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 #include "inline_invlist.c"
84 #include "utf8_strings.h"
86 #define RF_tainted 1 /* tainted information used? e.g. locale */
87 #define RF_warned 2 /* warned about big count? */
89 #define RF_utf8 8 /* Pattern contains multibyte chars? */
91 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
97 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
98 * call if there are no complications: i.e., if everything matchable is
99 * straight forward in the bitmap */
100 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
101 : ANYOF_BITMAP_TEST(p,*(c)))
107 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
108 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
110 #define HOPc(pos,off) \
111 (char *)(PL_reg_match_utf8 \
112 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
114 #define HOPBACKc(pos, off) \
115 (char*)(PL_reg_match_utf8\
116 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
117 : (pos - off >= PL_bostr) \
121 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
122 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
124 /* these are unrolled below in the CCC_TRY_XXX defined */
125 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
126 if (!CAT2(PL_utf8_,class)) { \
128 ENTER; save_re_context(); \
129 ok=CAT2(is_utf8_,class)((const U8*)str); \
130 PERL_UNUSED_VAR(ok); \
131 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
132 /* Doesn't do an assert to verify that is correct */
133 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
134 if (!CAT2(PL_utf8_,class)) { \
135 bool throw_away PERL_UNUSED_DECL; \
136 ENTER; save_re_context(); \
137 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
140 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
141 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
142 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
144 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
145 /* No asserts are done for some of these, in case called on a */ \
146 /* Unicode version in which they map to nothing */ \
147 LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
148 LOAD_UTF8_CHARCLASS_NO_CHECK(X_special_begin); \
149 LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
150 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* empty in most releases*/ \
151 LOAD_UTF8_CHARCLASS(X_L, HANGUL_CHOSEONG_KIYEOK_UTF8); \
152 LOAD_UTF8_CHARCLASS(X_LV_LVT_V, HANGUL_JUNGSEONG_FILLER_UTF8); \
153 LOAD_UTF8_CHARCLASS_NO_CHECK(X_RI); /* empty in many releases */ \
154 LOAD_UTF8_CHARCLASS(X_T, HANGUL_JONGSEONG_KIYEOK_UTF8); \
155 LOAD_UTF8_CHARCLASS(X_V, HANGUL_JUNGSEONG_FILLER_UTF8)
157 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
159 /* The actual code for CCC_TRY, which uses several variables from the routine
160 * it's callable from. It is designed to be the bulk of a case statement.
161 * FUNC is the macro or function to call on non-utf8 targets that indicate if
162 * nextchr matches the class.
163 * UTF8_TEST is the whole test string to use for utf8 targets
164 * LOAD is what to use to test, and if not present to load in the swash for the
166 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
168 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
169 * utf8 and a variant, load the swash if necessary and test using the utf8
170 * test. Advance to the next character if test is ok, otherwise fail; If not
171 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
172 * fails, or advance to the next character */
174 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
175 if (locinput >= PL_regeol) { \
178 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
179 LOAD_UTF8_CHARCLASS(CLASS, STR); \
180 if (POS_OR_NEG (UTF8_TEST)) { \
183 locinput += PL_utf8skip[nextchr]; \
184 nextchr = UCHARAT(locinput); \
187 if (POS_OR_NEG (FUNC(nextchr))) { \
190 nextchr = UCHARAT(++locinput); \
193 /* Handle the non-locale cases for a character class and its complement. It
194 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
195 * This is because that code fails when the test succeeds, so we want to have
196 * the test fail so that the code succeeds. The swash is stored in a
197 * predictable PL_ place */
198 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
201 _CCC_TRY_CODE( !, FUNC, \
202 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
203 (U8*)locinput, TRUE)), \
206 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
207 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
208 (U8*)locinput, TRUE)), \
211 /* Generate the case statements for both locale and non-locale character
212 * classes in regmatch for classes that don't have special unicode semantics.
213 * Locales don't use an immediate swash, but an intermediary special locale
214 * function that is called on the pointer to the current place in the input
215 * string. That function will resolve to needing the same swash. One might
216 * think that because we don't know what the locale will match, we shouldn't
217 * check with the swash loading function that it loaded properly; ie, that we
218 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
219 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
221 #define CCC_TRY(NAME, NNAME, FUNC, \
222 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
223 NAMEA, NNAMEA, FUNCA, \
226 PL_reg_flags |= RF_tainted; \
227 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
229 PL_reg_flags |= RF_tainted; \
230 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
233 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
236 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
237 nextchr = UCHARAT(++locinput); \
240 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
244 locinput += PL_utf8skip[nextchr]; \
245 nextchr = UCHARAT(locinput); \
248 nextchr = UCHARAT(++locinput); \
251 /* Generate the non-locale cases */ \
252 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
254 /* This is like CCC_TRY, but has an extra set of parameters for generating case
255 * statements to handle separate Unicode semantics nodes */
256 #define CCC_TRY_U(NAME, NNAME, FUNC, \
257 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
258 NAMEU, NNAMEU, FUNCU, \
259 NAMEA, NNAMEA, FUNCA, \
261 CCC_TRY(NAME, NNAME, FUNC, \
262 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
263 NAMEA, NNAMEA, FUNCA, \
265 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
267 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
269 /* for use after a quantifier and before an EXACT-like node -- japhy */
270 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
272 * NOTE that *nothing* that affects backtracking should be in here, specifically
273 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
274 * node that is in between two EXACT like nodes when ascertaining what the required
275 * "follow" character is. This should probably be moved to regex compile time
276 * although it may be done at run time beause of the REF possibility - more
277 * investigation required. -- demerphq
279 #define JUMPABLE(rn) ( \
281 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
283 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
284 OP(rn) == PLUS || OP(rn) == MINMOD || \
286 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
288 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
290 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
293 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
294 we don't need this definition. */
295 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
296 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
297 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
300 /* ... so we use this as its faster. */
301 #define IS_TEXT(rn) ( OP(rn)==EXACT )
302 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
303 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
304 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
309 Search for mandatory following text node; for lookahead, the text must
310 follow but for lookbehind (rn->flags != 0) we skip to the next step.
312 #define FIND_NEXT_IMPT(rn) STMT_START { \
313 while (JUMPABLE(rn)) { \
314 const OPCODE type = OP(rn); \
315 if (type == SUSPEND || PL_regkind[type] == CURLY) \
316 rn = NEXTOPER(NEXTOPER(rn)); \
317 else if (type == PLUS) \
319 else if (type == IFMATCH) \
320 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
321 else rn += NEXT_OFF(rn); \
326 static void restore_pos(pTHX_ void *arg);
328 #define REGCP_PAREN_ELEMS 3
329 #define REGCP_OTHER_ELEMS 3
330 #define REGCP_FRAME_ELEMS 1
331 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
332 * are needed for the regexp context stack bookkeeping. */
335 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
338 const int retval = PL_savestack_ix;
339 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
340 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
341 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
343 GET_RE_DEBUG_FLAGS_DECL;
345 PERL_ARGS_ASSERT_REGCPPUSH;
347 if (paren_elems_to_push < 0)
348 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
349 paren_elems_to_push);
351 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
352 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
353 " out of range (%lu-%ld)",
354 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
356 SSGROW(total_elems + REGCP_FRAME_ELEMS);
359 if ((int)PL_regsize > (int)parenfloor)
360 PerlIO_printf(Perl_debug_log,
361 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
366 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
367 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
368 SSPUSHINT(rex->offs[p].end);
369 SSPUSHINT(rex->offs[p].start);
370 SSPUSHINT(rex->offs[p].start_tmp);
371 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
372 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
374 (IV)rex->offs[p].start,
375 (IV)rex->offs[p].start_tmp,
379 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
380 SSPUSHINT(PL_regsize);
381 SSPUSHINT(rex->lastparen);
382 SSPUSHINT(rex->lastcloseparen);
383 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
388 /* These are needed since we do not localize EVAL nodes: */
389 #define REGCP_SET(cp) \
391 PerlIO_printf(Perl_debug_log, \
392 " Setting an EVAL scope, savestack=%"IVdf"\n", \
393 (IV)PL_savestack_ix)); \
396 #define REGCP_UNWIND(cp) \
398 if (cp != PL_savestack_ix) \
399 PerlIO_printf(Perl_debug_log, \
400 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
401 (IV)(cp), (IV)PL_savestack_ix)); \
404 #define UNWIND_PAREN(lp, lcp) \
405 for (n = rex->lastparen; n > lp; n--) \
406 rex->offs[n].end = -1; \
407 rex->lastparen = n; \
408 rex->lastcloseparen = lcp;
412 S_regcppop(pTHX_ regexp *rex)
417 GET_RE_DEBUG_FLAGS_DECL;
419 PERL_ARGS_ASSERT_REGCPPOP;
421 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
423 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
424 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
425 rex->lastcloseparen = SSPOPINT;
426 rex->lastparen = SSPOPINT;
427 PL_regsize = SSPOPINT;
429 i -= REGCP_OTHER_ELEMS;
430 /* Now restore the parentheses context. */
432 if (i || rex->lastparen + 1 <= rex->nparens)
433 PerlIO_printf(Perl_debug_log,
434 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
440 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
442 rex->offs[paren].start_tmp = SSPOPINT;
443 rex->offs[paren].start = SSPOPINT;
445 if (paren <= rex->lastparen)
446 rex->offs[paren].end = tmps;
447 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
448 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
450 (IV)rex->offs[paren].start,
451 (IV)rex->offs[paren].start_tmp,
452 (IV)rex->offs[paren].end,
453 (paren > rex->lastparen ? "(skipped)" : ""));
458 /* It would seem that the similar code in regtry()
459 * already takes care of this, and in fact it is in
460 * a better location to since this code can #if 0-ed out
461 * but the code in regtry() is needed or otherwise tests
462 * requiring null fields (pat.t#187 and split.t#{13,14}
463 * (as of patchlevel 7877) will fail. Then again,
464 * this code seems to be necessary or otherwise
465 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
466 * --jhi updated by dapm */
467 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
469 rex->offs[i].start = -1;
470 rex->offs[i].end = -1;
471 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
472 " \\%"UVuf": %s ..-1 undeffing\n",
474 (i > PL_regsize) ? "-1" : " "
480 /* restore the parens and associated vars at savestack position ix,
481 * but without popping the stack */
484 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
486 I32 tmpix = PL_savestack_ix;
487 PL_savestack_ix = ix;
489 PL_savestack_ix = tmpix;
492 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
495 * pregexec and friends
498 #ifndef PERL_IN_XSUB_RE
500 - pregexec - match a regexp against a string
503 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
504 char *strbeg, I32 minend, SV *screamer, U32 nosave)
505 /* stringarg: the point in the string at which to begin matching */
506 /* strend: pointer to null at end of string */
507 /* strbeg: real beginning of string */
508 /* minend: end of match must be >= minend bytes after stringarg. */
509 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
510 * itself is accessed via the pointers above */
511 /* nosave: For optimizations. */
513 PERL_ARGS_ASSERT_PREGEXEC;
516 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
517 nosave ? 0 : REXEC_COPY_STR);
522 * Need to implement the following flags for reg_anch:
524 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
526 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
527 * INTUIT_AUTORITATIVE_ML
528 * INTUIT_ONCE_NOML - Intuit can match in one location only.
531 * Another flag for this function: SECOND_TIME (so that float substrs
532 * with giant delta may be not rechecked).
535 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
537 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
538 Otherwise, only SvCUR(sv) is used to get strbeg. */
540 /* XXXX We assume that strpos is strbeg unless sv. */
542 /* XXXX Some places assume that there is a fixed substring.
543 An update may be needed if optimizer marks as "INTUITable"
544 RExen without fixed substrings. Similarly, it is assumed that
545 lengths of all the strings are no more than minlen, thus they
546 cannot come from lookahead.
547 (Or minlen should take into account lookahead.)
548 NOTE: Some of this comment is not correct. minlen does now take account
549 of lookahead/behind. Further research is required. -- demerphq
553 /* A failure to find a constant substring means that there is no need to make
554 an expensive call to REx engine, thus we celebrate a failure. Similarly,
555 finding a substring too deep into the string means that less calls to
556 regtry() should be needed.
558 REx compiler's optimizer found 4 possible hints:
559 a) Anchored substring;
561 c) Whether we are anchored (beginning-of-line or \G);
562 d) First node (of those at offset 0) which may distinguish positions;
563 We use a)b)d) and multiline-part of c), and try to find a position in the
564 string which does not contradict any of them.
567 /* Most of decisions we do here should have been done at compile time.
568 The nodes of the REx which we used for the search should have been
569 deleted from the finite automaton. */
572 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
573 char *strend, const U32 flags, re_scream_pos_data *data)
576 struct regexp *const prog = (struct regexp *)SvANY(rx);
578 /* Should be nonnegative! */
584 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
586 char *other_last = NULL; /* other substr checked before this */
587 char *check_at = NULL; /* check substr found at this pos */
588 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
589 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
590 RXi_GET_DECL(prog,progi);
592 const char * const i_strpos = strpos;
594 GET_RE_DEBUG_FLAGS_DECL;
596 PERL_ARGS_ASSERT_RE_INTUIT_START;
597 PERL_UNUSED_ARG(flags);
598 PERL_UNUSED_ARG(data);
600 RX_MATCH_UTF8_set(rx,utf8_target);
603 PL_reg_flags |= RF_utf8;
606 debug_start_match(rx, utf8_target, strpos, strend,
607 sv ? "Guessing start of match in sv for"
608 : "Guessing start of match in string for");
611 /* CHR_DIST() would be more correct here but it makes things slow. */
612 if (prog->minlen > strend - strpos) {
613 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
614 "String too short... [re_intuit_start]\n"));
618 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
621 if (!prog->check_utf8 && prog->check_substr)
622 to_utf8_substr(prog);
623 check = prog->check_utf8;
625 if (!prog->check_substr && prog->check_utf8)
626 to_byte_substr(prog);
627 check = prog->check_substr;
629 if (check == &PL_sv_undef) {
630 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
631 "Non-utf8 string cannot match utf8 check string\n"));
634 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
635 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
636 || ( (prog->extflags & RXf_ANCH_BOL)
637 && !multiline ) ); /* Check after \n? */
640 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
641 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
642 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
644 && (strpos != strbeg)) {
645 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
648 if (prog->check_offset_min == prog->check_offset_max &&
649 !(prog->extflags & RXf_CANY_SEEN)) {
650 /* Substring at constant offset from beg-of-str... */
653 s = HOP3c(strpos, prog->check_offset_min, strend);
656 slen = SvCUR(check); /* >= 1 */
658 if ( strend - s > slen || strend - s < slen - 1
659 || (strend - s == slen && strend[-1] != '\n')) {
660 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
663 /* Now should match s[0..slen-2] */
665 if (slen && (*SvPVX_const(check) != *s
667 && memNE(SvPVX_const(check), s, slen)))) {
669 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
673 else if (*SvPVX_const(check) != *s
674 || ((slen = SvCUR(check)) > 1
675 && memNE(SvPVX_const(check), s, slen)))
678 goto success_at_start;
681 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
683 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
684 end_shift = prog->check_end_shift;
687 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
688 - (SvTAIL(check) != 0);
689 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
691 if (end_shift < eshift)
695 else { /* Can match at random position */
698 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
699 end_shift = prog->check_end_shift;
701 /* end shift should be non negative here */
704 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
706 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
707 (IV)end_shift, RX_PRECOMP(prog));
711 /* Find a possible match in the region s..strend by looking for
712 the "check" substring in the region corrected by start/end_shift. */
715 I32 srch_start_shift = start_shift;
716 I32 srch_end_shift = end_shift;
719 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
720 srch_end_shift -= ((strbeg - s) - srch_start_shift);
721 srch_start_shift = strbeg - s;
723 DEBUG_OPTIMISE_MORE_r({
724 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
725 (IV)prog->check_offset_min,
726 (IV)srch_start_shift,
728 (IV)prog->check_end_shift);
731 if (prog->extflags & RXf_CANY_SEEN) {
732 start_point= (U8*)(s + srch_start_shift);
733 end_point= (U8*)(strend - srch_end_shift);
735 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
736 end_point= HOP3(strend, -srch_end_shift, strbeg);
738 DEBUG_OPTIMISE_MORE_r({
739 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
740 (int)(end_point - start_point),
741 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
745 s = fbm_instr( start_point, end_point,
746 check, multiline ? FBMrf_MULTILINE : 0);
748 /* Update the count-of-usability, remove useless subpatterns,
752 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
753 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
754 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
755 (s ? "Found" : "Did not find"),
756 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
757 ? "anchored" : "floating"),
760 (s ? " at offset " : "...\n") );
765 /* Finish the diagnostic message */
766 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
768 /* XXX dmq: first branch is for positive lookbehind...
769 Our check string is offset from the beginning of the pattern.
770 So we need to do any stclass tests offset forward from that
779 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
780 Start with the other substr.
781 XXXX no SCREAM optimization yet - and a very coarse implementation
782 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
783 *always* match. Probably should be marked during compile...
784 Probably it is right to do no SCREAM here...
787 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
788 : (prog->float_substr && prog->anchored_substr))
790 /* Take into account the "other" substring. */
791 /* XXXX May be hopelessly wrong for UTF... */
794 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
797 char * const last = HOP3c(s, -start_shift, strbeg);
799 char * const saved_s = s;
802 t = s - prog->check_offset_max;
803 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
805 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
810 t = HOP3c(t, prog->anchored_offset, strend);
811 if (t < other_last) /* These positions already checked */
813 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
816 /* XXXX It is not documented what units *_offsets are in.
817 We assume bytes, but this is clearly wrong.
818 Meaning this code needs to be carefully reviewed for errors.
822 /* On end-of-str: see comment below. */
823 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
824 if (must == &PL_sv_undef) {
826 DEBUG_r(must = prog->anchored_utf8); /* for debug */
831 HOP3(HOP3(last1, prog->anchored_offset, strend)
832 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
834 multiline ? FBMrf_MULTILINE : 0
837 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
838 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
839 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
840 (s ? "Found" : "Contradicts"),
841 quoted, RE_SV_TAIL(must));
846 if (last1 >= last2) {
847 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
848 ", giving up...\n"));
851 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
852 ", trying floating at offset %ld...\n",
853 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
854 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
855 s = HOP3c(last, 1, strend);
859 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
860 (long)(s - i_strpos)));
861 t = HOP3c(s, -prog->anchored_offset, strbeg);
862 other_last = HOP3c(s, 1, strend);
870 else { /* Take into account the floating substring. */
872 char * const saved_s = s;
875 t = HOP3c(s, -start_shift, strbeg);
877 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
878 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
879 last = HOP3c(t, prog->float_max_offset, strend);
880 s = HOP3c(t, prog->float_min_offset, strend);
883 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
884 must = utf8_target ? prog->float_utf8 : prog->float_substr;
885 /* fbm_instr() takes into account exact value of end-of-str
886 if the check is SvTAIL(ed). Since false positives are OK,
887 and end-of-str is not later than strend we are OK. */
888 if (must == &PL_sv_undef) {
890 DEBUG_r(must = prog->float_utf8); /* for debug message */
893 s = fbm_instr((unsigned char*)s,
894 (unsigned char*)last + SvCUR(must)
896 must, multiline ? FBMrf_MULTILINE : 0);
898 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
899 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
900 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
901 (s ? "Found" : "Contradicts"),
902 quoted, RE_SV_TAIL(must));
906 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
907 ", giving up...\n"));
910 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
911 ", trying anchored starting at offset %ld...\n",
912 (long)(saved_s + 1 - i_strpos)));
914 s = HOP3c(t, 1, strend);
918 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
919 (long)(s - i_strpos)));
920 other_last = s; /* Fix this later. --Hugo */
930 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
932 DEBUG_OPTIMISE_MORE_r(
933 PerlIO_printf(Perl_debug_log,
934 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
935 (IV)prog->check_offset_min,
936 (IV)prog->check_offset_max,
944 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
946 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
949 /* Fixed substring is found far enough so that the match
950 cannot start at strpos. */
952 if (ml_anch && t[-1] != '\n') {
953 /* Eventually fbm_*() should handle this, but often
954 anchored_offset is not 0, so this check will not be wasted. */
955 /* XXXX In the code below we prefer to look for "^" even in
956 presence of anchored substrings. And we search even
957 beyond the found float position. These pessimizations
958 are historical artefacts only. */
960 while (t < strend - prog->minlen) {
962 if (t < check_at - prog->check_offset_min) {
963 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
964 /* Since we moved from the found position,
965 we definitely contradict the found anchored
966 substr. Due to the above check we do not
967 contradict "check" substr.
968 Thus we can arrive here only if check substr
969 is float. Redo checking for "other"=="fixed".
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
973 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
974 goto do_other_anchored;
976 /* We don't contradict the found floating substring. */
977 /* XXXX Why not check for STCLASS? */
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
980 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
983 /* Position contradicts check-string */
984 /* XXXX probably better to look for check-string
985 than for "\n", so one should lower the limit for t? */
986 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
987 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
988 other_last = strpos = s = t + 1;
993 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
994 PL_colors[0], PL_colors[1]));
998 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
999 PL_colors[0], PL_colors[1]));
1003 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1006 /* The found string does not prohibit matching at strpos,
1007 - no optimization of calling REx engine can be performed,
1008 unless it was an MBOL and we are not after MBOL,
1009 or a future STCLASS check will fail this. */
1011 /* Even in this situation we may use MBOL flag if strpos is offset
1012 wrt the start of the string. */
1013 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1014 && (strpos != strbeg) && strpos[-1] != '\n'
1015 /* May be due to an implicit anchor of m{.*foo} */
1016 && !(prog->intflags & PREGf_IMPLICIT))
1021 DEBUG_EXECUTE_r( if (ml_anch)
1022 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1023 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1026 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1028 prog->check_utf8 /* Could be deleted already */
1029 && --BmUSEFUL(prog->check_utf8) < 0
1030 && (prog->check_utf8 == prog->float_utf8)
1032 prog->check_substr /* Could be deleted already */
1033 && --BmUSEFUL(prog->check_substr) < 0
1034 && (prog->check_substr == prog->float_substr)
1037 /* If flags & SOMETHING - do not do it many times on the same match */
1038 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1039 /* XXX Does the destruction order has to change with utf8_target? */
1040 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1041 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1042 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1043 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1044 check = NULL; /* abort */
1046 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1047 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1048 if (prog->intflags & PREGf_IMPLICIT)
1049 prog->extflags &= ~RXf_ANCH_MBOL;
1050 /* XXXX This is a remnant of the old implementation. It
1051 looks wasteful, since now INTUIT can use many
1052 other heuristics. */
1053 prog->extflags &= ~RXf_USE_INTUIT;
1054 /* XXXX What other flags might need to be cleared in this branch? */
1060 /* Last resort... */
1061 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1062 /* trie stclasses are too expensive to use here, we are better off to
1063 leave it to regmatch itself */
1064 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1065 /* minlen == 0 is possible if regstclass is \b or \B,
1066 and the fixed substr is ''$.
1067 Since minlen is already taken into account, s+1 is before strend;
1068 accidentally, minlen >= 1 guaranties no false positives at s + 1
1069 even for \b or \B. But (minlen? 1 : 0) below assumes that
1070 regstclass does not come from lookahead... */
1071 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1072 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1073 const U8* const str = (U8*)STRING(progi->regstclass);
1074 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1075 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1078 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1079 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1080 else if (prog->float_substr || prog->float_utf8)
1081 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1085 if (checked_upto < s)
1087 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1088 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1091 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1096 const char *what = NULL;
1098 if (endpos == strend) {
1099 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1100 "Could not match STCLASS...\n") );
1103 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1104 "This position contradicts STCLASS...\n") );
1105 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1107 checked_upto = HOPBACKc(endpos, start_shift);
1108 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1109 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1110 /* Contradict one of substrings */
1111 if (prog->anchored_substr || prog->anchored_utf8) {
1112 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1113 DEBUG_EXECUTE_r( what = "anchored" );
1115 s = HOP3c(t, 1, strend);
1116 if (s + start_shift + end_shift > strend) {
1117 /* XXXX Should be taken into account earlier? */
1118 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1119 "Could not match STCLASS...\n") );
1124 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1125 "Looking for %s substr starting at offset %ld...\n",
1126 what, (long)(s + start_shift - i_strpos)) );
1129 /* Have both, check_string is floating */
1130 if (t + start_shift >= check_at) /* Contradicts floating=check */
1131 goto retry_floating_check;
1132 /* Recheck anchored substring, but not floating... */
1136 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1137 "Looking for anchored substr starting at offset %ld...\n",
1138 (long)(other_last - i_strpos)) );
1139 goto do_other_anchored;
1141 /* Another way we could have checked stclass at the
1142 current position only: */
1147 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1148 "Looking for /%s^%s/m starting at offset %ld...\n",
1149 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1152 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1154 /* Check is floating substring. */
1155 retry_floating_check:
1156 t = check_at - start_shift;
1157 DEBUG_EXECUTE_r( what = "floating" );
1158 goto hop_and_restart;
1161 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1162 "By STCLASS: moving %ld --> %ld\n",
1163 (long)(t - i_strpos), (long)(s - i_strpos))
1167 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1168 "Does not contradict STCLASS...\n");
1173 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1174 PL_colors[4], (check ? "Guessed" : "Giving up"),
1175 PL_colors[5], (long)(s - i_strpos)) );
1178 fail_finish: /* Substring not found */
1179 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1180 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1182 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1183 PL_colors[4], PL_colors[5]));
1187 #define DECL_TRIE_TYPE(scan) \
1188 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1189 trie_type = ((scan->flags == EXACT) \
1190 ? (utf8_target ? trie_utf8 : trie_plain) \
1191 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1193 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1194 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1196 switch (trie_type) { \
1197 case trie_utf8_fold: \
1198 if ( foldlen>0 ) { \
1199 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1204 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1205 len = UTF8SKIP(uc); \
1206 skiplen = UNISKIP( uvc ); \
1207 foldlen -= skiplen; \
1208 uscan = foldbuf + skiplen; \
1211 case trie_latin_utf8_fold: \
1212 if ( foldlen>0 ) { \
1213 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1219 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1220 skiplen = UNISKIP( uvc ); \
1221 foldlen -= skiplen; \
1222 uscan = foldbuf + skiplen; \
1226 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1233 charid = trie->charmap[ uvc ]; \
1237 if (widecharmap) { \
1238 SV** const svpp = hv_fetch(widecharmap, \
1239 (char*)&uvc, sizeof(UV), 0); \
1241 charid = (U16)SvIV(*svpp); \
1246 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1250 && (ln == 1 || folder(s, pat_string, ln)) \
1251 && (!reginfo || regtry(reginfo, &s)) ) \
1257 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1259 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1265 #define REXEC_FBC_SCAN(CoDe) \
1267 while (s < strend) { \
1273 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1274 REXEC_FBC_UTF8_SCAN( \
1276 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1285 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1288 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1297 #define REXEC_FBC_TRYIT \
1298 if ((!reginfo || regtry(reginfo, &s))) \
1301 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1302 if (utf8_target) { \
1303 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1306 REXEC_FBC_CLASS_SCAN(CoNd); \
1309 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1310 if (utf8_target) { \
1312 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1315 REXEC_FBC_CLASS_SCAN(CoNd); \
1318 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1319 PL_reg_flags |= RF_tainted; \
1320 if (utf8_target) { \
1321 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1324 REXEC_FBC_CLASS_SCAN(CoNd); \
1327 #define DUMP_EXEC_POS(li,s,doutf8) \
1328 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1331 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1332 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1333 tmp = TEST_NON_UTF8(tmp); \
1334 REXEC_FBC_UTF8_SCAN( \
1335 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1344 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1345 if (s == PL_bostr) { \
1349 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1350 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1353 LOAD_UTF8_CHARCLASS_ALNUM(); \
1354 REXEC_FBC_UTF8_SCAN( \
1355 if (tmp == ! (TeSt2_UtF8)) { \
1364 /* The only difference between the BOUND and NBOUND cases is that
1365 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1366 * NBOUND. This is accomplished by passing it in either the if or else clause,
1367 * with the other one being empty */
1368 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1369 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1371 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1372 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1374 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1375 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1377 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1378 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1381 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1382 * be passed in completely with the variable name being tested, which isn't
1383 * such a clean interface, but this is easier to read than it was before. We
1384 * are looking for the boundary (or non-boundary between a word and non-word
1385 * character. The utf8 and non-utf8 cases have the same logic, but the details
1386 * must be different. Find the "wordness" of the character just prior to this
1387 * one, and compare it with the wordness of this one. If they differ, we have
1388 * a boundary. At the beginning of the string, pretend that the previous
1389 * character was a new-line */
1390 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1391 if (utf8_target) { \
1394 else { /* Not utf8 */ \
1395 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1396 tmp = TEST_NON_UTF8(tmp); \
1398 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1407 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1410 /* We know what class REx starts with. Try to find this position... */
1411 /* if reginfo is NULL, its a dryrun */
1412 /* annoyingly all the vars in this routine have different names from their counterparts
1413 in regmatch. /grrr */
1416 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1417 const char *strend, regmatch_info *reginfo)
1420 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1421 char *pat_string; /* The pattern's exactish string */
1422 char *pat_end; /* ptr to end char of pat_string */
1423 re_fold_t folder; /* Function for computing non-utf8 folds */
1424 const U8 *fold_array; /* array for folding ords < 256 */
1431 I32 tmp = 1; /* Scratch variable? */
1432 const bool utf8_target = PL_reg_match_utf8;
1433 UV utf8_fold_flags = 0;
1434 RXi_GET_DECL(prog,progi);
1436 PERL_ARGS_ASSERT_FIND_BYCLASS;
1438 /* We know what class it must start with. */
1442 if (utf8_target || OP(c) == ANYOFV) {
1443 STRLEN inclasslen = strend - s;
1444 REXEC_FBC_UTF8_CLASS_SCAN(
1445 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1448 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1453 if (tmp && (!reginfo || regtry(reginfo, &s)))
1461 if (UTF_PATTERN || utf8_target) {
1462 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1463 goto do_exactf_utf8;
1465 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1466 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1467 goto do_exactf_non_utf8; /* isn't dealt with by these */
1472 /* regcomp.c already folded this if pattern is in UTF-8 */
1473 utf8_fold_flags = 0;
1474 goto do_exactf_utf8;
1476 fold_array = PL_fold;
1478 goto do_exactf_non_utf8;
1481 if (UTF_PATTERN || utf8_target) {
1482 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1483 goto do_exactf_utf8;
1485 fold_array = PL_fold_locale;
1486 folder = foldEQ_locale;
1487 goto do_exactf_non_utf8;
1491 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1493 goto do_exactf_utf8;
1495 case EXACTFU_TRICKYFOLD:
1497 if (UTF_PATTERN || utf8_target) {
1498 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1499 goto do_exactf_utf8;
1502 /* Any 'ss' in the pattern should have been replaced by regcomp,
1503 * so we don't have to worry here about this single special case
1504 * in the Latin1 range */
1505 fold_array = PL_fold_latin1;
1506 folder = foldEQ_latin1;
1510 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1511 are no glitches with fold-length differences
1512 between the target string and pattern */
1514 /* The idea in the non-utf8 EXACTF* cases is to first find the
1515 * first character of the EXACTF* node and then, if necessary,
1516 * case-insensitively compare the full text of the node. c1 is the
1517 * first character. c2 is its fold. This logic will not work for
1518 * Unicode semantics and the german sharp ss, which hence should
1519 * not be compiled into a node that gets here. */
1520 pat_string = STRING(c);
1521 ln = STR_LEN(c); /* length to match in octets/bytes */
1523 /* We know that we have to match at least 'ln' bytes (which is the
1524 * same as characters, since not utf8). If we have to match 3
1525 * characters, and there are only 2 availabe, we know without
1526 * trying that it will fail; so don't start a match past the
1527 * required minimum number from the far end */
1528 e = HOP3c(strend, -((I32)ln), s);
1530 if (!reginfo && e < s) {
1531 e = s; /* Due to minlen logic of intuit() */
1535 c2 = fold_array[c1];
1536 if (c1 == c2) { /* If char and fold are the same */
1537 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1540 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1549 /* If one of the operands is in utf8, we can't use the simpler
1550 * folding above, due to the fact that many different characters
1551 * can have the same fold, or portion of a fold, or different-
1553 pat_string = STRING(c);
1554 ln = STR_LEN(c); /* length to match in octets/bytes */
1555 pat_end = pat_string + ln;
1556 lnc = (UTF_PATTERN) /* length to match in characters */
1557 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1560 /* We have 'lnc' characters to match in the pattern, but because of
1561 * multi-character folding, each character in the target can match
1562 * up to 3 characters (Unicode guarantees it will never exceed
1563 * this) if it is utf8-encoded; and up to 2 if not (based on the
1564 * fact that the Latin 1 folds are already determined, and the
1565 * only multi-char fold in that range is the sharp-s folding to
1566 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1567 * string character. Adjust lnc accordingly, rounding up, so that
1568 * if we need to match at least 4+1/3 chars, that really is 5. */
1569 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1570 lnc = (lnc + expansion - 1) / expansion;
1572 /* As in the non-UTF8 case, if we have to match 3 characters, and
1573 * only 2 are left, it's guaranteed to fail, so don't start a
1574 * match that would require us to go beyond the end of the string
1576 e = HOP3c(strend, -((I32)lnc), s);
1578 if (!reginfo && e < s) {
1579 e = s; /* Due to minlen logic of intuit() */
1582 /* XXX Note that we could recalculate e to stop the loop earlier,
1583 * as the worst case expansion above will rarely be met, and as we
1584 * go along we would usually find that e moves further to the left.
1585 * This would happen only after we reached the point in the loop
1586 * where if there were no expansion we should fail. Unclear if
1587 * worth the expense */
1590 char *my_strend= (char *)strend;
1591 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1592 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1593 && (!reginfo || regtry(reginfo, &s)) )
1597 s += (utf8_target) ? UTF8SKIP(s) : 1;
1602 PL_reg_flags |= RF_tainted;
1603 FBC_BOUND(isALNUM_LC,
1604 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1605 isALNUM_LC_utf8((U8*)s));
1608 PL_reg_flags |= RF_tainted;
1609 FBC_NBOUND(isALNUM_LC,
1610 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1611 isALNUM_LC_utf8((U8*)s));
1614 FBC_BOUND(isWORDCHAR,
1616 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1619 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1621 isWORDCHAR_A((U8*)s));
1624 FBC_NBOUND(isWORDCHAR,
1626 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1629 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1631 isWORDCHAR_A((U8*)s));
1634 FBC_BOUND(isWORDCHAR_L1,
1636 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1639 FBC_NBOUND(isWORDCHAR_L1,
1641 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1644 REXEC_FBC_CSCAN_TAINT(
1645 isALNUM_LC_utf8((U8*)s),
1650 REXEC_FBC_CSCAN_PRELOAD(
1651 LOAD_UTF8_CHARCLASS_ALNUM(),
1652 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1653 isWORDCHAR_L1((U8) *s)
1657 REXEC_FBC_CSCAN_PRELOAD(
1658 LOAD_UTF8_CHARCLASS_ALNUM(),
1659 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1664 /* Don't need to worry about utf8, as it can match only a single
1665 * byte invariant character */
1666 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1669 REXEC_FBC_CSCAN_PRELOAD(
1670 LOAD_UTF8_CHARCLASS_ALNUM(),
1671 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1672 ! isWORDCHAR_L1((U8) *s)
1676 REXEC_FBC_CSCAN_PRELOAD(
1677 LOAD_UTF8_CHARCLASS_ALNUM(),
1678 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1689 REXEC_FBC_CSCAN_TAINT(
1690 !isALNUM_LC_utf8((U8*)s),
1695 REXEC_FBC_CSCAN_PRELOAD(
1696 LOAD_UTF8_CHARCLASS_SPACE(),
1697 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1702 REXEC_FBC_CSCAN_PRELOAD(
1703 LOAD_UTF8_CHARCLASS_SPACE(),
1704 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1709 /* Don't need to worry about utf8, as it can match only a single
1710 * byte invariant character */
1711 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1714 REXEC_FBC_CSCAN_TAINT(
1715 isSPACE_LC_utf8((U8*)s),
1720 REXEC_FBC_CSCAN_PRELOAD(
1721 LOAD_UTF8_CHARCLASS_SPACE(),
1722 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1723 ! isSPACE_L1((U8) *s)
1727 REXEC_FBC_CSCAN_PRELOAD(
1728 LOAD_UTF8_CHARCLASS_SPACE(),
1729 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1740 REXEC_FBC_CSCAN_TAINT(
1741 !isSPACE_LC_utf8((U8*)s),
1746 REXEC_FBC_CSCAN_PRELOAD(
1747 LOAD_UTF8_CHARCLASS_DIGIT(),
1748 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1753 /* Don't need to worry about utf8, as it can match only a single
1754 * byte invariant character */
1755 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1758 REXEC_FBC_CSCAN_TAINT(
1759 isDIGIT_LC_utf8((U8*)s),
1764 REXEC_FBC_CSCAN_PRELOAD(
1765 LOAD_UTF8_CHARCLASS_DIGIT(),
1766 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1777 REXEC_FBC_CSCAN_TAINT(
1778 !isDIGIT_LC_utf8((U8*)s),
1785 is_LNBREAK_latin1(s)
1797 !is_VERTWS_latin1(s)
1803 is_HORIZWS_latin1(s)
1808 !is_HORIZWS_utf8(s),
1809 !is_HORIZWS_latin1(s)
1813 /* Don't need to worry about utf8, as it can match only a single
1814 * byte invariant character. The flag in this node type is the
1815 * class number to pass to _generic_isCC() to build a mask for
1816 * searching in PL_charclass[] */
1817 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1821 !_generic_isCC_A(*s, FLAGS(c)),
1822 !_generic_isCC_A(*s, FLAGS(c))
1830 /* what trie are we using right now */
1832 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1834 = (reg_trie_data*)progi->data->data[ aho->trie ];
1835 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1837 const char *last_start = strend - trie->minlen;
1839 const char *real_start = s;
1841 STRLEN maxlen = trie->maxlen;
1843 U8 **points; /* map of where we were in the input string
1844 when reading a given char. For ASCII this
1845 is unnecessary overhead as the relationship
1846 is always 1:1, but for Unicode, especially
1847 case folded Unicode this is not true. */
1848 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1852 GET_RE_DEBUG_FLAGS_DECL;
1854 /* We can't just allocate points here. We need to wrap it in
1855 * an SV so it gets freed properly if there is a croak while
1856 * running the match */
1859 sv_points=newSV(maxlen * sizeof(U8 *));
1860 SvCUR_set(sv_points,
1861 maxlen * sizeof(U8 *));
1862 SvPOK_on(sv_points);
1863 sv_2mortal(sv_points);
1864 points=(U8**)SvPV_nolen(sv_points );
1865 if ( trie_type != trie_utf8_fold
1866 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1869 bitmap=(U8*)trie->bitmap;
1871 bitmap=(U8*)ANYOF_BITMAP(c);
1873 /* this is the Aho-Corasick algorithm modified a touch
1874 to include special handling for long "unknown char"
1875 sequences. The basic idea being that we use AC as long
1876 as we are dealing with a possible matching char, when
1877 we encounter an unknown char (and we have not encountered
1878 an accepting state) we scan forward until we find a legal
1880 AC matching is basically that of trie matching, except
1881 that when we encounter a failing transition, we fall back
1882 to the current states "fail state", and try the current char
1883 again, a process we repeat until we reach the root state,
1884 state 1, or a legal transition. If we fail on the root state
1885 then we can either terminate if we have reached an accepting
1886 state previously, or restart the entire process from the beginning
1890 while (s <= last_start) {
1891 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1899 U8 *uscan = (U8*)NULL;
1900 U8 *leftmost = NULL;
1902 U32 accepted_word= 0;
1906 while ( state && uc <= (U8*)strend ) {
1908 U32 word = aho->states[ state ].wordnum;
1912 DEBUG_TRIE_EXECUTE_r(
1913 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1914 dump_exec_pos( (char *)uc, c, strend, real_start,
1915 (char *)uc, utf8_target );
1916 PerlIO_printf( Perl_debug_log,
1917 " Scanning for legal start char...\n");
1921 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1925 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1931 if (uc >(U8*)last_start) break;
1935 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1936 if (!leftmost || lpos < leftmost) {
1937 DEBUG_r(accepted_word=word);
1943 points[pointpos++ % maxlen]= uc;
1944 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1945 uscan, len, uvc, charid, foldlen,
1947 DEBUG_TRIE_EXECUTE_r({
1948 dump_exec_pos( (char *)uc, c, strend, real_start,
1950 PerlIO_printf(Perl_debug_log,
1951 " Charid:%3u CP:%4"UVxf" ",
1957 word = aho->states[ state ].wordnum;
1959 base = aho->states[ state ].trans.base;
1961 DEBUG_TRIE_EXECUTE_r({
1963 dump_exec_pos( (char *)uc, c, strend, real_start,
1965 PerlIO_printf( Perl_debug_log,
1966 "%sState: %4"UVxf", word=%"UVxf,
1967 failed ? " Fail transition to " : "",
1968 (UV)state, (UV)word);
1974 ( ((offset = base + charid
1975 - 1 - trie->uniquecharcount)) >= 0)
1976 && ((U32)offset < trie->lasttrans)
1977 && trie->trans[offset].check == state
1978 && (tmp=trie->trans[offset].next))
1980 DEBUG_TRIE_EXECUTE_r(
1981 PerlIO_printf( Perl_debug_log," - legal\n"));
1986 DEBUG_TRIE_EXECUTE_r(
1987 PerlIO_printf( Perl_debug_log," - fail\n"));
1989 state = aho->fail[state];
1993 /* we must be accepting here */
1994 DEBUG_TRIE_EXECUTE_r(
1995 PerlIO_printf( Perl_debug_log," - accepting\n"));
2004 if (!state) state = 1;
2007 if ( aho->states[ state ].wordnum ) {
2008 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2009 if (!leftmost || lpos < leftmost) {
2010 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2015 s = (char*)leftmost;
2016 DEBUG_TRIE_EXECUTE_r({
2018 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2019 (UV)accepted_word, (IV)(s - real_start)
2022 if (!reginfo || regtry(reginfo, &s)) {
2028 DEBUG_TRIE_EXECUTE_r({
2029 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2032 DEBUG_TRIE_EXECUTE_r(
2033 PerlIO_printf( Perl_debug_log,"No match.\n"));
2042 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2052 - regexec_flags - match a regexp against a string
2055 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2056 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2057 /* stringarg: the point in the string at which to begin matching */
2058 /* strend: pointer to null at end of string */
2059 /* strbeg: real beginning of string */
2060 /* minend: end of match must be >= minend bytes after stringarg. */
2061 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2062 * itself is accessed via the pointers above */
2063 /* data: May be used for some additional optimizations.
2064 Currently its only used, with a U32 cast, for transmitting
2065 the ganch offset when doing a /g match. This will change */
2066 /* nosave: For optimizations. */
2070 struct regexp *const prog = (struct regexp *)SvANY(rx);
2071 /*register*/ char *s;
2073 /*register*/ char *startpos = stringarg;
2074 I32 minlen; /* must match at least this many chars */
2075 I32 dontbother = 0; /* how many characters not to try at end */
2076 I32 end_shift = 0; /* Same for the end. */ /* CC */
2077 I32 scream_pos = -1; /* Internal iterator of scream. */
2078 char *scream_olds = NULL;
2079 const bool utf8_target = cBOOL(DO_UTF8(sv));
2081 RXi_GET_DECL(prog,progi);
2082 regmatch_info reginfo; /* create some info to pass to regtry etc */
2083 regexp_paren_pair *swap = NULL;
2084 GET_RE_DEBUG_FLAGS_DECL;
2086 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2087 PERL_UNUSED_ARG(data);
2089 /* Be paranoid... */
2090 if (prog == NULL || startpos == NULL) {
2091 Perl_croak(aTHX_ "NULL regexp parameter");
2095 multiline = prog->extflags & RXf_PMf_MULTILINE;
2096 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2098 RX_MATCH_UTF8_set(rx, utf8_target);
2100 debug_start_match(rx, utf8_target, startpos, strend,
2104 minlen = prog->minlen;
2106 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2107 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2108 "String too short [regexec_flags]...\n"));
2113 /* Check validity of program. */
2114 if (UCHARAT(progi->program) != REG_MAGIC) {
2115 Perl_croak(aTHX_ "corrupted regexp program");
2119 PL_reg_state.re_state_eval_setup_done = FALSE;
2123 PL_reg_flags |= RF_utf8;
2125 /* Mark beginning of line for ^ and lookbehind. */
2126 reginfo.bol = startpos; /* XXX not used ??? */
2130 /* Mark end of line for $ (and such) */
2133 /* see how far we have to get to not match where we matched before */
2134 reginfo.till = startpos+minend;
2136 /* If there is a "must appear" string, look for it. */
2139 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2141 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2142 reginfo.ganch = startpos + prog->gofs;
2143 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2144 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2145 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2147 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2148 && mg->mg_len >= 0) {
2149 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2150 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2151 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2153 if (prog->extflags & RXf_ANCH_GPOS) {
2154 if (s > reginfo.ganch)
2156 s = reginfo.ganch - prog->gofs;
2157 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2158 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2164 reginfo.ganch = strbeg + PTR2UV(data);
2165 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2166 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2168 } else { /* pos() not defined */
2169 reginfo.ganch = strbeg;
2170 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2171 "GPOS: reginfo.ganch = strbeg\n"));
2174 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2175 /* We have to be careful. If the previous successful match
2176 was from this regex we don't want a subsequent partially
2177 successful match to clobber the old results.
2178 So when we detect this possibility we add a swap buffer
2179 to the re, and switch the buffer each match. If we fail
2180 we switch it back, otherwise we leave it swapped.
2183 /* do we need a save destructor here for eval dies? */
2184 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2185 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2186 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2192 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2193 re_scream_pos_data d;
2195 d.scream_olds = &scream_olds;
2196 d.scream_pos = &scream_pos;
2197 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2199 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2200 goto phooey; /* not present */
2206 /* Simplest case: anchored match need be tried only once. */
2207 /* [unless only anchor is BOL and multiline is set] */
2208 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2209 if (s == startpos && regtry(®info, &startpos))
2211 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2212 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2217 dontbother = minlen - 1;
2218 end = HOP3c(strend, -dontbother, strbeg) - 1;
2219 /* for multiline we only have to try after newlines */
2220 if (prog->check_substr || prog->check_utf8) {
2221 /* because of the goto we can not easily reuse the macros for bifurcating the
2222 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2225 goto after_try_utf8;
2227 if (regtry(®info, &s)) {
2234 if (prog->extflags & RXf_USE_INTUIT) {
2235 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2244 } /* end search for check string in unicode */
2246 if (s == startpos) {
2247 goto after_try_latin;
2250 if (regtry(®info, &s)) {
2257 if (prog->extflags & RXf_USE_INTUIT) {
2258 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2267 } /* end search for check string in latin*/
2268 } /* end search for check string */
2269 else { /* search for newline */
2271 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2274 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2275 while (s <= end) { /* note it could be possible to match at the end of the string */
2276 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2277 if (regtry(®info, &s))
2281 } /* end search for newline */
2282 } /* end anchored/multiline check string search */
2284 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2286 /* the warning about reginfo.ganch being used without initialization
2287 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2288 and we only enter this block when the same bit is set. */
2289 char *tmp_s = reginfo.ganch - prog->gofs;
2291 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2296 /* Messy cases: unanchored match. */
2297 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2298 /* we have /x+whatever/ */
2299 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2304 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2305 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2306 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2311 DEBUG_EXECUTE_r( did_match = 1 );
2312 if (regtry(®info, &s)) goto got_it;
2314 while (s < strend && *s == ch)
2322 DEBUG_EXECUTE_r( did_match = 1 );
2323 if (regtry(®info, &s)) goto got_it;
2325 while (s < strend && *s == ch)
2330 DEBUG_EXECUTE_r(if (!did_match)
2331 PerlIO_printf(Perl_debug_log,
2332 "Did not find anchored character...\n")
2335 else if (prog->anchored_substr != NULL
2336 || prog->anchored_utf8 != NULL
2337 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2338 && prog->float_max_offset < strend - s)) {
2343 char *last1; /* Last position checked before */
2347 if (prog->anchored_substr || prog->anchored_utf8) {
2348 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2349 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2350 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2351 back_max = back_min = prog->anchored_offset;
2353 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2354 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2355 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2356 back_max = prog->float_max_offset;
2357 back_min = prog->float_min_offset;
2361 if (must == &PL_sv_undef)
2362 /* could not downgrade utf8 check substring, so must fail */
2368 last = HOP3c(strend, /* Cannot start after this */
2369 -(I32)(CHR_SVLEN(must)
2370 - (SvTAIL(must) != 0) + back_min), strbeg);
2373 last1 = HOPc(s, -1);
2375 last1 = s - 1; /* bogus */
2377 /* XXXX check_substr already used to find "s", can optimize if
2378 check_substr==must. */
2380 dontbother = end_shift;
2381 strend = HOPc(strend, -dontbother);
2382 while ( (s <= last) &&
2383 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2384 (unsigned char*)strend, must,
2385 multiline ? FBMrf_MULTILINE : 0)) ) {
2386 DEBUG_EXECUTE_r( did_match = 1 );
2387 if (HOPc(s, -back_max) > last1) {
2388 last1 = HOPc(s, -back_min);
2389 s = HOPc(s, -back_max);
2392 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2394 last1 = HOPc(s, -back_min);
2398 while (s <= last1) {
2399 if (regtry(®info, &s))
2405 while (s <= last1) {
2406 if (regtry(®info, &s))
2412 DEBUG_EXECUTE_r(if (!did_match) {
2413 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2414 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2415 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2416 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2417 ? "anchored" : "floating"),
2418 quoted, RE_SV_TAIL(must));
2422 else if ( (c = progi->regstclass) ) {
2424 const OPCODE op = OP(progi->regstclass);
2425 /* don't bother with what can't match */
2426 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2427 strend = HOPc(strend, -(minlen - 1));
2430 SV * const prop = sv_newmortal();
2431 regprop(prog, prop, c);
2433 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2435 PerlIO_printf(Perl_debug_log,
2436 "Matching stclass %.*s against %s (%d bytes)\n",
2437 (int)SvCUR(prop), SvPVX_const(prop),
2438 quoted, (int)(strend - s));
2441 if (find_byclass(prog, c, s, strend, ®info))
2443 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2447 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2454 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2455 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2456 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2458 little = SvPV_const(float_real, len);
2459 if (SvTAIL(float_real)) {
2460 /* This means that float_real contains an artificial \n on the end
2461 * due to the presence of something like this: /foo$/
2462 * where we can match both "foo" and "foo\n" at the end of the string.
2463 * So we have to compare the end of the string first against the float_real
2464 * without the \n and then against the full float_real with the string.
2465 * We have to watch out for cases where the string might be smaller
2466 * than the float_real or the float_real without the \n.
2468 char *checkpos= strend - len;
2470 PerlIO_printf(Perl_debug_log,
2471 "%sChecking for float_real.%s\n",
2472 PL_colors[4], PL_colors[5]));
2473 if (checkpos + 1 < strbeg) {
2474 /* can't match, even if we remove the trailing \n string is too short to match */
2476 PerlIO_printf(Perl_debug_log,
2477 "%sString shorter than required trailing substring, cannot match.%s\n",
2478 PL_colors[4], PL_colors[5]));
2480 } else if (memEQ(checkpos + 1, little, len - 1)) {
2481 /* can match, the end of the string matches without the "\n" */
2482 last = checkpos + 1;
2483 } else if (checkpos < strbeg) {
2484 /* cant match, string is too short when the "\n" is included */
2486 PerlIO_printf(Perl_debug_log,
2487 "%sString does not contain required trailing substring, cannot match.%s\n",
2488 PL_colors[4], PL_colors[5]));
2490 } else if (!multiline) {
2491 /* non multiline match, so compare with the "\n" at the end of the string */
2492 if (memEQ(checkpos, little, len)) {
2496 PerlIO_printf(Perl_debug_log,
2497 "%sString does not contain required trailing substring, cannot match.%s\n",
2498 PL_colors[4], PL_colors[5]));
2502 /* multiline match, so we have to search for a place where the full string is located */
2508 last = rninstr(s, strend, little, little + len);
2510 last = strend; /* matching "$" */
2513 /* at one point this block contained a comment which was probably
2514 * incorrect, which said that this was a "should not happen" case.
2515 * Even if it was true when it was written I am pretty sure it is
2516 * not anymore, so I have removed the comment and replaced it with
2519 PerlIO_printf(Perl_debug_log,
2520 "String does not contain required substring, cannot match.\n"
2524 dontbother = strend - last + prog->float_min_offset;
2526 if (minlen && (dontbother < minlen))
2527 dontbother = minlen - 1;
2528 strend -= dontbother; /* this one's always in bytes! */
2529 /* We don't know much -- general case. */
2532 if (regtry(®info, &s))
2541 if (regtry(®info, &s))
2543 } while (s++ < strend);
2553 PerlIO_printf(Perl_debug_log,
2554 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2560 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2562 if (PL_reg_state.re_state_eval_setup_done)
2563 restore_pos(aTHX_ prog);
2564 if (RXp_PAREN_NAMES(prog))
2565 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2567 /* make sure $`, $&, $', and $digit will work later */
2568 if ( !(flags & REXEC_NOT_FIRST) ) {
2569 if (flags & REXEC_COPY_STR) {
2570 #ifdef PERL_OLD_COPY_ON_WRITE
2572 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2574 PerlIO_printf(Perl_debug_log,
2575 "Copy on write: regexp capture, type %d\n",
2578 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2579 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2580 assert (SvPOKp(prog->saved_copy));
2581 prog->sublen = PL_regeol - strbeg;
2582 prog->suboffset = 0;
2583 prog->subcoffset = 0;
2588 I32 max = PL_regeol - strbeg;
2591 if ( (flags & REXEC_COPY_SKIP_POST)
2592 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2593 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2594 ) { /* don't copy $' part of string */
2597 /* calculate the right-most part of the string covered
2598 * by a capture. Due to look-ahead, this may be to
2599 * the right of $&, so we have to scan all captures */
2600 while (n <= prog->lastparen) {
2601 if (prog->offs[n].end > max)
2602 max = prog->offs[n].end;
2606 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2607 ? prog->offs[0].start
2609 assert(max >= 0 && max <= PL_regeol - strbeg);
2612 if ( (flags & REXEC_COPY_SKIP_PRE)
2613 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2614 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2615 ) { /* don't copy $` part of string */
2618 /* calculate the left-most part of the string covered
2619 * by a capture. Due to look-behind, this may be to
2620 * the left of $&, so we have to scan all captures */
2621 while (min && n <= prog->lastparen) {
2622 if ( prog->offs[n].start != -1
2623 && prog->offs[n].start < min)
2625 min = prog->offs[n].start;
2629 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2630 && min > prog->offs[0].end
2632 min = prog->offs[0].end;
2636 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2639 if (RX_MATCH_COPIED(rx)) {
2640 if (sublen > prog->sublen)
2642 (char*)saferealloc(prog->subbeg, sublen+1);
2645 prog->subbeg = (char*)safemalloc(sublen+1);
2646 Copy(strbeg + min, prog->subbeg, sublen, char);
2647 prog->subbeg[sublen] = '\0';
2648 prog->suboffset = min;
2649 prog->sublen = sublen;
2651 RX_MATCH_COPIED_on(rx);
2652 prog->subcoffset = prog->suboffset;
2653 if (prog->suboffset && utf8_target) {
2654 /* Convert byte offset to chars.
2655 * XXX ideally should only compute this if @-/@+
2656 * has been seen, a la PL_sawampersand ??? */
2658 /* If there's a direct correspondence between the
2659 * string which we're matching and the original SV,
2660 * then we can use the utf8 len cache associated with
2661 * the SV. In particular, it means that under //g,
2662 * sv_pos_b2u() will use the previously cached
2663 * position to speed up working out the new length of
2664 * subcoffset, rather than counting from the start of
2665 * the string each time. This stops
2666 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2667 * from going quadratic */
2668 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2669 sv_pos_b2u(sv, &(prog->subcoffset));
2671 prog->subcoffset = utf8_length((U8*)strbeg,
2672 (U8*)(strbeg+prog->suboffset));
2676 RX_MATCH_COPY_FREE(rx);
2677 prog->subbeg = strbeg;
2678 prog->suboffset = 0;
2679 prog->subcoffset = 0;
2680 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2687 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2688 PL_colors[4], PL_colors[5]));
2689 if (PL_reg_state.re_state_eval_setup_done)
2690 restore_pos(aTHX_ prog);
2692 /* we failed :-( roll it back */
2693 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2694 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2699 Safefree(prog->offs);
2707 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2708 * Do inc before dec, in case old and new rex are the same */
2709 #define SET_reg_curpm(Re2) \
2710 if (PL_reg_state.re_state_eval_setup_done) { \
2711 (void)ReREFCNT_inc(Re2); \
2712 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2713 PM_SETRE((PL_reg_curpm), (Re2)); \
2718 - regtry - try match at specific point
2720 STATIC I32 /* 0 failure, 1 success */
2721 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2725 REGEXP *const rx = reginfo->prog;
2726 regexp *const prog = (struct regexp *)SvANY(rx);
2727 RXi_GET_DECL(prog,progi);
2728 GET_RE_DEBUG_FLAGS_DECL;
2730 PERL_ARGS_ASSERT_REGTRY;
2732 reginfo->cutpoint=NULL;
2734 if ((prog->extflags & RXf_EVAL_SEEN)
2735 && !PL_reg_state.re_state_eval_setup_done)
2739 PL_reg_state.re_state_eval_setup_done = TRUE;
2741 /* Make $_ available to executed code. */
2742 if (reginfo->sv != DEFSV) {
2744 DEFSV_set(reginfo->sv);
2747 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2748 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2749 /* prepare for quick setting of pos */
2750 #ifdef PERL_OLD_COPY_ON_WRITE
2751 if (SvIsCOW(reginfo->sv))
2752 sv_force_normal_flags(reginfo->sv, 0);
2754 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2755 &PL_vtbl_mglob, NULL, 0);
2759 PL_reg_oldpos = mg->mg_len;
2760 SAVEDESTRUCTOR_X(restore_pos, prog);
2762 if (!PL_reg_curpm) {
2763 Newxz(PL_reg_curpm, 1, PMOP);
2766 SV* const repointer = &PL_sv_undef;
2767 /* this regexp is also owned by the new PL_reg_curpm, which
2768 will try to free it. */
2769 av_push(PL_regex_padav, repointer);
2770 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2771 PL_regex_pad = AvARRAY(PL_regex_padav);
2776 PL_reg_oldcurpm = PL_curpm;
2777 PL_curpm = PL_reg_curpm;
2778 if (RXp_MATCH_COPIED(prog)) {
2779 /* Here is a serious problem: we cannot rewrite subbeg,
2780 since it may be needed if this match fails. Thus
2781 $` inside (?{}) could fail... */
2782 PL_reg_oldsaved = prog->subbeg;
2783 PL_reg_oldsavedlen = prog->sublen;
2784 PL_reg_oldsavedoffset = prog->suboffset;
2785 PL_reg_oldsavedcoffset = prog->suboffset;
2786 #ifdef PERL_OLD_COPY_ON_WRITE
2787 PL_nrs = prog->saved_copy;
2789 RXp_MATCH_COPIED_off(prog);
2792 PL_reg_oldsaved = NULL;
2793 prog->subbeg = PL_bostr;
2794 prog->suboffset = 0;
2795 prog->subcoffset = 0;
2796 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2799 PL_reg_starttry = *startpos;
2801 prog->offs[0].start = *startpos - PL_bostr;
2802 PL_reginput = *startpos;
2803 prog->lastparen = 0;
2804 prog->lastcloseparen = 0;
2807 /* XXXX What this code is doing here?!!! There should be no need
2808 to do this again and again, prog->lastparen should take care of
2811 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2812 * Actually, the code in regcppop() (which Ilya may be meaning by
2813 * prog->lastparen), is not needed at all by the test suite
2814 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2815 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2816 * Meanwhile, this code *is* needed for the
2817 * above-mentioned test suite tests to succeed. The common theme
2818 * on those tests seems to be returning null fields from matches.
2819 * --jhi updated by dapm */
2821 if (prog->nparens) {
2822 regexp_paren_pair *pp = prog->offs;
2824 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2832 if (regmatch(reginfo, progi->program + 1)) {
2833 prog->offs[0].end = PL_reginput - PL_bostr;
2836 if (reginfo->cutpoint)
2837 *startpos= reginfo->cutpoint;
2838 REGCP_UNWIND(lastcp);
2843 #define sayYES goto yes
2844 #define sayNO goto no
2845 #define sayNO_SILENT goto no_silent
2847 /* we dont use STMT_START/END here because it leads to
2848 "unreachable code" warnings, which are bogus, but distracting. */
2849 #define CACHEsayNO \
2850 if (ST.cache_mask) \
2851 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2854 /* this is used to determine how far from the left messages like
2855 'failed...' are printed. It should be set such that messages
2856 are inline with the regop output that created them.
2858 #define REPORT_CODE_OFF 32
2861 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2862 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2864 #define SLAB_FIRST(s) (&(s)->states[0])
2865 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2867 /* grab a new slab and return the first slot in it */
2869 STATIC regmatch_state *
2872 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2875 regmatch_slab *s = PL_regmatch_slab->next;
2877 Newx(s, 1, regmatch_slab);
2878 s->prev = PL_regmatch_slab;
2880 PL_regmatch_slab->next = s;
2882 PL_regmatch_slab = s;
2883 return SLAB_FIRST(s);
2887 /* push a new state then goto it */
2889 #define PUSH_STATE_GOTO(state, node) \
2891 st->resume_state = state; \
2894 /* push a new state with success backtracking, then goto it */
2896 #define PUSH_YES_STATE_GOTO(state, node) \
2898 st->resume_state = state; \
2899 goto push_yes_state;
2905 regmatch() - main matching routine
2907 This is basically one big switch statement in a loop. We execute an op,
2908 set 'next' to point the next op, and continue. If we come to a point which
2909 we may need to backtrack to on failure such as (A|B|C), we push a
2910 backtrack state onto the backtrack stack. On failure, we pop the top
2911 state, and re-enter the loop at the state indicated. If there are no more
2912 states to pop, we return failure.
2914 Sometimes we also need to backtrack on success; for example /A+/, where
2915 after successfully matching one A, we need to go back and try to
2916 match another one; similarly for lookahead assertions: if the assertion
2917 completes successfully, we backtrack to the state just before the assertion
2918 and then carry on. In these cases, the pushed state is marked as
2919 'backtrack on success too'. This marking is in fact done by a chain of
2920 pointers, each pointing to the previous 'yes' state. On success, we pop to
2921 the nearest yes state, discarding any intermediate failure-only states.
2922 Sometimes a yes state is pushed just to force some cleanup code to be
2923 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2924 it to free the inner regex.
2926 Note that failure backtracking rewinds the cursor position, while
2927 success backtracking leaves it alone.
2929 A pattern is complete when the END op is executed, while a subpattern
2930 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2931 ops trigger the "pop to last yes state if any, otherwise return true"
2934 A common convention in this function is to use A and B to refer to the two
2935 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2936 the subpattern to be matched possibly multiple times, while B is the entire
2937 rest of the pattern. Variable and state names reflect this convention.
2939 The states in the main switch are the union of ops and failure/success of
2940 substates associated with with that op. For example, IFMATCH is the op
2941 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2942 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2943 successfully matched A and IFMATCH_A_fail is a state saying that we have
2944 just failed to match A. Resume states always come in pairs. The backtrack
2945 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2946 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2947 on success or failure.
2949 The struct that holds a backtracking state is actually a big union, with
2950 one variant for each major type of op. The variable st points to the
2951 top-most backtrack struct. To make the code clearer, within each
2952 block of code we #define ST to alias the relevant union.
2954 Here's a concrete example of a (vastly oversimplified) IFMATCH
2960 #define ST st->u.ifmatch
2962 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2963 ST.foo = ...; // some state we wish to save
2965 // push a yes backtrack state with a resume value of
2966 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2968 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2971 case IFMATCH_A: // we have successfully executed A; now continue with B
2973 bar = ST.foo; // do something with the preserved value
2976 case IFMATCH_A_fail: // A failed, so the assertion failed
2977 ...; // do some housekeeping, then ...
2978 sayNO; // propagate the failure
2985 For any old-timers reading this who are familiar with the old recursive
2986 approach, the code above is equivalent to:
2988 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2997 ...; // do some housekeeping, then ...
2998 sayNO; // propagate the failure
3001 The topmost backtrack state, pointed to by st, is usually free. If you
3002 want to claim it, populate any ST.foo fields in it with values you wish to
3003 save, then do one of
3005 PUSH_STATE_GOTO(resume_state, node);
3006 PUSH_YES_STATE_GOTO(resume_state, node);
3008 which sets that backtrack state's resume value to 'resume_state', pushes a
3009 new free entry to the top of the backtrack stack, then goes to 'node'.
3010 On backtracking, the free slot is popped, and the saved state becomes the
3011 new free state. An ST.foo field in this new top state can be temporarily
3012 accessed to retrieve values, but once the main loop is re-entered, it
3013 becomes available for reuse.
3015 Note that the depth of the backtrack stack constantly increases during the
3016 left-to-right execution of the pattern, rather than going up and down with
3017 the pattern nesting. For example the stack is at its maximum at Z at the
3018 end of the pattern, rather than at X in the following:
3020 /(((X)+)+)+....(Y)+....Z/
3022 The only exceptions to this are lookahead/behind assertions and the cut,
3023 (?>A), which pop all the backtrack states associated with A before
3026 Backtrack state structs are allocated in slabs of about 4K in size.
3027 PL_regmatch_state and st always point to the currently active state,
3028 and PL_regmatch_slab points to the slab currently containing
3029 PL_regmatch_state. The first time regmatch() is called, the first slab is
3030 allocated, and is never freed until interpreter destruction. When the slab
3031 is full, a new one is allocated and chained to the end. At exit from
3032 regmatch(), slabs allocated since entry are freed.
3037 #define DEBUG_STATE_pp(pp) \
3039 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3040 PerlIO_printf(Perl_debug_log, \
3041 " %*s"pp" %s%s%s%s%s\n", \
3043 PL_reg_name[st->resume_state], \
3044 ((st==yes_state||st==mark_state) ? "[" : ""), \
3045 ((st==yes_state) ? "Y" : ""), \
3046 ((st==mark_state) ? "M" : ""), \
3047 ((st==yes_state||st==mark_state) ? "]" : "") \
3052 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3057 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3058 const char *start, const char *end, const char *blurb)
3060 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3062 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3067 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3068 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3070 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3071 start, end - start, 60);
3073 PerlIO_printf(Perl_debug_log,
3074 "%s%s REx%s %s against %s\n",
3075 PL_colors[4], blurb, PL_colors[5], s0, s1);
3077 if (utf8_target||utf8_pat)
3078 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3079 utf8_pat ? "pattern" : "",
3080 utf8_pat && utf8_target ? " and " : "",
3081 utf8_target ? "string" : ""
3087 S_dump_exec_pos(pTHX_ const char *locinput,
3088 const regnode *scan,
3089 const char *loc_regeol,
3090 const char *loc_bostr,
3091 const char *loc_reg_starttry,
3092 const bool utf8_target)
3094 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3095 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3096 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3097 /* The part of the string before starttry has one color
3098 (pref0_len chars), between starttry and current
3099 position another one (pref_len - pref0_len chars),
3100 after the current position the third one.
3101 We assume that pref0_len <= pref_len, otherwise we
3102 decrease pref0_len. */
3103 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3104 ? (5 + taill) - l : locinput - loc_bostr;
3107 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3109 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3111 pref0_len = pref_len - (locinput - loc_reg_starttry);
3112 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3113 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3114 ? (5 + taill) - pref_len : loc_regeol - locinput);
3115 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3119 if (pref0_len > pref_len)
3120 pref0_len = pref_len;
3122 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3124 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3125 (locinput - pref_len),pref0_len, 60, 4, 5);
3127 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3128 (locinput - pref_len + pref0_len),
3129 pref_len - pref0_len, 60, 2, 3);
3131 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3132 locinput, loc_regeol - locinput, 10, 0, 1);
3134 const STRLEN tlen=len0+len1+len2;
3135 PerlIO_printf(Perl_debug_log,
3136 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3137 (IV)(locinput - loc_bostr),
3140 (docolor ? "" : "> <"),
3142 (int)(tlen > 19 ? 0 : 19 - tlen),
3149 /* reg_check_named_buff_matched()
3150 * Checks to see if a named buffer has matched. The data array of
3151 * buffer numbers corresponding to the buffer is expected to reside
3152 * in the regexp->data->data array in the slot stored in the ARG() of
3153 * node involved. Note that this routine doesn't actually care about the
3154 * name, that information is not preserved from compilation to execution.
3155 * Returns the index of the leftmost defined buffer with the given name
3156 * or 0 if non of the buffers matched.
3159 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3162 RXi_GET_DECL(rex,rexi);
3163 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3164 I32 *nums=(I32*)SvPVX(sv_dat);
3166 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3168 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3169 if ((I32)rex->lastparen >= nums[n] &&
3170 rex->offs[nums[n]].end != -1)
3179 /* free all slabs above current one - called during LEAVE_SCOPE */
3182 S_clear_backtrack_stack(pTHX_ void *p)
3184 regmatch_slab *s = PL_regmatch_slab->next;
3189 PL_regmatch_slab->next = NULL;
3191 regmatch_slab * const osl = s;
3198 STATIC I32 /* 0 failure, 1 success */
3199 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
3201 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3205 const bool utf8_target = PL_reg_match_utf8;
3206 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3207 REGEXP *rex_sv = reginfo->prog;
3208 regexp *rex = (struct regexp *)SvANY(rex_sv);
3209 RXi_GET_DECL(rex,rexi);
3211 /* the current state. This is a cached copy of PL_regmatch_state */
3213 /* cache heavy used fields of st in registers */
3216 U32 n = 0; /* general value; init to avoid compiler warning */
3217 I32 ln = 0; /* len or last; init to avoid compiler warning */
3218 char *locinput = PL_reginput;
3219 I32 nextchr; /* is always set to UCHARAT(locinput) */
3221 bool result = 0; /* return value of S_regmatch */
3222 int depth = 0; /* depth of backtrack stack */
3223 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3224 const U32 max_nochange_depth =
3225 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3226 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3227 regmatch_state *yes_state = NULL; /* state to pop to on success of
3229 /* mark_state piggy backs on the yes_state logic so that when we unwind
3230 the stack on success we can update the mark_state as we go */
3231 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3232 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3233 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3235 bool no_final = 0; /* prevent failure from backtracking? */
3236 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3237 char *startpoint = PL_reginput;
3238 SV *popmark = NULL; /* are we looking for a mark? */
3239 SV *sv_commit = NULL; /* last mark name seen in failure */
3240 SV *sv_yes_mark = NULL; /* last mark name we have seen
3241 during a successful match */
3242 U32 lastopen = 0; /* last open we saw */
3243 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3244 SV* const oreplsv = GvSV(PL_replgv);
3245 /* these three flags are set by various ops to signal information to
3246 * the very next op. They have a useful lifetime of exactly one loop
3247 * iteration, and are not preserved or restored by state pushes/pops
3249 bool sw = 0; /* the condition value in (?(cond)a|b) */
3250 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3251 int logical = 0; /* the following EVAL is:
3255 or the following IFMATCH/UNLESSM is:
3256 false: plain (?=foo)
3257 true: used as a condition: (?(?=foo))
3259 PAD* last_pad = NULL;
3261 I32 gimme = G_SCALAR;
3262 CV *caller_cv = NULL; /* who called us */
3263 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3264 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3267 GET_RE_DEBUG_FLAGS_DECL;
3270 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3271 multicall_oldcatch = 0;
3272 multicall_cv = NULL;
3274 PERL_UNUSED_VAR(multicall_cop);
3275 PERL_UNUSED_VAR(newsp);
3278 PERL_ARGS_ASSERT_REGMATCH;
3280 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3281 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3283 /* on first ever call to regmatch, allocate first slab */
3284 if (!PL_regmatch_slab) {
3285 Newx(PL_regmatch_slab, 1, regmatch_slab);
3286 PL_regmatch_slab->prev = NULL;
3287 PL_regmatch_slab->next = NULL;
3288 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3291 oldsave = PL_savestack_ix;
3292 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3293 SAVEVPTR(PL_regmatch_slab);
3294 SAVEVPTR(PL_regmatch_state);
3296 /* grab next free state slot */
3297 st = ++PL_regmatch_state;
3298 if (st > SLAB_LAST(PL_regmatch_slab))
3299 st = PL_regmatch_state = S_push_slab(aTHX);
3301 /* Note that nextchr is a byte even in UTF */
3302 nextchr = UCHARAT(locinput);
3304 while (scan != NULL) {
3307 SV * const prop = sv_newmortal();
3308 regnode *rnext=regnext(scan);
3309 DUMP_EXEC_POS( locinput, scan, utf8_target );
3310 regprop(rex, prop, scan);
3312 PerlIO_printf(Perl_debug_log,
3313 "%3"IVdf":%*s%s(%"IVdf")\n",
3314 (IV)(scan - rexi->program), depth*2, "",
3316 (PL_regkind[OP(scan)] == END || !rnext) ?
3317 0 : (IV)(rnext - rexi->program));
3320 next = scan + NEXT_OFF(scan);
3323 state_num = OP(scan);
3327 switch (state_num) {
3329 if (locinput == PL_bostr)
3331 /* reginfo->till = reginfo->bol; */
3336 if (locinput == PL_bostr ||
3337 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3343 if (locinput == PL_bostr)
3347 if (locinput == reginfo->ganch)
3352 /* update the startpoint */
3353 st->u.keeper.val = rex->offs[0].start;
3354 PL_reginput = locinput;
3355 rex->offs[0].start = locinput - PL_bostr;
3356 PUSH_STATE_GOTO(KEEPS_next, next);
3358 case KEEPS_next_fail:
3359 /* rollback the start point change */
3360 rex->offs[0].start = st->u.keeper.val;
3366 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3371 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3373 if (PL_regeol - locinput > 1)
3377 if (PL_regeol != locinput)
3381 if (!nextchr && locinput >= PL_regeol)
3384 locinput += PL_utf8skip[nextchr];
3385 if (locinput > PL_regeol)
3387 nextchr = UCHARAT(locinput);
3390 nextchr = UCHARAT(++locinput);
3393 if (!nextchr && locinput >= PL_regeol)
3395 nextchr = UCHARAT(++locinput);
3398 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3401 locinput += PL_utf8skip[nextchr];
3402 if (locinput > PL_regeol)
3404 nextchr = UCHARAT(locinput);
3407 nextchr = UCHARAT(++locinput);
3411 #define ST st->u.trie
3413 /* In this case the charclass data is available inline so
3414 we can fail fast without a lot of extra overhead.
3416 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3418 PerlIO_printf(Perl_debug_log,
3419 "%*s %sfailed to match trie start class...%s\n",
3420 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3423 assert(0); /* NOTREACHED */
3427 /* the basic plan of execution of the trie is:
3428 * At the beginning, run though all the states, and
3429 * find the longest-matching word. Also remember the position
3430 * of the shortest matching word. For example, this pattern:
3433 * when matched against the string "abcde", will generate
3434 * accept states for all words except 3, with the longest
3435 * matching word being 4, and the shortest being 1 (with
3436 * the position being after char 1 of the string).
3438 * Then for each matching word, in word order (i.e. 1,2,4,5),
3439 * we run the remainder of the pattern; on each try setting
3440 * the current position to the character following the word,
3441 * returning to try the next word on failure.
3443 * We avoid having to build a list of words at runtime by
3444 * using a compile-time structure, wordinfo[].prev, which
3445 * gives, for each word, the previous accepting word (if any).
3446 * In the case above it would contain the mappings 1->2, 2->0,
3447 * 3->0, 4->5, 5->1. We can use this table to generate, from
3448 * the longest word (4 above), a list of all words, by
3449 * following the list of prev pointers; this gives us the
3450 * unordered list 4,5,1,2. Then given the current word we have
3451 * just tried, we can go through the list and find the
3452 * next-biggest word to try (so if we just failed on word 2,
3453 * the next in the list is 4).
3455 * Since at runtime we don't record the matching position in
3456 * the string for each word, we have to work that out for
3457 * each word we're about to process. The wordinfo table holds
3458 * the character length of each word; given that we recorded
3459 * at the start: the position of the shortest word and its
3460 * length in chars, we just need to move the pointer the
3461 * difference between the two char lengths. Depending on
3462 * Unicode status and folding, that's cheap or expensive.
3464 * This algorithm is optimised for the case where are only a
3465 * small number of accept states, i.e. 0,1, or maybe 2.
3466 * With lots of accepts states, and having to try all of them,
3467 * it becomes quadratic on number of accept states to find all
3472 /* what type of TRIE am I? (utf8 makes this contextual) */
3473 DECL_TRIE_TYPE(scan);
3475 /* what trie are we using right now */
3476 reg_trie_data * const trie
3477 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3478 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3479 U32 state = trie->startstate;
3481 if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
3482 if (trie->states[ state ].wordnum) {
3484 PerlIO_printf(Perl_debug_log,
3485 "%*s %smatched empty string...%s\n",
3486 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3492 PerlIO_printf(Perl_debug_log,
3493 "%*s %sfailed to match trie start class...%s\n",
3494 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3501 U8 *uc = ( U8* )locinput;
3505 U8 *uscan = (U8*)NULL;
3506 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3507 U32 charcount = 0; /* how many input chars we have matched */
3508 U32 accepted = 0; /* have we seen any accepting states? */
3510 ST.jump = trie->jump;
3513 ST.longfold = FALSE; /* char longer if folded => it's harder */
3516 /* fully traverse the TRIE; note the position of the
3517 shortest accept state and the wordnum of the longest
3520 while ( state && uc <= (U8*)PL_regeol ) {
3521 U32 base = trie->states[ state ].trans.base;
3525 wordnum = trie->states[ state ].wordnum;
3527 if (wordnum) { /* it's an accept state */
3530 /* record first match position */
3532 ST.firstpos = (U8*)locinput;
3537 ST.firstchars = charcount;
3540 if (!ST.nextword || wordnum < ST.nextword)
3541 ST.nextword = wordnum;
3542 ST.topword = wordnum;
3545 DEBUG_TRIE_EXECUTE_r({
3546 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3547 PerlIO_printf( Perl_debug_log,
3548 "%*s %sState: %4"UVxf" Accepted: %c ",
3549 2+depth * 2, "", PL_colors[4],
3550 (UV)state, (accepted ? 'Y' : 'N'));
3553 /* read a char and goto next state */
3556 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3557 uscan, len, uvc, charid, foldlen,
3564 base + charid - 1 - trie->uniquecharcount)) >= 0)
3566 && ((U32)offset < trie->lasttrans)
3567 && trie->trans[offset].check == state)
3569 state = trie->trans[offset].next;
3580 DEBUG_TRIE_EXECUTE_r(
3581 PerlIO_printf( Perl_debug_log,
3582 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3583 charid, uvc, (UV)state, PL_colors[5] );
3589 /* calculate total number of accept states */
3594 w = trie->wordinfo[w].prev;
3597 ST.accepted = accepted;
3601 PerlIO_printf( Perl_debug_log,
3602 "%*s %sgot %"IVdf" possible matches%s\n",
3603 REPORT_CODE_OFF + depth * 2, "",
3604 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3606 goto trie_first_try; /* jump into the fail handler */
3608 assert(0); /* NOTREACHED */
3610 case TRIE_next_fail: /* we failed - try next alternative */
3612 REGCP_UNWIND(ST.cp);
3613 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3615 if (!--ST.accepted) {
3617 PerlIO_printf( Perl_debug_log,
3618 "%*s %sTRIE failed...%s\n",
3619 REPORT_CODE_OFF+depth*2, "",
3626 /* Find next-highest word to process. Note that this code
3627 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3630 U16 const nextword = ST.nextword;
3631 reg_trie_wordinfo * const wordinfo
3632 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3633 for (word=ST.topword; word; word=wordinfo[word].prev) {
3634 if (word > nextword && (!min || word < min))
3647 ST.lastparen = rex->lastparen;
3648 ST.lastcloseparen = rex->lastcloseparen;
3652 /* find start char of end of current word */
3654 U32 chars; /* how many chars to skip */
3655 U8 *uc = ST.firstpos;
3656 reg_trie_data * const trie
3657 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3659 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3661 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3665 /* the hard option - fold each char in turn and find
3666 * its folded length (which may be different */
3667 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3675 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3683 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3688 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3702 PL_reginput = (char *)uc;
3705 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
3706 ? ST.jump[ST.nextword]
3710 PerlIO_printf( Perl_debug_log,
3711 "%*s %sTRIE matched word #%d, continuing%s\n",
3712 REPORT_CODE_OFF+depth*2, "",
3719 if (ST.accepted > 1 || has_cutgroup) {
3720 PUSH_STATE_GOTO(TRIE_next, scan);
3721 assert(0); /* NOTREACHED */
3723 /* only one choice left - just continue */
3725 AV *const trie_words
3726 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3727 SV ** const tmp = av_fetch( trie_words,
3729 SV *sv= tmp ? sv_newmortal() : NULL;
3731 PerlIO_printf( Perl_debug_log,
3732 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3733 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3735 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3736 PL_colors[0], PL_colors[1],
3737 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3739 : "not compiled under -Dr",
3743 locinput = PL_reginput;
3744 nextchr = UCHARAT(locinput);
3745 continue; /* execute rest of RE */
3746 assert(0); /* NOTREACHED */
3750 char *s = STRING(scan);
3752 if (utf8_target != UTF_PATTERN) {
3753 /* The target and the pattern have differing utf8ness. */
3755 const char * const e = s + ln;
3758 /* The target is utf8, the pattern is not utf8. */
3763 if (NATIVE_TO_UNI(*(U8*)s) !=
3764 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3772 /* The target is not utf8, the pattern is utf8. */
3777 if (NATIVE_TO_UNI(*((U8*)l)) !=
3778 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3786 nextchr = UCHARAT(locinput);
3789 /* The target and the pattern have the same utf8ness. */
3790 /* Inline the first character, for speed. */
3791 if (UCHARAT(s) != nextchr)
3793 if (PL_regeol - locinput < ln)
3795 if (ln > 1 && memNE(s, locinput, ln))
3798 nextchr = UCHARAT(locinput);
3803 const U8 * fold_array;
3805 U32 fold_utf8_flags;
3807 PL_reg_flags |= RF_tainted;
3808 folder = foldEQ_locale;
3809 fold_array = PL_fold_locale;
3810 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3814 case EXACTFU_TRICKYFOLD:
3816 folder = foldEQ_latin1;
3817 fold_array = PL_fold_latin1;
3818 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
3822 folder = foldEQ_latin1;
3823 fold_array = PL_fold_latin1;
3824 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3829 fold_array = PL_fold;
3830 fold_utf8_flags = 0;
3836 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
3837 /* Either target or the pattern are utf8, or has the issue where
3838 * the fold lengths may differ. */
3839 const char * const l = locinput;
3840 char *e = PL_regeol;
3842 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3843 l, &e, 0, utf8_target, fold_utf8_flags))
3848 nextchr = UCHARAT(locinput);
3852 /* Neither the target nor the pattern are utf8 */
3853 if (UCHARAT(s) != nextchr &&
3854 UCHARAT(s) != fold_array[nextchr])
3858 if (PL_regeol - locinput < ln)
3860 if (ln > 1 && ! folder(s, locinput, ln))
3863 nextchr = UCHARAT(locinput);
3867 /* XXX Could improve efficiency by separating these all out using a
3868 * macro or in-line function. At that point regcomp.c would no longer
3869 * have to set the FLAGS fields of these */
3872 PL_reg_flags |= RF_tainted;
3880 /* was last char in word? */
3882 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3883 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3885 if (locinput == PL_bostr)
3888 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3890 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3892 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3893 ln = isALNUM_uni(ln);
3894 LOAD_UTF8_CHARCLASS_ALNUM();
3895 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3898 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3899 n = isALNUM_LC_utf8((U8*)locinput);
3904 /* Here the string isn't utf8, or is utf8 and only ascii
3905 * characters are to match \w. In the latter case looking at
3906 * the byte just prior to the current one may be just the final
3907 * byte of a multi-byte character. This is ok. There are two
3909 * 1) it is a single byte character, and then the test is doing
3910 * just what it's supposed to.
3911 * 2) it is a multi-byte character, in which case the final
3912 * byte is never mistakable for ASCII, and so the test
3913 * will say it is not a word character, which is the
3914 * correct answer. */
3915 ln = (locinput != PL_bostr) ?
3916 UCHARAT(locinput - 1) : '\n';
3917 switch (FLAGS(scan)) {
3918 case REGEX_UNICODE_CHARSET:
3919 ln = isWORDCHAR_L1(ln);
3920 n = isWORDCHAR_L1(nextchr);
3922 case REGEX_LOCALE_CHARSET:
3923 ln = isALNUM_LC(ln);
3924 n = isALNUM_LC(nextchr);
3926 case REGEX_DEPENDS_CHARSET:
3928 n = isALNUM(nextchr);
3930 case REGEX_ASCII_RESTRICTED_CHARSET:
3931 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3932 ln = isWORDCHAR_A(ln);
3933 n = isWORDCHAR_A(nextchr);
3936 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3940 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3942 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3947 if (utf8_target || state_num == ANYOFV) {
3948 STRLEN inclasslen = PL_regeol - locinput;
3949 if (locinput >= PL_regeol)
3952 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3954 locinput += inclasslen;
3955 nextchr = UCHARAT(locinput);
3960 nextchr = UCHARAT(locinput);
3961 if (!nextchr && locinput >= PL_regeol)
3963 if (!REGINCLASS(rex, scan, (U8*)locinput))
3965 nextchr = UCHARAT(++locinput);
3969 /* Special char classes - The defines start on line 129 or so */
3970 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3971 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3972 ALNUMU, NALNUMU, isWORDCHAR_L1,
3973 ALNUMA, NALNUMA, isWORDCHAR_A,
3976 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3977 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3978 SPACEU, NSPACEU, isSPACE_L1,
3979 SPACEA, NSPACEA, isSPACE_A,
3982 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3983 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3984 DIGITA, NDIGITA, isDIGIT_A,
3988 if (locinput >= PL_regeol || ! _generic_isCC_A(nextchr, FLAGS(scan))) {
3991 /* Matched a utf8-invariant, so don't have to worry about utf8 */
3992 nextchr = UCHARAT(++locinput);
3995 if (locinput >= PL_regeol || _generic_isCC_A(nextchr, FLAGS(scan))) {
3999 locinput += PL_utf8skip[nextchr];
4000 nextchr = UCHARAT(locinput);
4003 nextchr = UCHARAT(++locinput);
4007 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4008 a Unicode extended Grapheme Cluster */
4009 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4010 extended Grapheme Cluster is:
4013 | Prepend* Begin Extend*
4016 Begin is: ( Special_Begin | ! Control )
4017 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4018 Extend is: ( Grapheme_Extend | Spacing_Mark )
4019 Control is: [ GCB_Control CR LF ]
4020 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4022 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4025 Begin is ( Regular_Begin + Special Begin )
4027 It turns out that 98.4% of all Unicode code points match
4028 Regular_Begin. Doing it this way eliminates a table match in
4029 the previouls implementation for almost all Unicode code points.
4031 There is a subtlety with Prepend* which showed up in testing.
4032 Note that the Begin, and only the Begin is required in:
4033 | Prepend* Begin Extend*
4034 Also, Begin contains '! Control'. A Prepend must be a
4035 '! Control', which means it must also be a Begin. What it
4036 comes down to is that if we match Prepend* and then find no
4037 suitable Begin afterwards, that if we backtrack the last
4038 Prepend, that one will be a suitable Begin.
4041 if (locinput >= PL_regeol)
4043 if (! utf8_target) {
4045 /* Match either CR LF or '.', as all the other possibilities
4047 locinput++; /* Match the . or CR */
4048 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4050 && locinput < PL_regeol
4051 && UCHARAT(locinput) == '\n') locinput++;
4055 /* Utf8: See if is ( CR LF ); already know that locinput <
4056 * PL_regeol, so locinput+1 is in bounds */
4057 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
4061 /* In case have to backtrack to beginning, then match '.' */
4062 char *starting = locinput;
4064 /* In case have to backtrack the last prepend */
4065 char *previous_prepend = 0;
4067 LOAD_UTF8_CHARCLASS_GCB();
4069 /* Match (prepend)*, but don't bother trying if empty (as
4070 * being set to _undef indicates) */
4071 if (PL_utf8_X_prepend != &PL_sv_undef) {
4072 while (locinput < PL_regeol
4073 && swash_fetch(PL_utf8_X_prepend,
4074 (U8*)locinput, utf8_target))
4076 previous_prepend = locinput;
4077 locinput += UTF8SKIP(locinput);
4081 /* As noted above, if we matched a prepend character, but
4082 * the next thing won't match, back off the last prepend we
4083 * matched, as it is guaranteed to match the begin */
4084 if (previous_prepend
4085 && (locinput >= PL_regeol
4086 || ! swash_fetch(PL_utf8_X_regular_begin,
4087 (U8*)locinput, utf8_target)))
4089 locinput = previous_prepend;
4092 /* Note that here we know PL_regeol > locinput, as we
4093 * tested that upon input to this switch case, and if we
4094 * moved locinput forward, we tested the result just above
4095 * and it either passed, or we backed off so that it will
4097 if (swash_fetch(PL_utf8_X_regular_begin,
4098 (U8*)locinput, utf8_target)) {
4099 locinput += UTF8SKIP(locinput);
4101 else if (! swash_fetch(PL_utf8_X_special_begin,
4102 (U8*)locinput, utf8_target))
4105 /* Here did not match the required 'Begin' in the
4106 * second term. So just match the very first
4107 * character, the '.' of the final term of the regex */
4108 locinput = starting + UTF8SKIP(starting);
4112 /* Here is a special begin. It can be composed of
4113 * several individual characters. One possibility is
4115 if (swash_fetch(PL_utf8_X_RI,
4116 (U8*)locinput, utf8_target))
4118 locinput += UTF8SKIP(locinput);
4119 while (locinput < PL_regeol
4120 && swash_fetch(PL_utf8_X_RI,
4121 (U8*)locinput, utf8_target))
4123 locinput += UTF8SKIP(locinput);
4125 } else /* Another possibility is T+ */
4126 if (swash_fetch(PL_utf8_X_T,
4127 (U8*)locinput, utf8_target))
4129 locinput += UTF8SKIP(locinput);
4130 while (locinput < PL_regeol
4131 && swash_fetch(PL_utf8_X_T,
4132 (U8*)locinput, utf8_target))
4134 locinput += UTF8SKIP(locinput);
4138 /* Here, neither RI+ nor T+; must be some other
4139 * Hangul. That means it is one of the others: L,
4140 * LV, LVT or V, and matches:
4141 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4144 while (locinput < PL_regeol
4145 && swash_fetch(PL_utf8_X_L,
4146 (U8*)locinput, utf8_target))
4148 locinput += UTF8SKIP(locinput);
4151 /* Here, have exhausted L*. If the next character
4152 * is not an LV, LVT nor V, it means we had to have
4153 * at least one L, so matches L+ in the original
4154 * equation, we have a complete hangul syllable.
4157 if (locinput < PL_regeol
4158 && swash_fetch(PL_utf8_X_LV_LVT_V,
4159 (U8*)locinput, utf8_target))
4162 /* Otherwise keep going. Must be LV, LVT or V.
4164 if (is_utf8_X_LVT((U8*)locinput)) {
4165 locinput += UTF8SKIP(locinput);
4168 /* Must be V or LV. Take it, then match
4170 locinput += UTF8SKIP(locinput);
4171 while (locinput < PL_regeol
4172 && swash_fetch(PL_utf8_X_V,
4176 locinput += UTF8SKIP(locinput);
4180 /* And any of LV, LVT, or V can be followed
4182 while (locinput < PL_regeol
4183 && swash_fetch(PL_utf8_X_T,
4187 locinput += UTF8SKIP(locinput);
4193 /* Match any extender */
4194 while (locinput < PL_regeol
4195 && swash_fetch(PL_utf8_X_extend,
4196 (U8*)locinput, utf8_target))
4198 locinput += UTF8SKIP(locinput);
4202 if (locinput > PL_regeol) sayNO;
4204 nextchr = UCHARAT(locinput);
4208 { /* The capture buffer cases. The ones beginning with N for the
4209 named buffers just convert to the equivalent numbered and
4210 pretend they were called as the corresponding numbered buffer
4212 /* don't initialize these in the declaration, it makes C++
4217 const U8 *fold_array;
4220 PL_reg_flags |= RF_tainted;
4221 folder = foldEQ_locale;
4222 fold_array = PL_fold_locale;
4224 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4228 folder = foldEQ_latin1;
4229 fold_array = PL_fold_latin1;
4231 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4235 folder = foldEQ_latin1;
4236 fold_array = PL_fold_latin1;
4238 utf8_fold_flags = 0;
4243 fold_array = PL_fold;
4245 utf8_fold_flags = 0;
4252 utf8_fold_flags = 0;
4255 /* For the named back references, find the corresponding buffer
4257 n = reg_check_named_buff_matched(rex,scan);
4262 goto do_nref_ref_common;
4265 PL_reg_flags |= RF_tainted;
4266 folder = foldEQ_locale;
4267 fold_array = PL_fold_locale;
4268 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4272 folder = foldEQ_latin1;
4273 fold_array = PL_fold_latin1;
4274 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4278 folder = foldEQ_latin1;
4279 fold_array = PL_fold_latin1;
4280 utf8_fold_flags = 0;
4285 fold_array = PL_fold;
4286 utf8_fold_flags = 0;
4292 utf8_fold_flags = 0;
4296 n = ARG(scan); /* which paren pair */
4299 ln = rex->offs[n].start;
4300 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4301 if (rex->lastparen < n || ln == -1)
4302 sayNO; /* Do not match unless seen CLOSEn. */
4303 if (ln == rex->offs[n].end)
4307 if (type != REF /* REF can do byte comparison */
4308 && (utf8_target || type == REFFU))
4309 { /* XXX handle REFFL better */
4310 char * limit = PL_regeol;
4312 /* This call case insensitively compares the entire buffer
4313 * at s, with the current input starting at locinput, but
4314 * not going off the end given by PL_regeol, and returns in
4315 * limit upon success, how much of the current input was
4317 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4318 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4323 nextchr = UCHARAT(locinput);
4327 /* Not utf8: Inline the first character, for speed. */
4328 if (UCHARAT(s) != nextchr &&
4330 UCHARAT(s) != fold_array[nextchr]))
4332 ln = rex->offs[n].end - ln;
4333 if (locinput + ln > PL_regeol)
4335 if (ln > 1 && (type == REF
4336 ? memNE(s, locinput, ln)
4337 : ! folder(s, locinput, ln)))
4340 nextchr = UCHARAT(locinput);
4350 #define ST st->u.eval
4355 regexp_internal *rei;
4356 regnode *startpoint;
4359 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4360 if (cur_eval && cur_eval->locinput==locinput) {
4361 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4362 Perl_croak(aTHX_ "Infinite recursion in regex");
4363 if ( ++nochange_depth > max_nochange_depth )
4365 "Pattern subroutine nesting without pos change"
4366 " exceeded limit in regex");
4373 if (OP(scan)==GOSUB) {
4374 startpoint = scan + ARG2L(scan);
4375 ST.close_paren = ARG(scan);
4377 startpoint = rei->program+1;
4380 goto eval_recurse_doit;
4381 assert(0); /* NOTREACHED */
4382 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4383 if (cur_eval && cur_eval->locinput==locinput) {
4384 if ( ++nochange_depth > max_nochange_depth )
4385 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4390 /* execute the code in the {...} */
4394 OP * const oop = PL_op;
4395 COP * const ocurcop = PL_curcop;
4397 char *saved_regeol = PL_regeol;
4398 struct re_save_state saved_state;
4401 /* save *all* paren positions */
4403 REGCP_SET(runops_cp);
4405 /* To not corrupt the existing regex state while executing the
4406 * eval we would normally put it on the save stack, like with
4407 * save_re_context. However, re-evals have a weird scoping so we
4408 * can't just add ENTER/LEAVE here. With that, things like
4410 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4412 * would break, as they expect the localisation to be unwound
4413 * only when the re-engine backtracks through the bit that
4416 * What we do instead is just saving the state in a local c
4419 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4421 PL_reg_state.re_reparsing = FALSE;
4424 caller_cv = find_runcv(NULL);
4428 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4429 newcv = ((struct regexp *)SvANY(
4430 (REGEXP*)(rexi->data->data[n])
4433 nop = (OP*)rexi->data->data[n+1];
4435 else if (rexi->data->what[n] == 'l') { /* literal code */
4437 nop = (OP*)rexi->data->data[n];
4438 assert(CvDEPTH(newcv));
4441 /* literal with own CV */
4442 assert(rexi->data->what[n] == 'L');
4443 newcv = rex->qr_anoncv;
4444 nop = (OP*)rexi->data->data[n];
4447 /* normally if we're about to execute code from the same
4448 * CV that we used previously, we just use the existing
4449 * CX stack entry. However, its possible that in the
4450 * meantime we may have backtracked, popped from the save
4451 * stack, and undone the SAVECOMPPAD(s) associated with
4452 * PUSH_MULTICALL; in which case PL_comppad no longer
4453 * points to newcv's pad. */
4454 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4456 I32 depth = (newcv == caller_cv) ? 0 : 1;
4457 if (last_pushed_cv) {
4458 CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
4461 PUSH_MULTICALL_WITHDEPTH(newcv, depth);
4463 last_pushed_cv = newcv;
4465 last_pad = PL_comppad;
4467 /* the initial nextstate you would normally execute
4468 * at the start of an eval (which would cause error
4469 * messages to come from the eval), may be optimised
4470 * away from the execution path in the regex code blocks;
4471 * so manually set PL_curcop to it initially */
4473 OP *o = cUNOPx(nop)->op_first;
4474 assert(o->op_type == OP_NULL);
4475 if (o->op_targ == OP_SCOPE) {
4476 o = cUNOPo->op_first;
4479 assert(o->op_targ == OP_LEAVE);
4480 o = cUNOPo->op_first;
4481 assert(o->op_type == OP_ENTER);
4485 if (o->op_type != OP_STUB) {
4486 assert( o->op_type == OP_NEXTSTATE
4487 || o->op_type == OP_DBSTATE
4488 || (o->op_type == OP_NULL
4489 && ( o->op_targ == OP_NEXTSTATE
4490 || o->op_targ == OP_DBSTATE
4494 PL_curcop = (COP*)o;
4499 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4500 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4502 rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4505 SV *sv_mrk = get_sv("REGMARK", 1);
4506 sv_setsv(sv_mrk, sv_yes_mark);
4509 /* we don't use MULTICALL here as we want to call the
4510 * first op of the block of interest, rather than the
4511 * first op of the sub */
4514 CALLRUNOPS(aTHX); /* Scalar context. */
4517 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4523 /* before restoring everything, evaluate the returned
4524 * value, so that 'uninit' warnings don't use the wrong
4525 * PL_op or pad. Also need to process any magic vars
4526 * (e.g. $1) *before* parentheses are restored */
4531 if (logical == 0) /* (?{})/ */
4532 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4533 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4534 sw = cBOOL(SvTRUE(ret));
4537 else { /* /(??{}) */
4538 /* if its overloaded, let the regex compiler handle
4539 * it; otherwise extract regex, or stringify */
4540 if (!SvAMAGIC(ret)) {
4544 if (SvTYPE(sv) == SVt_REGEXP)
4545 re_sv = (REGEXP*) sv;
4546 else if (SvSMAGICAL(sv)) {
4547 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
4549 re_sv = (REGEXP *) mg->mg_obj;
4552 /* force any magic, undef warnings here */
4554 ret = sv_mortalcopy(ret);
4555 (void) SvPV_force_nolen(ret);
4561 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4563 /* *** Note that at this point we don't restore
4564 * PL_comppad, (or pop the CxSUB) on the assumption it may
4565 * be used again soon. This is safe as long as nothing
4566 * in the regexp code uses the pad ! */
4568 PL_curcop = ocurcop;
4569 PL_regeol = saved_regeol;
4570 S_regcp_restore(aTHX_ rex, runops_cp);
4576 /* only /(??{})/ from now on */
4579 /* extract RE object from returned value; compiling if
4583 re_sv = reg_temp_copy(NULL, re_sv);
4587 const I32 osize = PL_regsize;
4589 if (SvUTF8(ret) && IN_BYTES) {
4590 /* In use 'bytes': make a copy of the octet
4591 * sequence, but without the flag on */
4593 const char *const p = SvPV(ret, len);
4594 ret = newSVpvn_flags(p, len, SVs_TEMP);
4596 if (rex->intflags & PREGf_USE_RE_EVAL)
4597 pm_flags |= PMf_USE_RE_EVAL;
4599 /* if we got here, it should be an engine which
4600 * supports compiling code blocks and stuff */
4601 assert(rex->engine && rex->engine->op_comp);
4602 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
4603 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
4604 rex->engine, NULL, NULL,
4605 /* copy /msix etc to inner pattern */
4610 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4612 /* This isn't a first class regexp. Instead, it's
4613 caching a regexp onto an existing, Perl visible
4615 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
4618 /* safe to do now that any $1 etc has been
4619 * interpolated into the new pattern string and
4621 S_regcp_restore(aTHX_ rex, runops_cp);
4623 re = (struct regexp *)SvANY(re_sv);
4625 RXp_MATCH_COPIED_off(re);
4626 re->subbeg = rex->subbeg;
4627 re->sublen = rex->sublen;
4628 re->suboffset = rex->suboffset;
4629 re->subcoffset = rex->subcoffset;
4632 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4633 "Matching embedded");
4635 startpoint = rei->program + 1;
4636 ST.close_paren = 0; /* only used for GOSUB */
4638 eval_recurse_doit: /* Share code with GOSUB below this line */
4639 /* run the pattern returned from (??{...}) */
4640 ST.cp = regcppush(rex, 0); /* Save *all* the positions. */
4641 REGCP_SET(ST.lastcp);
4644 re->lastcloseparen = 0;
4646 PL_reginput = locinput;
4649 /* XXXX This is too dramatic a measure... */
4652 ST.toggle_reg_flags = PL_reg_flags;
4654 PL_reg_flags |= RF_utf8;
4656 PL_reg_flags &= ~RF_utf8;
4657 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4659 ST.prev_rex = rex_sv;
4660 ST.prev_curlyx = cur_curlyx;
4662 SET_reg_curpm(rex_sv);
4667 ST.prev_eval = cur_eval;
4669 /* now continue from first node in postoned RE */
4670 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4671 assert(0); /* NOTREACHED */
4674 case EVAL_AB: /* cleanup after a successful (??{A})B */
4675 /* note: this is called twice; first after popping B, then A */
4676 PL_reg_flags ^= ST.toggle_reg_flags;
4677 rex_sv = ST.prev_rex;
4678 SET_reg_curpm(rex_sv);
4679 rex = (struct regexp *)SvANY(rex_sv);
4680 rexi = RXi_GET(rex);
4682 cur_eval = ST.prev_eval;
4683 cur_curlyx = ST.prev_curlyx;
4685 /* XXXX This is too dramatic a measure... */
4687 if ( nochange_depth )
4692 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4693 /* note: this is called twice; first after popping B, then A */
4694 PL_reg_flags ^= ST.toggle_reg_flags;
4695 rex_sv = ST.prev_rex;
4696 SET_reg_curpm(rex_sv);
4697 rex = (struct regexp *)SvANY(rex_sv);
4698 rexi = RXi_GET(rex);
4700 PL_reginput = locinput;
4701 REGCP_UNWIND(ST.lastcp);
4703 cur_eval = ST.prev_eval;
4704 cur_curlyx = ST.prev_curlyx;
4705 /* XXXX This is too dramatic a measure... */
4707 if ( nochange_depth )
4713 n = ARG(scan); /* which paren pair */
4714 rex->offs[n].start_tmp = locinput - PL_bostr;
4717 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
4718 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; regsize=%"UVuf"\n",
4722 (IV)rex->offs[n].start_tmp,
4728 /* XXX really need to log other places start/end are set too */
4729 #define CLOSE_CAPTURE \
4730 rex->offs[n].start = rex->offs[n].start_tmp; \
4731 rex->offs[n].end = locinput - PL_bostr; \
4732 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
4733 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
4735 PTR2UV(rex->offs), \
4737 (IV)rex->offs[n].start, \
4738 (IV)rex->offs[n].end \
4742 n = ARG(scan); /* which paren pair */
4744 /*if (n > PL_regsize)
4746 if (n > rex->lastparen)
4748 rex->lastcloseparen = n;
4749 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4757 cursor && OP(cursor)!=END;
4758 cursor=regnext(cursor))
4760 if ( OP(cursor)==CLOSE ){
4762 if ( n <= lastopen ) {
4764 /*if (n > PL_regsize)
4766 if (n > rex->lastparen)
4768 rex->lastcloseparen = n;
4769 if ( n == ARG(scan) || (cur_eval &&
4770 cur_eval->u.eval.close_paren == n))
4779 n = ARG(scan); /* which paren pair */
4780 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
4783 /* reg_check_named_buff_matched returns 0 for no match */
4784 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4788 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4794 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4796 next = NEXTOPER(NEXTOPER(scan));
4798 next = scan + ARG(scan);
4799 if (OP(next) == IFTHEN) /* Fake one. */
4800 next = NEXTOPER(NEXTOPER(next));
4804 logical = scan->flags;
4807 /*******************************************************************
4809 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4810 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4811 STAR/PLUS/CURLY/CURLYN are used instead.)
4813 A*B is compiled as <CURLYX><A><WHILEM><B>
4815 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4816 state, which contains the current count, initialised to -1. It also sets
4817 cur_curlyx to point to this state, with any previous value saved in the
4820 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4821 since the pattern may possibly match zero times (i.e. it's a while {} loop
4822 rather than a do {} while loop).
4824 Each entry to WHILEM represents a successful match of A. The count in the
4825 CURLYX block is incremented, another WHILEM state is pushed, and execution
4826 passes to A or B depending on greediness and the current count.
4828 For example, if matching against the string a1a2a3b (where the aN are
4829 substrings that match /A/), then the match progresses as follows: (the
4830 pushed states are interspersed with the bits of strings matched so far):
4833 <CURLYX cnt=0><WHILEM>
4834 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4835 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4836 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4837 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4839 (Contrast this with something like CURLYM, which maintains only a single
4843 a1 <CURLYM cnt=1> a2
4844 a1 a2 <CURLYM cnt=2> a3
4845 a1 a2 a3 <CURLYM cnt=3> b
4848 Each WHILEM state block marks a point to backtrack to upon partial failure
4849 of A or B, and also contains some minor state data related to that
4850 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4851 overall state, such as the count, and pointers to the A and B ops.
4853 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4854 must always point to the *current* CURLYX block, the rules are:
4856 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4857 and set cur_curlyx to point the new block.
4859 When popping the CURLYX block after a successful or unsuccessful match,
4860 restore the previous cur_curlyx.
4862 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4863 to the outer one saved in the CURLYX block.
4865 When popping the WHILEM block after a successful or unsuccessful B match,
4866 restore the previous cur_curlyx.
4868 Here's an example for the pattern (AI* BI)*BO
4869 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4872 curlyx backtrack stack
4873 ------ ---------------
4875 CO <CO prev=NULL> <WO>
4876 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4877 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4878 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4880 At this point the pattern succeeds, and we work back down the stack to
4881 clean up, restoring as we go:
4883 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4884 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4885 CO <CO prev=NULL> <WO>
4888 *******************************************************************/
4890 #define ST st->u.curlyx
4892 case CURLYX: /* start of /A*B/ (for complex A) */
4894 /* No need to save/restore up to this paren */
4895 I32 parenfloor = scan->flags;
4897 assert(next); /* keep Coverity happy */
4898 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4901 /* XXXX Probably it is better to teach regpush to support
4902 parenfloor > PL_regsize... */
4903 if (parenfloor > (I32)rex->lastparen)
4904 parenfloor = rex->lastparen; /* Pessimization... */
4906 ST.prev_curlyx= cur_curlyx;
4908 ST.cp = PL_savestack_ix;
4910 /* these fields contain the state of the current curly.
4911 * they are accessed by subsequent WHILEMs */
4912 ST.parenfloor = parenfloor;
4917 ST.count = -1; /* this will be updated by WHILEM */
4918 ST.lastloc = NULL; /* this will be updated by WHILEM */
4920 PL_reginput = locinput;
4921 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4922 assert(0); /* NOTREACHED */
4925 case CURLYX_end: /* just finished matching all of A*B */
4926 cur_curlyx = ST.prev_curlyx;
4928 assert(0); /* NOTREACHED */
4930 case CURLYX_end_fail: /* just failed to match all of A*B */
4932 cur_curlyx = ST.prev_curlyx;
4934 assert(0); /* NOTREACHED */
4938 #define ST st->u.whilem
4940 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4942 /* see the discussion above about CURLYX/WHILEM */
4944 int min = ARG1(cur_curlyx->u.curlyx.me);
4945 int max = ARG2(cur_curlyx->u.curlyx.me);
4946 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4948 assert(cur_curlyx); /* keep Coverity happy */
4949 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4950 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4951 ST.cache_offset = 0;
4954 PL_reginput = locinput;
4956 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4957 "%*s whilem: matched %ld out of %d..%d\n",
4958 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4961 /* First just match a string of min A's. */
4964 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
4965 cur_curlyx->u.curlyx.lastloc = locinput;
4966 REGCP_SET(ST.lastcp);
4968 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4969 assert(0); /* NOTREACHED */
4972 /* If degenerate A matches "", assume A done. */
4974 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4975 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4976 "%*s whilem: empty match detected, trying continuation...\n",
4977 REPORT_CODE_OFF+depth*2, "")
4979 goto do_whilem_B_max;
4982 /* super-linear cache processing */
4986 if (!PL_reg_maxiter) {
4987 /* start the countdown: Postpone detection until we
4988 * know the match is not *that* much linear. */
4989 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4990 /* possible overflow for long strings and many CURLYX's */
4991 if (PL_reg_maxiter < 0)
4992 PL_reg_maxiter = I32_MAX;
4993 PL_reg_leftiter = PL_reg_maxiter;
4996 if (PL_reg_leftiter-- == 0) {
4997 /* initialise cache */
4998 const I32 size = (PL_reg_maxiter + 7)/8;
4999 if (PL_reg_poscache) {
5000 if ((I32)PL_reg_poscache_size < size) {
5001 Renew(PL_reg_poscache, size, char);
5002 PL_reg_poscache_size = size;
5004 Zero(PL_reg_poscache, size, char);
5007 PL_reg_poscache_size = size;
5008 Newxz(PL_reg_poscache, size, char);
5010 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5011 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5012 PL_colors[4], PL_colors[5])
5016 if (PL_reg_leftiter < 0) {
5017 /* have we already failed at this position? */
5019 offset = (scan->flags & 0xf) - 1
5020 + (locinput - PL_bostr) * (scan->flags>>4);
5021 mask = 1 << (offset % 8);
5023 if (PL_reg_poscache[offset] & mask) {
5024 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5025 "%*s whilem: (cache) already tried at this position...\n",
5026 REPORT_CODE_OFF+depth*2, "")
5028 sayNO; /* cache records failure */
5030 ST.cache_offset = offset;
5031 ST.cache_mask = mask;
5035 /* Prefer B over A for minimal matching. */
5037 if (cur_curlyx->u.curlyx.minmod) {
5038 ST.save_curlyx = cur_curlyx;
5039 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5040 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor);
5041 REGCP_SET(ST.lastcp);
5042 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
5043 assert(0); /* NOTREACHED */
5046 /* Prefer A over B for maximal matching. */
5048 if (n < max) { /* More greed allowed? */
5049 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5050 cur_curlyx->u.curlyx.lastloc = locinput;
5051 REGCP_SET(ST.lastcp);
5052 PUSH_STATE_GOTO(WHILEM_A_max, A);
5053 assert(0); /* NOTREACHED */
5055 goto do_whilem_B_max;
5057 assert(0); /* NOTREACHED */
5059 case WHILEM_B_min: /* just matched B in a minimal match */
5060 case WHILEM_B_max: /* just matched B in a maximal match */
5061 cur_curlyx = ST.save_curlyx;
5063 assert(0); /* NOTREACHED */
5065 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5066 cur_curlyx = ST.save_curlyx;
5067 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5068 cur_curlyx->u.curlyx.count--;
5070 assert(0); /* NOTREACHED */
5072 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5074 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5075 REGCP_UNWIND(ST.lastcp);
5077 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5078 cur_curlyx->u.curlyx.count--;
5080 assert(0); /* NOTREACHED */
5082 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5083 REGCP_UNWIND(ST.lastcp);
5084 regcppop(rex); /* Restore some previous $<digit>s? */
5085 PL_reginput = locinput;
5086 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5087 "%*s whilem: failed, trying continuation...\n",
5088 REPORT_CODE_OFF+depth*2, "")
5091 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5092 && ckWARN(WARN_REGEXP)
5093 && !(PL_reg_flags & RF_warned))
5095 PL_reg_flags |= RF_warned;
5096 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5097 "Complex regular subexpression recursion limit (%d) "
5103 ST.save_curlyx = cur_curlyx;
5104 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5105 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
5106 assert(0); /* NOTREACHED */
5108 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5109 cur_curlyx = ST.save_curlyx;
5110 REGCP_UNWIND(ST.lastcp);
5113 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5114 /* Maximum greed exceeded */
5115 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5116 && ckWARN(WARN_REGEXP)
5117 && !(PL_reg_flags & RF_warned))
5119 PL_reg_flags |= RF_warned;
5120 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5121 "Complex regular subexpression recursion "
5122 "limit (%d) exceeded",
5125 cur_curlyx->u.curlyx.count--;
5129 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5130 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5132 /* Try grabbing another A and see if it helps. */
5133 PL_reginput = locinput;
5134 cur_curlyx->u.curlyx.lastloc = locinput;
5135 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5136 REGCP_SET(ST.lastcp);
5137 PUSH_STATE_GOTO(WHILEM_A_min,
5138 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
5139 assert(0); /* NOTREACHED */
5142 #define ST st->u.branch
5144 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5145 next = scan + ARG(scan);
5148 scan = NEXTOPER(scan);
5151 case BRANCH: /* /(...|A|...)/ */
5152 scan = NEXTOPER(scan); /* scan now points to inner node */
5153 ST.lastparen = rex->lastparen;
5154 ST.lastcloseparen = rex->lastcloseparen;
5155 ST.next_branch = next;
5157 PL_reginput = locinput;
5159 /* Now go into the branch */
5161 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
5163 PUSH_STATE_GOTO(BRANCH_next, scan);
5165 assert(0); /* NOTREACHED */
5167 PL_reginput = locinput;
5168 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5169 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5170 PUSH_STATE_GOTO(CUTGROUP_next,next);
5171 assert(0); /* NOTREACHED */
5172 case CUTGROUP_next_fail:
5175 if (st->u.mark.mark_name)
5176 sv_commit = st->u.mark.mark_name;
5178 assert(0); /* NOTREACHED */
5181 assert(0); /* NOTREACHED */
5182 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5187 REGCP_UNWIND(ST.cp);
5188 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5189 scan = ST.next_branch;
5190 /* no more branches? */
5191 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5193 PerlIO_printf( Perl_debug_log,
5194 "%*s %sBRANCH failed...%s\n",
5195 REPORT_CODE_OFF+depth*2, "",
5201 continue; /* execute next BRANCH[J] op */
5202 assert(0); /* NOTREACHED */
5209 #define ST st->u.curlym
5211 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5213 /* This is an optimisation of CURLYX that enables us to push
5214 * only a single backtracking state, no matter how many matches
5215 * there are in {m,n}. It relies on the pattern being constant
5216 * length, with no parens to influence future backrefs
5220 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5222 ST.lastparen = rex->lastparen;
5223 ST.lastcloseparen = rex->lastcloseparen;
5225 /* if paren positive, emulate an OPEN/CLOSE around A */
5227 U32 paren = ST.me->flags;
5228 if (paren > PL_regsize)
5230 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5238 ST.c1 = CHRTEST_UNINIT;
5241 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5244 curlym_do_A: /* execute the A in /A{m,n}B/ */
5245 PL_reginput = locinput;
5246 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
5247 assert(0); /* NOTREACHED */
5249 case CURLYM_A: /* we've just matched an A */
5250 locinput = st->locinput;
5251 nextchr = UCHARAT(locinput);
5254 /* after first match, determine A's length: u.curlym.alen */
5255 if (ST.count == 1) {
5256 if (PL_reg_match_utf8) {
5258 while (s < PL_reginput) {
5264 ST.alen = PL_reginput - locinput;
5267 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5270 PerlIO_printf(Perl_debug_log,
5271 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5272 (int)(REPORT_CODE_OFF+(depth*2)), "",
5273 (IV) ST.count, (IV)ST.alen)
5276 locinput = PL_reginput;
5278 if (cur_eval && cur_eval->u.eval.close_paren &&
5279 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5283 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5284 if ( max == REG_INFTY || ST.count < max )
5285 goto curlym_do_A; /* try to match another A */
5287 goto curlym_do_B; /* try to match B */
5289 case CURLYM_A_fail: /* just failed to match an A */
5290 REGCP_UNWIND(ST.cp);
5292 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5293 || (cur_eval && cur_eval->u.eval.close_paren &&
5294 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5297 curlym_do_B: /* execute the B in /A{m,n}B/ */
5298 PL_reginput = locinput;
5299 if (ST.c1 == CHRTEST_UNINIT) {
5300 /* calculate c1 and c2 for possible match of 1st char
5301 * following curly */
5302 ST.c1 = ST.c2 = CHRTEST_VOID;
5303 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5304 regnode *text_node = ST.B;
5305 if (! HAS_TEXT(text_node))
5306 FIND_NEXT_IMPT(text_node);
5309 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5311 But the former is redundant in light of the latter.
5313 if this changes back then the macro for
5314 IS_TEXT and friends need to change.
5316 if (PL_regkind[OP(text_node)] == EXACT)
5319 ST.c1 = (U8)*STRING(text_node);
5320 switch (OP(text_node)) {
5321 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5324 case EXACTFU_TRICKYFOLD:
5325 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5326 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5327 default: ST.c2 = ST.c1;
5334 PerlIO_printf(Perl_debug_log,
5335 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5336 (int)(REPORT_CODE_OFF+(depth*2)),
5339 if (ST.c1 != CHRTEST_VOID
5340 && UCHARAT(PL_reginput) != ST.c1
5341 && UCHARAT(PL_reginput) != ST.c2)
5343 /* simulate B failing */
5345 PerlIO_printf(Perl_debug_log,
5346 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5347 (int)(REPORT_CODE_OFF+(depth*2)),"",
5350 state_num = CURLYM_B_fail;
5351 goto reenter_switch;
5355 /* emulate CLOSE: mark current A as captured */
5356 I32 paren = ST.me->flags;
5358 rex->offs[paren].start
5359 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5360 rex->offs[paren].end = PL_reginput - PL_bostr;
5361 if ((U32)paren > rex->lastparen)
5362 rex->lastparen = paren;
5363 rex->lastcloseparen = paren;
5366 rex->offs[paren].end = -1;
5367 if (cur_eval && cur_eval->u.eval.close_paren &&
5368 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5377 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5378 assert(0); /* NOTREACHED */
5380 case CURLYM_B_fail: /* just failed to match a B */
5381 REGCP_UNWIND(ST.cp);
5382 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5384 I32 max = ARG2(ST.me);
5385 if (max != REG_INFTY && ST.count == max)
5387 goto curlym_do_A; /* try to match a further A */
5389 /* backtrack one A */
5390 if (ST.count == ARG1(ST.me) /* min */)
5393 locinput = HOPc(locinput, -ST.alen);
5394 goto curlym_do_B; /* try to match B */
5397 #define ST st->u.curly
5399 #define CURLY_SETPAREN(paren, success) \
5402 rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5403 rex->offs[paren].end = locinput - PL_bostr; \
5404 if (paren > rex->lastparen) \
5405 rex->lastparen = paren; \
5406 rex->lastcloseparen = paren; \
5409 rex->offs[paren].end = -1; \
5410 rex->lastparen = ST.lastparen; \
5411 rex->lastcloseparen = ST.lastcloseparen; \
5415 case STAR: /* /A*B/ where A is width 1 */
5419 scan = NEXTOPER(scan);
5421 case PLUS: /* /A+B/ where A is width 1 */
5425 scan = NEXTOPER(scan);
5427 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5428 ST.paren = scan->flags; /* Which paren to set */
5429 ST.lastparen = rex->lastparen;
5430 ST.lastcloseparen = rex->lastcloseparen;
5431 if (ST.paren > PL_regsize)
5432 PL_regsize = ST.paren;
5433 ST.min = ARG1(scan); /* min to match */
5434 ST.max = ARG2(scan); /* max to match */
5435 if (cur_eval && cur_eval->u.eval.close_paren &&
5436 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5440 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5442 case CURLY: /* /A{m,n}B/ where A is width 1 */
5444 ST.min = ARG1(scan); /* min to match */
5445 ST.max = ARG2(scan); /* max to match */
5446 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5449 * Lookahead to avoid useless match attempts
5450 * when we know what character comes next.
5452 * Used to only do .*x and .*?x, but now it allows
5453 * for )'s, ('s and (?{ ... })'s to be in the way
5454 * of the quantifier and the EXACT-like node. -- japhy
5457 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5459 if (HAS_TEXT(next) || JUMPABLE(next)) {
5461 regnode *text_node = next;
5463 if (! HAS_TEXT(text_node))
5464 FIND_NEXT_IMPT(text_node);
5466 if (! HAS_TEXT(text_node))
5467 ST.c1 = ST.c2 = CHRTEST_VOID;
5469 if ( PL_regkind[OP(text_node)] != EXACT ) {
5470 ST.c1 = ST.c2 = CHRTEST_VOID;
5471 goto assume_ok_easy;
5474 s = (U8*)STRING(text_node);
5476 /* Currently we only get here when
5478 PL_rekind[OP(text_node)] == EXACT
5480 if this changes back then the macro for IS_TEXT and
5481 friends need to change. */
5484 switch (OP(text_node)) {
5485 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5488 case EXACTFU_TRICKYFOLD:
5489 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5490 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5491 default: ST.c2 = ST.c1; break;
5494 else { /* UTF_PATTERN */
5495 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5497 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
5499 to_utf8_fold((U8*)s, tmpbuf, &ulen);
5500 ST.c1 = ST.c2 = utf8n_to_uvchr(tmpbuf, UTF8_MAXLEN, 0,
5504 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5511 ST.c1 = ST.c2 = CHRTEST_VOID;
5516 PL_reginput = locinput;
5519 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5522 locinput = PL_reginput;
5524 if (ST.c1 == CHRTEST_VOID)
5525 goto curly_try_B_min;
5527 ST.oldloc = locinput;
5529 /* set ST.maxpos to the furthest point along the
5530 * string that could possibly match */
5531 if (ST.max == REG_INFTY) {
5532 ST.maxpos = PL_regeol - 1;
5534 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5537 else if (utf8_target) {
5538 int m = ST.max - ST.min;
5539 for (ST.maxpos = locinput;
5540 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5541 ST.maxpos += UTF8SKIP(ST.maxpos);
5544 ST.maxpos = locinput + ST.max - ST.min;
5545 if (ST.maxpos >= PL_regeol)
5546 ST.maxpos = PL_regeol - 1;
5548 goto curly_try_B_min_known;
5552 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5553 locinput = PL_reginput;
5554 if (ST.count < ST.min)
5556 if ((ST.count > ST.min)
5557 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5559 /* A{m,n} must come at the end of the string, there's
5560 * no point in backing off ... */
5562 /* ...except that $ and \Z can match before *and* after
5563 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5564 We may back off by one in this case. */
5565 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5569 goto curly_try_B_max;
5571 assert(0); /* NOTREACHED */
5574 case CURLY_B_min_known_fail:
5575 /* failed to find B in a non-greedy match where c1,c2 valid */
5577 PL_reginput = locinput; /* Could be reset... */
5578 REGCP_UNWIND(ST.cp);
5580 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5582 /* Couldn't or didn't -- move forward. */
5583 ST.oldloc = locinput;
5585 locinput += UTF8SKIP(locinput);
5589 curly_try_B_min_known:
5590 /* find the next place where 'B' could work, then call B */
5594 n = (ST.oldloc == locinput) ? 0 : 1;
5595 if (ST.c1 == ST.c2) {
5597 /* set n to utf8_distance(oldloc, locinput) */
5598 while (locinput <= ST.maxpos &&
5599 utf8n_to_uvchr((U8*)locinput,
5600 UTF8_MAXBYTES, &len,
5601 uniflags) != (UV)ST.c1) {
5607 /* set n to utf8_distance(oldloc, locinput) */
5608 while (locinput <= ST.maxpos) {
5610 const UV c = utf8n_to_uvchr((U8*)locinput,
5611 UTF8_MAXBYTES, &len,
5613 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5621 if (ST.c1 == ST.c2) {
5622 while (locinput <= ST.maxpos &&
5623 UCHARAT(locinput) != ST.c1)
5627 while (locinput <= ST.maxpos
5628 && UCHARAT(locinput) != ST.c1
5629 && UCHARAT(locinput) != ST.c2)
5632 n = locinput - ST.oldloc;
5634 if (locinput > ST.maxpos)
5636 /* PL_reginput == oldloc now */
5639 if (regrepeat(rex, ST.A, n, depth) < n)
5642 PL_reginput = locinput;
5643 CURLY_SETPAREN(ST.paren, ST.count);
5644 if (cur_eval && cur_eval->u.eval.close_paren &&
5645 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5648 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5650 assert(0); /* NOTREACHED */
5653 case CURLY_B_min_fail:
5654 /* failed to find B in a non-greedy match where c1,c2 invalid */
5656 REGCP_UNWIND(ST.cp);
5658 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5660 /* failed -- move forward one */
5661 PL_reginput = locinput;
5662 if (regrepeat(rex, ST.A, 1, depth)) {
5664 locinput = PL_reginput;
5665 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5666 ST.count > 0)) /* count overflow ? */
5669 CURLY_SETPAREN(ST.paren, ST.count);
5670 if (cur_eval && cur_eval->u.eval.close_paren &&
5671 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5674 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5678 assert(0); /* NOTREACHED */
5682 /* a successful greedy match: now try to match B */
5683 if (cur_eval && cur_eval->u.eval.close_paren &&
5684 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5689 if (ST.c1 != CHRTEST_VOID)
5690 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5691 UTF8_MAXBYTES, 0, uniflags)
5692 : (UV) UCHARAT(PL_reginput);
5693 /* If it could work, try it. */
5694 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5695 CURLY_SETPAREN(ST.paren, ST.count);
5696 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5697 assert(0); /* NOTREACHED */
5701 case CURLY_B_max_fail:
5702 /* failed to find B in a greedy match */
5704 REGCP_UNWIND(ST.cp);
5706 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5709 if (--ST.count < ST.min)
5711 PL_reginput = locinput = HOPc(locinput, -1);
5712 goto curly_try_B_max;
5719 /* we've just finished A in /(??{A})B/; now continue with B */
5720 st->u.eval.toggle_reg_flags
5721 = cur_eval->u.eval.toggle_reg_flags;
5722 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5724 st->u.eval.prev_rex = rex_sv; /* inner */
5725 st->u.eval.cp = regcppush(rex, 0); /* Save *all* the positions. */
5726 rex_sv = cur_eval->u.eval.prev_rex;
5727 SET_reg_curpm(rex_sv);
5728 rex = (struct regexp *)SvANY(rex_sv);
5729 rexi = RXi_GET(rex);
5730 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5732 REGCP_SET(st->u.eval.lastcp);
5733 PL_reginput = locinput;
5735 /* Restore parens of the outer rex without popping the
5737 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp);
5739 st->u.eval.prev_eval = cur_eval;
5740 cur_eval = cur_eval->u.eval.prev_eval;
5742 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5743 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5744 if ( nochange_depth )
5747 PUSH_YES_STATE_GOTO(EVAL_AB,
5748 st->u.eval.prev_eval->u.eval.B); /* match B */
5751 if (locinput < reginfo->till) {
5752 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5753 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5755 (long)(locinput - PL_reg_starttry),
5756 (long)(reginfo->till - PL_reg_starttry),
5759 sayNO_SILENT; /* Cannot match: too short. */
5761 PL_reginput = locinput; /* put where regtry can find it */
5762 sayYES; /* Success! */
5764 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5766 PerlIO_printf(Perl_debug_log,
5767 "%*s %ssubpattern success...%s\n",
5768 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5769 PL_reginput = locinput; /* put where regtry can find it */
5770 sayYES; /* Success! */
5773 #define ST st->u.ifmatch
5775 case SUSPEND: /* (?>A) */
5777 PL_reginput = locinput;
5780 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5782 goto ifmatch_trivial_fail_test;
5784 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5786 ifmatch_trivial_fail_test:
5788 char * const s = HOPBACKc(locinput, scan->flags);
5793 sw = 1 - cBOOL(ST.wanted);
5797 next = scan + ARG(scan);
5805 PL_reginput = locinput;
5809 ST.logical = logical;
5810 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5812 /* execute body of (?...A) */
5813 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5814 assert(0); /* NOTREACHED */
5816 case IFMATCH_A_fail: /* body of (?...A) failed */
5817 ST.wanted = !ST.wanted;
5820 case IFMATCH_A: /* body of (?...A) succeeded */
5822 sw = cBOOL(ST.wanted);
5824 else if (!ST.wanted)
5827 if (OP(ST.me) == SUSPEND)
5828 locinput = PL_reginput;
5830 locinput = PL_reginput = st->locinput;
5831 nextchr = UCHARAT(locinput);
5833 scan = ST.me + ARG(ST.me);
5836 continue; /* execute B */
5841 next = scan + ARG(scan);
5846 reginfo->cutpoint = PL_regeol;
5849 PL_reginput = locinput;
5851 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5852 PUSH_STATE_GOTO(COMMIT_next,next);
5853 assert(0); /* NOTREACHED */
5854 case COMMIT_next_fail:
5859 assert(0); /* NOTREACHED */
5861 #define ST st->u.mark
5863 ST.prev_mark = mark_state;
5864 ST.mark_name = sv_commit = sv_yes_mark
5865 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5867 ST.mark_loc = PL_reginput = locinput;
5868 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5869 assert(0); /* NOTREACHED */
5870 case MARKPOINT_next:
5871 mark_state = ST.prev_mark;
5873 assert(0); /* NOTREACHED */
5874 case MARKPOINT_next_fail:
5875 if (popmark && sv_eq(ST.mark_name,popmark))
5877 if (ST.mark_loc > startpoint)
5878 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5879 popmark = NULL; /* we found our mark */
5880 sv_commit = ST.mark_name;
5883 PerlIO_printf(Perl_debug_log,
5884 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5885 REPORT_CODE_OFF+depth*2, "",
5886 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5889 mark_state = ST.prev_mark;
5890 sv_yes_mark = mark_state ?
5891 mark_state->u.mark.mark_name : NULL;
5893 assert(0); /* NOTREACHED */
5895 PL_reginput = locinput;
5897 /* (*SKIP) : if we fail we cut here*/
5898 ST.mark_name = NULL;
5899 ST.mark_loc = locinput;
5900 PUSH_STATE_GOTO(SKIP_next,next);
5902 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5903 otherwise do nothing. Meaning we need to scan
5905 regmatch_state *cur = mark_state;
5906 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5909 if ( sv_eq( cur->u.mark.mark_name,
5912 ST.mark_name = find;
5913 PUSH_STATE_GOTO( SKIP_next, next );
5915 cur = cur->u.mark.prev_mark;
5918 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5920 case SKIP_next_fail:
5922 /* (*CUT:NAME) - Set up to search for the name as we
5923 collapse the stack*/
5924 popmark = ST.mark_name;
5926 /* (*CUT) - No name, we cut here.*/
5927 if (ST.mark_loc > startpoint)
5928 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5929 /* but we set sv_commit to latest mark_name if there
5930 is one so they can test to see how things lead to this
5933 sv_commit=mark_state->u.mark.mark_name;
5937 assert(0); /* NOTREACHED */
5940 if ((n=is_LNBREAK(locinput,utf8_target))) {
5942 nextchr = UCHARAT(locinput);
5947 #define CASE_CLASS(nAmE) \
5949 if (locinput >= PL_regeol) \
5951 if ((n=is_##nAmE(locinput,utf8_target))) { \
5953 nextchr = UCHARAT(locinput); \
5958 if (locinput >= PL_regeol) \
5960 if ((n=is_##nAmE(locinput,utf8_target))) { \
5963 locinput += UTF8SKIP(locinput); \
5964 nextchr = UCHARAT(locinput); \
5969 CASE_CLASS(HORIZWS);
5973 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5974 PTR2UV(scan), OP(scan));
5975 Perl_croak(aTHX_ "regexp memory corruption");
5979 /* switch break jumps here */
5980 scan = next; /* prepare to execute the next op and ... */
5981 continue; /* ... jump back to the top, reusing st */
5982 assert(0); /* NOTREACHED */
5985 /* push a state that backtracks on success */
5986 st->u.yes.prev_yes_state = yes_state;
5990 /* push a new regex state, then continue at scan */
5992 regmatch_state *newst;
5995 regmatch_state *cur = st;
5996 regmatch_state *curyes = yes_state;
5998 regmatch_slab *slab = PL_regmatch_slab;
5999 for (;curd > -1;cur--,curd--) {
6000 if (cur < SLAB_FIRST(slab)) {
6002 cur = SLAB_LAST(slab);
6004 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6005 REPORT_CODE_OFF + 2 + depth * 2,"",
6006 curd, PL_reg_name[cur->resume_state],
6007 (curyes == cur) ? "yes" : ""
6010 curyes = cur->u.yes.prev_yes_state;
6013 DEBUG_STATE_pp("push")
6016 st->locinput = locinput;
6018 if (newst > SLAB_LAST(PL_regmatch_slab))
6019 newst = S_push_slab(aTHX);
6020 PL_regmatch_state = newst;
6022 locinput = PL_reginput;
6023 nextchr = UCHARAT(locinput);
6026 assert(0); /* NOTREACHED */
6031 * We get here only if there's trouble -- normally "case END" is
6032 * the terminating point.
6034 Perl_croak(aTHX_ "corrupted regexp pointers");
6040 /* we have successfully completed a subexpression, but we must now
6041 * pop to the state marked by yes_state and continue from there */
6042 assert(st != yes_state);
6044 while (st != yes_state) {
6046 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6047 PL_regmatch_slab = PL_regmatch_slab->prev;
6048 st = SLAB_LAST(PL_regmatch_slab);
6052 DEBUG_STATE_pp("pop (no final)");
6054 DEBUG_STATE_pp("pop (yes)");
6060 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6061 || yes_state > SLAB_LAST(PL_regmatch_slab))
6063 /* not in this slab, pop slab */
6064 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6065 PL_regmatch_slab = PL_regmatch_slab->prev;
6066 st = SLAB_LAST(PL_regmatch_slab);
6068 depth -= (st - yes_state);
6071 yes_state = st->u.yes.prev_yes_state;
6072 PL_regmatch_state = st;
6075 locinput= st->locinput;
6076 nextchr = UCHARAT(locinput);
6078 state_num = st->resume_state + no_final;
6079 goto reenter_switch;
6082 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6083 PL_colors[4], PL_colors[5]));
6085 if (PL_reg_state.re_state_eval_setup_done) {
6086 /* each successfully executed (?{...}) block does the equivalent of
6087 * local $^R = do {...}
6088 * When popping the save stack, all these locals would be undone;
6089 * bypass this by setting the outermost saved $^R to the latest
6091 if (oreplsv != GvSV(PL_replgv))
6092 sv_setsv(oreplsv, GvSV(PL_replgv));
6099 PerlIO_printf(Perl_debug_log,
6100 "%*s %sfailed...%s\n",
6101 REPORT_CODE_OFF+depth*2, "",
6102 PL_colors[4], PL_colors[5])
6114 /* there's a previous state to backtrack to */
6116 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6117 PL_regmatch_slab = PL_regmatch_slab->prev;
6118 st = SLAB_LAST(PL_regmatch_slab);
6120 PL_regmatch_state = st;
6121 locinput= st->locinput;
6122 nextchr = UCHARAT(locinput);
6124 DEBUG_STATE_pp("pop");
6126 if (yes_state == st)
6127 yes_state = st->u.yes.prev_yes_state;
6129 state_num = st->resume_state + 1; /* failure = success + 1 */
6130 goto reenter_switch;
6135 if (rex->intflags & PREGf_VERBARG_SEEN) {
6136 SV *sv_err = get_sv("REGERROR", 1);
6137 SV *sv_mrk = get_sv("REGMARK", 1);
6139 sv_commit = &PL_sv_no;
6141 sv_yes_mark = &PL_sv_yes;
6144 sv_commit = &PL_sv_yes;
6145 sv_yes_mark = &PL_sv_no;
6147 sv_setsv(sv_err, sv_commit);
6148 sv_setsv(sv_mrk, sv_yes_mark);
6152 if (last_pushed_cv) {
6155 PERL_UNUSED_VAR(SP);
6158 /* clean up; in particular, free all slabs above current one */
6159 LEAVE_SCOPE(oldsave);
6165 - regrepeat - repeatedly match something simple, report how many
6168 * [This routine now assumes that it will only match on things of length 1.
6169 * That was true before, but now we assume scan - reginput is the count,
6170 * rather than incrementing count on every character. [Er, except utf8.]]
6173 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
6178 char *loceol = PL_regeol;
6180 bool utf8_target = PL_reg_match_utf8;
6183 PERL_UNUSED_ARG(depth);
6186 PERL_ARGS_ASSERT_REGREPEAT;
6189 if (max == REG_INFTY)
6191 else if (max < loceol - scan)
6192 loceol = scan + max;
6197 while (scan < loceol && hardcount < max && *scan != '\n') {
6198 scan += UTF8SKIP(scan);
6202 while (scan < loceol && *scan != '\n')
6209 while (scan < loceol && hardcount < max) {
6210 scan += UTF8SKIP(scan);
6221 /* To get here, EXACTish nodes must have *byte* length == 1. That
6222 * means they match only characters in the string that can be expressed
6223 * as a single byte. For non-utf8 strings, that means a simple match.
6224 * For utf8 strings, the character matched must be an invariant, or
6225 * downgradable to a single byte. The pattern's utf8ness is
6226 * irrelevant, as since it's a single byte, it either isn't utf8, or if
6227 * it is, it's an invariant */
6230 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6232 if (! utf8_target || UNI_IS_INVARIANT(c)) {
6233 while (scan < loceol && UCHARAT(scan) == c) {
6239 /* Here, the string is utf8, and the pattern char is different
6240 * in utf8 than not, so can't compare them directly. Outside the
6241 * loop, find the two utf8 bytes that represent c, and then
6242 * look for those in sequence in the utf8 string */
6243 U8 high = UTF8_TWO_BYTE_HI(c);
6244 U8 low = UTF8_TWO_BYTE_LO(c);
6247 while (hardcount < max
6248 && scan + 1 < loceol
6249 && UCHARAT(scan) == high
6250 && UCHARAT(scan + 1) == low)
6258 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6262 PL_reg_flags |= RF_tainted;
6263 utf8_flags = FOLDEQ_UTF8_LOCALE;
6271 case EXACTFU_TRICKYFOLD:
6273 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6275 /* The comments for the EXACT case above apply as well to these fold
6280 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
6282 if (utf8_target || OP(p) == EXACTFU_SS) { /* Use full Unicode fold matching */
6283 char *tmpeol = loceol;
6284 while (hardcount < max
6285 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6286 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
6293 /* XXX Note that the above handles properly the German sharp s in
6294 * the pattern matching ss in the string. But it doesn't handle
6295 * properly cases where the string contains say 'LIGATURE ff' and
6296 * the pattern is 'f+'. This would require, say, a new function or
6297 * revised interface to foldEQ_utf8(), in which the maximum number
6298 * of characters to match could be passed and it would return how
6299 * many actually did. This is just one of many cases where
6300 * multi-char folds don't work properly, and so the fix is being
6306 /* Here, the string isn't utf8 and c is a single byte; and either
6307 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6308 * doesn't affect c. Can just do simple comparisons for exact or
6311 case EXACTF: folded = PL_fold[c]; break;
6313 case EXACTFU_TRICKYFOLD:
6314 case EXACTFU: folded = PL_fold_latin1[c]; break;
6315 case EXACTFL: folded = PL_fold_locale[c]; break;
6316 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6318 while (scan < loceol &&
6319 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6327 if (utf8_target || OP(p) == ANYOFV) {
6330 inclasslen = loceol - scan;
6331 while (hardcount < max
6332 && ((inclasslen = loceol - scan) > 0)
6333 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6339 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6347 LOAD_UTF8_CHARCLASS_ALNUM();
6348 while (hardcount < max && scan < loceol &&
6349 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6351 scan += UTF8SKIP(scan);
6355 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6363 while (scan < loceol && isALNUM((U8) *scan)) {
6368 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6373 PL_reg_flags |= RF_tainted;
6376 while (hardcount < max && scan < loceol &&
6377 isALNUM_LC_utf8((U8*)scan)) {
6378 scan += UTF8SKIP(scan);
6382 while (scan < loceol && isALNUM_LC(*scan))
6392 LOAD_UTF8_CHARCLASS_ALNUM();
6393 while (hardcount < max && scan < loceol &&
6394 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6396 scan += UTF8SKIP(scan);
6400 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6407 goto utf8_Nwordchar;
6408 while (scan < loceol && ! isALNUM((U8) *scan)) {
6414 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6420 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6421 scan += UTF8SKIP(scan);
6425 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6432 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6433 scan += UTF8SKIP(scan);
6437 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6443 PL_reg_flags |= RF_tainted;
6446 while (hardcount < max && scan < loceol &&
6447 !isALNUM_LC_utf8((U8*)scan)) {
6448 scan += UTF8SKIP(scan);
6452 while (scan < loceol && !isALNUM_LC(*scan))
6462 LOAD_UTF8_CHARCLASS_SPACE();
6463 while (hardcount < max && scan < loceol &&
6465 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6467 scan += UTF8SKIP(scan);
6473 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6482 while (scan < loceol && isSPACE((U8) *scan)) {
6487 while (scan < loceol && isSPACE_A((U8) *scan)) {
6492 PL_reg_flags |= RF_tainted;
6495 while (hardcount < max && scan < loceol &&
6496 isSPACE_LC_utf8((U8*)scan)) {
6497 scan += UTF8SKIP(scan);
6501 while (scan < loceol && isSPACE_LC(*scan))
6511 LOAD_UTF8_CHARCLASS_SPACE();
6512 while (hardcount < max && scan < loceol &&
6514 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6516 scan += UTF8SKIP(scan);
6522 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6531 while (scan < loceol && ! isSPACE((U8) *scan)) {
6537 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6538 scan += UTF8SKIP(scan);
6542 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6548 PL_reg_flags |= RF_tainted;
6551 while (hardcount < max && scan < loceol &&
6552 !isSPACE_LC_utf8((U8*)scan)) {
6553 scan += UTF8SKIP(scan);
6557 while (scan < loceol && !isSPACE_LC(*scan))
6564 LOAD_UTF8_CHARCLASS_DIGIT();
6565 while (hardcount < max && scan < loceol &&
6566 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6567 scan += UTF8SKIP(scan);
6571 while (scan < loceol && isDIGIT(*scan))
6576 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6581 PL_reg_flags |= RF_tainted;
6584 while (hardcount < max && scan < loceol &&
6585 isDIGIT_LC_utf8((U8*)scan)) {
6586 scan += UTF8SKIP(scan);
6590 while (scan < loceol && isDIGIT_LC(*scan))
6597 LOAD_UTF8_CHARCLASS_DIGIT();
6598 while (hardcount < max && scan < loceol &&
6599 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6600 scan += UTF8SKIP(scan);
6604 while (scan < loceol && !isDIGIT(*scan))
6610 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6611 scan += UTF8SKIP(scan);
6615 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6621 PL_reg_flags |= RF_tainted;
6624 while (hardcount < max && scan < loceol &&
6625 !isDIGIT_LC_utf8((U8*)scan)) {
6626 scan += UTF8SKIP(scan);
6630 while (scan < loceol && !isDIGIT_LC(*scan))
6637 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6643 LNBREAK can match two latin chars, which is ok,
6644 because we have a null terminated string, but we
6645 have to use hardcount in this situation
6647 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6656 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6661 while (scan < loceol && is_HORIZWS_latin1(scan))
6668 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6669 scan += UTF8SKIP(scan);
6673 while (scan < loceol && !is_HORIZWS_latin1(scan))
6681 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6686 while (scan < loceol && is_VERTWS_latin1(scan))
6694 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6695 scan += UTF8SKIP(scan);
6699 while (scan < loceol && !is_VERTWS_latin1(scan))
6705 default: /* Called on something of 0 width. */
6706 break; /* So match right here or not at all. */
6712 c = scan - PL_reginput;
6716 GET_RE_DEBUG_FLAGS_DECL;
6718 SV * const prop = sv_newmortal();
6719 regprop(prog, prop, p);
6720 PerlIO_printf(Perl_debug_log,
6721 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6722 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6730 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6732 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
6733 create a copy so that changes the caller makes won't change the shared one
6736 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6738 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6739 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp, altsvp));
6744 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6746 /* Returns the swash for the input 'node' in the regex 'prog'.
6747 * If <doinit> is true, will attempt to create the swash if not already
6749 * If <listsvp> is non-null, will return the swash initialization string in
6751 * If <altsvp> is non-null, will return the alternates to the regular swash
6753 * Tied intimately to how regcomp.c sets up the data structure */
6761 RXi_GET_DECL(prog,progi);
6762 const struct reg_data * const data = prog ? progi->data : NULL;
6764 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
6766 assert(ANYOF_NONBITMAP(node));
6768 if (data && data->count) {
6769 const U32 n = ARG(node);
6771 if (data->what[n] == 's') {
6772 SV * const rv = MUTABLE_SV(data->data[n]);
6773 AV * const av = MUTABLE_AV(SvRV(rv));
6774 SV **const ary = AvARRAY(av);
6775 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
6777 si = *ary; /* ary[0] = the string to initialize the swash with */
6779 /* Elements 3 and 4 are either both present or both absent. [3] is
6780 * any inversion list generated at compile time; [4] indicates if
6781 * that inversion list has any user-defined properties in it. */
6782 if (av_len(av) >= 3) {
6785 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
6792 /* Element [1] is reserved for the set-up swash. If already there,
6793 * return it; if not, create it and store it there */
6794 if (SvROK(ary[1])) {
6797 else if (si && doinit) {
6799 sw = _core_swash_init("utf8", /* the utf8 package */
6803 0, /* not from tr/// */
6806 (void)av_store(av, 1, sw);
6809 /* Element [2] is for any multi-char folds. Note that is a
6810 * fundamentally flawed design, because can't backtrack and try
6811 * again. See [perl #89774] */
6812 if (SvTYPE(ary[2]) == SVt_PVAV) {
6819 SV* matches_string = newSVpvn("", 0);
6821 /* Use the swash, if any, which has to have incorporated into it all
6823 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
6824 && (si && si != &PL_sv_undef))
6827 /* If no swash, use the input initialization string, if available */
6828 sv_catsv(matches_string, si);
6831 /* Add the inversion list to whatever we have. This may have come from
6832 * the swash, or from an input parameter */
6834 sv_catsv(matches_string, _invlist_contents(invlist));
6836 *listsvp = matches_string;
6846 - reginclass - determine if a character falls into a character class
6848 n is the ANYOF regnode
6849 p is the target string
6850 lenp is pointer to the maximum number of bytes of how far to go in p
6851 (This is assumed wthout checking to always be at least the current
6853 utf8_target tells whether p is in UTF-8.
6855 Returns true if matched; false otherwise. If lenp is not NULL, on return
6856 from a successful match, the value it points to will be updated to how many
6857 bytes in p were matched. If there was no match, the value is undefined,
6858 possibly changed from the input.
6860 Note that this can be a synthetic start class, a combination of various
6861 nodes, so things you think might be mutually exclusive, such as locale,
6862 aren't. It can match both locale and non-locale
6867 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6870 const char flags = ANYOF_FLAGS(n);
6876 PERL_ARGS_ASSERT_REGINCLASS;
6878 /* If c is not already the code point, get it */
6879 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6880 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6881 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6882 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6883 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6884 * UTF8_ALLOW_FFFF */
6885 if (c_len == (STRLEN)-1)
6886 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6892 /* Use passed in max length, or one character if none passed in or less
6893 * than one character. And assume will match just one character. This is
6894 * overwritten later if matched more. */
6896 maxlen = (*lenp > c_len) ? *lenp : c_len;
6904 /* If this character is potentially in the bitmap, check it */
6906 if (ANYOF_BITMAP_TEST(n, c))
6908 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6915 else if (flags & ANYOF_LOCALE) {
6916 PL_reg_flags |= RF_tainted;
6918 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6919 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6923 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6924 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6925 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6926 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6927 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6928 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6929 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6930 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6931 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6932 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6933 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6934 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII_LC(c)) ||
6935 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII_LC(c)) ||
6936 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6937 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6938 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6939 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6940 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6941 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6942 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6943 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6944 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6945 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6946 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6947 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6948 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6949 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6950 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6951 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6952 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK_LC(c)) ||
6953 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK_LC(c))
6954 ) /* How's that for a conditional? */
6961 /* If the bitmap didn't (or couldn't) match, and something outside the
6962 * bitmap could match, try that. Locale nodes specifiy completely the
6963 * behavior of code points in the bit map (otherwise, a utf8 target would
6964 * cause them to be treated as Unicode and not locale), except in
6965 * the very unlikely event when this node is a synthetic start class, which
6966 * could be a combination of locale and non-locale nodes. So allow locale
6967 * to match for the synthetic start class, which will give a false
6968 * positive that will be resolved when the match is done again as not part
6969 * of the synthetic start class */
6971 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6972 match = TRUE; /* Everything above 255 matches */
6974 else if (ANYOF_NONBITMAP(n)
6975 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6978 || (! (flags & ANYOF_LOCALE))
6979 || (flags & ANYOF_IS_SYNTHETIC)))))
6982 SV * const sw = core_regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6990 /* Not utf8. Convert as much of the string as available up
6991 * to the limit of how far the (single) character in the
6992 * pattern can possibly match (no need to go further). If
6993 * the node is a straight ANYOF or not folding, it can't
6994 * match more than one. Otherwise, It can match up to how
6995 * far a single char can fold to. Since not utf8, each
6996 * character is a single byte, so the max it can be in
6997 * bytes is the same as the max it can be in characters */
6998 STRLEN len = (OP(n) == ANYOF
6999 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
7001 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
7003 : UTF8_MAX_FOLD_CHAR_EXPAND;
7004 utf8_p = bytes_to_utf8(p, &len);
7007 if (swash_fetch(sw, utf8_p, TRUE))
7009 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
7011 /* Here, we need to test if the fold of the target string
7012 * matches. The non-multi char folds have all been moved to
7013 * the compilation phase, and the multi-char folds have
7014 * been stored by regcomp into 'av'; we linearly check to
7015 * see if any match the target string (folded). We know
7016 * that the originals were each one character, but we don't
7017 * currently know how many characters/bytes each folded to,
7018 * except we do know that there are small limits imposed by
7019 * Unicode. XXX A performance enhancement would be to have
7020 * regcomp.c store the max number of chars/bytes that are
7021 * in an av entry, as, say the 0th element. Even better
7022 * would be to have a hash of the few characters that can
7023 * start a multi-char fold to the max number of chars of
7026 * If there is a match, we will need to advance (if lenp is
7027 * specified) the match pointer in the target string. But
7028 * what we are comparing here isn't that string directly,
7029 * but its fold, whose length may differ from the original.
7030 * As we go along in constructing the fold, therefore, we
7031 * create a map so that we know how many bytes in the
7032 * source to advance given that we have matched a certain
7033 * number of bytes in the fold. This map is stored in
7034 * 'map_fold_len_back'. Let n mean the number of bytes in
7035 * the fold of the first character that we are folding.
7036 * Then map_fold_len_back[n] is set to the number of bytes
7037 * in that first character. Similarly let m be the
7038 * corresponding number for the second character to be
7039 * folded. Then map_fold_len_back[n+m] is set to the
7040 * number of bytes occupied by the first two source
7041 * characters. ... */
7042 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
7043 U8 folded[UTF8_MAXBYTES_CASE+1];
7044 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
7045 STRLEN total_foldlen = 0; /* num bytes in fold of all
7048 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
7050 /* Here, only need to fold the first char of the target
7051 * string. It the source wasn't utf8, is 1 byte long */
7052 to_utf8_fold(utf8_p, folded, &foldlen);
7053 total_foldlen = foldlen;
7054 map_fold_len_back[foldlen] = (utf8_target)
7060 /* Here, need to fold more than the first char. Do so
7061 * up to the limits */
7062 U8* source_ptr = utf8_p; /* The source for the fold
7065 U8* folded_ptr = folded;
7066 U8* e = utf8_p + maxlen; /* Can't go beyond last
7067 available byte in the
7071 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
7075 /* Fold the next character */
7076 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
7077 STRLEN this_char_foldlen;
7078 to_utf8_fold(source_ptr,
7080 &this_char_foldlen);
7082 /* Bail if it would exceed the byte limit for
7083 * folding a single char. */
7084 if (this_char_foldlen + folded_ptr - folded >
7090 /* Add the fold of this character */
7091 Copy(this_char_folded,
7095 source_ptr += UTF8SKIP(source_ptr);
7096 folded_ptr += this_char_foldlen;
7097 total_foldlen = folded_ptr - folded;
7099 /* Create map from the number of bytes in the fold
7100 * back to the number of bytes in the source. If
7101 * the source isn't utf8, the byte count is just
7102 * the number of characters so far */
7103 map_fold_len_back[total_foldlen]
7105 ? source_ptr - utf8_p
7112 /* Do the linear search to see if the fold is in the list
7113 * of multi-char folds. */
7116 for (i = 0; i <= av_len(av); i++) {
7117 SV* const sv = *av_fetch(av, i, FALSE);
7119 const char * const s = SvPV_const(sv, len);
7121 if (len <= total_foldlen
7122 && memEQ(s, (char*)folded, len)
7124 /* If 0, means matched a partial char. See
7126 && map_fold_len_back[len])
7129 /* Advance the target string ptr to account for
7130 * this fold, but have to translate from the
7131 * folded length to the corresponding source
7134 *lenp = map_fold_len_back[len];
7143 /* If we allocated a string above, free it */
7144 if (! utf8_target) Safefree(utf8_p);
7148 if (UNICODE_IS_SUPER(c)
7149 && (flags & ANYOF_WARN_SUPER)
7150 && ckWARN_d(WARN_NON_UNICODE))
7152 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7153 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7157 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7158 return cBOOL(flags & ANYOF_INVERT) ^ match;
7162 S_reghop3(U8 *s, I32 off, const U8* lim)
7164 /* return the position 'off' UTF-8 characters away from 's', forward if
7165 * 'off' >= 0, backwards if negative. But don't go outside of position
7166 * 'lim', which better be < s if off < 0 */
7170 PERL_ARGS_ASSERT_REGHOP3;
7173 while (off-- && s < lim) {
7174 /* XXX could check well-formedness here */
7179 while (off++ && s > lim) {
7181 if (UTF8_IS_CONTINUED(*s)) {
7182 while (s > lim && UTF8_IS_CONTINUATION(*s))
7185 /* XXX could check well-formedness here */
7192 /* there are a bunch of places where we use two reghop3's that should
7193 be replaced with this routine. but since thats not done yet
7194 we ifdef it out - dmq
7197 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7201 PERL_ARGS_ASSERT_REGHOP4;
7204 while (off-- && s < rlim) {
7205 /* XXX could check well-formedness here */
7210 while (off++ && s > llim) {
7212 if (UTF8_IS_CONTINUED(*s)) {
7213 while (s > llim && UTF8_IS_CONTINUATION(*s))
7216 /* XXX could check well-formedness here */
7224 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7228 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7231 while (off-- && s < lim) {
7232 /* XXX could check well-formedness here */
7239 while (off++ && s > lim) {
7241 if (UTF8_IS_CONTINUED(*s)) {
7242 while (s > lim && UTF8_IS_CONTINUATION(*s))
7245 /* XXX could check well-formedness here */
7254 restore_pos(pTHX_ void *arg)
7257 regexp * const rex = (regexp *)arg;
7258 if (PL_reg_state.re_state_eval_setup_done) {
7259 if (PL_reg_oldsaved) {
7260 rex->subbeg = PL_reg_oldsaved;
7261 rex->sublen = PL_reg_oldsavedlen;
7262 rex->suboffset = PL_reg_oldsavedoffset;
7263 rex->subcoffset = PL_reg_oldsavedcoffset;
7264 #ifdef PERL_OLD_COPY_ON_WRITE
7265 rex->saved_copy = PL_nrs;
7267 RXp_MATCH_COPIED_on(rex);
7269 PL_reg_magic->mg_len = PL_reg_oldpos;
7270 PL_reg_state.re_state_eval_setup_done = FALSE;
7271 PL_curpm = PL_reg_oldcurpm;
7276 S_to_utf8_substr(pTHX_ register regexp *prog)
7280 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7283 if (prog->substrs->data[i].substr
7284 && !prog->substrs->data[i].utf8_substr) {
7285 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7286 prog->substrs->data[i].utf8_substr = sv;
7287 sv_utf8_upgrade(sv);
7288 if (SvVALID(prog->substrs->data[i].substr)) {
7289 if (SvTAIL(prog->substrs->data[i].substr)) {
7290 /* Trim the trailing \n that fbm_compile added last
7292 SvCUR_set(sv, SvCUR(sv) - 1);
7293 /* Whilst this makes the SV technically "invalid" (as its
7294 buffer is no longer followed by "\0") when fbm_compile()
7295 adds the "\n" back, a "\0" is restored. */
7296 fbm_compile(sv, FBMcf_TAIL);
7300 if (prog->substrs->data[i].substr == prog->check_substr)
7301 prog->check_utf8 = sv;
7307 S_to_byte_substr(pTHX_ register regexp *prog)
7312 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7315 if (prog->substrs->data[i].utf8_substr
7316 && !prog->substrs->data[i].substr) {
7317 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7318 if (sv_utf8_downgrade(sv, TRUE)) {
7319 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7320 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7321 /* Trim the trailing \n that fbm_compile added last
7323 SvCUR_set(sv, SvCUR(sv) - 1);
7324 fbm_compile(sv, FBMcf_TAIL);
7332 prog->substrs->data[i].substr = sv;
7333 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7334 prog->check_substr = sv;
7341 * c-indentation-style: bsd
7343 * indent-tabs-mode: nil
7346 * ex: set ts=8 sts=4 sw=4 et: