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 PERL_UNUSED_DECL; \
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)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || 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)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || 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, %i < 0",
357 paren_elems_to_push);
359 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
360 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
361 " out of range (%lu-%ld)",
362 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
364 SSGROW(total_elems + REGCP_FRAME_ELEMS);
366 for (p = PL_regsize; p > parenfloor; p--) {
367 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
368 SSPUSHINT(PL_regoffs[p].end);
369 SSPUSHINT(PL_regoffs[p].start);
370 SSPUSHPTR(PL_reg_start_tmp[p]);
372 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
373 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
374 (UV)p, (IV)PL_regoffs[p].start,
375 (IV)(PL_reg_start_tmp[p] - PL_bostr),
376 (IV)PL_regoffs[p].end
379 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
380 SSPUSHPTR(PL_regoffs);
381 SSPUSHINT(PL_regsize);
382 SSPUSHINT(*PL_reglastparen);
383 SSPUSHINT(*PL_reglastcloseparen);
384 SSPUSHPTR(PL_reginput);
385 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
390 /* These are needed since we do not localize EVAL nodes: */
391 #define REGCP_SET(cp) \
393 PerlIO_printf(Perl_debug_log, \
394 " Setting an EVAL scope, savestack=%"IVdf"\n", \
395 (IV)PL_savestack_ix)); \
398 #define REGCP_UNWIND(cp) \
400 if (cp != PL_savestack_ix) \
401 PerlIO_printf(Perl_debug_log, \
402 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
403 (IV)(cp), (IV)PL_savestack_ix)); \
407 S_regcppop(pTHX_ const regexp *rex)
412 GET_RE_DEBUG_FLAGS_DECL;
414 PERL_ARGS_ASSERT_REGCPPOP;
416 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
418 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
419 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
420 input = (char *) SSPOPPTR;
421 *PL_reglastcloseparen = SSPOPINT;
422 *PL_reglastparen = SSPOPINT;
423 PL_regsize = SSPOPINT;
424 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
426 i -= REGCP_OTHER_ELEMS;
427 /* Now restore the parentheses context. */
428 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
430 U32 paren = (U32)SSPOPINT;
431 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
432 PL_regoffs[paren].start = SSPOPINT;
434 if (paren <= *PL_reglastparen)
435 PL_regoffs[paren].end = tmps;
437 PerlIO_printf(Perl_debug_log,
438 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
439 (UV)paren, (IV)PL_regoffs[paren].start,
440 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
441 (IV)PL_regoffs[paren].end,
442 (paren > *PL_reglastparen ? "(no)" : ""));
446 if (*PL_reglastparen + 1 <= rex->nparens) {
447 PerlIO_printf(Perl_debug_log,
448 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
449 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
453 /* It would seem that the similar code in regtry()
454 * already takes care of this, and in fact it is in
455 * a better location to since this code can #if 0-ed out
456 * but the code in regtry() is needed or otherwise tests
457 * requiring null fields (pat.t#187 and split.t#{13,14}
458 * (as of patchlevel 7877) will fail. Then again,
459 * this code seems to be necessary or otherwise
460 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
461 * --jhi updated by dapm */
462 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
464 PL_regoffs[i].start = -1;
465 PL_regoffs[i].end = -1;
471 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
474 * pregexec and friends
477 #ifndef PERL_IN_XSUB_RE
479 - pregexec - match a regexp against a string
482 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
483 char *strbeg, I32 minend, SV *screamer, U32 nosave)
484 /* strend: pointer to null at end of string */
485 /* strbeg: real beginning of string */
486 /* minend: end of match must be >=minend after stringarg. */
487 /* nosave: For optimizations. */
489 PERL_ARGS_ASSERT_PREGEXEC;
492 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
493 nosave ? 0 : REXEC_COPY_STR);
498 * Need to implement the following flags for reg_anch:
500 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
502 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
503 * INTUIT_AUTORITATIVE_ML
504 * INTUIT_ONCE_NOML - Intuit can match in one location only.
507 * Another flag for this function: SECOND_TIME (so that float substrs
508 * with giant delta may be not rechecked).
511 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
513 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
514 Otherwise, only SvCUR(sv) is used to get strbeg. */
516 /* XXXX We assume that strpos is strbeg unless sv. */
518 /* XXXX Some places assume that there is a fixed substring.
519 An update may be needed if optimizer marks as "INTUITable"
520 RExen without fixed substrings. Similarly, it is assumed that
521 lengths of all the strings are no more than minlen, thus they
522 cannot come from lookahead.
523 (Or minlen should take into account lookahead.)
524 NOTE: Some of this comment is not correct. minlen does now take account
525 of lookahead/behind. Further research is required. -- demerphq
529 /* A failure to find a constant substring means that there is no need to make
530 an expensive call to REx engine, thus we celebrate a failure. Similarly,
531 finding a substring too deep into the string means that less calls to
532 regtry() should be needed.
534 REx compiler's optimizer found 4 possible hints:
535 a) Anchored substring;
537 c) Whether we are anchored (beginning-of-line or \G);
538 d) First node (of those at offset 0) which may distinguish positions;
539 We use a)b)d) and multiline-part of c), and try to find a position in the
540 string which does not contradict any of them.
543 /* Most of decisions we do here should have been done at compile time.
544 The nodes of the REx which we used for the search should have been
545 deleted from the finite automaton. */
548 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
549 char *strend, const U32 flags, re_scream_pos_data *data)
552 struct regexp *const prog = (struct regexp *)SvANY(rx);
553 register I32 start_shift = 0;
554 /* Should be nonnegative! */
555 register I32 end_shift = 0;
560 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
562 register char *other_last = NULL; /* other substr checked before this */
563 char *check_at = NULL; /* check substr found at this pos */
564 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
565 RXi_GET_DECL(prog,progi);
567 const char * const i_strpos = strpos;
569 GET_RE_DEBUG_FLAGS_DECL;
571 PERL_ARGS_ASSERT_RE_INTUIT_START;
573 RX_MATCH_UTF8_set(rx,utf8_target);
576 PL_reg_flags |= RF_utf8;
579 debug_start_match(rx, utf8_target, strpos, strend,
580 sv ? "Guessing start of match in sv for"
581 : "Guessing start of match in string for");
584 /* CHR_DIST() would be more correct here but it makes things slow. */
585 if (prog->minlen > strend - strpos) {
586 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
587 "String too short... [re_intuit_start]\n"));
591 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
594 if (!prog->check_utf8 && prog->check_substr)
595 to_utf8_substr(prog);
596 check = prog->check_utf8;
598 if (!prog->check_substr && prog->check_utf8)
599 to_byte_substr(prog);
600 check = prog->check_substr;
602 if (check == &PL_sv_undef) {
603 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
604 "Non-utf8 string cannot match utf8 check string\n"));
607 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
608 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
609 || ( (prog->extflags & RXf_ANCH_BOL)
610 && !multiline ) ); /* Check after \n? */
613 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
614 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
615 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
617 && (strpos != strbeg)) {
618 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
621 if (prog->check_offset_min == prog->check_offset_max &&
622 !(prog->extflags & RXf_CANY_SEEN)) {
623 /* Substring at constant offset from beg-of-str... */
626 s = HOP3c(strpos, prog->check_offset_min, strend);
629 slen = SvCUR(check); /* >= 1 */
631 if ( strend - s > slen || strend - s < slen - 1
632 || (strend - s == slen && strend[-1] != '\n')) {
633 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
636 /* Now should match s[0..slen-2] */
638 if (slen && (*SvPVX_const(check) != *s
640 && memNE(SvPVX_const(check), s, slen)))) {
642 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
646 else if (*SvPVX_const(check) != *s
647 || ((slen = SvCUR(check)) > 1
648 && memNE(SvPVX_const(check), s, slen)))
651 goto success_at_start;
654 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
656 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
657 end_shift = prog->check_end_shift;
660 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
661 - (SvTAIL(check) != 0);
662 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
664 if (end_shift < eshift)
668 else { /* Can match at random position */
671 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
672 end_shift = prog->check_end_shift;
674 /* end shift should be non negative here */
677 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
679 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
680 (IV)end_shift, RX_PRECOMP(prog));
684 /* Find a possible match in the region s..strend by looking for
685 the "check" substring in the region corrected by start/end_shift. */
688 I32 srch_start_shift = start_shift;
689 I32 srch_end_shift = end_shift;
690 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
691 srch_end_shift -= ((strbeg - s) - srch_start_shift);
692 srch_start_shift = strbeg - s;
694 DEBUG_OPTIMISE_MORE_r({
695 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
696 (IV)prog->check_offset_min,
697 (IV)srch_start_shift,
699 (IV)prog->check_end_shift);
702 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
703 I32 p = -1; /* Internal iterator of scream. */
704 I32 * const pp = data ? data->scream_pos : &p;
708 assert(SvMAGICAL(sv));
709 mg = mg_find(sv, PERL_MAGIC_study);
712 if (mg->mg_private == 1) {
713 found = ((U8 *)mg->mg_ptr)[BmRARE(check)] != (U8)~0;
714 } else if (mg->mg_private == 2) {
715 found = ((U16 *)mg->mg_ptr)[BmRARE(check)] != (U16)~0;
717 assert (mg->mg_private == 4);
718 found = ((U32 *)mg->mg_ptr)[BmRARE(check)] != (U32)~0;
722 || ( BmRARE(check) == '\n'
723 && (BmPREVIOUS(check) == SvCUR(check) - 1)
725 s = screaminstr(sv, check,
726 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
729 /* we may be pointing at the wrong string */
730 if (s && RXp_MATCH_COPIED(prog))
731 s = strbeg + (s - SvPVX_const(sv));
733 *data->scream_olds = s;
738 if (prog->extflags & RXf_CANY_SEEN) {
739 start_point= (U8*)(s + srch_start_shift);
740 end_point= (U8*)(strend - srch_end_shift);
742 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
743 end_point= HOP3(strend, -srch_end_shift, strbeg);
745 DEBUG_OPTIMISE_MORE_r({
746 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
747 (int)(end_point - start_point),
748 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
752 s = fbm_instr( start_point, end_point,
753 check, multiline ? FBMrf_MULTILINE : 0);
756 /* Update the count-of-usability, remove useless subpatterns,
760 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
761 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
762 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
763 (s ? "Found" : "Did not find"),
764 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
765 ? "anchored" : "floating"),
768 (s ? " at offset " : "...\n") );
773 /* Finish the diagnostic message */
774 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
776 /* XXX dmq: first branch is for positive lookbehind...
777 Our check string is offset from the beginning of the pattern.
778 So we need to do any stclass tests offset forward from that
787 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
788 Start with the other substr.
789 XXXX no SCREAM optimization yet - and a very coarse implementation
790 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
791 *always* match. Probably should be marked during compile...
792 Probably it is right to do no SCREAM here...
795 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
796 : (prog->float_substr && prog->anchored_substr))
798 /* Take into account the "other" substring. */
799 /* XXXX May be hopelessly wrong for UTF... */
802 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
805 char * const last = HOP3c(s, -start_shift, strbeg);
807 char * const saved_s = s;
810 t = s - prog->check_offset_max;
811 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
813 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
818 t = HOP3c(t, prog->anchored_offset, strend);
819 if (t < other_last) /* These positions already checked */
821 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
824 /* XXXX It is not documented what units *_offsets are in.
825 We assume bytes, but this is clearly wrong.
826 Meaning this code needs to be carefully reviewed for errors.
830 /* On end-of-str: see comment below. */
831 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
832 if (must == &PL_sv_undef) {
834 DEBUG_r(must = prog->anchored_utf8); /* for debug */
839 HOP3(HOP3(last1, prog->anchored_offset, strend)
840 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
842 multiline ? FBMrf_MULTILINE : 0
845 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
846 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
847 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
848 (s ? "Found" : "Contradicts"),
849 quoted, RE_SV_TAIL(must));
854 if (last1 >= last2) {
855 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
856 ", giving up...\n"));
859 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
860 ", trying floating at offset %ld...\n",
861 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
862 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
863 s = HOP3c(last, 1, strend);
867 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
868 (long)(s - i_strpos)));
869 t = HOP3c(s, -prog->anchored_offset, strbeg);
870 other_last = HOP3c(s, 1, strend);
878 else { /* Take into account the floating substring. */
880 char * const saved_s = s;
883 t = HOP3c(s, -start_shift, strbeg);
885 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
886 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
887 last = HOP3c(t, prog->float_max_offset, strend);
888 s = HOP3c(t, prog->float_min_offset, strend);
891 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
892 must = utf8_target ? prog->float_utf8 : prog->float_substr;
893 /* fbm_instr() takes into account exact value of end-of-str
894 if the check is SvTAIL(ed). Since false positives are OK,
895 and end-of-str is not later than strend we are OK. */
896 if (must == &PL_sv_undef) {
898 DEBUG_r(must = prog->float_utf8); /* for debug message */
901 s = fbm_instr((unsigned char*)s,
902 (unsigned char*)last + SvCUR(must)
904 must, multiline ? FBMrf_MULTILINE : 0);
906 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
907 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
908 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
909 (s ? "Found" : "Contradicts"),
910 quoted, RE_SV_TAIL(must));
914 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
915 ", giving up...\n"));
918 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
919 ", trying anchored starting at offset %ld...\n",
920 (long)(saved_s + 1 - i_strpos)));
922 s = HOP3c(t, 1, strend);
926 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
927 (long)(s - i_strpos)));
928 other_last = s; /* Fix this later. --Hugo */
938 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
940 DEBUG_OPTIMISE_MORE_r(
941 PerlIO_printf(Perl_debug_log,
942 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
943 (IV)prog->check_offset_min,
944 (IV)prog->check_offset_max,
952 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
954 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
957 /* Fixed substring is found far enough so that the match
958 cannot start at strpos. */
960 if (ml_anch && t[-1] != '\n') {
961 /* Eventually fbm_*() should handle this, but often
962 anchored_offset is not 0, so this check will not be wasted. */
963 /* XXXX In the code below we prefer to look for "^" even in
964 presence of anchored substrings. And we search even
965 beyond the found float position. These pessimizations
966 are historical artefacts only. */
968 while (t < strend - prog->minlen) {
970 if (t < check_at - prog->check_offset_min) {
971 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
972 /* Since we moved from the found position,
973 we definitely contradict the found anchored
974 substr. Due to the above check we do not
975 contradict "check" substr.
976 Thus we can arrive here only if check substr
977 is float. Redo checking for "other"=="fixed".
980 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
981 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
982 goto do_other_anchored;
984 /* We don't contradict the found floating substring. */
985 /* XXXX Why not check for STCLASS? */
987 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
988 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
991 /* Position contradicts check-string */
992 /* XXXX probably better to look for check-string
993 than for "\n", so one should lower the limit for t? */
994 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
995 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
996 other_last = strpos = s = t + 1;
1001 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1002 PL_colors[0], PL_colors[1]));
1006 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1007 PL_colors[0], PL_colors[1]));
1011 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1014 /* The found string does not prohibit matching at strpos,
1015 - no optimization of calling REx engine can be performed,
1016 unless it was an MBOL and we are not after MBOL,
1017 or a future STCLASS check will fail this. */
1019 /* Even in this situation we may use MBOL flag if strpos is offset
1020 wrt the start of the string. */
1021 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1022 && (strpos != strbeg) && strpos[-1] != '\n'
1023 /* May be due to an implicit anchor of m{.*foo} */
1024 && !(prog->intflags & PREGf_IMPLICIT))
1029 DEBUG_EXECUTE_r( if (ml_anch)
1030 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1031 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1034 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1036 prog->check_utf8 /* Could be deleted already */
1037 && --BmUSEFUL(prog->check_utf8) < 0
1038 && (prog->check_utf8 == prog->float_utf8)
1040 prog->check_substr /* Could be deleted already */
1041 && --BmUSEFUL(prog->check_substr) < 0
1042 && (prog->check_substr == prog->float_substr)
1045 /* If flags & SOMETHING - do not do it many times on the same match */
1046 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1047 /* XXX Does the destruction order has to change with utf8_target? */
1048 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1049 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1050 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1051 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1052 check = NULL; /* abort */
1054 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1055 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1056 if (prog->intflags & PREGf_IMPLICIT)
1057 prog->extflags &= ~RXf_ANCH_MBOL;
1058 /* XXXX This is a remnant of the old implementation. It
1059 looks wasteful, since now INTUIT can use many
1060 other heuristics. */
1061 prog->extflags &= ~RXf_USE_INTUIT;
1062 /* XXXX What other flags might need to be cleared in this branch? */
1068 /* Last resort... */
1069 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1070 /* trie stclasses are too expensive to use here, we are better off to
1071 leave it to regmatch itself */
1072 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1073 /* minlen == 0 is possible if regstclass is \b or \B,
1074 and the fixed substr is ''$.
1075 Since minlen is already taken into account, s+1 is before strend;
1076 accidentally, minlen >= 1 guaranties no false positives at s + 1
1077 even for \b or \B. But (minlen? 1 : 0) below assumes that
1078 regstclass does not come from lookahead... */
1079 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1080 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1081 const U8* const str = (U8*)STRING(progi->regstclass);
1082 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1083 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1086 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1087 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1088 else if (prog->float_substr || prog->float_utf8)
1089 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1093 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1094 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1097 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1100 const char *what = NULL;
1102 if (endpos == strend) {
1103 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1104 "Could not match STCLASS...\n") );
1107 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1108 "This position contradicts STCLASS...\n") );
1109 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1111 /* Contradict one of substrings */
1112 if (prog->anchored_substr || prog->anchored_utf8) {
1113 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1114 DEBUG_EXECUTE_r( what = "anchored" );
1116 s = HOP3c(t, 1, strend);
1117 if (s + start_shift + end_shift > strend) {
1118 /* XXXX Should be taken into account earlier? */
1119 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1120 "Could not match STCLASS...\n") );
1125 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1126 "Looking for %s substr starting at offset %ld...\n",
1127 what, (long)(s + start_shift - i_strpos)) );
1130 /* Have both, check_string is floating */
1131 if (t + start_shift >= check_at) /* Contradicts floating=check */
1132 goto retry_floating_check;
1133 /* Recheck anchored substring, but not floating... */
1137 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1138 "Looking for anchored substr starting at offset %ld...\n",
1139 (long)(other_last - i_strpos)) );
1140 goto do_other_anchored;
1142 /* Another way we could have checked stclass at the
1143 current position only: */
1148 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1149 "Looking for /%s^%s/m starting at offset %ld...\n",
1150 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1153 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1155 /* Check is floating substring. */
1156 retry_floating_check:
1157 t = check_at - start_shift;
1158 DEBUG_EXECUTE_r( what = "floating" );
1159 goto hop_and_restart;
1162 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1163 "By STCLASS: moving %ld --> %ld\n",
1164 (long)(t - i_strpos), (long)(s - i_strpos))
1168 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1169 "Does not contradict STCLASS...\n");
1174 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1175 PL_colors[4], (check ? "Guessed" : "Giving up"),
1176 PL_colors[5], (long)(s - i_strpos)) );
1179 fail_finish: /* Substring not found */
1180 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1181 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1183 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1184 PL_colors[4], PL_colors[5]));
1188 #define DECL_TRIE_TYPE(scan) \
1189 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1190 trie_type = ((scan->flags == EXACT) \
1191 ? (utf8_target ? trie_utf8 : trie_plain) \
1192 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1194 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1195 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1197 switch (trie_type) { \
1198 case trie_utf8_fold: \
1199 if ( foldlen>0 ) { \
1200 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1205 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1206 len = UTF8SKIP(uc); \
1207 skiplen = UNISKIP( uvc ); \
1208 foldlen -= skiplen; \
1209 uscan = foldbuf + skiplen; \
1212 case trie_latin_utf8_fold: \
1213 if ( foldlen>0 ) { \
1214 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1220 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1221 skiplen = UNISKIP( uvc ); \
1222 foldlen -= skiplen; \
1223 uscan = foldbuf + skiplen; \
1227 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1234 charid = trie->charmap[ uvc ]; \
1238 if (widecharmap) { \
1239 SV** const svpp = hv_fetch(widecharmap, \
1240 (char*)&uvc, sizeof(UV), 0); \
1242 charid = (U16)SvIV(*svpp); \
1247 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1251 && (ln == 1 || folder(s, pat_string, ln)) \
1252 && (!reginfo || regtry(reginfo, &s)) ) \
1258 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1260 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1266 #define REXEC_FBC_SCAN(CoDe) \
1268 while (s < strend) { \
1274 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1275 REXEC_FBC_UTF8_SCAN( \
1277 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1286 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1289 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1298 #define REXEC_FBC_TRYIT \
1299 if ((!reginfo || regtry(reginfo, &s))) \
1302 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1303 if (utf8_target) { \
1304 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1307 REXEC_FBC_CLASS_SCAN(CoNd); \
1310 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1311 if (utf8_target) { \
1313 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1316 REXEC_FBC_CLASS_SCAN(CoNd); \
1319 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1320 PL_reg_flags |= RF_tainted; \
1321 if (utf8_target) { \
1322 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1325 REXEC_FBC_CLASS_SCAN(CoNd); \
1328 #define DUMP_EXEC_POS(li,s,doutf8) \
1329 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1332 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1333 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1334 tmp = TEST_NON_UTF8(tmp); \
1335 REXEC_FBC_UTF8_SCAN( \
1336 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1345 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1346 if (s == PL_bostr) { \
1350 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1351 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1354 LOAD_UTF8_CHARCLASS_ALNUM(); \
1355 REXEC_FBC_UTF8_SCAN( \
1356 if (tmp == ! (TeSt2_UtF8)) { \
1365 /* The only difference between the BOUND and NBOUND cases is that
1366 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1367 * NBOUND. This is accomplished by passing it in either the if or else clause,
1368 * with the other one being empty */
1369 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1370 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1372 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1373 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1375 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1376 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1378 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1379 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1382 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1383 * be passed in completely with the variable name being tested, which isn't
1384 * such a clean interface, but this is easier to read than it was before. We
1385 * are looking for the boundary (or non-boundary between a word and non-word
1386 * character. The utf8 and non-utf8 cases have the same logic, but the details
1387 * must be different. Find the "wordness" of the character just prior to this
1388 * one, and compare it with the wordness of this one. If they differ, we have
1389 * a boundary. At the beginning of the string, pretend that the previous
1390 * character was a new-line */
1391 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1392 if (utf8_target) { \
1395 else { /* Not utf8 */ \
1396 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1397 tmp = TEST_NON_UTF8(tmp); \
1399 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1408 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1411 /* We know what class REx starts with. Try to find this position... */
1412 /* if reginfo is NULL, its a dryrun */
1413 /* annoyingly all the vars in this routine have different names from their counterparts
1414 in regmatch. /grrr */
1417 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1418 const char *strend, regmatch_info *reginfo)
1421 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1422 char *pat_string; /* The pattern's exactish string */
1423 char *pat_end; /* ptr to end char of pat_string */
1424 re_fold_t folder; /* Function for computing non-utf8 folds */
1425 const U8 *fold_array; /* array for folding ords < 256 */
1428 register STRLEN uskip;
1432 register I32 tmp = 1; /* Scratch variable? */
1433 register const bool utf8_target = PL_reg_match_utf8;
1434 UV utf8_fold_flags = 0;
1435 RXi_GET_DECL(prog,progi);
1437 PERL_ARGS_ASSERT_FIND_BYCLASS;
1439 /* We know what class it must start with. */
1443 if (utf8_target || OP(c) == ANYOFV) {
1444 STRLEN inclasslen = strend - s;
1445 REXEC_FBC_UTF8_CLASS_SCAN(
1446 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1449 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1454 if (tmp && (!reginfo || regtry(reginfo, &s)))
1462 if (UTF_PATTERN || utf8_target) {
1463 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1464 goto do_exactf_utf8;
1466 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1467 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1468 goto do_exactf_non_utf8; /* isn't dealt with by these */
1473 /* regcomp.c already folded this if pattern is in UTF-8 */
1474 utf8_fold_flags = 0;
1475 goto do_exactf_utf8;
1477 fold_array = PL_fold;
1479 goto do_exactf_non_utf8;
1482 if (UTF_PATTERN || utf8_target) {
1483 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1484 goto do_exactf_utf8;
1486 fold_array = PL_fold_locale;
1487 folder = foldEQ_locale;
1488 goto do_exactf_non_utf8;
1492 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1494 goto do_exactf_utf8;
1496 case EXACTFU_TRICKYFOLD:
1498 if (UTF_PATTERN || utf8_target) {
1499 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1500 goto do_exactf_utf8;
1503 /* Any 'ss' in the pattern should have been replaced by regcomp,
1504 * so we don't have to worry here about this single special case
1505 * in the Latin1 range */
1506 fold_array = PL_fold_latin1;
1507 folder = foldEQ_latin1;
1511 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1512 are no glitches with fold-length differences
1513 between the target string and pattern */
1515 /* The idea in the non-utf8 EXACTF* cases is to first find the
1516 * first character of the EXACTF* node and then, if necessary,
1517 * case-insensitively compare the full text of the node. c1 is the
1518 * first character. c2 is its fold. This logic will not work for
1519 * Unicode semantics and the german sharp ss, which hence should
1520 * not be compiled into a node that gets here. */
1521 pat_string = STRING(c);
1522 ln = STR_LEN(c); /* length to match in octets/bytes */
1524 /* We know that we have to match at least 'ln' bytes (which is the
1525 * same as characters, since not utf8). If we have to match 3
1526 * characters, and there are only 2 availabe, we know without
1527 * trying that it will fail; so don't start a match past the
1528 * required minimum number from the far end */
1529 e = HOP3c(strend, -((I32)ln), s);
1531 if (!reginfo && e < s) {
1532 e = s; /* Due to minlen logic of intuit() */
1536 c2 = fold_array[c1];
1537 if (c1 == c2) { /* If char and fold are the same */
1538 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1541 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1550 /* If one of the operands is in utf8, we can't use the simpler
1551 * folding above, due to the fact that many different characters
1552 * can have the same fold, or portion of a fold, or different-
1554 pat_string = STRING(c);
1555 ln = STR_LEN(c); /* length to match in octets/bytes */
1556 pat_end = pat_string + ln;
1557 lnc = (UTF_PATTERN) /* length to match in characters */
1558 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1561 /* We have 'lnc' characters to match in the pattern, but because of
1562 * multi-character folding, each character in the target can match
1563 * up to 3 characters (Unicode guarantees it will never exceed
1564 * this) if it is utf8-encoded; and up to 2 if not (based on the
1565 * fact that the Latin 1 folds are already determined, and the
1566 * only multi-char fold in that range is the sharp-s folding to
1567 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1568 * string character. Adjust lnc accordingly, rounding up, so that
1569 * if we need to match at least 4+1/3 chars, that really is 5. */
1570 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1571 lnc = (lnc + expansion - 1) / expansion;
1573 /* As in the non-UTF8 case, if we have to match 3 characters, and
1574 * only 2 are left, it's guaranteed to fail, so don't start a
1575 * match that would require us to go beyond the end of the string
1577 e = HOP3c(strend, -((I32)lnc), s);
1579 if (!reginfo && e < s) {
1580 e = s; /* Due to minlen logic of intuit() */
1583 /* XXX Note that we could recalculate e to stop the loop earlier,
1584 * as the worst case expansion above will rarely be met, and as we
1585 * go along we would usually find that e moves further to the left.
1586 * This would happen only after we reached the point in the loop
1587 * where if there were no expansion we should fail. Unclear if
1588 * worth the expense */
1591 char *my_strend= (char *)strend;
1592 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1593 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1594 && (!reginfo || regtry(reginfo, &s)) )
1598 s += (utf8_target) ? UTF8SKIP(s) : 1;
1603 PL_reg_flags |= RF_tainted;
1604 FBC_BOUND(isALNUM_LC,
1605 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1606 isALNUM_LC_utf8((U8*)s));
1609 PL_reg_flags |= RF_tainted;
1610 FBC_NBOUND(isALNUM_LC,
1611 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1612 isALNUM_LC_utf8((U8*)s));
1615 FBC_BOUND(isWORDCHAR,
1617 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1620 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1622 isWORDCHAR_A((U8*)s));
1625 FBC_NBOUND(isWORDCHAR,
1627 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1630 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1632 isWORDCHAR_A((U8*)s));
1635 FBC_BOUND(isWORDCHAR_L1,
1637 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1640 FBC_NBOUND(isWORDCHAR_L1,
1642 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1645 REXEC_FBC_CSCAN_TAINT(
1646 isALNUM_LC_utf8((U8*)s),
1651 REXEC_FBC_CSCAN_PRELOAD(
1652 LOAD_UTF8_CHARCLASS_ALNUM(),
1653 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1654 isWORDCHAR_L1((U8) *s)
1658 REXEC_FBC_CSCAN_PRELOAD(
1659 LOAD_UTF8_CHARCLASS_ALNUM(),
1660 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1665 /* Don't need to worry about utf8, as it can match only a single
1666 * byte invariant character */
1667 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1670 REXEC_FBC_CSCAN_PRELOAD(
1671 LOAD_UTF8_CHARCLASS_ALNUM(),
1672 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1673 ! isWORDCHAR_L1((U8) *s)
1677 REXEC_FBC_CSCAN_PRELOAD(
1678 LOAD_UTF8_CHARCLASS_ALNUM(),
1679 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1690 REXEC_FBC_CSCAN_TAINT(
1691 !isALNUM_LC_utf8((U8*)s),
1696 REXEC_FBC_CSCAN_PRELOAD(
1697 LOAD_UTF8_CHARCLASS_SPACE(),
1698 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1703 REXEC_FBC_CSCAN_PRELOAD(
1704 LOAD_UTF8_CHARCLASS_SPACE(),
1705 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1710 /* Don't need to worry about utf8, as it can match only a single
1711 * byte invariant character */
1712 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1715 REXEC_FBC_CSCAN_TAINT(
1716 isSPACE_LC_utf8((U8*)s),
1721 REXEC_FBC_CSCAN_PRELOAD(
1722 LOAD_UTF8_CHARCLASS_SPACE(),
1723 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1724 ! isSPACE_L1((U8) *s)
1728 REXEC_FBC_CSCAN_PRELOAD(
1729 LOAD_UTF8_CHARCLASS_SPACE(),
1730 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1741 REXEC_FBC_CSCAN_TAINT(
1742 !isSPACE_LC_utf8((U8*)s),
1747 REXEC_FBC_CSCAN_PRELOAD(
1748 LOAD_UTF8_CHARCLASS_DIGIT(),
1749 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1754 /* Don't need to worry about utf8, as it can match only a single
1755 * byte invariant character */
1756 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1759 REXEC_FBC_CSCAN_TAINT(
1760 isDIGIT_LC_utf8((U8*)s),
1765 REXEC_FBC_CSCAN_PRELOAD(
1766 LOAD_UTF8_CHARCLASS_DIGIT(),
1767 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1778 REXEC_FBC_CSCAN_TAINT(
1779 !isDIGIT_LC_utf8((U8*)s),
1786 is_LNBREAK_latin1(s)
1798 !is_VERTWS_latin1(s)
1804 is_HORIZWS_latin1(s)
1809 !is_HORIZWS_utf8(s),
1810 !is_HORIZWS_latin1(s)
1817 /* what trie are we using right now */
1819 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1821 = (reg_trie_data*)progi->data->data[ aho->trie ];
1822 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1824 const char *last_start = strend - trie->minlen;
1826 const char *real_start = s;
1828 STRLEN maxlen = trie->maxlen;
1830 U8 **points; /* map of where we were in the input string
1831 when reading a given char. For ASCII this
1832 is unnecessary overhead as the relationship
1833 is always 1:1, but for Unicode, especially
1834 case folded Unicode this is not true. */
1835 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1839 GET_RE_DEBUG_FLAGS_DECL;
1841 /* We can't just allocate points here. We need to wrap it in
1842 * an SV so it gets freed properly if there is a croak while
1843 * running the match */
1846 sv_points=newSV(maxlen * sizeof(U8 *));
1847 SvCUR_set(sv_points,
1848 maxlen * sizeof(U8 *));
1849 SvPOK_on(sv_points);
1850 sv_2mortal(sv_points);
1851 points=(U8**)SvPV_nolen(sv_points );
1852 if ( trie_type != trie_utf8_fold
1853 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1856 bitmap=(U8*)trie->bitmap;
1858 bitmap=(U8*)ANYOF_BITMAP(c);
1860 /* this is the Aho-Corasick algorithm modified a touch
1861 to include special handling for long "unknown char"
1862 sequences. The basic idea being that we use AC as long
1863 as we are dealing with a possible matching char, when
1864 we encounter an unknown char (and we have not encountered
1865 an accepting state) we scan forward until we find a legal
1867 AC matching is basically that of trie matching, except
1868 that when we encounter a failing transition, we fall back
1869 to the current states "fail state", and try the current char
1870 again, a process we repeat until we reach the root state,
1871 state 1, or a legal transition. If we fail on the root state
1872 then we can either terminate if we have reached an accepting
1873 state previously, or restart the entire process from the beginning
1877 while (s <= last_start) {
1878 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1886 U8 *uscan = (U8*)NULL;
1887 U8 *leftmost = NULL;
1889 U32 accepted_word= 0;
1893 while ( state && uc <= (U8*)strend ) {
1895 U32 word = aho->states[ state ].wordnum;
1899 DEBUG_TRIE_EXECUTE_r(
1900 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1901 dump_exec_pos( (char *)uc, c, strend, real_start,
1902 (char *)uc, utf8_target );
1903 PerlIO_printf( Perl_debug_log,
1904 " Scanning for legal start char...\n");
1908 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1912 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1918 if (uc >(U8*)last_start) break;
1922 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1923 if (!leftmost || lpos < leftmost) {
1924 DEBUG_r(accepted_word=word);
1930 points[pointpos++ % maxlen]= uc;
1931 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1932 uscan, len, uvc, charid, foldlen,
1934 DEBUG_TRIE_EXECUTE_r({
1935 dump_exec_pos( (char *)uc, c, strend, real_start,
1937 PerlIO_printf(Perl_debug_log,
1938 " Charid:%3u CP:%4"UVxf" ",
1944 word = aho->states[ state ].wordnum;
1946 base = aho->states[ state ].trans.base;
1948 DEBUG_TRIE_EXECUTE_r({
1950 dump_exec_pos( (char *)uc, c, strend, real_start,
1952 PerlIO_printf( Perl_debug_log,
1953 "%sState: %4"UVxf", word=%"UVxf,
1954 failed ? " Fail transition to " : "",
1955 (UV)state, (UV)word);
1961 ( ((offset = base + charid
1962 - 1 - trie->uniquecharcount)) >= 0)
1963 && ((U32)offset < trie->lasttrans)
1964 && trie->trans[offset].check == state
1965 && (tmp=trie->trans[offset].next))
1967 DEBUG_TRIE_EXECUTE_r(
1968 PerlIO_printf( Perl_debug_log," - legal\n"));
1973 DEBUG_TRIE_EXECUTE_r(
1974 PerlIO_printf( Perl_debug_log," - fail\n"));
1976 state = aho->fail[state];
1980 /* we must be accepting here */
1981 DEBUG_TRIE_EXECUTE_r(
1982 PerlIO_printf( Perl_debug_log," - accepting\n"));
1991 if (!state) state = 1;
1994 if ( aho->states[ state ].wordnum ) {
1995 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1996 if (!leftmost || lpos < leftmost) {
1997 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2002 s = (char*)leftmost;
2003 DEBUG_TRIE_EXECUTE_r({
2005 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2006 (UV)accepted_word, (IV)(s - real_start)
2009 if (!reginfo || regtry(reginfo, &s)) {
2015 DEBUG_TRIE_EXECUTE_r({
2016 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2019 DEBUG_TRIE_EXECUTE_r(
2020 PerlIO_printf( Perl_debug_log,"No match.\n"));
2029 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2039 - regexec_flags - match a regexp against a string
2042 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2043 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2044 /* strend: pointer to null at end of string */
2045 /* strbeg: real beginning of string */
2046 /* minend: end of match must be >=minend after stringarg. */
2047 /* data: May be used for some additional optimizations.
2048 Currently its only used, with a U32 cast, for transmitting
2049 the ganch offset when doing a /g match. This will change */
2050 /* nosave: For optimizations. */
2053 struct regexp *const prog = (struct regexp *)SvANY(rx);
2054 /*register*/ char *s;
2055 register regnode *c;
2056 /*register*/ char *startpos = stringarg;
2057 I32 minlen; /* must match at least this many chars */
2058 I32 dontbother = 0; /* how many characters not to try at end */
2059 I32 end_shift = 0; /* Same for the end. */ /* CC */
2060 I32 scream_pos = -1; /* Internal iterator of scream. */
2061 char *scream_olds = NULL;
2062 const bool utf8_target = cBOOL(DO_UTF8(sv));
2064 RXi_GET_DECL(prog,progi);
2065 regmatch_info reginfo; /* create some info to pass to regtry etc */
2066 regexp_paren_pair *swap = NULL;
2067 GET_RE_DEBUG_FLAGS_DECL;
2069 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2070 PERL_UNUSED_ARG(data);
2072 /* Be paranoid... */
2073 if (prog == NULL || startpos == NULL) {
2074 Perl_croak(aTHX_ "NULL regexp parameter");
2078 multiline = prog->extflags & RXf_PMf_MULTILINE;
2079 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2081 RX_MATCH_UTF8_set(rx, utf8_target);
2083 debug_start_match(rx, utf8_target, startpos, strend,
2087 minlen = prog->minlen;
2089 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2090 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2091 "String too short [regexec_flags]...\n"));
2096 /* Check validity of program. */
2097 if (UCHARAT(progi->program) != REG_MAGIC) {
2098 Perl_croak(aTHX_ "corrupted regexp program");
2102 PL_reg_eval_set = 0;
2106 PL_reg_flags |= RF_utf8;
2108 /* Mark beginning of line for ^ and lookbehind. */
2109 reginfo.bol = startpos; /* XXX not used ??? */
2113 /* Mark end of line for $ (and such) */
2116 /* see how far we have to get to not match where we matched before */
2117 reginfo.till = startpos+minend;
2119 /* If there is a "must appear" string, look for it. */
2122 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2124 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2125 reginfo.ganch = startpos + prog->gofs;
2126 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2127 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2128 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2130 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2131 && mg->mg_len >= 0) {
2132 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2133 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2134 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2136 if (prog->extflags & RXf_ANCH_GPOS) {
2137 if (s > reginfo.ganch)
2139 s = reginfo.ganch - prog->gofs;
2140 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2141 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2147 reginfo.ganch = strbeg + PTR2UV(data);
2148 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2149 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2151 } else { /* pos() not defined */
2152 reginfo.ganch = strbeg;
2153 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2154 "GPOS: reginfo.ganch = strbeg\n"));
2157 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2158 /* We have to be careful. If the previous successful match
2159 was from this regex we don't want a subsequent partially
2160 successful match to clobber the old results.
2161 So when we detect this possibility we add a swap buffer
2162 to the re, and switch the buffer each match. If we fail
2163 we switch it back, otherwise we leave it swapped.
2166 /* do we need a save destructor here for eval dies? */
2167 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2169 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2170 re_scream_pos_data d;
2172 d.scream_olds = &scream_olds;
2173 d.scream_pos = &scream_pos;
2174 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2176 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2177 goto phooey; /* not present */
2183 /* Simplest case: anchored match need be tried only once. */
2184 /* [unless only anchor is BOL and multiline is set] */
2185 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2186 if (s == startpos && regtry(®info, &startpos))
2188 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2189 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2194 dontbother = minlen - 1;
2195 end = HOP3c(strend, -dontbother, strbeg) - 1;
2196 /* for multiline we only have to try after newlines */
2197 if (prog->check_substr || prog->check_utf8) {
2198 /* because of the goto we can not easily reuse the macros for bifurcating the
2199 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2202 goto after_try_utf8;
2204 if (regtry(®info, &s)) {
2211 if (prog->extflags & RXf_USE_INTUIT) {
2212 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2221 } /* end search for check string in unicode */
2223 if (s == startpos) {
2224 goto after_try_latin;
2227 if (regtry(®info, &s)) {
2234 if (prog->extflags & RXf_USE_INTUIT) {
2235 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2244 } /* end search for check string in latin*/
2245 } /* end search for check string */
2246 else { /* search for newline */
2248 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2251 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2252 while (s <= end) { /* note it could be possible to match at the end of the string */
2253 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2254 if (regtry(®info, &s))
2258 } /* end search for newline */
2259 } /* end anchored/multiline check string search */
2261 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2263 /* the warning about reginfo.ganch being used without initialization
2264 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2265 and we only enter this block when the same bit is set. */
2266 char *tmp_s = reginfo.ganch - prog->gofs;
2268 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2273 /* Messy cases: unanchored match. */
2274 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2275 /* we have /x+whatever/ */
2276 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2281 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2282 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2283 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2288 DEBUG_EXECUTE_r( did_match = 1 );
2289 if (regtry(®info, &s)) goto got_it;
2291 while (s < strend && *s == ch)
2299 DEBUG_EXECUTE_r( did_match = 1 );
2300 if (regtry(®info, &s)) goto got_it;
2302 while (s < strend && *s == ch)
2307 DEBUG_EXECUTE_r(if (!did_match)
2308 PerlIO_printf(Perl_debug_log,
2309 "Did not find anchored character...\n")
2312 else if (prog->anchored_substr != NULL
2313 || prog->anchored_utf8 != NULL
2314 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2315 && prog->float_max_offset < strend - s)) {
2320 char *last1; /* Last position checked before */
2324 if (prog->anchored_substr || prog->anchored_utf8) {
2325 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2326 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2327 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2328 back_max = back_min = prog->anchored_offset;
2330 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2331 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2332 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2333 back_max = prog->float_max_offset;
2334 back_min = prog->float_min_offset;
2338 if (must == &PL_sv_undef)
2339 /* could not downgrade utf8 check substring, so must fail */
2345 last = HOP3c(strend, /* Cannot start after this */
2346 -(I32)(CHR_SVLEN(must)
2347 - (SvTAIL(must) != 0) + back_min), strbeg);
2350 last1 = HOPc(s, -1);
2352 last1 = s - 1; /* bogus */
2354 /* XXXX check_substr already used to find "s", can optimize if
2355 check_substr==must. */
2357 dontbother = end_shift;
2358 strend = HOPc(strend, -dontbother);
2359 while ( (s <= last) &&
2360 ((flags & REXEC_SCREAM) && SvSCREAM(sv)
2361 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2362 end_shift, &scream_pos, 0))
2363 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2364 (unsigned char*)strend, must,
2365 multiline ? FBMrf_MULTILINE : 0))) ) {
2366 /* we may be pointing at the wrong string */
2367 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2368 s = strbeg + (s - SvPVX_const(sv));
2369 DEBUG_EXECUTE_r( did_match = 1 );
2370 if (HOPc(s, -back_max) > last1) {
2371 last1 = HOPc(s, -back_min);
2372 s = HOPc(s, -back_max);
2375 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2377 last1 = HOPc(s, -back_min);
2381 while (s <= last1) {
2382 if (regtry(®info, &s))
2388 while (s <= last1) {
2389 if (regtry(®info, &s))
2395 DEBUG_EXECUTE_r(if (!did_match) {
2396 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2397 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2398 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2399 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2400 ? "anchored" : "floating"),
2401 quoted, RE_SV_TAIL(must));
2405 else if ( (c = progi->regstclass) ) {
2407 const OPCODE op = OP(progi->regstclass);
2408 /* don't bother with what can't match */
2409 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2410 strend = HOPc(strend, -(minlen - 1));
2413 SV * const prop = sv_newmortal();
2414 regprop(prog, prop, c);
2416 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2418 PerlIO_printf(Perl_debug_log,
2419 "Matching stclass %.*s against %s (%d bytes)\n",
2420 (int)SvCUR(prop), SvPVX_const(prop),
2421 quoted, (int)(strend - s));
2424 if (find_byclass(prog, c, s, strend, ®info))
2426 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2430 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2435 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2436 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2437 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2439 if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) {
2440 last = screaminstr(sv, float_real, s - strbeg,
2441 end_shift, &scream_pos, 1); /* last one */
2443 last = scream_olds; /* Only one occurrence. */
2444 /* we may be pointing at the wrong string */
2445 else if (RXp_MATCH_COPIED(prog))
2446 s = strbeg + (s - SvPVX_const(sv));
2450 const char * const little = SvPV_const(float_real, len);
2451 if (SvTAIL(float_real)) {
2452 /* This means that float_real contains an artificial \n on the end
2453 * due to the presence of something like this: /foo$/
2454 * where we can match both "foo" and "foo\n" at the end of the string.
2455 * So we have to compare the end of the string first against the float_real
2456 * without the \n and then against the full float_real with the string.
2457 * We have to watch out for cases where the string might be smaller
2458 * than the float_real or the float_real without the \n.
2460 char *checkpos= strend - len;
2462 PerlIO_printf(Perl_debug_log,
2463 "%sChecking for float_real.%s\n",
2464 PL_colors[4], PL_colors[5]));
2465 if (checkpos + 1 < strbeg) {
2466 /* can't match, even if we remove the trailing \n string is too short to match */
2468 PerlIO_printf(Perl_debug_log,
2469 "%sString shorter than required trailing substring, cannot match.%s\n",
2470 PL_colors[4], PL_colors[5]));
2472 } else if (memEQ(checkpos + 1, little, len - 1)) {
2473 /* can match, the end of the string matches without the "\n" */
2474 last = checkpos + 1;
2475 } else if (checkpos < strbeg) {
2476 /* cant match, string is too short when the "\n" is included */
2478 PerlIO_printf(Perl_debug_log,
2479 "%sString does not contain required trailing substring, cannot match.%s\n",
2480 PL_colors[4], PL_colors[5]));
2482 } else if (!multiline) {
2483 /* non multiline match, so compare with the "\n" at the end of the string */
2484 if (memEQ(checkpos, little, len)) {
2488 PerlIO_printf(Perl_debug_log,
2489 "%sString does not contain required trailing substring, cannot match.%s\n",
2490 PL_colors[4], PL_colors[5]));
2494 /* multiline match, so we have to search for a place where the full string is located */
2500 last = rninstr(s, strend, little, little + len);
2502 last = strend; /* matching "$" */
2506 /* at one point this block contained a comment which was probably
2507 * incorrect, which said that this was a "should not happen" case.
2508 * Even if it was true when it was written I am pretty sure it is
2509 * not anymore, so I have removed the comment and replaced it with
2512 PerlIO_printf(Perl_debug_log,
2513 "String does not contain required substring, cannot match.\n"
2517 dontbother = strend - last + prog->float_min_offset;
2519 if (minlen && (dontbother < minlen))
2520 dontbother = minlen - 1;
2521 strend -= dontbother; /* this one's always in bytes! */
2522 /* We don't know much -- general case. */
2525 if (regtry(®info, &s))
2534 if (regtry(®info, &s))
2536 } while (s++ < strend);
2545 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2547 if (PL_reg_eval_set)
2548 restore_pos(aTHX_ prog);
2549 if (RXp_PAREN_NAMES(prog))
2550 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2552 /* make sure $`, $&, $', and $digit will work later */
2553 if ( !(flags & REXEC_NOT_FIRST) ) {
2554 RX_MATCH_COPY_FREE(rx);
2555 if (flags & REXEC_COPY_STR) {
2556 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2557 #ifdef PERL_OLD_COPY_ON_WRITE
2559 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2561 PerlIO_printf(Perl_debug_log,
2562 "Copy on write: regexp capture, type %d\n",
2565 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2566 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2567 assert (SvPOKp(prog->saved_copy));
2571 RX_MATCH_COPIED_on(rx);
2572 s = savepvn(strbeg, i);
2578 prog->subbeg = strbeg;
2579 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2586 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2587 PL_colors[4], PL_colors[5]));
2588 if (PL_reg_eval_set)
2589 restore_pos(aTHX_ prog);
2591 /* we failed :-( roll it back */
2592 Safefree(prog->offs);
2601 - regtry - try match at specific point
2603 STATIC I32 /* 0 failure, 1 success */
2604 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2608 REGEXP *const rx = reginfo->prog;
2609 regexp *const prog = (struct regexp *)SvANY(rx);
2610 RXi_GET_DECL(prog,progi);
2611 GET_RE_DEBUG_FLAGS_DECL;
2613 PERL_ARGS_ASSERT_REGTRY;
2615 reginfo->cutpoint=NULL;
2617 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2620 PL_reg_eval_set = RS_init;
2621 DEBUG_EXECUTE_r(DEBUG_s(
2622 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2623 (IV)(PL_stack_sp - PL_stack_base));
2626 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2627 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2629 /* Apparently this is not needed, judging by wantarray. */
2630 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2631 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2634 /* Make $_ available to executed code. */
2635 if (reginfo->sv != DEFSV) {
2637 DEFSV_set(reginfo->sv);
2640 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2641 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2642 /* prepare for quick setting of pos */
2643 #ifdef PERL_OLD_COPY_ON_WRITE
2644 if (SvIsCOW(reginfo->sv))
2645 sv_force_normal_flags(reginfo->sv, 0);
2647 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2648 &PL_vtbl_mglob, NULL, 0);
2652 PL_reg_oldpos = mg->mg_len;
2653 SAVEDESTRUCTOR_X(restore_pos, prog);
2655 if (!PL_reg_curpm) {
2656 Newxz(PL_reg_curpm, 1, PMOP);
2659 SV* const repointer = &PL_sv_undef;
2660 /* this regexp is also owned by the new PL_reg_curpm, which
2661 will try to free it. */
2662 av_push(PL_regex_padav, repointer);
2663 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2664 PL_regex_pad = AvARRAY(PL_regex_padav);
2669 /* It seems that non-ithreads works both with and without this code.
2670 So for efficiency reasons it seems best not to have the code
2671 compiled when it is not needed. */
2672 /* This is safe against NULLs: */
2673 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2674 /* PM_reg_curpm owns a reference to this regexp. */
2675 (void)ReREFCNT_inc(rx);
2677 PM_SETRE(PL_reg_curpm, rx);
2678 PL_reg_oldcurpm = PL_curpm;
2679 PL_curpm = PL_reg_curpm;
2680 if (RXp_MATCH_COPIED(prog)) {
2681 /* Here is a serious problem: we cannot rewrite subbeg,
2682 since it may be needed if this match fails. Thus
2683 $` inside (?{}) could fail... */
2684 PL_reg_oldsaved = prog->subbeg;
2685 PL_reg_oldsavedlen = prog->sublen;
2686 #ifdef PERL_OLD_COPY_ON_WRITE
2687 PL_nrs = prog->saved_copy;
2689 RXp_MATCH_COPIED_off(prog);
2692 PL_reg_oldsaved = NULL;
2693 prog->subbeg = PL_bostr;
2694 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2696 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2697 prog->offs[0].start = *startpos - PL_bostr;
2698 PL_reginput = *startpos;
2699 PL_reglastparen = &prog->lastparen;
2700 PL_reglastcloseparen = &prog->lastcloseparen;
2701 prog->lastparen = 0;
2702 prog->lastcloseparen = 0;
2704 PL_regoffs = prog->offs;
2705 if (PL_reg_start_tmpl <= prog->nparens) {
2706 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2707 if(PL_reg_start_tmp)
2708 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2710 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2713 /* XXXX What this code is doing here?!!! There should be no need
2714 to do this again and again, PL_reglastparen should take care of
2717 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2718 * Actually, the code in regcppop() (which Ilya may be meaning by
2719 * PL_reglastparen), is not needed at all by the test suite
2720 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2721 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2722 * Meanwhile, this code *is* needed for the
2723 * above-mentioned test suite tests to succeed. The common theme
2724 * on those tests seems to be returning null fields from matches.
2725 * --jhi updated by dapm */
2727 if (prog->nparens) {
2728 regexp_paren_pair *pp = PL_regoffs;
2730 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2738 if (regmatch(reginfo, progi->program + 1)) {
2739 PL_regoffs[0].end = PL_reginput - PL_bostr;
2742 if (reginfo->cutpoint)
2743 *startpos= reginfo->cutpoint;
2744 REGCP_UNWIND(lastcp);
2749 #define sayYES goto yes
2750 #define sayNO goto no
2751 #define sayNO_SILENT goto no_silent
2753 /* we dont use STMT_START/END here because it leads to
2754 "unreachable code" warnings, which are bogus, but distracting. */
2755 #define CACHEsayNO \
2756 if (ST.cache_mask) \
2757 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2760 /* this is used to determine how far from the left messages like
2761 'failed...' are printed. It should be set such that messages
2762 are inline with the regop output that created them.
2764 #define REPORT_CODE_OFF 32
2767 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2768 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2770 #define SLAB_FIRST(s) (&(s)->states[0])
2771 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2773 /* grab a new slab and return the first slot in it */
2775 STATIC regmatch_state *
2778 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2781 regmatch_slab *s = PL_regmatch_slab->next;
2783 Newx(s, 1, regmatch_slab);
2784 s->prev = PL_regmatch_slab;
2786 PL_regmatch_slab->next = s;
2788 PL_regmatch_slab = s;
2789 return SLAB_FIRST(s);
2793 /* push a new state then goto it */
2795 #define PUSH_STATE_GOTO(state, node) \
2797 st->resume_state = state; \
2800 /* push a new state with success backtracking, then goto it */
2802 #define PUSH_YES_STATE_GOTO(state, node) \
2804 st->resume_state = state; \
2805 goto push_yes_state;
2811 regmatch() - main matching routine
2813 This is basically one big switch statement in a loop. We execute an op,
2814 set 'next' to point the next op, and continue. If we come to a point which
2815 we may need to backtrack to on failure such as (A|B|C), we push a
2816 backtrack state onto the backtrack stack. On failure, we pop the top
2817 state, and re-enter the loop at the state indicated. If there are no more
2818 states to pop, we return failure.
2820 Sometimes we also need to backtrack on success; for example /A+/, where
2821 after successfully matching one A, we need to go back and try to
2822 match another one; similarly for lookahead assertions: if the assertion
2823 completes successfully, we backtrack to the state just before the assertion
2824 and then carry on. In these cases, the pushed state is marked as
2825 'backtrack on success too'. This marking is in fact done by a chain of
2826 pointers, each pointing to the previous 'yes' state. On success, we pop to
2827 the nearest yes state, discarding any intermediate failure-only states.
2828 Sometimes a yes state is pushed just to force some cleanup code to be
2829 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2830 it to free the inner regex.
2832 Note that failure backtracking rewinds the cursor position, while
2833 success backtracking leaves it alone.
2835 A pattern is complete when the END op is executed, while a subpattern
2836 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2837 ops trigger the "pop to last yes state if any, otherwise return true"
2840 A common convention in this function is to use A and B to refer to the two
2841 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2842 the subpattern to be matched possibly multiple times, while B is the entire
2843 rest of the pattern. Variable and state names reflect this convention.
2845 The states in the main switch are the union of ops and failure/success of
2846 substates associated with with that op. For example, IFMATCH is the op
2847 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2848 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2849 successfully matched A and IFMATCH_A_fail is a state saying that we have
2850 just failed to match A. Resume states always come in pairs. The backtrack
2851 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2852 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2853 on success or failure.
2855 The struct that holds a backtracking state is actually a big union, with
2856 one variant for each major type of op. The variable st points to the
2857 top-most backtrack struct. To make the code clearer, within each
2858 block of code we #define ST to alias the relevant union.
2860 Here's a concrete example of a (vastly oversimplified) IFMATCH
2866 #define ST st->u.ifmatch
2868 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2869 ST.foo = ...; // some state we wish to save
2871 // push a yes backtrack state with a resume value of
2872 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2874 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2877 case IFMATCH_A: // we have successfully executed A; now continue with B
2879 bar = ST.foo; // do something with the preserved value
2882 case IFMATCH_A_fail: // A failed, so the assertion failed
2883 ...; // do some housekeeping, then ...
2884 sayNO; // propagate the failure
2891 For any old-timers reading this who are familiar with the old recursive
2892 approach, the code above is equivalent to:
2894 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2903 ...; // do some housekeeping, then ...
2904 sayNO; // propagate the failure
2907 The topmost backtrack state, pointed to by st, is usually free. If you
2908 want to claim it, populate any ST.foo fields in it with values you wish to
2909 save, then do one of
2911 PUSH_STATE_GOTO(resume_state, node);
2912 PUSH_YES_STATE_GOTO(resume_state, node);
2914 which sets that backtrack state's resume value to 'resume_state', pushes a
2915 new free entry to the top of the backtrack stack, then goes to 'node'.
2916 On backtracking, the free slot is popped, and the saved state becomes the
2917 new free state. An ST.foo field in this new top state can be temporarily
2918 accessed to retrieve values, but once the main loop is re-entered, it
2919 becomes available for reuse.
2921 Note that the depth of the backtrack stack constantly increases during the
2922 left-to-right execution of the pattern, rather than going up and down with
2923 the pattern nesting. For example the stack is at its maximum at Z at the
2924 end of the pattern, rather than at X in the following:
2926 /(((X)+)+)+....(Y)+....Z/
2928 The only exceptions to this are lookahead/behind assertions and the cut,
2929 (?>A), which pop all the backtrack states associated with A before
2932 Backtrack state structs are allocated in slabs of about 4K in size.
2933 PL_regmatch_state and st always point to the currently active state,
2934 and PL_regmatch_slab points to the slab currently containing
2935 PL_regmatch_state. The first time regmatch() is called, the first slab is
2936 allocated, and is never freed until interpreter destruction. When the slab
2937 is full, a new one is allocated and chained to the end. At exit from
2938 regmatch(), slabs allocated since entry are freed.
2943 #define DEBUG_STATE_pp(pp) \
2945 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2946 PerlIO_printf(Perl_debug_log, \
2947 " %*s"pp" %s%s%s%s%s\n", \
2949 PL_reg_name[st->resume_state], \
2950 ((st==yes_state||st==mark_state) ? "[" : ""), \
2951 ((st==yes_state) ? "Y" : ""), \
2952 ((st==mark_state) ? "M" : ""), \
2953 ((st==yes_state||st==mark_state) ? "]" : "") \
2958 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2963 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2964 const char *start, const char *end, const char *blurb)
2966 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2968 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2973 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2974 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2976 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2977 start, end - start, 60);
2979 PerlIO_printf(Perl_debug_log,
2980 "%s%s REx%s %s against %s\n",
2981 PL_colors[4], blurb, PL_colors[5], s0, s1);
2983 if (utf8_target||utf8_pat)
2984 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2985 utf8_pat ? "pattern" : "",
2986 utf8_pat && utf8_target ? " and " : "",
2987 utf8_target ? "string" : ""
2993 S_dump_exec_pos(pTHX_ const char *locinput,
2994 const regnode *scan,
2995 const char *loc_regeol,
2996 const char *loc_bostr,
2997 const char *loc_reg_starttry,
2998 const bool utf8_target)
3000 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3001 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3002 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3003 /* The part of the string before starttry has one color
3004 (pref0_len chars), between starttry and current
3005 position another one (pref_len - pref0_len chars),
3006 after the current position the third one.
3007 We assume that pref0_len <= pref_len, otherwise we
3008 decrease pref0_len. */
3009 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3010 ? (5 + taill) - l : locinput - loc_bostr;
3013 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3015 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3017 pref0_len = pref_len - (locinput - loc_reg_starttry);
3018 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3019 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3020 ? (5 + taill) - pref_len : loc_regeol - locinput);
3021 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3025 if (pref0_len > pref_len)
3026 pref0_len = pref_len;
3028 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3030 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3031 (locinput - pref_len),pref0_len, 60, 4, 5);
3033 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3034 (locinput - pref_len + pref0_len),
3035 pref_len - pref0_len, 60, 2, 3);
3037 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3038 locinput, loc_regeol - locinput, 10, 0, 1);
3040 const STRLEN tlen=len0+len1+len2;
3041 PerlIO_printf(Perl_debug_log,
3042 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3043 (IV)(locinput - loc_bostr),
3046 (docolor ? "" : "> <"),
3048 (int)(tlen > 19 ? 0 : 19 - tlen),
3055 /* reg_check_named_buff_matched()
3056 * Checks to see if a named buffer has matched. The data array of
3057 * buffer numbers corresponding to the buffer is expected to reside
3058 * in the regexp->data->data array in the slot stored in the ARG() of
3059 * node involved. Note that this routine doesn't actually care about the
3060 * name, that information is not preserved from compilation to execution.
3061 * Returns the index of the leftmost defined buffer with the given name
3062 * or 0 if non of the buffers matched.
3065 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3068 RXi_GET_DECL(rex,rexi);
3069 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3070 I32 *nums=(I32*)SvPVX(sv_dat);
3072 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3074 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3075 if ((I32)*PL_reglastparen >= nums[n] &&
3076 PL_regoffs[nums[n]].end != -1)
3085 /* free all slabs above current one - called during LEAVE_SCOPE */
3088 S_clear_backtrack_stack(pTHX_ void *p)
3090 regmatch_slab *s = PL_regmatch_slab->next;
3095 PL_regmatch_slab->next = NULL;
3097 regmatch_slab * const osl = s;
3104 #define SETREX(Re1,Re2) \
3105 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
3108 STATIC I32 /* 0 failure, 1 success */
3109 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3111 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3115 register const bool utf8_target = PL_reg_match_utf8;
3116 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3117 REGEXP *rex_sv = reginfo->prog;
3118 regexp *rex = (struct regexp *)SvANY(rex_sv);
3119 RXi_GET_DECL(rex,rexi);
3121 /* the current state. This is a cached copy of PL_regmatch_state */
3122 register regmatch_state *st;
3123 /* cache heavy used fields of st in registers */
3124 register regnode *scan;
3125 register regnode *next;
3126 register U32 n = 0; /* general value; init to avoid compiler warning */
3127 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3128 register char *locinput = PL_reginput;
3129 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3131 bool result = 0; /* return value of S_regmatch */
3132 int depth = 0; /* depth of backtrack stack */
3133 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3134 const U32 max_nochange_depth =
3135 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3136 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3137 regmatch_state *yes_state = NULL; /* state to pop to on success of
3139 /* mark_state piggy backs on the yes_state logic so that when we unwind
3140 the stack on success we can update the mark_state as we go */
3141 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3142 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3143 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3145 bool no_final = 0; /* prevent failure from backtracking? */
3146 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3147 char *startpoint = PL_reginput;
3148 SV *popmark = NULL; /* are we looking for a mark? */
3149 SV *sv_commit = NULL; /* last mark name seen in failure */
3150 SV *sv_yes_mark = NULL; /* last mark name we have seen
3151 during a successful match */
3152 U32 lastopen = 0; /* last open we saw */
3153 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3154 SV* const oreplsv = GvSV(PL_replgv);
3155 /* these three flags are set by various ops to signal information to
3156 * the very next op. They have a useful lifetime of exactly one loop
3157 * iteration, and are not preserved or restored by state pushes/pops
3159 bool sw = 0; /* the condition value in (?(cond)a|b) */
3160 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3161 int logical = 0; /* the following EVAL is:
3165 or the following IFMATCH/UNLESSM is:
3166 false: plain (?=foo)
3167 true: used as a condition: (?(?=foo))
3170 GET_RE_DEBUG_FLAGS_DECL;
3173 PERL_ARGS_ASSERT_REGMATCH;
3175 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3176 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3178 /* on first ever call to regmatch, allocate first slab */
3179 if (!PL_regmatch_slab) {
3180 Newx(PL_regmatch_slab, 1, regmatch_slab);
3181 PL_regmatch_slab->prev = NULL;
3182 PL_regmatch_slab->next = NULL;
3183 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3186 oldsave = PL_savestack_ix;
3187 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3188 SAVEVPTR(PL_regmatch_slab);
3189 SAVEVPTR(PL_regmatch_state);
3191 /* grab next free state slot */
3192 st = ++PL_regmatch_state;
3193 if (st > SLAB_LAST(PL_regmatch_slab))
3194 st = PL_regmatch_state = S_push_slab(aTHX);
3196 /* Note that nextchr is a byte even in UTF */
3197 nextchr = UCHARAT(locinput);
3199 while (scan != NULL) {
3202 SV * const prop = sv_newmortal();
3203 regnode *rnext=regnext(scan);
3204 DUMP_EXEC_POS( locinput, scan, utf8_target );
3205 regprop(rex, prop, scan);
3207 PerlIO_printf(Perl_debug_log,
3208 "%3"IVdf":%*s%s(%"IVdf")\n",
3209 (IV)(scan - rexi->program), depth*2, "",
3211 (PL_regkind[OP(scan)] == END || !rnext) ?
3212 0 : (IV)(rnext - rexi->program));
3215 next = scan + NEXT_OFF(scan);
3218 state_num = OP(scan);
3222 assert(PL_reglastparen == &rex->lastparen);
3223 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3224 assert(PL_regoffs == rex->offs);
3226 switch (state_num) {
3228 if (locinput == PL_bostr)
3230 /* reginfo->till = reginfo->bol; */
3235 if (locinput == PL_bostr ||
3236 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3242 if (locinput == PL_bostr)
3246 if (locinput == reginfo->ganch)
3251 /* update the startpoint */
3252 st->u.keeper.val = PL_regoffs[0].start;
3253 PL_reginput = locinput;
3254 PL_regoffs[0].start = locinput - PL_bostr;
3255 PUSH_STATE_GOTO(KEEPS_next, next);
3257 case KEEPS_next_fail:
3258 /* rollback the start point change */
3259 PL_regoffs[0].start = st->u.keeper.val;
3265 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3270 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3272 if (PL_regeol - locinput > 1)
3276 if (PL_regeol != locinput)
3280 if (!nextchr && locinput >= PL_regeol)
3283 locinput += PL_utf8skip[nextchr];
3284 if (locinput > PL_regeol)
3286 nextchr = UCHARAT(locinput);
3289 nextchr = UCHARAT(++locinput);
3292 if (!nextchr && locinput >= PL_regeol)
3294 nextchr = UCHARAT(++locinput);
3297 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3300 locinput += PL_utf8skip[nextchr];
3301 if (locinput > PL_regeol)
3303 nextchr = UCHARAT(locinput);
3306 nextchr = UCHARAT(++locinput);
3310 #define ST st->u.trie
3312 /* In this case the charclass data is available inline so
3313 we can fail fast without a lot of extra overhead.
3315 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3317 PerlIO_printf(Perl_debug_log,
3318 "%*s %sfailed to match trie start class...%s\n",
3319 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3326 /* the basic plan of execution of the trie is:
3327 * At the beginning, run though all the states, and
3328 * find the longest-matching word. Also remember the position
3329 * of the shortest matching word. For example, this pattern:
3332 * when matched against the string "abcde", will generate
3333 * accept states for all words except 3, with the longest
3334 * matching word being 4, and the shortest being 1 (with
3335 * the position being after char 1 of the string).
3337 * Then for each matching word, in word order (i.e. 1,2,4,5),
3338 * we run the remainder of the pattern; on each try setting
3339 * the current position to the character following the word,
3340 * returning to try the next word on failure.
3342 * We avoid having to build a list of words at runtime by
3343 * using a compile-time structure, wordinfo[].prev, which
3344 * gives, for each word, the previous accepting word (if any).
3345 * In the case above it would contain the mappings 1->2, 2->0,
3346 * 3->0, 4->5, 5->1. We can use this table to generate, from
3347 * the longest word (4 above), a list of all words, by
3348 * following the list of prev pointers; this gives us the
3349 * unordered list 4,5,1,2. Then given the current word we have
3350 * just tried, we can go through the list and find the
3351 * next-biggest word to try (so if we just failed on word 2,
3352 * the next in the list is 4).
3354 * Since at runtime we don't record the matching position in
3355 * the string for each word, we have to work that out for
3356 * each word we're about to process. The wordinfo table holds
3357 * the character length of each word; given that we recorded
3358 * at the start: the position of the shortest word and its
3359 * length in chars, we just need to move the pointer the
3360 * difference between the two char lengths. Depending on
3361 * Unicode status and folding, that's cheap or expensive.
3363 * This algorithm is optimised for the case where are only a
3364 * small number of accept states, i.e. 0,1, or maybe 2.
3365 * With lots of accepts states, and having to try all of them,
3366 * it becomes quadratic on number of accept states to find all
3371 /* what type of TRIE am I? (utf8 makes this contextual) */
3372 DECL_TRIE_TYPE(scan);
3374 /* what trie are we using right now */
3375 reg_trie_data * const trie
3376 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3377 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3378 U32 state = trie->startstate;
3380 if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
3381 if (trie->states[ state ].wordnum) {
3383 PerlIO_printf(Perl_debug_log,
3384 "%*s %smatched empty string...%s\n",
3385 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3391 PerlIO_printf(Perl_debug_log,
3392 "%*s %sfailed to match trie start class...%s\n",
3393 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3400 U8 *uc = ( U8* )locinput;
3404 U8 *uscan = (U8*)NULL;
3405 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3406 U32 charcount = 0; /* how many input chars we have matched */
3407 U32 accepted = 0; /* have we seen any accepting states? */
3410 ST.jump = trie->jump;
3413 ST.longfold = FALSE; /* char longer if folded => it's harder */
3416 /* fully traverse the TRIE; note the position of the
3417 shortest accept state and the wordnum of the longest
3420 while ( state && uc <= (U8*)PL_regeol ) {
3421 U32 base = trie->states[ state ].trans.base;
3425 wordnum = trie->states[ state ].wordnum;
3427 if (wordnum) { /* it's an accept state */
3430 /* record first match position */
3432 ST.firstpos = (U8*)locinput;
3437 ST.firstchars = charcount;
3440 if (!ST.nextword || wordnum < ST.nextword)
3441 ST.nextword = wordnum;
3442 ST.topword = wordnum;
3445 DEBUG_TRIE_EXECUTE_r({
3446 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3447 PerlIO_printf( Perl_debug_log,
3448 "%*s %sState: %4"UVxf" Accepted: %c ",
3449 2+depth * 2, "", PL_colors[4],
3450 (UV)state, (accepted ? 'Y' : 'N'));
3453 /* read a char and goto next state */
3456 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3457 uscan, len, uvc, charid, foldlen,
3464 base + charid - 1 - trie->uniquecharcount)) >= 0)
3466 && ((U32)offset < trie->lasttrans)
3467 && trie->trans[offset].check == state)
3469 state = trie->trans[offset].next;
3480 DEBUG_TRIE_EXECUTE_r(
3481 PerlIO_printf( Perl_debug_log,
3482 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3483 charid, uvc, (UV)state, PL_colors[5] );
3489 /* calculate total number of accept states */
3494 w = trie->wordinfo[w].prev;
3497 ST.accepted = accepted;
3501 PerlIO_printf( Perl_debug_log,
3502 "%*s %sgot %"IVdf" possible matches%s\n",
3503 REPORT_CODE_OFF + depth * 2, "",
3504 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3506 goto trie_first_try; /* jump into the fail handler */
3510 case TRIE_next_fail: /* we failed - try next alternative */
3512 REGCP_UNWIND(ST.cp);
3513 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3514 PL_regoffs[n].end = -1;
3515 *PL_reglastparen = n;
3517 if (!--ST.accepted) {
3519 PerlIO_printf( Perl_debug_log,
3520 "%*s %sTRIE failed...%s\n",
3521 REPORT_CODE_OFF+depth*2, "",
3528 /* Find next-highest word to process. Note that this code
3529 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3530 register U16 min = 0;
3532 register U16 const nextword = ST.nextword;
3533 register reg_trie_wordinfo * const wordinfo
3534 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3535 for (word=ST.topword; word; word=wordinfo[word].prev) {
3536 if (word > nextword && (!min || word < min))
3549 ST.lastparen = *PL_reglastparen;
3553 /* find start char of end of current word */
3555 U32 chars; /* how many chars to skip */
3556 U8 *uc = ST.firstpos;
3557 reg_trie_data * const trie
3558 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3560 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3562 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3566 /* the hard option - fold each char in turn and find
3567 * its folded length (which may be different */
3568 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3576 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3584 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3589 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3603 PL_reginput = (char *)uc;
3606 scan = (ST.jump && ST.jump[ST.nextword])
3607 ? ST.me + ST.jump[ST.nextword]
3611 PerlIO_printf( Perl_debug_log,
3612 "%*s %sTRIE matched word #%d, continuing%s\n",
3613 REPORT_CODE_OFF+depth*2, "",
3620 if (ST.accepted > 1 || has_cutgroup) {
3621 PUSH_STATE_GOTO(TRIE_next, scan);
3624 /* only one choice left - just continue */
3626 AV *const trie_words
3627 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3628 SV ** const tmp = av_fetch( trie_words,
3630 SV *sv= tmp ? sv_newmortal() : NULL;
3632 PerlIO_printf( Perl_debug_log,
3633 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3634 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3636 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3637 PL_colors[0], PL_colors[1],
3638 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3640 : "not compiled under -Dr",
3644 locinput = PL_reginput;
3645 nextchr = UCHARAT(locinput);
3646 continue; /* execute rest of RE */
3651 char *s = STRING(scan);
3653 if (utf8_target != UTF_PATTERN) {
3654 /* The target and the pattern have differing utf8ness. */
3656 const char * const e = s + ln;
3659 /* The target is utf8, the pattern is not utf8. */
3664 if (NATIVE_TO_UNI(*(U8*)s) !=
3665 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3673 /* The target is not utf8, the pattern is utf8. */
3678 if (NATIVE_TO_UNI(*((U8*)l)) !=
3679 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3687 nextchr = UCHARAT(locinput);
3690 /* The target and the pattern have the same utf8ness. */
3691 /* Inline the first character, for speed. */
3692 if (UCHARAT(s) != nextchr)
3694 if (PL_regeol - locinput < ln)
3696 if (ln > 1 && memNE(s, locinput, ln))
3699 nextchr = UCHARAT(locinput);
3704 const U8 * fold_array;
3706 U32 fold_utf8_flags;
3708 PL_reg_flags |= RF_tainted;
3709 folder = foldEQ_locale;
3710 fold_array = PL_fold_locale;
3711 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3715 case EXACTFU_TRICKYFOLD:
3717 folder = foldEQ_latin1;
3718 fold_array = PL_fold_latin1;
3719 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3723 folder = foldEQ_latin1;
3724 fold_array = PL_fold_latin1;
3725 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3730 fold_array = PL_fold;
3731 fold_utf8_flags = 0;
3737 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
3738 /* Either target or the pattern are utf8, or has the issue where
3739 * the fold lengths may differ. */
3740 const char * const l = locinput;
3741 char *e = PL_regeol;
3743 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3744 l, &e, 0, utf8_target, fold_utf8_flags))
3749 nextchr = UCHARAT(locinput);
3753 /* Neither the target nor the pattern are utf8 */
3754 if (UCHARAT(s) != nextchr &&
3755 UCHARAT(s) != fold_array[nextchr])
3759 if (PL_regeol - locinput < ln)
3761 if (ln > 1 && ! folder(s, locinput, ln))
3764 nextchr = UCHARAT(locinput);
3768 /* XXX Could improve efficiency by separating these all out using a
3769 * macro or in-line function. At that point regcomp.c would no longer
3770 * have to set the FLAGS fields of these */
3773 PL_reg_flags |= RF_tainted;
3781 /* was last char in word? */
3783 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3784 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3786 if (locinput == PL_bostr)
3789 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3791 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3793 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3794 ln = isALNUM_uni(ln);
3795 LOAD_UTF8_CHARCLASS_ALNUM();
3796 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3799 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3800 n = isALNUM_LC_utf8((U8*)locinput);
3805 /* Here the string isn't utf8, or is utf8 and only ascii
3806 * characters are to match \w. In the latter case looking at
3807 * the byte just prior to the current one may be just the final
3808 * byte of a multi-byte character. This is ok. There are two
3810 * 1) it is a single byte character, and then the test is doing
3811 * just what it's supposed to.
3812 * 2) it is a multi-byte character, in which case the final
3813 * byte is never mistakable for ASCII, and so the test
3814 * will say it is not a word character, which is the
3815 * correct answer. */
3816 ln = (locinput != PL_bostr) ?
3817 UCHARAT(locinput - 1) : '\n';
3818 switch (FLAGS(scan)) {
3819 case REGEX_UNICODE_CHARSET:
3820 ln = isWORDCHAR_L1(ln);
3821 n = isWORDCHAR_L1(nextchr);
3823 case REGEX_LOCALE_CHARSET:
3824 ln = isALNUM_LC(ln);
3825 n = isALNUM_LC(nextchr);
3827 case REGEX_DEPENDS_CHARSET:
3829 n = isALNUM(nextchr);
3831 case REGEX_ASCII_RESTRICTED_CHARSET:
3832 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3833 ln = isWORDCHAR_A(ln);
3834 n = isWORDCHAR_A(nextchr);
3837 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3841 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3843 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3848 if (utf8_target || state_num == ANYOFV) {
3849 STRLEN inclasslen = PL_regeol - locinput;
3850 if (locinput >= PL_regeol)
3853 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3855 locinput += inclasslen;
3856 nextchr = UCHARAT(locinput);
3861 nextchr = UCHARAT(locinput);
3862 if (!nextchr && locinput >= PL_regeol)
3864 if (!REGINCLASS(rex, scan, (U8*)locinput))
3866 nextchr = UCHARAT(++locinput);
3870 /* Special char classes - The defines start on line 129 or so */
3871 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3872 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3873 ALNUMU, NALNUMU, isWORDCHAR_L1,
3874 ALNUMA, NALNUMA, isWORDCHAR_A,
3877 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3878 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3879 SPACEU, NSPACEU, isSPACE_L1,
3880 SPACEA, NSPACEA, isSPACE_A,
3883 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3884 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3885 DIGITA, NDIGITA, isDIGIT_A,
3888 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3889 a Unicode extended Grapheme Cluster */
3890 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3891 extended Grapheme Cluster is:
3894 | Prepend* Begin Extend*
3897 Begin is (Hangul-syllable | ! Control)
3898 Extend is (Grapheme_Extend | Spacing_Mark)
3899 Control is [ GCB_Control CR LF ]
3901 The discussion below shows how the code for CLUMP is derived
3902 from this regex. Note that most of these concepts are from
3903 property values of the Grapheme Cluster Boundary (GCB) property.
3904 No code point can have multiple property values for a given
3905 property. Thus a code point in Prepend can't be in Control, but
3906 it must be in !Control. This is why Control above includes
3907 GCB_Control plus CR plus LF. The latter two are used in the GCB
3908 property separately, and so can't be in GCB_Control, even though
3909 they logically are controls. Control is not the same as gc=cc,
3910 but includes format and other characters as well.
3912 The Unicode definition of Hangul-syllable is:
3914 | (L* ( ( V | LV ) V* | LVT ) T*)
3917 Each of these is a value for the GCB property, and hence must be
3918 disjoint, so the order they are tested is immaterial, so the
3919 above can safely be changed to
3922 | (L* ( LVT | ( V | LV ) V*) T*)
3924 The last two terms can be combined like this:
3926 | (( LVT | ( V | LV ) V*) T*))
3928 And refactored into this:
3929 L* (L | LVT T* | V V* T* | LV V* T*)
3931 That means that if we have seen any L's at all we can quit
3932 there, but if the next character is an LVT, a V, or an LV we
3935 There is a subtlety with Prepend* which showed up in testing.
3936 Note that the Begin, and only the Begin is required in:
3937 | Prepend* Begin Extend*
3938 Also, Begin contains '! Control'. A Prepend must be a
3939 '! Control', which means it must also be a Begin. What it
3940 comes down to is that if we match Prepend* and then find no
3941 suitable Begin afterwards, that if we backtrack the last
3942 Prepend, that one will be a suitable Begin.
3945 if (locinput >= PL_regeol)
3947 if (! utf8_target) {
3949 /* Match either CR LF or '.', as all the other possibilities
3951 locinput++; /* Match the . or CR */
3952 if (nextchr == '\r' /* And if it was CR, and the next is LF,
3954 && locinput < PL_regeol
3955 && UCHARAT(locinput) == '\n') locinput++;
3959 /* Utf8: See if is ( CR LF ); already know that locinput <
3960 * PL_regeol, so locinput+1 is in bounds */
3961 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3965 /* In case have to backtrack to beginning, then match '.' */
3966 char *starting = locinput;
3968 /* In case have to backtrack the last prepend */
3969 char *previous_prepend = 0;
3971 LOAD_UTF8_CHARCLASS_GCB();
3973 /* Match (prepend)* */
3974 while (locinput < PL_regeol
3975 && swash_fetch(PL_utf8_X_prepend,
3976 (U8*)locinput, utf8_target))
3978 previous_prepend = locinput;
3979 locinput += UTF8SKIP(locinput);
3982 /* As noted above, if we matched a prepend character, but
3983 * the next thing won't match, back off the last prepend we
3984 * matched, as it is guaranteed to match the begin */
3985 if (previous_prepend
3986 && (locinput >= PL_regeol
3987 || ! swash_fetch(PL_utf8_X_begin,
3988 (U8*)locinput, utf8_target)))
3990 locinput = previous_prepend;
3993 /* Note that here we know PL_regeol > locinput, as we
3994 * tested that upon input to this switch case, and if we
3995 * moved locinput forward, we tested the result just above
3996 * and it either passed, or we backed off so that it will
3998 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
4000 /* Here did not match the required 'Begin' in the
4001 * second term. So just match the very first
4002 * character, the '.' of the final term of the regex */
4003 locinput = starting + UTF8SKIP(starting);
4006 /* Here is the beginning of a character that can have
4007 * an extender. It is either a hangul syllable, or a
4009 if (swash_fetch(PL_utf8_X_non_hangul,
4010 (U8*)locinput, utf8_target))
4013 /* Here not a Hangul syllable, must be a
4014 * ('! * Control') */
4015 locinput += UTF8SKIP(locinput);
4018 /* Here is a Hangul syllable. It can be composed
4019 * of several individual characters. One
4020 * possibility is T+ */
4021 if (swash_fetch(PL_utf8_X_T,
4022 (U8*)locinput, utf8_target))
4024 while (locinput < PL_regeol
4025 && swash_fetch(PL_utf8_X_T,
4026 (U8*)locinput, utf8_target))
4028 locinput += UTF8SKIP(locinput);
4032 /* Here, not T+, but is a Hangul. That means
4033 * it is one of the others: L, LV, LVT or V,
4035 * L* (L | LVT T* | V V* T* | LV V* T*) */
4038 while (locinput < PL_regeol
4039 && swash_fetch(PL_utf8_X_L,
4040 (U8*)locinput, utf8_target))
4042 locinput += UTF8SKIP(locinput);
4045 /* Here, have exhausted L*. If the next
4046 * character is not an LV, LVT nor V, it means
4047 * we had to have at least one L, so matches L+
4048 * in the original equation, we have a complete
4049 * hangul syllable. Are done. */
4051 if (locinput < PL_regeol
4052 && swash_fetch(PL_utf8_X_LV_LVT_V,
4053 (U8*)locinput, utf8_target))
4056 /* Otherwise keep going. Must be LV, LVT
4057 * or V. See if LVT */
4058 if (swash_fetch(PL_utf8_X_LVT,
4059 (U8*)locinput, utf8_target))
4061 locinput += UTF8SKIP(locinput);
4064 /* Must be V or LV. Take it, then
4066 locinput += UTF8SKIP(locinput);
4067 while (locinput < PL_regeol
4068 && swash_fetch(PL_utf8_X_V,
4069 (U8*)locinput, utf8_target))
4071 locinput += UTF8SKIP(locinput);
4075 /* And any of LV, LVT, or V can be followed
4077 while (locinput < PL_regeol
4078 && swash_fetch(PL_utf8_X_T,
4082 locinput += UTF8SKIP(locinput);
4088 /* Match any extender */
4089 while (locinput < PL_regeol
4090 && swash_fetch(PL_utf8_X_extend,
4091 (U8*)locinput, utf8_target))
4093 locinput += UTF8SKIP(locinput);
4097 if (locinput > PL_regeol) sayNO;
4099 nextchr = UCHARAT(locinput);
4103 { /* The capture buffer cases. The ones beginning with N for the
4104 named buffers just convert to the equivalent numbered and
4105 pretend they were called as the corresponding numbered buffer
4107 /* don't initialize these in the declaration, it makes C++
4112 const U8 *fold_array;
4115 PL_reg_flags |= RF_tainted;
4116 folder = foldEQ_locale;
4117 fold_array = PL_fold_locale;
4119 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4123 folder = foldEQ_latin1;
4124 fold_array = PL_fold_latin1;
4126 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4130 folder = foldEQ_latin1;
4131 fold_array = PL_fold_latin1;
4133 utf8_fold_flags = 0;
4138 fold_array = PL_fold;
4140 utf8_fold_flags = 0;
4147 utf8_fold_flags = 0;
4150 /* For the named back references, find the corresponding buffer
4152 n = reg_check_named_buff_matched(rex,scan);
4157 goto do_nref_ref_common;
4160 PL_reg_flags |= RF_tainted;
4161 folder = foldEQ_locale;
4162 fold_array = PL_fold_locale;
4163 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4167 folder = foldEQ_latin1;
4168 fold_array = PL_fold_latin1;
4169 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4173 folder = foldEQ_latin1;
4174 fold_array = PL_fold_latin1;
4175 utf8_fold_flags = 0;
4180 fold_array = PL_fold;
4181 utf8_fold_flags = 0;
4187 utf8_fold_flags = 0;
4191 n = ARG(scan); /* which paren pair */
4194 ln = PL_regoffs[n].start;
4195 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4196 if (*PL_reglastparen < n || ln == -1)
4197 sayNO; /* Do not match unless seen CLOSEn. */
4198 if (ln == PL_regoffs[n].end)
4202 if (type != REF /* REF can do byte comparison */
4203 && (utf8_target || type == REFFU))
4204 { /* XXX handle REFFL better */
4205 char * limit = PL_regeol;
4207 /* This call case insensitively compares the entire buffer
4208 * at s, with the current input starting at locinput, but
4209 * not going off the end given by PL_regeol, and returns in
4210 * limit upon success, how much of the current input was
4212 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4213 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4218 nextchr = UCHARAT(locinput);
4222 /* Not utf8: Inline the first character, for speed. */
4223 if (UCHARAT(s) != nextchr &&
4225 UCHARAT(s) != fold_array[nextchr]))
4227 ln = PL_regoffs[n].end - ln;
4228 if (locinput + ln > PL_regeol)
4230 if (ln > 1 && (type == REF
4231 ? memNE(s, locinput, ln)
4232 : ! folder(s, locinput, ln)))
4235 nextchr = UCHARAT(locinput);
4245 #define ST st->u.eval
4250 regexp_internal *rei;
4251 regnode *startpoint;
4254 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4255 if (cur_eval && cur_eval->locinput==locinput) {
4256 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4257 Perl_croak(aTHX_ "Infinite recursion in regex");
4258 if ( ++nochange_depth > max_nochange_depth )
4260 "Pattern subroutine nesting without pos change"
4261 " exceeded limit in regex");
4268 (void)ReREFCNT_inc(rex_sv);
4269 if (OP(scan)==GOSUB) {
4270 startpoint = scan + ARG2L(scan);
4271 ST.close_paren = ARG(scan);
4273 startpoint = rei->program+1;
4276 goto eval_recurse_doit;
4278 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4279 if (cur_eval && cur_eval->locinput==locinput) {
4280 if ( ++nochange_depth > max_nochange_depth )
4281 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4286 /* execute the code in the {...} */
4288 SV ** const before = SP;
4289 OP_4tree * const oop = PL_op;
4290 COP * const ocurcop = PL_curcop;
4292 char *saved_regeol = PL_regeol;
4293 struct re_save_state saved_state;
4295 /* To not corrupt the existing regex state while executing the
4296 * eval we would normally put it on the save stack, like with
4297 * save_re_context. However, re-evals have a weird scoping so we
4298 * can't just add ENTER/LEAVE here. With that, things like
4300 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4302 * would break, as they expect the localisation to be unwound
4303 * only when the re-engine backtracks through the bit that
4306 * What we do instead is just saving the state in a local c
4309 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4312 PL_op = (OP_4tree*)rexi->data->data[n];
4313 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4314 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4315 /* wrap the call in two SAVECOMPPADs. This ensures that
4316 * when the save stack is eventually unwound, all the
4317 * accumulated SAVEt_CLEARSV's will be processed with
4318 * interspersed SAVEt_COMPPAD's to ensure that lexicals
4319 * are cleared in the right pad */
4321 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4322 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4325 SV *sv_mrk = get_sv("REGMARK", 1);
4326 sv_setsv(sv_mrk, sv_yes_mark);
4329 CALLRUNOPS(aTHX); /* Scalar context. */
4332 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4338 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4342 PAD_RESTORE_LOCAL(old_comppad);
4343 PL_curcop = ocurcop;
4344 PL_regeol = saved_regeol;
4347 sv_setsv(save_scalar(PL_replgv), ret);
4351 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4354 /* extract RE object from returned value; compiling if
4360 SV *const sv = SvRV(ret);
4362 if (SvTYPE(sv) == SVt_REGEXP) {
4364 } else if (SvSMAGICAL(sv)) {
4365 mg = mg_find(sv, PERL_MAGIC_qr);
4368 } else if (SvTYPE(ret) == SVt_REGEXP) {
4370 } else if (SvSMAGICAL(ret)) {
4371 if (SvGMAGICAL(ret)) {
4372 /* I don't believe that there is ever qr magic
4374 assert(!mg_find(ret, PERL_MAGIC_qr));
4375 sv_unmagic(ret, PERL_MAGIC_qr);
4378 mg = mg_find(ret, PERL_MAGIC_qr);
4379 /* testing suggests mg only ends up non-NULL for
4380 scalars who were upgraded and compiled in the
4381 else block below. In turn, this is only
4382 triggered in the "postponed utf8 string" tests
4388 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4392 rx = reg_temp_copy(NULL, rx);
4396 const I32 osize = PL_regsize;
4399 assert (SvUTF8(ret));
4400 } else if (SvUTF8(ret)) {
4401 /* Not doing UTF-8, despite what the SV says. Is
4402 this only if we're trapped in use 'bytes'? */
4403 /* Make a copy of the octet sequence, but without
4404 the flag on, as the compiler now honours the
4405 SvUTF8 flag on ret. */
4407 const char *const p = SvPV(ret, len);
4408 ret = newSVpvn_flags(p, len, SVs_TEMP);
4410 rx = CALLREGCOMP(ret, pm_flags);
4412 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4414 /* This isn't a first class regexp. Instead, it's
4415 caching a regexp onto an existing, Perl visible
4417 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4422 re = (struct regexp *)SvANY(rx);
4424 RXp_MATCH_COPIED_off(re);
4425 re->subbeg = rex->subbeg;
4426 re->sublen = rex->sublen;
4429 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4430 "Matching embedded");
4432 startpoint = rei->program + 1;
4433 ST.close_paren = 0; /* only used for GOSUB */
4434 /* borrowed from regtry */
4435 if (PL_reg_start_tmpl <= re->nparens) {
4436 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4437 if(PL_reg_start_tmp)
4438 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4440 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4443 eval_recurse_doit: /* Share code with GOSUB below this line */
4444 /* run the pattern returned from (??{...}) */
4445 ST.cp = regcppush(0); /* Save *all* the positions. */
4446 REGCP_SET(ST.lastcp);
4448 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4450 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4451 PL_reglastparen = &re->lastparen;
4452 PL_reglastcloseparen = &re->lastcloseparen;
4454 re->lastcloseparen = 0;
4456 PL_reginput = locinput;
4459 /* XXXX This is too dramatic a measure... */
4462 ST.toggle_reg_flags = PL_reg_flags;
4464 PL_reg_flags |= RF_utf8;
4466 PL_reg_flags &= ~RF_utf8;
4467 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4469 ST.prev_rex = rex_sv;
4470 ST.prev_curlyx = cur_curlyx;
4471 SETREX(rex_sv,re_sv);
4476 ST.prev_eval = cur_eval;
4478 /* now continue from first node in postoned RE */
4479 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4482 /* logical is 1, /(?(?{...})X|Y)/ */
4483 sw = cBOOL(SvTRUE(ret));
4488 case EVAL_AB: /* cleanup after a successful (??{A})B */
4489 /* note: this is called twice; first after popping B, then A */
4490 PL_reg_flags ^= ST.toggle_reg_flags;
4491 ReREFCNT_dec(rex_sv);
4492 SETREX(rex_sv,ST.prev_rex);
4493 rex = (struct regexp *)SvANY(rex_sv);
4494 rexi = RXi_GET(rex);
4496 cur_eval = ST.prev_eval;
4497 cur_curlyx = ST.prev_curlyx;
4499 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4500 PL_reglastparen = &rex->lastparen;
4501 PL_reglastcloseparen = &rex->lastcloseparen;
4502 /* also update PL_regoffs */
4503 PL_regoffs = rex->offs;
4505 /* XXXX This is too dramatic a measure... */
4507 if ( nochange_depth )
4512 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4513 /* note: this is called twice; first after popping B, then A */
4514 PL_reg_flags ^= ST.toggle_reg_flags;
4515 ReREFCNT_dec(rex_sv);
4516 SETREX(rex_sv,ST.prev_rex);
4517 rex = (struct regexp *)SvANY(rex_sv);
4518 rexi = RXi_GET(rex);
4519 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4520 PL_reglastparen = &rex->lastparen;
4521 PL_reglastcloseparen = &rex->lastcloseparen;
4523 PL_reginput = locinput;
4524 REGCP_UNWIND(ST.lastcp);
4526 cur_eval = ST.prev_eval;
4527 cur_curlyx = ST.prev_curlyx;
4528 /* XXXX This is too dramatic a measure... */
4530 if ( nochange_depth )
4536 n = ARG(scan); /* which paren pair */
4537 PL_reg_start_tmp[n] = locinput;
4543 n = ARG(scan); /* which paren pair */
4544 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4545 PL_regoffs[n].end = locinput - PL_bostr;
4546 /*if (n > PL_regsize)
4548 if (n > *PL_reglastparen)
4549 *PL_reglastparen = n;
4550 *PL_reglastcloseparen = n;
4551 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4559 cursor && OP(cursor)!=END;
4560 cursor=regnext(cursor))
4562 if ( OP(cursor)==CLOSE ){
4564 if ( n <= lastopen ) {
4566 = PL_reg_start_tmp[n] - PL_bostr;
4567 PL_regoffs[n].end = locinput - PL_bostr;
4568 /*if (n > PL_regsize)
4570 if (n > *PL_reglastparen)
4571 *PL_reglastparen = n;
4572 *PL_reglastcloseparen = n;
4573 if ( n == ARG(scan) || (cur_eval &&
4574 cur_eval->u.eval.close_paren == n))
4583 n = ARG(scan); /* which paren pair */
4584 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4587 /* reg_check_named_buff_matched returns 0 for no match */
4588 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4592 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4598 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4600 next = NEXTOPER(NEXTOPER(scan));
4602 next = scan + ARG(scan);
4603 if (OP(next) == IFTHEN) /* Fake one. */
4604 next = NEXTOPER(NEXTOPER(next));
4608 logical = scan->flags;
4611 /*******************************************************************
4613 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4614 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4615 STAR/PLUS/CURLY/CURLYN are used instead.)
4617 A*B is compiled as <CURLYX><A><WHILEM><B>
4619 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4620 state, which contains the current count, initialised to -1. It also sets
4621 cur_curlyx to point to this state, with any previous value saved in the
4624 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4625 since the pattern may possibly match zero times (i.e. it's a while {} loop
4626 rather than a do {} while loop).
4628 Each entry to WHILEM represents a successful match of A. The count in the
4629 CURLYX block is incremented, another WHILEM state is pushed, and execution
4630 passes to A or B depending on greediness and the current count.
4632 For example, if matching against the string a1a2a3b (where the aN are
4633 substrings that match /A/), then the match progresses as follows: (the
4634 pushed states are interspersed with the bits of strings matched so far):
4637 <CURLYX cnt=0><WHILEM>
4638 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4639 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4640 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4641 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4643 (Contrast this with something like CURLYM, which maintains only a single
4647 a1 <CURLYM cnt=1> a2
4648 a1 a2 <CURLYM cnt=2> a3
4649 a1 a2 a3 <CURLYM cnt=3> b
4652 Each WHILEM state block marks a point to backtrack to upon partial failure
4653 of A or B, and also contains some minor state data related to that
4654 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4655 overall state, such as the count, and pointers to the A and B ops.
4657 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4658 must always point to the *current* CURLYX block, the rules are:
4660 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4661 and set cur_curlyx to point the new block.
4663 When popping the CURLYX block after a successful or unsuccessful match,
4664 restore the previous cur_curlyx.
4666 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4667 to the outer one saved in the CURLYX block.
4669 When popping the WHILEM block after a successful or unsuccessful B match,
4670 restore the previous cur_curlyx.
4672 Here's an example for the pattern (AI* BI)*BO
4673 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4676 curlyx backtrack stack
4677 ------ ---------------
4679 CO <CO prev=NULL> <WO>
4680 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4681 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4682 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4684 At this point the pattern succeeds, and we work back down the stack to
4685 clean up, restoring as we go:
4687 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4688 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4689 CO <CO prev=NULL> <WO>
4692 *******************************************************************/
4694 #define ST st->u.curlyx
4696 case CURLYX: /* start of /A*B/ (for complex A) */
4698 /* No need to save/restore up to this paren */
4699 I32 parenfloor = scan->flags;
4701 assert(next); /* keep Coverity happy */
4702 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4705 /* XXXX Probably it is better to teach regpush to support
4706 parenfloor > PL_regsize... */
4707 if (parenfloor > (I32)*PL_reglastparen)
4708 parenfloor = *PL_reglastparen; /* Pessimization... */
4710 ST.prev_curlyx= cur_curlyx;
4712 ST.cp = PL_savestack_ix;
4714 /* these fields contain the state of the current curly.
4715 * they are accessed by subsequent WHILEMs */
4716 ST.parenfloor = parenfloor;
4721 ST.count = -1; /* this will be updated by WHILEM */
4722 ST.lastloc = NULL; /* this will be updated by WHILEM */
4724 PL_reginput = locinput;
4725 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4729 case CURLYX_end: /* just finished matching all of A*B */
4730 cur_curlyx = ST.prev_curlyx;
4734 case CURLYX_end_fail: /* just failed to match all of A*B */
4736 cur_curlyx = ST.prev_curlyx;
4742 #define ST st->u.whilem
4744 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4746 /* see the discussion above about CURLYX/WHILEM */
4748 int min = ARG1(cur_curlyx->u.curlyx.me);
4749 int max = ARG2(cur_curlyx->u.curlyx.me);
4750 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4752 assert(cur_curlyx); /* keep Coverity happy */
4753 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4754 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4755 ST.cache_offset = 0;
4758 PL_reginput = locinput;
4760 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4761 "%*s whilem: matched %ld out of %d..%d\n",
4762 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4765 /* First just match a string of min A's. */
4768 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4769 cur_curlyx->u.curlyx.lastloc = locinput;
4770 REGCP_SET(ST.lastcp);
4772 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4776 /* If degenerate A matches "", assume A done. */
4778 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4779 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4780 "%*s whilem: empty match detected, trying continuation...\n",
4781 REPORT_CODE_OFF+depth*2, "")
4783 goto do_whilem_B_max;
4786 /* super-linear cache processing */
4790 if (!PL_reg_maxiter) {
4791 /* start the countdown: Postpone detection until we
4792 * know the match is not *that* much linear. */
4793 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4794 /* possible overflow for long strings and many CURLYX's */
4795 if (PL_reg_maxiter < 0)
4796 PL_reg_maxiter = I32_MAX;
4797 PL_reg_leftiter = PL_reg_maxiter;
4800 if (PL_reg_leftiter-- == 0) {
4801 /* initialise cache */
4802 const I32 size = (PL_reg_maxiter + 7)/8;
4803 if (PL_reg_poscache) {
4804 if ((I32)PL_reg_poscache_size < size) {
4805 Renew(PL_reg_poscache, size, char);
4806 PL_reg_poscache_size = size;
4808 Zero(PL_reg_poscache, size, char);
4811 PL_reg_poscache_size = size;
4812 Newxz(PL_reg_poscache, size, char);
4814 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4815 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4816 PL_colors[4], PL_colors[5])
4820 if (PL_reg_leftiter < 0) {
4821 /* have we already failed at this position? */
4823 offset = (scan->flags & 0xf) - 1
4824 + (locinput - PL_bostr) * (scan->flags>>4);
4825 mask = 1 << (offset % 8);
4827 if (PL_reg_poscache[offset] & mask) {
4828 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4829 "%*s whilem: (cache) already tried at this position...\n",
4830 REPORT_CODE_OFF+depth*2, "")
4832 sayNO; /* cache records failure */
4834 ST.cache_offset = offset;
4835 ST.cache_mask = mask;
4839 /* Prefer B over A for minimal matching. */
4841 if (cur_curlyx->u.curlyx.minmod) {
4842 ST.save_curlyx = cur_curlyx;
4843 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4844 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4845 REGCP_SET(ST.lastcp);
4846 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4850 /* Prefer A over B for maximal matching. */
4852 if (n < max) { /* More greed allowed? */
4853 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4854 cur_curlyx->u.curlyx.lastloc = locinput;
4855 REGCP_SET(ST.lastcp);
4856 PUSH_STATE_GOTO(WHILEM_A_max, A);
4859 goto do_whilem_B_max;
4863 case WHILEM_B_min: /* just matched B in a minimal match */
4864 case WHILEM_B_max: /* just matched B in a maximal match */
4865 cur_curlyx = ST.save_curlyx;
4869 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4870 cur_curlyx = ST.save_curlyx;
4871 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4872 cur_curlyx->u.curlyx.count--;
4876 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4878 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4879 REGCP_UNWIND(ST.lastcp);
4881 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4882 cur_curlyx->u.curlyx.count--;
4886 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4887 REGCP_UNWIND(ST.lastcp);
4888 regcppop(rex); /* Restore some previous $<digit>s? */
4889 PL_reginput = locinput;
4890 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4891 "%*s whilem: failed, trying continuation...\n",
4892 REPORT_CODE_OFF+depth*2, "")
4895 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4896 && ckWARN(WARN_REGEXP)
4897 && !(PL_reg_flags & RF_warned))
4899 PL_reg_flags |= RF_warned;
4900 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4901 "Complex regular subexpression recursion limit (%d) "
4907 ST.save_curlyx = cur_curlyx;
4908 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4909 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4912 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4913 cur_curlyx = ST.save_curlyx;
4914 REGCP_UNWIND(ST.lastcp);
4917 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4918 /* Maximum greed exceeded */
4919 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4920 && ckWARN(WARN_REGEXP)
4921 && !(PL_reg_flags & RF_warned))
4923 PL_reg_flags |= RF_warned;
4924 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4925 "Complex regular subexpression recursion "
4926 "limit (%d) exceeded",
4929 cur_curlyx->u.curlyx.count--;
4933 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4934 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4936 /* Try grabbing another A and see if it helps. */
4937 PL_reginput = locinput;
4938 cur_curlyx->u.curlyx.lastloc = locinput;
4939 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4940 REGCP_SET(ST.lastcp);
4941 PUSH_STATE_GOTO(WHILEM_A_min,
4942 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4946 #define ST st->u.branch
4948 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4949 next = scan + ARG(scan);
4952 scan = NEXTOPER(scan);
4955 case BRANCH: /* /(...|A|...)/ */
4956 scan = NEXTOPER(scan); /* scan now points to inner node */
4957 ST.lastparen = *PL_reglastparen;
4958 ST.next_branch = next;
4960 PL_reginput = locinput;
4962 /* Now go into the branch */
4964 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4966 PUSH_STATE_GOTO(BRANCH_next, scan);
4970 PL_reginput = locinput;
4971 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4972 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4973 PUSH_STATE_GOTO(CUTGROUP_next,next);
4975 case CUTGROUP_next_fail:
4978 if (st->u.mark.mark_name)
4979 sv_commit = st->u.mark.mark_name;
4985 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4990 REGCP_UNWIND(ST.cp);
4991 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4992 PL_regoffs[n].end = -1;
4993 *PL_reglastparen = n;
4994 /*dmq: *PL_reglastcloseparen = n; */
4995 scan = ST.next_branch;
4996 /* no more branches? */
4997 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4999 PerlIO_printf( Perl_debug_log,
5000 "%*s %sBRANCH failed...%s\n",
5001 REPORT_CODE_OFF+depth*2, "",
5007 continue; /* execute next BRANCH[J] op */
5015 #define ST st->u.curlym
5017 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5019 /* This is an optimisation of CURLYX that enables us to push
5020 * only a single backtracking state, no matter how many matches
5021 * there are in {m,n}. It relies on the pattern being constant
5022 * length, with no parens to influence future backrefs
5026 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5028 /* if paren positive, emulate an OPEN/CLOSE around A */
5030 U32 paren = ST.me->flags;
5031 if (paren > PL_regsize)
5033 if (paren > *PL_reglastparen)
5034 *PL_reglastparen = paren;
5035 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5043 ST.c1 = CHRTEST_UNINIT;
5046 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5049 curlym_do_A: /* execute the A in /A{m,n}B/ */
5050 PL_reginput = locinput;
5051 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
5054 case CURLYM_A: /* we've just matched an A */
5055 locinput = st->locinput;
5056 nextchr = UCHARAT(locinput);
5059 /* after first match, determine A's length: u.curlym.alen */
5060 if (ST.count == 1) {
5061 if (PL_reg_match_utf8) {
5063 while (s < PL_reginput) {
5069 ST.alen = PL_reginput - locinput;
5072 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5075 PerlIO_printf(Perl_debug_log,
5076 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5077 (int)(REPORT_CODE_OFF+(depth*2)), "",
5078 (IV) ST.count, (IV)ST.alen)
5081 locinput = PL_reginput;
5083 if (cur_eval && cur_eval->u.eval.close_paren &&
5084 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5088 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5089 if ( max == REG_INFTY || ST.count < max )
5090 goto curlym_do_A; /* try to match another A */
5092 goto curlym_do_B; /* try to match B */
5094 case CURLYM_A_fail: /* just failed to match an A */
5095 REGCP_UNWIND(ST.cp);
5097 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5098 || (cur_eval && cur_eval->u.eval.close_paren &&
5099 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5102 curlym_do_B: /* execute the B in /A{m,n}B/ */
5103 PL_reginput = locinput;
5104 if (ST.c1 == CHRTEST_UNINIT) {
5105 /* calculate c1 and c2 for possible match of 1st char
5106 * following curly */
5107 ST.c1 = ST.c2 = CHRTEST_VOID;
5108 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5109 regnode *text_node = ST.B;
5110 if (! HAS_TEXT(text_node))
5111 FIND_NEXT_IMPT(text_node);
5114 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5116 But the former is redundant in light of the latter.
5118 if this changes back then the macro for
5119 IS_TEXT and friends need to change.
5121 if (PL_regkind[OP(text_node)] == EXACT)
5124 ST.c1 = (U8)*STRING(text_node);
5125 switch (OP(text_node)) {
5126 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5129 case EXACTFU_TRICKYFOLD:
5130 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5131 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5132 default: ST.c2 = ST.c1;
5139 PerlIO_printf(Perl_debug_log,
5140 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5141 (int)(REPORT_CODE_OFF+(depth*2)),
5144 if (ST.c1 != CHRTEST_VOID
5145 && UCHARAT(PL_reginput) != ST.c1
5146 && UCHARAT(PL_reginput) != ST.c2)
5148 /* simulate B failing */
5150 PerlIO_printf(Perl_debug_log,
5151 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5152 (int)(REPORT_CODE_OFF+(depth*2)),"",
5155 state_num = CURLYM_B_fail;
5156 goto reenter_switch;
5160 /* mark current A as captured */
5161 I32 paren = ST.me->flags;
5163 PL_regoffs[paren].start
5164 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5165 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5166 /*dmq: *PL_reglastcloseparen = paren; */
5169 PL_regoffs[paren].end = -1;
5170 if (cur_eval && cur_eval->u.eval.close_paren &&
5171 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5180 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5183 case CURLYM_B_fail: /* just failed to match a B */
5184 REGCP_UNWIND(ST.cp);
5186 I32 max = ARG2(ST.me);
5187 if (max != REG_INFTY && ST.count == max)
5189 goto curlym_do_A; /* try to match a further A */
5191 /* backtrack one A */
5192 if (ST.count == ARG1(ST.me) /* min */)
5195 locinput = HOPc(locinput, -ST.alen);
5196 goto curlym_do_B; /* try to match B */
5199 #define ST st->u.curly
5201 #define CURLY_SETPAREN(paren, success) \
5204 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5205 PL_regoffs[paren].end = locinput - PL_bostr; \
5206 *PL_reglastcloseparen = paren; \
5209 PL_regoffs[paren].end = -1; \
5212 case STAR: /* /A*B/ where A is width 1 */
5216 scan = NEXTOPER(scan);
5218 case PLUS: /* /A+B/ where A is width 1 */
5222 scan = NEXTOPER(scan);
5224 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5225 ST.paren = scan->flags; /* Which paren to set */
5226 if (ST.paren > PL_regsize)
5227 PL_regsize = ST.paren;
5228 if (ST.paren > *PL_reglastparen)
5229 *PL_reglastparen = ST.paren;
5230 ST.min = ARG1(scan); /* min to match */
5231 ST.max = ARG2(scan); /* max to match */
5232 if (cur_eval && cur_eval->u.eval.close_paren &&
5233 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5237 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5239 case CURLY: /* /A{m,n}B/ where A is width 1 */
5241 ST.min = ARG1(scan); /* min to match */
5242 ST.max = ARG2(scan); /* max to match */
5243 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5246 * Lookahead to avoid useless match attempts
5247 * when we know what character comes next.
5249 * Used to only do .*x and .*?x, but now it allows
5250 * for )'s, ('s and (?{ ... })'s to be in the way
5251 * of the quantifier and the EXACT-like node. -- japhy
5254 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5256 if (HAS_TEXT(next) || JUMPABLE(next)) {
5258 regnode *text_node = next;
5260 if (! HAS_TEXT(text_node))
5261 FIND_NEXT_IMPT(text_node);
5263 if (! HAS_TEXT(text_node))
5264 ST.c1 = ST.c2 = CHRTEST_VOID;
5266 if ( PL_regkind[OP(text_node)] != EXACT ) {
5267 ST.c1 = ST.c2 = CHRTEST_VOID;
5268 goto assume_ok_easy;
5271 s = (U8*)STRING(text_node);
5273 /* Currently we only get here when
5275 PL_rekind[OP(text_node)] == EXACT
5277 if this changes back then the macro for IS_TEXT and
5278 friends need to change. */
5281 switch (OP(text_node)) {
5282 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5285 case EXACTFU_TRICKYFOLD:
5286 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5287 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5288 default: ST.c2 = ST.c1; break;
5291 else { /* UTF_PATTERN */
5292 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5294 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5296 to_utf8_fold((U8*)s, tmpbuf, &ulen);
5297 ST.c1 = ST.c2 = utf8n_to_uvchr(tmpbuf, UTF8_MAXLEN, 0,
5301 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5308 ST.c1 = ST.c2 = CHRTEST_VOID;
5313 PL_reginput = locinput;
5316 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5319 locinput = PL_reginput;
5321 if (ST.c1 == CHRTEST_VOID)
5322 goto curly_try_B_min;
5324 ST.oldloc = locinput;
5326 /* set ST.maxpos to the furthest point along the
5327 * string that could possibly match */
5328 if (ST.max == REG_INFTY) {
5329 ST.maxpos = PL_regeol - 1;
5331 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5334 else if (utf8_target) {
5335 int m = ST.max - ST.min;
5336 for (ST.maxpos = locinput;
5337 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5338 ST.maxpos += UTF8SKIP(ST.maxpos);
5341 ST.maxpos = locinput + ST.max - ST.min;
5342 if (ST.maxpos >= PL_regeol)
5343 ST.maxpos = PL_regeol - 1;
5345 goto curly_try_B_min_known;
5349 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5350 locinput = PL_reginput;
5351 if (ST.count < ST.min)
5353 if ((ST.count > ST.min)
5354 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5356 /* A{m,n} must come at the end of the string, there's
5357 * no point in backing off ... */
5359 /* ...except that $ and \Z can match before *and* after
5360 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5361 We may back off by one in this case. */
5362 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5366 goto curly_try_B_max;
5371 case CURLY_B_min_known_fail:
5372 /* failed to find B in a non-greedy match where c1,c2 valid */
5373 if (ST.paren && ST.count)
5374 PL_regoffs[ST.paren].end = -1;
5376 PL_reginput = locinput; /* Could be reset... */
5377 REGCP_UNWIND(ST.cp);
5378 /* Couldn't or didn't -- move forward. */
5379 ST.oldloc = locinput;
5381 locinput += UTF8SKIP(locinput);
5385 curly_try_B_min_known:
5386 /* find the next place where 'B' could work, then call B */
5390 n = (ST.oldloc == locinput) ? 0 : 1;
5391 if (ST.c1 == ST.c2) {
5393 /* set n to utf8_distance(oldloc, locinput) */
5394 while (locinput <= ST.maxpos &&
5395 utf8n_to_uvchr((U8*)locinput,
5396 UTF8_MAXBYTES, &len,
5397 uniflags) != (UV)ST.c1) {
5403 /* set n to utf8_distance(oldloc, locinput) */
5404 while (locinput <= ST.maxpos) {
5406 const UV c = utf8n_to_uvchr((U8*)locinput,
5407 UTF8_MAXBYTES, &len,
5409 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5417 if (ST.c1 == ST.c2) {
5418 while (locinput <= ST.maxpos &&
5419 UCHARAT(locinput) != ST.c1)
5423 while (locinput <= ST.maxpos
5424 && UCHARAT(locinput) != ST.c1
5425 && UCHARAT(locinput) != ST.c2)
5428 n = locinput - ST.oldloc;
5430 if (locinput > ST.maxpos)
5432 /* PL_reginput == oldloc now */
5435 if (regrepeat(rex, ST.A, n, depth) < n)
5438 PL_reginput = locinput;
5439 CURLY_SETPAREN(ST.paren, ST.count);
5440 if (cur_eval && cur_eval->u.eval.close_paren &&
5441 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5444 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5449 case CURLY_B_min_fail:
5450 /* failed to find B in a non-greedy match where c1,c2 invalid */
5451 if (ST.paren && ST.count)
5452 PL_regoffs[ST.paren].end = -1;
5454 REGCP_UNWIND(ST.cp);
5455 /* failed -- move forward one */
5456 PL_reginput = locinput;
5457 if (regrepeat(rex, ST.A, 1, depth)) {
5459 locinput = PL_reginput;
5460 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5461 ST.count > 0)) /* count overflow ? */
5464 CURLY_SETPAREN(ST.paren, ST.count);
5465 if (cur_eval && cur_eval->u.eval.close_paren &&
5466 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5469 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5477 /* a successful greedy match: now try to match B */
5478 if (cur_eval && cur_eval->u.eval.close_paren &&
5479 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5484 if (ST.c1 != CHRTEST_VOID)
5485 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5486 UTF8_MAXBYTES, 0, uniflags)
5487 : (UV) UCHARAT(PL_reginput);
5488 /* If it could work, try it. */
5489 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5490 CURLY_SETPAREN(ST.paren, ST.count);
5491 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5496 case CURLY_B_max_fail:
5497 /* failed to find B in a greedy match */
5498 if (ST.paren && ST.count)
5499 PL_regoffs[ST.paren].end = -1;
5501 REGCP_UNWIND(ST.cp);
5503 if (--ST.count < ST.min)
5505 PL_reginput = locinput = HOPc(locinput, -1);
5506 goto curly_try_B_max;
5513 /* we've just finished A in /(??{A})B/; now continue with B */
5515 st->u.eval.toggle_reg_flags
5516 = cur_eval->u.eval.toggle_reg_flags;
5517 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5519 st->u.eval.prev_rex = rex_sv; /* inner */
5520 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5521 rex = (struct regexp *)SvANY(rex_sv);
5522 rexi = RXi_GET(rex);
5523 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5524 (void)ReREFCNT_inc(rex_sv);
5525 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5527 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5528 PL_reglastparen = &rex->lastparen;
5529 PL_reglastcloseparen = &rex->lastcloseparen;
5531 REGCP_SET(st->u.eval.lastcp);
5532 PL_reginput = locinput;
5534 /* Restore parens of the outer rex without popping the
5536 tmpix = PL_savestack_ix;
5537 PL_savestack_ix = cur_eval->u.eval.lastcp;
5539 PL_savestack_ix = tmpix;
5541 st->u.eval.prev_eval = cur_eval;
5542 cur_eval = cur_eval->u.eval.prev_eval;
5544 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5545 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5546 if ( nochange_depth )
5549 PUSH_YES_STATE_GOTO(EVAL_AB,
5550 st->u.eval.prev_eval->u.eval.B); /* match B */
5553 if (locinput < reginfo->till) {
5554 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5555 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5557 (long)(locinput - PL_reg_starttry),
5558 (long)(reginfo->till - PL_reg_starttry),
5561 sayNO_SILENT; /* Cannot match: too short. */
5563 PL_reginput = locinput; /* put where regtry can find it */
5564 sayYES; /* Success! */
5566 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5568 PerlIO_printf(Perl_debug_log,
5569 "%*s %ssubpattern success...%s\n",
5570 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5571 PL_reginput = locinput; /* put where regtry can find it */
5572 sayYES; /* Success! */
5575 #define ST st->u.ifmatch
5577 case SUSPEND: /* (?>A) */
5579 PL_reginput = locinput;
5582 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5584 goto ifmatch_trivial_fail_test;
5586 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5588 ifmatch_trivial_fail_test:
5590 char * const s = HOPBACKc(locinput, scan->flags);
5595 sw = 1 - cBOOL(ST.wanted);
5599 next = scan + ARG(scan);
5607 PL_reginput = locinput;
5611 ST.logical = logical;
5612 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5614 /* execute body of (?...A) */
5615 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5618 case IFMATCH_A_fail: /* body of (?...A) failed */
5619 ST.wanted = !ST.wanted;
5622 case IFMATCH_A: /* body of (?...A) succeeded */
5624 sw = cBOOL(ST.wanted);
5626 else if (!ST.wanted)
5629 if (OP(ST.me) == SUSPEND)
5630 locinput = PL_reginput;
5632 locinput = PL_reginput = st->locinput;
5633 nextchr = UCHARAT(locinput);
5635 scan = ST.me + ARG(ST.me);
5638 continue; /* execute B */
5643 next = scan + ARG(scan);
5648 reginfo->cutpoint = PL_regeol;
5651 PL_reginput = locinput;
5653 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5654 PUSH_STATE_GOTO(COMMIT_next,next);
5656 case COMMIT_next_fail:
5663 #define ST st->u.mark
5665 ST.prev_mark = mark_state;
5666 ST.mark_name = sv_commit = sv_yes_mark
5667 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5669 ST.mark_loc = PL_reginput = locinput;
5670 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5672 case MARKPOINT_next:
5673 mark_state = ST.prev_mark;
5676 case MARKPOINT_next_fail:
5677 if (popmark && sv_eq(ST.mark_name,popmark))
5679 if (ST.mark_loc > startpoint)
5680 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5681 popmark = NULL; /* we found our mark */
5682 sv_commit = ST.mark_name;
5685 PerlIO_printf(Perl_debug_log,
5686 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5687 REPORT_CODE_OFF+depth*2, "",
5688 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5691 mark_state = ST.prev_mark;
5692 sv_yes_mark = mark_state ?
5693 mark_state->u.mark.mark_name : NULL;
5697 PL_reginput = locinput;
5699 /* (*SKIP) : if we fail we cut here*/
5700 ST.mark_name = NULL;
5701 ST.mark_loc = locinput;
5702 PUSH_STATE_GOTO(SKIP_next,next);
5704 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5705 otherwise do nothing. Meaning we need to scan
5707 regmatch_state *cur = mark_state;
5708 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5711 if ( sv_eq( cur->u.mark.mark_name,
5714 ST.mark_name = find;
5715 PUSH_STATE_GOTO( SKIP_next, next );
5717 cur = cur->u.mark.prev_mark;
5720 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5722 case SKIP_next_fail:
5724 /* (*CUT:NAME) - Set up to search for the name as we
5725 collapse the stack*/
5726 popmark = ST.mark_name;
5728 /* (*CUT) - No name, we cut here.*/
5729 if (ST.mark_loc > startpoint)
5730 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5731 /* but we set sv_commit to latest mark_name if there
5732 is one so they can test to see how things lead to this
5735 sv_commit=mark_state->u.mark.mark_name;
5742 if ((n=is_LNBREAK(locinput,utf8_target))) {
5744 nextchr = UCHARAT(locinput);
5749 #define CASE_CLASS(nAmE) \
5751 if (locinput >= PL_regeol) \
5753 if ((n=is_##nAmE(locinput,utf8_target))) { \
5755 nextchr = UCHARAT(locinput); \
5760 if (locinput >= PL_regeol) \
5762 if ((n=is_##nAmE(locinput,utf8_target))) { \
5765 locinput += UTF8SKIP(locinput); \
5766 nextchr = UCHARAT(locinput); \
5771 CASE_CLASS(HORIZWS);
5775 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5776 PTR2UV(scan), OP(scan));
5777 Perl_croak(aTHX_ "regexp memory corruption");
5781 /* switch break jumps here */
5782 scan = next; /* prepare to execute the next op and ... */
5783 continue; /* ... jump back to the top, reusing st */
5787 /* push a state that backtracks on success */
5788 st->u.yes.prev_yes_state = yes_state;
5792 /* push a new regex state, then continue at scan */
5794 regmatch_state *newst;
5797 regmatch_state *cur = st;
5798 regmatch_state *curyes = yes_state;
5800 regmatch_slab *slab = PL_regmatch_slab;
5801 for (;curd > -1;cur--,curd--) {
5802 if (cur < SLAB_FIRST(slab)) {
5804 cur = SLAB_LAST(slab);
5806 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5807 REPORT_CODE_OFF + 2 + depth * 2,"",
5808 curd, PL_reg_name[cur->resume_state],
5809 (curyes == cur) ? "yes" : ""
5812 curyes = cur->u.yes.prev_yes_state;
5815 DEBUG_STATE_pp("push")
5818 st->locinput = locinput;
5820 if (newst > SLAB_LAST(PL_regmatch_slab))
5821 newst = S_push_slab(aTHX);
5822 PL_regmatch_state = newst;
5824 locinput = PL_reginput;
5825 nextchr = UCHARAT(locinput);
5833 * We get here only if there's trouble -- normally "case END" is
5834 * the terminating point.
5836 Perl_croak(aTHX_ "corrupted regexp pointers");
5842 /* we have successfully completed a subexpression, but we must now
5843 * pop to the state marked by yes_state and continue from there */
5844 assert(st != yes_state);
5846 while (st != yes_state) {
5848 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5849 PL_regmatch_slab = PL_regmatch_slab->prev;
5850 st = SLAB_LAST(PL_regmatch_slab);
5854 DEBUG_STATE_pp("pop (no final)");
5856 DEBUG_STATE_pp("pop (yes)");
5862 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5863 || yes_state > SLAB_LAST(PL_regmatch_slab))
5865 /* not in this slab, pop slab */
5866 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5867 PL_regmatch_slab = PL_regmatch_slab->prev;
5868 st = SLAB_LAST(PL_regmatch_slab);
5870 depth -= (st - yes_state);
5873 yes_state = st->u.yes.prev_yes_state;
5874 PL_regmatch_state = st;
5877 locinput= st->locinput;
5878 nextchr = UCHARAT(locinput);
5880 state_num = st->resume_state + no_final;
5881 goto reenter_switch;
5884 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5885 PL_colors[4], PL_colors[5]));
5887 if (PL_reg_eval_set) {
5888 /* each successfully executed (?{...}) block does the equivalent of
5889 * local $^R = do {...}
5890 * When popping the save stack, all these locals would be undone;
5891 * bypass this by setting the outermost saved $^R to the latest
5893 if (oreplsv != GvSV(PL_replgv))
5894 sv_setsv(oreplsv, GvSV(PL_replgv));
5901 PerlIO_printf(Perl_debug_log,
5902 "%*s %sfailed...%s\n",
5903 REPORT_CODE_OFF+depth*2, "",
5904 PL_colors[4], PL_colors[5])
5916 /* there's a previous state to backtrack to */
5918 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5919 PL_regmatch_slab = PL_regmatch_slab->prev;
5920 st = SLAB_LAST(PL_regmatch_slab);
5922 PL_regmatch_state = st;
5923 locinput= st->locinput;
5924 nextchr = UCHARAT(locinput);
5926 DEBUG_STATE_pp("pop");
5928 if (yes_state == st)
5929 yes_state = st->u.yes.prev_yes_state;
5931 state_num = st->resume_state + 1; /* failure = success + 1 */
5932 goto reenter_switch;
5937 if (rex->intflags & PREGf_VERBARG_SEEN) {
5938 SV *sv_err = get_sv("REGERROR", 1);
5939 SV *sv_mrk = get_sv("REGMARK", 1);
5941 sv_commit = &PL_sv_no;
5943 sv_yes_mark = &PL_sv_yes;
5946 sv_commit = &PL_sv_yes;
5947 sv_yes_mark = &PL_sv_no;
5949 sv_setsv(sv_err, sv_commit);
5950 sv_setsv(sv_mrk, sv_yes_mark);
5953 /* clean up; in particular, free all slabs above current one */
5954 LEAVE_SCOPE(oldsave);
5960 - regrepeat - repeatedly match something simple, report how many
5963 * [This routine now assumes that it will only match on things of length 1.
5964 * That was true before, but now we assume scan - reginput is the count,
5965 * rather than incrementing count on every character. [Er, except utf8.]]
5968 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5971 register char *scan;
5973 register char *loceol = PL_regeol;
5974 register I32 hardcount = 0;
5975 register bool utf8_target = PL_reg_match_utf8;
5978 PERL_UNUSED_ARG(depth);
5981 PERL_ARGS_ASSERT_REGREPEAT;
5984 if (max == REG_INFTY)
5986 else if (max < loceol - scan)
5987 loceol = scan + max;
5992 while (scan < loceol && hardcount < max && *scan != '\n') {
5993 scan += UTF8SKIP(scan);
5997 while (scan < loceol && *scan != '\n')
6004 while (scan < loceol && hardcount < max) {
6005 scan += UTF8SKIP(scan);
6016 /* To get here, EXACTish nodes must have *byte* length == 1. That
6017 * means they match only characters in the string that can be expressed
6018 * as a single byte. For non-utf8 strings, that means a simple match.
6019 * For utf8 strings, the character matched must be an invariant, or
6020 * downgradable to a single byte. The pattern's utf8ness is
6021 * irrelevant, as since it's a single byte, it either isn't utf8, or if
6022 * it is, it's an invariant */
6025 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6027 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6028 while (scan < loceol && UCHARAT(scan) == c) {
6034 /* Here, the string is utf8, and the pattern char is different
6035 * in utf8 than not, so can't compare them directly. Outside the
6036 * loop, find the two utf8 bytes that represent c, and then
6037 * look for those in sequence in the utf8 string */
6038 U8 high = UTF8_TWO_BYTE_HI(c);
6039 U8 low = UTF8_TWO_BYTE_LO(c);
6042 while (hardcount < max
6043 && scan + 1 < loceol
6044 && UCHARAT(scan) == high
6045 && UCHARAT(scan + 1) == low)
6053 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6057 PL_reg_flags |= RF_tainted;
6058 utf8_flags = FOLDEQ_UTF8_LOCALE;
6066 case EXACTFU_TRICKYFOLD:
6068 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6070 /* The comments for the EXACT case above apply as well to these fold
6075 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6077 if (utf8_target || OP(p) == EXACTFU_SS) { /* Use full Unicode fold matching */
6078 char *tmpeol = loceol;
6079 while (hardcount < max
6080 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6081 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6088 /* XXX Note that the above handles properly the German sharp s in
6089 * the pattern matching ss in the string. But it doesn't handle
6090 * properly cases where the string contains say 'LIGATURE ff' and
6091 * the pattern is 'f+'. This would require, say, a new function or
6092 * revised interface to foldEQ_utf8(), in which the maximum number
6093 * of characters to match could be passed and it would return how
6094 * many actually did. This is just one of many cases where
6095 * multi-char folds don't work properly, and so the fix is being
6101 /* Here, the string isn't utf8 and c is a single byte; and either
6102 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6103 * doesn't affect c. Can just do simple comparisons for exact or
6106 case EXACTF: folded = PL_fold[c]; break;
6108 case EXACTFU_TRICKYFOLD:
6109 case EXACTFU: folded = PL_fold_latin1[c]; break;
6110 case EXACTFL: folded = PL_fold_locale[c]; break;
6111 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6113 while (scan < loceol &&
6114 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6122 if (utf8_target || OP(p) == ANYOFV) {
6125 inclasslen = loceol - scan;
6126 while (hardcount < max
6127 && ((inclasslen = loceol - scan) > 0)
6128 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6134 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6142 LOAD_UTF8_CHARCLASS_ALNUM();
6143 while (hardcount < max && scan < loceol &&
6144 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6146 scan += UTF8SKIP(scan);
6150 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6158 while (scan < loceol && isALNUM((U8) *scan)) {
6163 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6168 PL_reg_flags |= RF_tainted;
6171 while (hardcount < max && scan < loceol &&
6172 isALNUM_LC_utf8((U8*)scan)) {
6173 scan += UTF8SKIP(scan);
6177 while (scan < loceol && isALNUM_LC(*scan))
6187 LOAD_UTF8_CHARCLASS_ALNUM();
6188 while (hardcount < max && scan < loceol &&
6189 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6191 scan += UTF8SKIP(scan);
6195 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6202 goto utf8_Nwordchar;
6203 while (scan < loceol && ! isALNUM((U8) *scan)) {
6209 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6210 scan += UTF8SKIP(scan);
6214 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6220 PL_reg_flags |= RF_tainted;
6223 while (hardcount < max && scan < loceol &&
6224 !isALNUM_LC_utf8((U8*)scan)) {
6225 scan += UTF8SKIP(scan);
6229 while (scan < loceol && !isALNUM_LC(*scan))
6239 LOAD_UTF8_CHARCLASS_SPACE();
6240 while (hardcount < max && scan < loceol &&
6242 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6244 scan += UTF8SKIP(scan);
6250 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6259 while (scan < loceol && isSPACE((U8) *scan)) {
6264 while (scan < loceol && isSPACE_A((U8) *scan)) {
6269 PL_reg_flags |= RF_tainted;
6272 while (hardcount < max && scan < loceol &&
6273 isSPACE_LC_utf8((U8*)scan)) {
6274 scan += UTF8SKIP(scan);
6278 while (scan < loceol && isSPACE_LC(*scan))
6288 LOAD_UTF8_CHARCLASS_SPACE();
6289 while (hardcount < max && scan < loceol &&
6291 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6293 scan += UTF8SKIP(scan);
6299 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6308 while (scan < loceol && ! isSPACE((U8) *scan)) {
6314 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6315 scan += UTF8SKIP(scan);
6319 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6325 PL_reg_flags |= RF_tainted;
6328 while (hardcount < max && scan < loceol &&
6329 !isSPACE_LC_utf8((U8*)scan)) {
6330 scan += UTF8SKIP(scan);
6334 while (scan < loceol && !isSPACE_LC(*scan))
6341 LOAD_UTF8_CHARCLASS_DIGIT();
6342 while (hardcount < max && scan < loceol &&
6343 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6344 scan += UTF8SKIP(scan);
6348 while (scan < loceol && isDIGIT(*scan))
6353 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6358 PL_reg_flags |= RF_tainted;
6361 while (hardcount < max && scan < loceol &&
6362 isDIGIT_LC_utf8((U8*)scan)) {
6363 scan += UTF8SKIP(scan);
6367 while (scan < loceol && isDIGIT_LC(*scan))
6374 LOAD_UTF8_CHARCLASS_DIGIT();
6375 while (hardcount < max && scan < loceol &&
6376 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6377 scan += UTF8SKIP(scan);
6381 while (scan < loceol && !isDIGIT(*scan))
6387 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6388 scan += UTF8SKIP(scan);
6392 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6398 PL_reg_flags |= RF_tainted;
6401 while (hardcount < max && scan < loceol &&
6402 !isDIGIT_LC_utf8((U8*)scan)) {
6403 scan += UTF8SKIP(scan);
6407 while (scan < loceol && !isDIGIT_LC(*scan))
6414 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6420 LNBREAK can match two latin chars, which is ok,
6421 because we have a null terminated string, but we
6422 have to use hardcount in this situation
6424 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6433 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6438 while (scan < loceol && is_HORIZWS_latin1(scan))
6445 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6446 scan += UTF8SKIP(scan);
6450 while (scan < loceol && !is_HORIZWS_latin1(scan))
6458 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6463 while (scan < loceol && is_VERTWS_latin1(scan))
6471 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6472 scan += UTF8SKIP(scan);
6476 while (scan < loceol && !is_VERTWS_latin1(scan))
6482 default: /* Called on something of 0 width. */
6483 break; /* So match right here or not at all. */
6489 c = scan - PL_reginput;
6493 GET_RE_DEBUG_FLAGS_DECL;
6495 SV * const prop = sv_newmortal();
6496 regprop(prog, prop, p);
6497 PerlIO_printf(Perl_debug_log,
6498 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6499 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6507 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6509 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
6510 create a copy so that changes the caller makes won't change the shared one
6513 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6515 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6516 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp, altsvp));
6521 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6523 /* Returns the swash for the input 'node' in the regex 'prog'.
6524 * If <doinit> is true, will attempt to create the swash if not already
6526 * If <listsvp> is non-null, will return the swash initialization string in
6528 * If <altsvp> is non-null, will return the alternates to the regular swash
6530 * Tied intimately to how regcomp.c sets up the data structure */
6538 RXi_GET_DECL(prog,progi);
6539 const struct reg_data * const data = prog ? progi->data : NULL;
6541 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
6543 assert(ANYOF_NONBITMAP(node));
6545 if (data && data->count) {
6546 const U32 n = ARG(node);
6548 if (data->what[n] == 's') {
6549 SV * const rv = MUTABLE_SV(data->data[n]);
6550 AV * const av = MUTABLE_AV(SvRV(rv));
6551 SV **const ary = AvARRAY(av);
6552 bool invlist_has_user_defined_property;
6554 si = *ary; /* ary[0] = the string to initialize the swash with */
6556 /* Elements 3 and 4 are either both present or both absent. [3] is
6557 * any inversion list generated at compile time; [4] indicates if
6558 * that inversion list has any user-defined properties in it. */
6559 if (av_len(av) >= 3) {
6561 invlist_has_user_defined_property = cBOOL(SvUV(ary[4]));
6565 invlist_has_user_defined_property = FALSE;
6568 /* Element [1] is reserved for the set-up swash. If already there,
6569 * return it; if not, create it and store it there */
6570 if (SvROK(ary[1])) {
6573 else if (si && doinit) {
6575 sw = _core_swash_init("utf8", /* the utf8 package */
6579 0, /* not from tr/// */
6580 FALSE, /* is error if can't find
6583 invlist_has_user_defined_property);
6584 (void)av_store(av, 1, sw);
6587 /* Element [2] is for any multi-char folds. Note that is a
6588 * fundamentally flawed design, because can't backtrack and try
6589 * again. See [perl #89774] */
6590 if (SvTYPE(ary[2]) == SVt_PVAV) {
6597 SV* matches_string = newSVpvn("", 0);
6600 /* Use the swash, if any, which has to have incorporated into it all
6604 && SvTYPE(SvRV(sw)) == SVt_PVHV
6605 && (invlistsvp = hv_fetchs(MUTABLE_HV(SvRV(sw)), "INVLIST", FALSE)))
6607 invlist = *invlistsvp;
6609 else if (si && si != &PL_sv_undef) {
6611 /* If no swash, use the input nitialization string, if available */
6612 sv_catsv(matches_string, si);
6615 /* Add the inversion list to whatever we have. This may have come from
6616 * the swash, or from an input parameter */
6618 sv_catsv(matches_string, _invlist_contents(invlist));
6620 *listsvp = matches_string;
6630 - reginclass - determine if a character falls into a character class
6632 n is the ANYOF regnode
6633 p is the target string
6634 lenp is pointer to the maximum number of bytes of how far to go in p
6635 (This is assumed wthout checking to always be at least the current
6637 utf8_target tells whether p is in UTF-8.
6639 Returns true if matched; false otherwise. If lenp is not NULL, on return
6640 from a successful match, the value it points to will be updated to how many
6641 bytes in p were matched. If there was no match, the value is undefined,
6642 possibly changed from the input.
6644 Note that this can be a synthetic start class, a combination of various
6645 nodes, so things you think might be mutually exclusive, such as locale,
6646 aren't. It can match both locale and non-locale
6651 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6654 const char flags = ANYOF_FLAGS(n);
6660 PERL_ARGS_ASSERT_REGINCLASS;
6662 /* If c is not already the code point, get it */
6663 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6664 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6665 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6666 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6667 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6668 * UTF8_ALLOW_FFFF */
6669 if (c_len == (STRLEN)-1)
6670 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6676 /* Use passed in max length, or one character if none passed in or less
6677 * than one character. And assume will match just one character. This is
6678 * overwritten later if matched more. */
6680 maxlen = (*lenp > c_len) ? *lenp : c_len;
6688 /* If this character is potentially in the bitmap, check it */
6690 if (ANYOF_BITMAP_TEST(n, c))
6692 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6699 else if (flags & ANYOF_LOCALE) {
6700 PL_reg_flags |= RF_tainted;
6702 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6703 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6707 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6708 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6709 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6710 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6711 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6712 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6713 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6714 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6715 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6716 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6717 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6718 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII_LC(c)) ||
6719 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII_LC(c)) ||
6720 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6721 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6722 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6723 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6724 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6725 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6726 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6727 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6728 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6729 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6730 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6731 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6732 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6733 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6734 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6735 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6736 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK_LC(c)) ||
6737 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK_LC(c))
6738 ) /* How's that for a conditional? */
6745 /* If the bitmap didn't (or couldn't) match, and something outside the
6746 * bitmap could match, try that. Locale nodes specifiy completely the
6747 * behavior of code points in the bit map (otherwise, a utf8 target would
6748 * cause them to be treated as Unicode and not locale), except in
6749 * the very unlikely event when this node is a synthetic start class, which
6750 * could be a combination of locale and non-locale nodes. So allow locale
6751 * to match for the synthetic start class, which will give a false
6752 * positive that will be resolved when the match is done again as not part
6753 * of the synthetic start class */
6755 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6756 match = TRUE; /* Everything above 255 matches */
6758 else if (ANYOF_NONBITMAP(n)
6759 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6762 || (! (flags & ANYOF_LOCALE))
6763 || (flags & ANYOF_IS_SYNTHETIC)))))
6766 SV * const sw = core_regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6774 /* Not utf8. Convert as much of the string as available up
6775 * to the limit of how far the (single) character in the
6776 * pattern can possibly match (no need to go further). If
6777 * the node is a straight ANYOF or not folding, it can't
6778 * match more than one. Otherwise, It can match up to how
6779 * far a single char can fold to. Since not utf8, each
6780 * character is a single byte, so the max it can be in
6781 * bytes is the same as the max it can be in characters */
6782 STRLEN len = (OP(n) == ANYOF
6783 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6785 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6787 : UTF8_MAX_FOLD_CHAR_EXPAND;
6788 utf8_p = bytes_to_utf8(p, &len);
6791 if (swash_fetch(sw, utf8_p, TRUE))
6793 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6795 /* Here, we need to test if the fold of the target string
6796 * matches. The non-multi char folds have all been moved to
6797 * the compilation phase, and the multi-char folds have
6798 * been stored by regcomp into 'av'; we linearly check to
6799 * see if any match the target string (folded). We know
6800 * that the originals were each one character, but we don't
6801 * currently know how many characters/bytes each folded to,
6802 * except we do know that there are small limits imposed by
6803 * Unicode. XXX A performance enhancement would be to have
6804 * regcomp.c store the max number of chars/bytes that are
6805 * in an av entry, as, say the 0th element. Even better
6806 * would be to have a hash of the few characters that can
6807 * start a multi-char fold to the max number of chars of
6810 * If there is a match, we will need to advance (if lenp is
6811 * specified) the match pointer in the target string. But
6812 * what we are comparing here isn't that string directly,
6813 * but its fold, whose length may differ from the original.
6814 * As we go along in constructing the fold, therefore, we
6815 * create a map so that we know how many bytes in the
6816 * source to advance given that we have matched a certain
6817 * number of bytes in the fold. This map is stored in
6818 * 'map_fold_len_back'. Let n mean the number of bytes in
6819 * the fold of the first character that we are folding.
6820 * Then map_fold_len_back[n] is set to the number of bytes
6821 * in that first character. Similarly let m be the
6822 * corresponding number for the second character to be
6823 * folded. Then map_fold_len_back[n+m] is set to the
6824 * number of bytes occupied by the first two source
6825 * characters. ... */
6826 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6827 U8 folded[UTF8_MAXBYTES_CASE+1];
6828 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6829 STRLEN total_foldlen = 0; /* num bytes in fold of all
6832 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6834 /* Here, only need to fold the first char of the target
6835 * string. It the source wasn't utf8, is 1 byte long */
6836 to_utf8_fold(utf8_p, folded, &foldlen);
6837 total_foldlen = foldlen;
6838 map_fold_len_back[foldlen] = (utf8_target)
6844 /* Here, need to fold more than the first char. Do so
6845 * up to the limits */
6846 U8* source_ptr = utf8_p; /* The source for the fold
6849 U8* folded_ptr = folded;
6850 U8* e = utf8_p + maxlen; /* Can't go beyond last
6851 available byte in the
6855 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6859 /* Fold the next character */
6860 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6861 STRLEN this_char_foldlen;
6862 to_utf8_fold(source_ptr,
6864 &this_char_foldlen);
6866 /* Bail if it would exceed the byte limit for
6867 * folding a single char. */
6868 if (this_char_foldlen + folded_ptr - folded >
6874 /* Add the fold of this character */
6875 Copy(this_char_folded,
6879 source_ptr += UTF8SKIP(source_ptr);
6880 folded_ptr += this_char_foldlen;
6881 total_foldlen = folded_ptr - folded;
6883 /* Create map from the number of bytes in the fold
6884 * back to the number of bytes in the source. If
6885 * the source isn't utf8, the byte count is just
6886 * the number of characters so far */
6887 map_fold_len_back[total_foldlen]
6889 ? source_ptr - utf8_p
6896 /* Do the linear search to see if the fold is in the list
6897 * of multi-char folds. */
6900 for (i = 0; i <= av_len(av); i++) {
6901 SV* const sv = *av_fetch(av, i, FALSE);
6903 const char * const s = SvPV_const(sv, len);
6905 if (len <= total_foldlen
6906 && memEQ(s, (char*)folded, len)
6908 /* If 0, means matched a partial char. See
6910 && map_fold_len_back[len])
6913 /* Advance the target string ptr to account for
6914 * this fold, but have to translate from the
6915 * folded length to the corresponding source
6918 *lenp = map_fold_len_back[len];
6927 /* If we allocated a string above, free it */
6928 if (! utf8_target) Safefree(utf8_p);
6933 return (flags & ANYOF_INVERT) ? !match : match;
6937 S_reghop3(U8 *s, I32 off, const U8* lim)
6939 /* return the position 'off' UTF-8 characters away from 's', forward if
6940 * 'off' >= 0, backwards if negative. But don't go outside of position
6941 * 'lim', which better be < s if off < 0 */
6945 PERL_ARGS_ASSERT_REGHOP3;
6948 while (off-- && s < lim) {
6949 /* XXX could check well-formedness here */
6954 while (off++ && s > lim) {
6956 if (UTF8_IS_CONTINUED(*s)) {
6957 while (s > lim && UTF8_IS_CONTINUATION(*s))
6960 /* XXX could check well-formedness here */
6967 /* there are a bunch of places where we use two reghop3's that should
6968 be replaced with this routine. but since thats not done yet
6969 we ifdef it out - dmq
6972 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6976 PERL_ARGS_ASSERT_REGHOP4;
6979 while (off-- && s < rlim) {
6980 /* XXX could check well-formedness here */
6985 while (off++ && s > llim) {
6987 if (UTF8_IS_CONTINUED(*s)) {
6988 while (s > llim && UTF8_IS_CONTINUATION(*s))
6991 /* XXX could check well-formedness here */
6999 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7003 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7006 while (off-- && s < lim) {
7007 /* XXX could check well-formedness here */
7014 while (off++ && s > lim) {
7016 if (UTF8_IS_CONTINUED(*s)) {
7017 while (s > lim && UTF8_IS_CONTINUATION(*s))
7020 /* XXX could check well-formedness here */
7029 restore_pos(pTHX_ void *arg)
7032 regexp * const rex = (regexp *)arg;
7033 if (PL_reg_eval_set) {
7034 if (PL_reg_oldsaved) {
7035 rex->subbeg = PL_reg_oldsaved;
7036 rex->sublen = PL_reg_oldsavedlen;
7037 #ifdef PERL_OLD_COPY_ON_WRITE
7038 rex->saved_copy = PL_nrs;
7040 RXp_MATCH_COPIED_on(rex);
7042 PL_reg_magic->mg_len = PL_reg_oldpos;
7043 PL_reg_eval_set = 0;
7044 PL_curpm = PL_reg_oldcurpm;
7049 S_to_utf8_substr(pTHX_ register regexp *prog)
7053 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7056 if (prog->substrs->data[i].substr
7057 && !prog->substrs->data[i].utf8_substr) {
7058 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7059 prog->substrs->data[i].utf8_substr = sv;
7060 sv_utf8_upgrade(sv);
7061 if (SvVALID(prog->substrs->data[i].substr)) {
7062 if (SvTAIL(prog->substrs->data[i].substr)) {
7063 /* Trim the trailing \n that fbm_compile added last
7065 SvCUR_set(sv, SvCUR(sv) - 1);
7066 /* Whilst this makes the SV technically "invalid" (as its
7067 buffer is no longer followed by "\0") when fbm_compile()
7068 adds the "\n" back, a "\0" is restored. */
7069 fbm_compile(sv, FBMcf_TAIL);
7073 if (prog->substrs->data[i].substr == prog->check_substr)
7074 prog->check_utf8 = sv;
7080 S_to_byte_substr(pTHX_ register regexp *prog)
7085 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7088 if (prog->substrs->data[i].utf8_substr
7089 && !prog->substrs->data[i].substr) {
7090 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7091 if (sv_utf8_downgrade(sv, TRUE)) {
7092 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7093 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7094 /* Trim the trailing \n that fbm_compile added last
7096 SvCUR_set(sv, SvCUR(sv) - 1);
7097 fbm_compile(sv, FBMcf_TAIL);
7105 prog->substrs->data[i].substr = sv;
7106 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7107 prog->check_substr = sv;
7114 * c-indentation-style: bsd
7116 * indent-tabs-mode: t
7119 * ex: set ts=8 sts=4 sw=4 noet: