop = sv_newmortal();
6431 regprop(prog, prop, p);
6432 PerlIO_printf(Perl_debug_log,
6433 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6434 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6442 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6444 - regclass_swash - prepare the utf8 swash
6448 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6454 RXi_GET_DECL(prog,progi);
6455 const struct reg_data * const data = prog ? progi->data : NULL;
6457 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6459 assert(ANYOF_NONBITMAP(node));
6461 if (data && data->count) {
6462 const U32 n = ARG(node);
6464 if (data->what[n] == 's') {
6465 SV * const rv = MUTABLE_SV(data->data[n]);
6466 AV * const av = MUTABLE_AV(SvRV(rv));
6467 SV **const ary = AvARRAY(av);
6470 /* See the end of regcomp.c:S_regclass() for
6471 * documentation of these array elements. */
6474 a = SvROK(ary[1]) ? &ary[1] : NULL;
6475 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6479 else if (si && doinit) {
6480 sw = swash_init("utf8", "", si, 1, 0);
6481 (void)av_store(av, 1, sw);
6498 - reginclass - determine if a character falls into a character class
6500 n is the ANYOF regnode
6501 p is the target string
6502 lenp is pointer to the maximum number of bytes of how far to go in p
6503 (This is assumed wthout checking to always be at least the current
6505 utf8_target tells whether p is in UTF-8.
6507 Returns true if matched; false otherwise. If lenp is not NULL, on return
6508 from a successful match, the value it points to will be updated to how many
6509 bytes in p were matched. If there was no match, the value is undefined,
6510 possibly changed from the input.
6512 Note that this can be a synthetic start class, a combination of various
6513 nodes, so things you think might be mutually exclusive, such as locale,
6514 aren't. It can match both locale and non-locale
6519 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6522 const char flags = ANYOF_FLAGS(n);
6528 PERL_ARGS_ASSERT_REGINCLASS;
6530 /* If c is not already the code point, get it */
6531 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6532 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6533 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6534 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6535 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6536 * UTF8_ALLOW_FFFF */
6537 if (c_len == (STRLEN)-1)
6538 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6544 /* Use passed in max length, or one character if none passed in or less
6545 * than one character. And assume will match just one character. This is
6546 * overwritten later if matched more. */
6548 maxlen = (*lenp > c_len) ? *lenp : c_len;
6556 /* If this character is potentially in the bitmap, check it */
6558 if (ANYOF_BITMAP_TEST(n, c))
6560 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6567 else if (flags & ANYOF_LOCALE) {
6568 PL_reg_flags |= RF_tainted;
6570 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6571 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6575 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6576 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6577 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6578 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6579 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6580 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6581 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6582 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6583 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6584 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6585 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6586 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6587 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6588 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6589 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6590 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6591 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6592 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6593 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6594 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6595 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6596 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6597 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6598 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6599 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6600 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6601 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6602 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6603 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6604 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6605 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6606 ) /* How's that for a conditional? */
6613 /* If the bitmap didn't (or couldn't) match, and something outside the
6614 * bitmap could match, try that. Locale nodes specifiy completely the
6615 * behavior of code points in the bit map (otherwise, a utf8 target would
6616 * cause them to be treated as Unicode and not locale), except in
6617 * the very unlikely event when this node is a synthetic start class, which
6618 * could be a combination of locale and non-locale nodes. So allow locale
6619 * to match for the synthetic start class, which will give a false
6620 * positive that will be resolved when the match is done again as not part
6621 * of the synthetic start class */
6623 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6624 match = TRUE; /* Everything above 255 matches */
6626 else if (ANYOF_NONBITMAP(n)
6627 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6630 || (! (flags & ANYOF_LOCALE))
6631 || (flags & ANYOF_IS_SYNTHETIC)))))
6634 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6642 /* Not utf8. Convert as much of the string as available up
6643 * to the limit of how far the (single) character in the
6644 * pattern can possibly match (no need to go further). If
6645 * the node is a straight ANYOF or not folding, it can't
6646 * match more than one. Otherwise, It can match up to how
6647 * far a single char can fold to. Since not utf8, each
6648 * character is a single byte, so the max it can be in
6649 * bytes is the same as the max it can be in characters */
6650 STRLEN len = (OP(n) == ANYOF
6651 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6653 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6655 : UTF8_MAX_FOLD_CHAR_EXPAND;
6656 utf8_p = bytes_to_utf8(p, &len);
6659 if (swash_fetch(sw, utf8_p, TRUE))
6661 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6663 /* Here, we need to test if the fold of the target string
6664 * matches. The non-multi char folds have all been moved to
6665 * the compilation phase, and the multi-char folds have
6666 * been stored by regcomp into 'av'; we linearly check to
6667 * see if any match the target string (folded). We know
6668 * that the originals were each one character, but we don't
6669 * currently know how many characters/bytes each folded to,
6670 * except we do know that there are small limits imposed by
6671 * Unicode. XXX A performance enhancement would be to have
6672 * regcomp.c store the max number of chars/bytes that are
6673 * in an av entry, as, say the 0th element. Even better
6674 * would be to have a hash of the few characters that can
6675 * start a multi-char fold to the max number of chars of
6678 * If there is a match, we will need to advance (if lenp is
6679 * specified) the match pointer in the target string. But
6680 * what we are comparing here isn't that string directly,
6681 * but its fold, whose length may differ from the original.
6682 * As we go along in constructing the fold, therefore, we
6683 * create a map so that we know how many bytes in the
6684 * source to advance given that we have matched a certain
6685 * number of bytes in the fold. This map is stored in
6686 * 'map_fold_len_back'. Let n mean the number of bytes in
6687 * the fold of the first character that we are folding.
6688 * Then map_fold_len_back[n] is set to the number of bytes
6689 * in that first character. Similarly let m be the
6690 * corresponding number for the second character to be
6691 * folded. Then map_fold_len_back[n+m] is set to the
6692 * number of bytes occupied by the first two source
6693 * characters. ... */
6694 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6695 U8 folded[UTF8_MAXBYTES_CASE+1];
6696 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6697 STRLEN total_foldlen = 0; /* num bytes in fold of all
6700 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6702 /* Here, only need to fold the first char of the target
6703 * string. It the source wasn't utf8, is 1 byte long */
6704 to_utf8_fold(utf8_p, folded, &foldlen);
6705 total_foldlen = foldlen;
6706 map_fold_len_back[foldlen] = (utf8_target)
6712 /* Here, need to fold more than the first char. Do so
6713 * up to the limits */
6714 U8* source_ptr = utf8_p; /* The source for the fold
6717 U8* folded_ptr = folded;
6718 U8* e = utf8_p + maxlen; /* Can't go beyond last
6719 available byte in the
6723 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6727 /* Fold the next character */
6728 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6729 STRLEN this_char_foldlen;
6730 to_utf8_fold(source_ptr,
6732 &this_char_foldlen);
6734 /* Bail if it would exceed the byte limit for
6735 * folding a single char. */
6736 if (this_char_foldlen + folded_ptr - folded >
6742 /* Add the fold of this character */
6743 Copy(this_char_folded,
6747 source_ptr += UTF8SKIP(source_ptr);
6748 folded_ptr += this_char_foldlen;
6749 total_foldlen = folded_ptr - folded;
6751 /* Create map from the number of bytes in the fold
6752 * back to the number of bytes in the source. If
6753 * the source isn't utf8, the byte count is just
6754 * the number of characters so far */
6755 map_fold_len_back[total_foldlen]
6757 ? source_ptr - utf8_p
6764 /* Do the linear search to see if the fold is in the list
6765 * of multi-char folds. */
6768 for (i = 0; i <= av_len(av); i++) {
6769 SV* const sv = *av_fetch(av, i, FALSE);
6771 const char * const s = SvPV_const(sv, len);
6773 if (len <= total_foldlen
6774 && memEQ(s, (char*)folded, len)
6776 /* If 0, means matched a partial char. See
6778 && map_fold_len_back[len])
6781 /* Advance the target string ptr to account for
6782 * this fold, but have to translate from the
6783 * folded length to the corresponding source
6786 *lenp = map_fold_len_back[len];
6795 /* If we allocated a string above, free it */
6796 if (! utf8_target) Safefree(utf8_p);
6801 return (flags & ANYOF_INVERT) ? !match : match;
6805 S_reghop3(U8 *s, I32 off, const U8* lim)
6809 PERL_ARGS_ASSERT_REGHOP3;
6812 while (off-- && s < lim) {
6813 /* XXX could check well-formedness here */
6818 while (off++ && s > lim) {
6820 if (UTF8_IS_CONTINUED(*s)) {
6821 while (s > lim && UTF8_IS_CONTINUATION(*s))
6824 /* XXX could check well-formedness here */
6831 /* there are a bunch of places where we use two reghop3's that should
6832 be replaced with this routine. but since thats not done yet
6833 we ifdef it out - dmq
6836 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6840 PERL_ARGS_ASSERT_REGHOP4;
6843 while (off-- && s < rlim) {
6844 /* XXX could check well-formedness here */
6849 while (off++ && s > llim) {
6851 if (UTF8_IS_CONTINUED(*s)) {
6852 while (s > llim && UTF8_IS_CONTINUATION(*s))
6855 /* XXX could check well-formedness here */
6863 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6867 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6870 while (off-- && s < lim) {
6871 /* XXX could check well-formedness here */
6878 while (off++ && s > lim) {
6880 if (UTF8_IS_CONTINUED(*s)) {
6881 while (s > lim && UTF8_IS_CONTINUATION(*s))
6884 /* XXX could check well-formedness here */
6893 restore_pos(pTHX_ void *arg)
6896 regexp * const rex = (regexp *)arg;
6897 if (PL_reg_eval_set) {
6898 if (PL_reg_oldsaved) {
6899 rex->subbeg = PL_reg_oldsaved;
6900 rex->sublen = PL_reg_oldsavedlen;
6901 #ifdef PERL_OLD_COPY_ON_WRITE
6902 rex->saved_copy = PL_nrs;
6904 RXp_MATCH_COPIED_on(rex);
6906 PL_reg_magic->mg_len = PL_reg_oldpos;
6907 PL_reg_eval_set = 0;
6908 PL_curpm = PL_reg_oldcurpm;
6913 S_to_utf8_substr(pTHX_ register regexp *prog)
6917 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6920 if (prog->substrs->data[i].substr
6921 && !prog->substrs->data[i].utf8_substr) {
6922 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6923 prog->substrs->data[i].utf8_substr = sv;
6924 sv_utf8_upgrade(sv);
6925 if (SvVALID(prog->substrs->data[i].substr)) {
6926 if (SvTAIL(prog->substrs->data[i].substr)) {
6927 /* Trim the trailing \n that fbm_compile added last
6929 SvCUR_set(sv, SvCUR(sv) - 1);
6930 /* Whilst this makes the SV technically "invalid" (as its
6931 buffer is no longer followed by "\0") when fbm_compile()
6932 adds the "\n" back, a "\0" is restored. */
6933 fbm_compile(sv, FBMcf_TAIL);
6937 if (prog->substrs->data[i].substr == prog->check_substr)
6938 prog->check_utf8 = sv;
6944 S_to_byte_substr(pTHX_ register regexp *prog)
6949 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6952 if (prog->substrs->data[i].utf8_substr
6953 && !prog->substrs->data[i].substr) {
6954 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6955 if (sv_utf8_downgrade(sv, TRUE)) {
6956 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6957 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
6958 /* Trim the trailing \n that fbm_compile added last
6960 SvCUR_set(sv, SvCUR(sv) - 1);
6961 fbm_compile(sv, FBMcf_TAIL);
6969 prog->substrs->data[i].substr = sv;
6970 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6971 prog->check_substr = sv;
6978 * c-indentation-style: bsd
6980 * indent-tabs-mode: t
6983 * ex: set ts=8 sts=4 sw=4 noet: