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
40 /* At least one required character in the target string is expressible only in
42 static const char* const non_utf8_target_but_utf8_required
43 = "Can't match, because target string needs to be in UTF-8\n";
45 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
46 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
51 * pregcomp and pregexec -- regsub and regerror are not used in perl
53 * Copyright (c) 1986 by University of Toronto.
54 * Written by Henry Spencer. Not derived from licensed software.
56 * Permission is granted to anyone to use this software for any
57 * purpose on any computer system, and to redistribute it freely,
58 * subject to the following restrictions:
60 * 1. The author is not responsible for the consequences of use of
61 * this software, no matter how awful, even if they arise
64 * 2. The origin of this software must not be misrepresented, either
65 * by explicit claim or by omission.
67 * 3. Altered versions must be plainly marked as such, and must not
68 * be misrepresented as being the original software.
70 **** Alterations to Henry's code are...
72 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
73 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
74 **** by Larry Wall and others
76 **** You may distribute under the terms of either the GNU General Public
77 **** License or the Artistic License, as specified in the README file.
79 * Beware that some of this code is subtly aware of the way operator
80 * precedence is structured in regular expressions. Serious changes in
81 * regular-expression syntax might require a total rethink.
84 #define PERL_IN_REGEXEC_C
87 #ifdef PERL_IN_XSUB_RE
93 #include "inline_invlist.c"
94 #include "unicode_constants.h"
96 #define RF_tainted 1 /* tainted information used? e.g. locale */
97 #define RF_warned 2 /* warned about big count? */
99 #define RF_utf8 8 /* Pattern contains multibyte chars? */
101 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
103 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
106 #define STATIC static
109 /* Valid for non-utf8 strings: avoids the reginclass
110 * call if there are no complications: i.e., if everything matchable is
111 * straight forward in the bitmap */
112 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \
113 : ANYOF_BITMAP_TEST(p,*(c)))
119 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
120 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
122 #define HOPc(pos,off) \
123 (char *)(PL_reg_match_utf8 \
124 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
126 #define HOPBACKc(pos, off) \
127 (char*)(PL_reg_match_utf8\
128 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
129 : (pos - off >= PL_bostr) \
133 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
134 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
137 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
138 #define NEXTCHR_IS_EOS (nextchr < 0)
140 #define SET_nextchr \
141 nextchr = ((locinput < PL_regeol) ? UCHARAT(locinput) : NEXTCHR_EOS)
143 #define SET_locinput(p) \
148 /* these are unrolled below in the CCC_TRY_XXX defined */
149 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
150 if (!CAT2(PL_utf8_,class)) { \
152 ENTER; save_re_context(); \
153 ok=CAT2(is_utf8_,class)((const U8*)str); \
154 PERL_UNUSED_VAR(ok); \
155 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
156 /* Doesn't do an assert to verify that is correct */
157 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
158 if (!CAT2(PL_utf8_,class)) { \
159 bool throw_away PERL_UNUSED_DECL; \
160 ENTER; save_re_context(); \
161 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
162 PERL_UNUSED_VAR(throw_away); \
165 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
166 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
167 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
169 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
170 /* No asserts are done for some of these, in case called on a */ \
171 /* Unicode version in which they map to nothing */ \
172 LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
173 LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
175 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
177 /* The actual code for CCC_TRY, which uses several variables from the routine
178 * it's callable from. It is designed to be the bulk of a case statement.
179 * FUNC is the macro or function to call on non-utf8 targets that indicate if
180 * nextchr matches the class.
181 * UTF8_TEST is the whole test string to use for utf8 targets
182 * LOAD is what to use to test, and if not present to load in the swash for the
184 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
186 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
187 * utf8 and a variant, load the swash if necessary and test using the utf8
188 * test. Advance to the next character if test is ok, otherwise fail; If not
189 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
190 * fails, or advance to the next character */
192 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
193 if (NEXTCHR_IS_EOS) { \
196 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
197 LOAD_UTF8_CHARCLASS(CLASS, STR); \
198 if (POS_OR_NEG (UTF8_TEST)) { \
202 else if (POS_OR_NEG (FUNC(nextchr))) { \
205 goto increment_locinput;
207 /* Handle the non-locale cases for a character class and its complement. It
208 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
209 * This is because that code fails when the test succeeds, so we want to have
210 * the test fail so that the code succeeds. The swash is stored in a
211 * predictable PL_ place */
212 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
215 _CCC_TRY_CODE( !, FUNC, \
216 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
217 (U8*)locinput, TRUE)), \
220 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
221 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
222 (U8*)locinput, TRUE)), \
225 /* Generate the case statements for both locale and non-locale character
226 * classes in regmatch for classes that don't have special unicode semantics.
227 * Locales don't use an immediate swash, but an intermediary special locale
228 * function that is called on the pointer to the current place in the input
229 * string. That function will resolve to needing the same swash. One might
230 * think that because we don't know what the locale will match, we shouldn't
231 * check with the swash loading function that it loaded properly; ie, that we
232 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
233 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
235 #define CCC_TRY(NAME, NNAME, FUNC, \
236 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
237 NAMEA, NNAMEA, FUNCA, \
240 PL_reg_flags |= RF_tainted; \
241 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
243 PL_reg_flags |= RF_tainted; \
244 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
247 if (NEXTCHR_IS_EOS || ! FUNCA(nextchr)) { \
250 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
254 if (NEXTCHR_IS_EOS || FUNCA(nextchr)) { \
257 goto increment_locinput; \
258 /* Generate the non-locale cases */ \
259 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
261 /* This is like CCC_TRY, but has an extra set of parameters for generating case
262 * statements to handle separate Unicode semantics nodes */
263 #define CCC_TRY_U(NAME, NNAME, FUNC, \
264 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
265 NAMEU, NNAMEU, FUNCU, \
266 NAMEA, NNAMEA, FUNCA, \
268 CCC_TRY(NAME, NNAME, FUNC, \
269 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
270 NAMEA, NNAMEA, FUNCA, \
272 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
274 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
276 /* for use after a quantifier and before an EXACT-like node -- japhy */
277 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
279 * NOTE that *nothing* that affects backtracking should be in here, specifically
280 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
281 * node that is in between two EXACT like nodes when ascertaining what the required
282 * "follow" character is. This should probably be moved to regex compile time
283 * although it may be done at run time beause of the REF possibility - more
284 * investigation required. -- demerphq
286 #define JUMPABLE(rn) ( \
288 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
290 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
291 OP(rn) == PLUS || OP(rn) == MINMOD || \
293 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
295 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
297 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
300 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
301 we don't need this definition. */
302 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
303 #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 )
304 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
307 /* ... so we use this as its faster. */
308 #define IS_TEXT(rn) ( OP(rn)==EXACT )
309 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
310 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
311 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
316 Search for mandatory following text node; for lookahead, the text must
317 follow but for lookbehind (rn->flags != 0) we skip to the next step.
319 #define FIND_NEXT_IMPT(rn) STMT_START { \
320 while (JUMPABLE(rn)) { \
321 const OPCODE type = OP(rn); \
322 if (type == SUSPEND || PL_regkind[type] == CURLY) \
323 rn = NEXTOPER(NEXTOPER(rn)); \
324 else if (type == PLUS) \
326 else if (type == IFMATCH) \
327 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
328 else rn += NEXT_OFF(rn); \
333 static void restore_pos(pTHX_ void *arg);
335 #define REGCP_PAREN_ELEMS 3
336 #define REGCP_OTHER_ELEMS 3
337 #define REGCP_FRAME_ELEMS 1
338 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
339 * are needed for the regexp context stack bookkeeping. */
342 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
345 const int retval = PL_savestack_ix;
346 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
347 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
348 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
350 GET_RE_DEBUG_FLAGS_DECL;
352 PERL_ARGS_ASSERT_REGCPPUSH;
354 if (paren_elems_to_push < 0)
355 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
356 paren_elems_to_push);
358 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
359 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
360 " out of range (%lu-%ld)",
361 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
363 SSGROW(total_elems + REGCP_FRAME_ELEMS);
366 if ((int)PL_regsize > (int)parenfloor)
367 PerlIO_printf(Perl_debug_log,
368 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
373 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
374 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
375 SSPUSHINT(rex->offs[p].end);
376 SSPUSHINT(rex->offs[p].start);
377 SSPUSHINT(rex->offs[p].start_tmp);
378 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
379 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
381 (IV)rex->offs[p].start,
382 (IV)rex->offs[p].start_tmp,
386 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
387 SSPUSHINT(PL_regsize);
388 SSPUSHINT(rex->lastparen);
389 SSPUSHINT(rex->lastcloseparen);
390 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
395 /* These are needed since we do not localize EVAL nodes: */
396 #define REGCP_SET(cp) \
398 PerlIO_printf(Perl_debug_log, \
399 " Setting an EVAL scope, savestack=%"IVdf"\n", \
400 (IV)PL_savestack_ix)); \
403 #define REGCP_UNWIND(cp) \
405 if (cp != PL_savestack_ix) \
406 PerlIO_printf(Perl_debug_log, \
407 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
408 (IV)(cp), (IV)PL_savestack_ix)); \
411 #define UNWIND_PAREN(lp, lcp) \
412 for (n = rex->lastparen; n > lp; n--) \
413 rex->offs[n].end = -1; \
414 rex->lastparen = n; \
415 rex->lastcloseparen = lcp;
419 S_regcppop(pTHX_ regexp *rex)
424 GET_RE_DEBUG_FLAGS_DECL;
426 PERL_ARGS_ASSERT_REGCPPOP;
428 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
430 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
431 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
432 rex->lastcloseparen = SSPOPINT;
433 rex->lastparen = SSPOPINT;
434 PL_regsize = SSPOPINT;
436 i -= REGCP_OTHER_ELEMS;
437 /* Now restore the parentheses context. */
439 if (i || rex->lastparen + 1 <= rex->nparens)
440 PerlIO_printf(Perl_debug_log,
441 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
447 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
449 rex->offs[paren].start_tmp = SSPOPINT;
450 rex->offs[paren].start = SSPOPINT;
452 if (paren <= rex->lastparen)
453 rex->offs[paren].end = tmps;
454 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
455 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
457 (IV)rex->offs[paren].start,
458 (IV)rex->offs[paren].start_tmp,
459 (IV)rex->offs[paren].end,
460 (paren > rex->lastparen ? "(skipped)" : ""));
465 /* It would seem that the similar code in regtry()
466 * already takes care of this, and in fact it is in
467 * a better location to since this code can #if 0-ed out
468 * but the code in regtry() is needed or otherwise tests
469 * requiring null fields (pat.t#187 and split.t#{13,14}
470 * (as of patchlevel 7877) will fail. Then again,
471 * this code seems to be necessary or otherwise
472 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
473 * --jhi updated by dapm */
474 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
476 rex->offs[i].start = -1;
477 rex->offs[i].end = -1;
478 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
479 " \\%"UVuf": %s ..-1 undeffing\n",
481 (i > PL_regsize) ? "-1" : " "
487 /* restore the parens and associated vars at savestack position ix,
488 * but without popping the stack */
491 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
493 I32 tmpix = PL_savestack_ix;
494 PL_savestack_ix = ix;
496 PL_savestack_ix = tmpix;
499 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
502 * pregexec and friends
505 #ifndef PERL_IN_XSUB_RE
507 - pregexec - match a regexp against a string
510 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
511 char *strbeg, I32 minend, SV *screamer, U32 nosave)
512 /* stringarg: the point in the string at which to begin matching */
513 /* strend: pointer to null at end of string */
514 /* strbeg: real beginning of string */
515 /* minend: end of match must be >= minend bytes after stringarg. */
516 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
517 * itself is accessed via the pointers above */
518 /* nosave: For optimizations. */
520 PERL_ARGS_ASSERT_PREGEXEC;
523 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
524 nosave ? 0 : REXEC_COPY_STR);
529 * Need to implement the following flags for reg_anch:
531 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
533 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
534 * INTUIT_AUTORITATIVE_ML
535 * INTUIT_ONCE_NOML - Intuit can match in one location only.
538 * Another flag for this function: SECOND_TIME (so that float substrs
539 * with giant delta may be not rechecked).
542 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
544 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
545 Otherwise, only SvCUR(sv) is used to get strbeg. */
547 /* XXXX We assume that strpos is strbeg unless sv. */
549 /* XXXX Some places assume that there is a fixed substring.
550 An update may be needed if optimizer marks as "INTUITable"
551 RExen without fixed substrings. Similarly, it is assumed that
552 lengths of all the strings are no more than minlen, thus they
553 cannot come from lookahead.
554 (Or minlen should take into account lookahead.)
555 NOTE: Some of this comment is not correct. minlen does now take account
556 of lookahead/behind. Further research is required. -- demerphq
560 /* A failure to find a constant substring means that there is no need to make
561 an expensive call to REx engine, thus we celebrate a failure. Similarly,
562 finding a substring too deep into the string means that less calls to
563 regtry() should be needed.
565 REx compiler's optimizer found 4 possible hints:
566 a) Anchored substring;
568 c) Whether we are anchored (beginning-of-line or \G);
569 d) First node (of those at offset 0) which may distinguish positions;
570 We use a)b)d) and multiline-part of c), and try to find a position in the
571 string which does not contradict any of them.
574 /* Most of decisions we do here should have been done at compile time.
575 The nodes of the REx which we used for the search should have been
576 deleted from the finite automaton. */
579 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
580 char *strend, const U32 flags, re_scream_pos_data *data)
583 struct regexp *const prog = ReANY(rx);
585 /* Should be nonnegative! */
591 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
593 char *other_last = NULL; /* other substr checked before this */
594 char *check_at = NULL; /* check substr found at this pos */
595 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
596 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
597 RXi_GET_DECL(prog,progi);
599 const char * const i_strpos = strpos;
601 GET_RE_DEBUG_FLAGS_DECL;
603 PERL_ARGS_ASSERT_RE_INTUIT_START;
604 PERL_UNUSED_ARG(flags);
605 PERL_UNUSED_ARG(data);
607 RX_MATCH_UTF8_set(rx,utf8_target);
610 PL_reg_flags |= RF_utf8;
613 debug_start_match(rx, utf8_target, strpos, strend,
614 sv ? "Guessing start of match in sv for"
615 : "Guessing start of match in string for");
618 /* CHR_DIST() would be more correct here but it makes things slow. */
619 if (prog->minlen > strend - strpos) {
620 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
621 "String too short... [re_intuit_start]\n"));
625 /* XXX we need to pass strbeg as a separate arg: the following is
626 * guesswork and can be wrong... */
627 if (sv && SvPOK(sv)) {
628 char * p = SvPVX(sv);
629 STRLEN cur = SvCUR(sv);
630 if (p <= strpos && strpos < p + cur) {
632 assert(p <= strend && strend <= p + cur);
635 strbeg = strend - cur;
642 if (!prog->check_utf8 && prog->check_substr)
643 to_utf8_substr(prog);
644 check = prog->check_utf8;
646 if (!prog->check_substr && prog->check_utf8) {
647 if (! to_byte_substr(prog)) {
648 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
651 check = prog->check_substr;
653 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
654 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
655 || ( (prog->extflags & RXf_ANCH_BOL)
656 && !multiline ) ); /* Check after \n? */
659 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
660 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
661 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
663 && (strpos != strbeg)) {
664 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
667 if (prog->check_offset_min == prog->check_offset_max
668 && !(prog->extflags & RXf_CANY_SEEN)
669 && ! multiline) /* /m can cause \n's to match that aren't
670 accounted for in the string max length.
671 See [perl #115242] */
673 /* Substring at constant offset from beg-of-str... */
676 s = HOP3c(strpos, prog->check_offset_min, strend);
679 slen = SvCUR(check); /* >= 1 */
681 if ( strend - s > slen || strend - s < slen - 1
682 || (strend - s == slen && strend[-1] != '\n')) {
683 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
686 /* Now should match s[0..slen-2] */
688 if (slen && (*SvPVX_const(check) != *s
690 && memNE(SvPVX_const(check), s, slen)))) {
692 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
696 else if (*SvPVX_const(check) != *s
697 || ((slen = SvCUR(check)) > 1
698 && memNE(SvPVX_const(check), s, slen)))
701 goto success_at_start;
704 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
706 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
707 end_shift = prog->check_end_shift;
710 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
711 - (SvTAIL(check) != 0);
712 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
714 if (end_shift < eshift)
718 else { /* Can match at random position */
721 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
722 end_shift = prog->check_end_shift;
724 /* end shift should be non negative here */
727 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
729 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
730 (IV)end_shift, RX_PRECOMP(prog));
734 /* Find a possible match in the region s..strend by looking for
735 the "check" substring in the region corrected by start/end_shift. */
738 I32 srch_start_shift = start_shift;
739 I32 srch_end_shift = end_shift;
742 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
743 srch_end_shift -= ((strbeg - s) - srch_start_shift);
744 srch_start_shift = strbeg - s;
746 DEBUG_OPTIMISE_MORE_r({
747 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
748 (IV)prog->check_offset_min,
749 (IV)srch_start_shift,
751 (IV)prog->check_end_shift);
754 if (prog->extflags & RXf_CANY_SEEN) {
755 start_point= (U8*)(s + srch_start_shift);
756 end_point= (U8*)(strend - srch_end_shift);
758 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
759 end_point= HOP3(strend, -srch_end_shift, strbeg);
761 DEBUG_OPTIMISE_MORE_r({
762 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
763 (int)(end_point - start_point),
764 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
768 s = fbm_instr( start_point, end_point,
769 check, multiline ? FBMrf_MULTILINE : 0);
771 /* Update the count-of-usability, remove useless subpatterns,
775 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
776 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
777 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
778 (s ? "Found" : "Did not find"),
779 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
780 ? "anchored" : "floating"),
783 (s ? " at offset " : "...\n") );
788 /* Finish the diagnostic message */
789 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
791 /* XXX dmq: first branch is for positive lookbehind...
792 Our check string is offset from the beginning of the pattern.
793 So we need to do any stclass tests offset forward from that
802 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
803 Start with the other substr.
804 XXXX no SCREAM optimization yet - and a very coarse implementation
805 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
806 *always* match. Probably should be marked during compile...
807 Probably it is right to do no SCREAM here...
810 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
811 : (prog->float_substr && prog->anchored_substr))
813 /* Take into account the "other" substring. */
814 /* XXXX May be hopelessly wrong for UTF... */
817 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
820 char * const last = HOP3c(s, -start_shift, strbeg);
822 char * const saved_s = s;
825 t = s - prog->check_offset_max;
826 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
828 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
833 t = HOP3c(t, prog->anchored_offset, strend);
834 if (t < other_last) /* These positions already checked */
836 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
839 /* XXXX It is not documented what units *_offsets are in.
840 We assume bytes, but this is clearly wrong.
841 Meaning this code needs to be carefully reviewed for errors.
845 /* On end-of-str: see comment below. */
846 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
847 if (must == &PL_sv_undef) {
849 DEBUG_r(must = prog->anchored_utf8); /* for debug */
854 HOP3(HOP3(last1, prog->anchored_offset, strend)
855 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
857 multiline ? FBMrf_MULTILINE : 0
860 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
861 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
862 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
863 (s ? "Found" : "Contradicts"),
864 quoted, RE_SV_TAIL(must));
869 if (last1 >= last2) {
870 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
871 ", giving up...\n"));
874 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
875 ", trying floating at offset %ld...\n",
876 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
877 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
878 s = HOP3c(last, 1, strend);
882 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
883 (long)(s - i_strpos)));
884 t = HOP3c(s, -prog->anchored_offset, strbeg);
885 other_last = HOP3c(s, 1, strend);
893 else { /* Take into account the floating substring. */
895 char * const saved_s = s;
898 t = HOP3c(s, -start_shift, strbeg);
900 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
901 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
902 last = HOP3c(t, prog->float_max_offset, strend);
903 s = HOP3c(t, prog->float_min_offset, strend);
906 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
907 must = utf8_target ? prog->float_utf8 : prog->float_substr;
908 /* fbm_instr() takes into account exact value of end-of-str
909 if the check is SvTAIL(ed). Since false positives are OK,
910 and end-of-str is not later than strend we are OK. */
911 if (must == &PL_sv_undef) {
913 DEBUG_r(must = prog->float_utf8); /* for debug message */
916 s = fbm_instr((unsigned char*)s,
917 (unsigned char*)last + SvCUR(must)
919 must, multiline ? FBMrf_MULTILINE : 0);
921 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
922 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
923 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
924 (s ? "Found" : "Contradicts"),
925 quoted, RE_SV_TAIL(must));
929 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
930 ", giving up...\n"));
933 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
934 ", trying anchored starting at offset %ld...\n",
935 (long)(saved_s + 1 - i_strpos)));
937 s = HOP3c(t, 1, strend);
941 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
942 (long)(s - i_strpos)));
943 other_last = s; /* Fix this later. --Hugo */
953 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
955 DEBUG_OPTIMISE_MORE_r(
956 PerlIO_printf(Perl_debug_log,
957 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
958 (IV)prog->check_offset_min,
959 (IV)prog->check_offset_max,
967 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
969 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
972 /* Fixed substring is found far enough so that the match
973 cannot start at strpos. */
975 if (ml_anch && t[-1] != '\n') {
976 /* Eventually fbm_*() should handle this, but often
977 anchored_offset is not 0, so this check will not be wasted. */
978 /* XXXX In the code below we prefer to look for "^" even in
979 presence of anchored substrings. And we search even
980 beyond the found float position. These pessimizations
981 are historical artefacts only. */
983 while (t < strend - prog->minlen) {
985 if (t < check_at - prog->check_offset_min) {
986 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
987 /* Since we moved from the found position,
988 we definitely contradict the found anchored
989 substr. Due to the above check we do not
990 contradict "check" substr.
991 Thus we can arrive here only if check substr
992 is float. Redo checking for "other"=="fixed".
995 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
996 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
997 goto do_other_anchored;
999 /* We don't contradict the found floating substring. */
1000 /* XXXX Why not check for STCLASS? */
1002 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1003 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1006 /* Position contradicts check-string */
1007 /* XXXX probably better to look for check-string
1008 than for "\n", so one should lower the limit for t? */
1009 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1010 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1011 other_last = strpos = s = t + 1;
1016 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1017 PL_colors[0], PL_colors[1]));
1021 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1022 PL_colors[0], PL_colors[1]));
1026 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1029 /* The found string does not prohibit matching at strpos,
1030 - no optimization of calling REx engine can be performed,
1031 unless it was an MBOL and we are not after MBOL,
1032 or a future STCLASS check will fail this. */
1034 /* Even in this situation we may use MBOL flag if strpos is offset
1035 wrt the start of the string. */
1036 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1037 && (strpos != strbeg) && strpos[-1] != '\n'
1038 /* May be due to an implicit anchor of m{.*foo} */
1039 && !(prog->intflags & PREGf_IMPLICIT))
1044 DEBUG_EXECUTE_r( if (ml_anch)
1045 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1046 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1049 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1051 prog->check_utf8 /* Could be deleted already */
1052 && --BmUSEFUL(prog->check_utf8) < 0
1053 && (prog->check_utf8 == prog->float_utf8)
1055 prog->check_substr /* Could be deleted already */
1056 && --BmUSEFUL(prog->check_substr) < 0
1057 && (prog->check_substr == prog->float_substr)
1060 /* If flags & SOMETHING - do not do it many times on the same match */
1061 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1062 /* XXX Does the destruction order has to change with utf8_target? */
1063 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1064 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1065 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1066 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1067 check = NULL; /* abort */
1069 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1070 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1071 if (prog->intflags & PREGf_IMPLICIT)
1072 prog->extflags &= ~RXf_ANCH_MBOL;
1073 /* XXXX This is a remnant of the old implementation. It
1074 looks wasteful, since now INTUIT can use many
1075 other heuristics. */
1076 prog->extflags &= ~RXf_USE_INTUIT;
1077 /* XXXX What other flags might need to be cleared in this branch? */
1083 /* Last resort... */
1084 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1085 /* trie stclasses are too expensive to use here, we are better off to
1086 leave it to regmatch itself */
1087 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1088 /* minlen == 0 is possible if regstclass is \b or \B,
1089 and the fixed substr is ''$.
1090 Since minlen is already taken into account, s+1 is before strend;
1091 accidentally, minlen >= 1 guaranties no false positives at s + 1
1092 even for \b or \B. But (minlen? 1 : 0) below assumes that
1093 regstclass does not come from lookahead... */
1094 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1095 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1096 const U8* const str = (U8*)STRING(progi->regstclass);
1097 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1098 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1101 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1102 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1103 else if (prog->float_substr || prog->float_utf8)
1104 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1108 if (checked_upto < s)
1110 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1111 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1114 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1119 const char *what = NULL;
1121 if (endpos == strend) {
1122 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1123 "Could not match STCLASS...\n") );
1126 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1127 "This position contradicts STCLASS...\n") );
1128 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1130 checked_upto = HOPBACKc(endpos, start_shift);
1131 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1132 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1133 /* Contradict one of substrings */
1134 if (prog->anchored_substr || prog->anchored_utf8) {
1135 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1136 DEBUG_EXECUTE_r( what = "anchored" );
1138 s = HOP3c(t, 1, strend);
1139 if (s + start_shift + end_shift > strend) {
1140 /* XXXX Should be taken into account earlier? */
1141 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1142 "Could not match STCLASS...\n") );
1147 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1148 "Looking for %s substr starting at offset %ld...\n",
1149 what, (long)(s + start_shift - i_strpos)) );
1152 /* Have both, check_string is floating */
1153 if (t + start_shift >= check_at) /* Contradicts floating=check */
1154 goto retry_floating_check;
1155 /* Recheck anchored substring, but not floating... */
1159 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1160 "Looking for anchored substr starting at offset %ld...\n",
1161 (long)(other_last - i_strpos)) );
1162 goto do_other_anchored;
1164 /* Another way we could have checked stclass at the
1165 current position only: */
1170 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1171 "Looking for /%s^%s/m starting at offset %ld...\n",
1172 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1175 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1177 /* Check is floating substring. */
1178 retry_floating_check:
1179 t = check_at - start_shift;
1180 DEBUG_EXECUTE_r( what = "floating" );
1181 goto hop_and_restart;
1184 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1185 "By STCLASS: moving %ld --> %ld\n",
1186 (long)(t - i_strpos), (long)(s - i_strpos))
1190 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1191 "Does not contradict STCLASS...\n");
1196 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1197 PL_colors[4], (check ? "Guessed" : "Giving up"),
1198 PL_colors[5], (long)(s - i_strpos)) );
1201 fail_finish: /* Substring not found */
1202 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1203 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1205 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1206 PL_colors[4], PL_colors[5]));
1210 #define DECL_TRIE_TYPE(scan) \
1211 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1212 trie_type = ((scan->flags == EXACT) \
1213 ? (utf8_target ? trie_utf8 : trie_plain) \
1214 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1216 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1217 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1219 switch (trie_type) { \
1220 case trie_utf8_fold: \
1221 if ( foldlen>0 ) { \
1222 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1227 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1228 len = UTF8SKIP(uc); \
1229 skiplen = UNISKIP( uvc ); \
1230 foldlen -= skiplen; \
1231 uscan = foldbuf + skiplen; \
1234 case trie_latin_utf8_fold: \
1235 if ( foldlen>0 ) { \
1236 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1242 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1243 skiplen = UNISKIP( uvc ); \
1244 foldlen -= skiplen; \
1245 uscan = foldbuf + skiplen; \
1249 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1256 charid = trie->charmap[ uvc ]; \
1260 if (widecharmap) { \
1261 SV** const svpp = hv_fetch(widecharmap, \
1262 (char*)&uvc, sizeof(UV), 0); \
1264 charid = (U16)SvIV(*svpp); \
1269 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1273 && (ln == 1 || folder(s, pat_string, ln)) \
1274 && (!reginfo || regtry(reginfo, &s)) ) \
1280 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1282 while (s < strend && s + (uskip = UTF8SKIP(s)) <= strend) { \
1288 #define REXEC_FBC_SCAN(CoDe) \
1290 while (s < strend) { \
1296 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1297 REXEC_FBC_UTF8_SCAN( \
1299 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1308 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1311 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1320 #define REXEC_FBC_TRYIT \
1321 if ((!reginfo || regtry(reginfo, &s))) \
1324 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1325 if (utf8_target) { \
1326 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1329 REXEC_FBC_CLASS_SCAN(CoNd); \
1332 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1333 if (utf8_target) { \
1335 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1338 REXEC_FBC_CLASS_SCAN(CoNd); \
1341 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1342 PL_reg_flags |= RF_tainted; \
1343 if (utf8_target) { \
1344 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1347 REXEC_FBC_CLASS_SCAN(CoNd); \
1350 #define DUMP_EXEC_POS(li,s,doutf8) \
1351 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1354 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1355 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1356 tmp = TEST_NON_UTF8(tmp); \
1357 REXEC_FBC_UTF8_SCAN( \
1358 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1367 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1368 if (s == PL_bostr) { \
1372 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1373 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1376 LOAD_UTF8_CHARCLASS_ALNUM(); \
1377 REXEC_FBC_UTF8_SCAN( \
1378 if (tmp == ! (TeSt2_UtF8)) { \
1387 /* The only difference between the BOUND and NBOUND cases is that
1388 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1389 * NBOUND. This is accomplished by passing it in either the if or else clause,
1390 * with the other one being empty */
1391 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1392 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1394 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1395 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1397 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1398 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1400 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1401 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1404 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1405 * be passed in completely with the variable name being tested, which isn't
1406 * such a clean interface, but this is easier to read than it was before. We
1407 * are looking for the boundary (or non-boundary between a word and non-word
1408 * character. The utf8 and non-utf8 cases have the same logic, but the details
1409 * must be different. Find the "wordness" of the character just prior to this
1410 * one, and compare it with the wordness of this one. If they differ, we have
1411 * a boundary. At the beginning of the string, pretend that the previous
1412 * character was a new-line */
1413 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1414 if (utf8_target) { \
1417 else { /* Not utf8 */ \
1418 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1419 tmp = TEST_NON_UTF8(tmp); \
1421 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1430 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1433 /* We know what class REx starts with. Try to find this position... */
1434 /* if reginfo is NULL, its a dryrun */
1435 /* annoyingly all the vars in this routine have different names from their counterparts
1436 in regmatch. /grrr */
1439 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1440 const char *strend, regmatch_info *reginfo)
1443 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1444 char *pat_string; /* The pattern's exactish string */
1445 char *pat_end; /* ptr to end char of pat_string */
1446 re_fold_t folder; /* Function for computing non-utf8 folds */
1447 const U8 *fold_array; /* array for folding ords < 256 */
1454 I32 tmp = 1; /* Scratch variable? */
1455 const bool utf8_target = PL_reg_match_utf8;
1456 UV utf8_fold_flags = 0;
1457 RXi_GET_DECL(prog,progi);
1459 PERL_ARGS_ASSERT_FIND_BYCLASS;
1461 /* We know what class it must start with. */
1465 REXEC_FBC_UTF8_CLASS_SCAN(
1466 reginclass(prog, c, (U8*)s, utf8_target));
1469 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1474 if (tmp && (!reginfo || regtry(reginfo, &s)))
1482 if (UTF_PATTERN || utf8_target) {
1483 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1484 goto do_exactf_utf8;
1486 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1487 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1488 goto do_exactf_non_utf8; /* isn't dealt with by these */
1493 /* regcomp.c already folded this if pattern is in UTF-8 */
1494 utf8_fold_flags = 0;
1495 goto do_exactf_utf8;
1497 fold_array = PL_fold;
1499 goto do_exactf_non_utf8;
1502 if (UTF_PATTERN || utf8_target) {
1503 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1504 goto do_exactf_utf8;
1506 fold_array = PL_fold_locale;
1507 folder = foldEQ_locale;
1508 goto do_exactf_non_utf8;
1512 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1514 goto do_exactf_utf8;
1516 case EXACTFU_TRICKYFOLD:
1518 if (UTF_PATTERN || utf8_target) {
1519 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1520 goto do_exactf_utf8;
1523 /* Any 'ss' in the pattern should have been replaced by regcomp,
1524 * so we don't have to worry here about this single special case
1525 * in the Latin1 range */
1526 fold_array = PL_fold_latin1;
1527 folder = foldEQ_latin1;
1531 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1532 are no glitches with fold-length differences
1533 between the target string and pattern */
1535 /* The idea in the non-utf8 EXACTF* cases is to first find the
1536 * first character of the EXACTF* node and then, if necessary,
1537 * case-insensitively compare the full text of the node. c1 is the
1538 * first character. c2 is its fold. This logic will not work for
1539 * Unicode semantics and the german sharp ss, which hence should
1540 * not be compiled into a node that gets here. */
1541 pat_string = STRING(c);
1542 ln = STR_LEN(c); /* length to match in octets/bytes */
1544 /* We know that we have to match at least 'ln' bytes (which is the
1545 * same as characters, since not utf8). If we have to match 3
1546 * characters, and there are only 2 availabe, we know without
1547 * trying that it will fail; so don't start a match past the
1548 * required minimum number from the far end */
1549 e = HOP3c(strend, -((I32)ln), s);
1551 if (!reginfo && e < s) {
1552 e = s; /* Due to minlen logic of intuit() */
1556 c2 = fold_array[c1];
1557 if (c1 == c2) { /* If char and fold are the same */
1558 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1561 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1570 /* If one of the operands is in utf8, we can't use the simpler
1571 * folding above, due to the fact that many different characters
1572 * can have the same fold, or portion of a fold, or different-
1574 pat_string = STRING(c);
1575 ln = STR_LEN(c); /* length to match in octets/bytes */
1576 pat_end = pat_string + ln;
1577 lnc = (UTF_PATTERN) /* length to match in characters */
1578 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1581 /* We have 'lnc' characters to match in the pattern, but because of
1582 * multi-character folding, each character in the target can match
1583 * up to 3 characters (Unicode guarantees it will never exceed
1584 * this) if it is utf8-encoded; and up to 2 if not (based on the
1585 * fact that the Latin 1 folds are already determined, and the
1586 * only multi-char fold in that range is the sharp-s folding to
1587 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1588 * string character. Adjust lnc accordingly, rounding up, so that
1589 * if we need to match at least 4+1/3 chars, that really is 5. */
1590 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1591 lnc = (lnc + expansion - 1) / expansion;
1593 /* As in the non-UTF8 case, if we have to match 3 characters, and
1594 * only 2 are left, it's guaranteed to fail, so don't start a
1595 * match that would require us to go beyond the end of the string
1597 e = HOP3c(strend, -((I32)lnc), s);
1599 if (!reginfo && e < s) {
1600 e = s; /* Due to minlen logic of intuit() */
1603 /* XXX Note that we could recalculate e to stop the loop earlier,
1604 * as the worst case expansion above will rarely be met, and as we
1605 * go along we would usually find that e moves further to the left.
1606 * This would happen only after we reached the point in the loop
1607 * where if there were no expansion we should fail. Unclear if
1608 * worth the expense */
1611 char *my_strend= (char *)strend;
1612 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1613 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1614 && (!reginfo || regtry(reginfo, &s)) )
1618 s += (utf8_target) ? UTF8SKIP(s) : 1;
1623 PL_reg_flags |= RF_tainted;
1624 FBC_BOUND(isALNUM_LC,
1625 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1626 isALNUM_LC_utf8((U8*)s));
1629 PL_reg_flags |= RF_tainted;
1630 FBC_NBOUND(isALNUM_LC,
1631 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1632 isALNUM_LC_utf8((U8*)s));
1635 FBC_BOUND(isWORDCHAR,
1637 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1640 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1642 isWORDCHAR_A((U8*)s));
1645 FBC_NBOUND(isWORDCHAR,
1647 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1650 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1652 isWORDCHAR_A((U8*)s));
1655 FBC_BOUND(isWORDCHAR_L1,
1657 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1660 FBC_NBOUND(isWORDCHAR_L1,
1662 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1665 REXEC_FBC_CSCAN_TAINT(
1666 isALNUM_LC_utf8((U8*)s),
1671 REXEC_FBC_CSCAN_PRELOAD(
1672 LOAD_UTF8_CHARCLASS_ALNUM(),
1673 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1674 isWORDCHAR_L1((U8) *s)
1678 REXEC_FBC_CSCAN_PRELOAD(
1679 LOAD_UTF8_CHARCLASS_ALNUM(),
1680 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1685 /* Don't need to worry about utf8, as it can match only a single
1686 * byte invariant character */
1687 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1690 REXEC_FBC_CSCAN_PRELOAD(
1691 LOAD_UTF8_CHARCLASS_ALNUM(),
1692 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1693 ! isWORDCHAR_L1((U8) *s)
1697 REXEC_FBC_CSCAN_PRELOAD(
1698 LOAD_UTF8_CHARCLASS_ALNUM(),
1699 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1710 REXEC_FBC_CSCAN_TAINT(
1711 !isALNUM_LC_utf8((U8*)s),
1716 REXEC_FBC_CSCAN_PRELOAD(
1717 LOAD_UTF8_CHARCLASS_SPACE(),
1718 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1723 REXEC_FBC_CSCAN_PRELOAD(
1724 LOAD_UTF8_CHARCLASS_SPACE(),
1725 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1730 /* Don't need to worry about utf8, as it can match only a single
1731 * byte invariant character */
1732 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1735 REXEC_FBC_CSCAN_TAINT(
1736 isSPACE_LC_utf8((U8*)s),
1741 REXEC_FBC_CSCAN_PRELOAD(
1742 LOAD_UTF8_CHARCLASS_SPACE(),
1743 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1744 ! isSPACE_L1((U8) *s)
1748 REXEC_FBC_CSCAN_PRELOAD(
1749 LOAD_UTF8_CHARCLASS_SPACE(),
1750 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1761 REXEC_FBC_CSCAN_TAINT(
1762 !isSPACE_LC_utf8((U8*)s),
1767 REXEC_FBC_CSCAN_PRELOAD(
1768 LOAD_UTF8_CHARCLASS_DIGIT(),
1769 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1774 /* Don't need to worry about utf8, as it can match only a single
1775 * byte invariant character */
1776 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1779 REXEC_FBC_CSCAN_TAINT(
1780 isDIGIT_LC_utf8((U8*)s),
1785 REXEC_FBC_CSCAN_PRELOAD(
1786 LOAD_UTF8_CHARCLASS_DIGIT(),
1787 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1798 REXEC_FBC_CSCAN_TAINT(
1799 !isDIGIT_LC_utf8((U8*)s),
1805 is_LNBREAK_utf8_safe(s, strend),
1806 is_LNBREAK_latin1_safe(s, strend)
1811 is_VERTWS_utf8_safe(s, strend),
1812 is_VERTWS_latin1_safe(s, strend)
1817 !is_VERTWS_utf8_safe(s, strend),
1818 !is_VERTWS_latin1_safe(s, strend)
1823 is_HORIZWS_utf8_safe(s, strend),
1824 is_HORIZWS_latin1_safe(s, strend)
1829 !is_HORIZWS_utf8_safe(s, strend),
1830 !is_HORIZWS_latin1_safe(s, strend)
1834 /* Don't need to worry about utf8, as it can match only a single
1835 * byte invariant character. The flag in this node type is the
1836 * class number to pass to _generic_isCC() to build a mask for
1837 * searching in PL_charclass[] */
1838 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1842 !_generic_isCC_A(*s, FLAGS(c)),
1843 !_generic_isCC_A(*s, FLAGS(c))
1851 /* what trie are we using right now */
1853 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1855 = (reg_trie_data*)progi->data->data[ aho->trie ];
1856 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1858 const char *last_start = strend - trie->minlen;
1860 const char *real_start = s;
1862 STRLEN maxlen = trie->maxlen;
1864 U8 **points; /* map of where we were in the input string
1865 when reading a given char. For ASCII this
1866 is unnecessary overhead as the relationship
1867 is always 1:1, but for Unicode, especially
1868 case folded Unicode this is not true. */
1869 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1873 GET_RE_DEBUG_FLAGS_DECL;
1875 /* We can't just allocate points here. We need to wrap it in
1876 * an SV so it gets freed properly if there is a croak while
1877 * running the match */
1880 sv_points=newSV(maxlen * sizeof(U8 *));
1881 SvCUR_set(sv_points,
1882 maxlen * sizeof(U8 *));
1883 SvPOK_on(sv_points);
1884 sv_2mortal(sv_points);
1885 points=(U8**)SvPV_nolen(sv_points );
1886 if ( trie_type != trie_utf8_fold
1887 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1890 bitmap=(U8*)trie->bitmap;
1892 bitmap=(U8*)ANYOF_BITMAP(c);
1894 /* this is the Aho-Corasick algorithm modified a touch
1895 to include special handling for long "unknown char"
1896 sequences. The basic idea being that we use AC as long
1897 as we are dealing with a possible matching char, when
1898 we encounter an unknown char (and we have not encountered
1899 an accepting state) we scan forward until we find a legal
1901 AC matching is basically that of trie matching, except
1902 that when we encounter a failing transition, we fall back
1903 to the current states "fail state", and try the current char
1904 again, a process we repeat until we reach the root state,
1905 state 1, or a legal transition. If we fail on the root state
1906 then we can either terminate if we have reached an accepting
1907 state previously, or restart the entire process from the beginning
1911 while (s <= last_start) {
1912 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1920 U8 *uscan = (U8*)NULL;
1921 U8 *leftmost = NULL;
1923 U32 accepted_word= 0;
1927 while ( state && uc <= (U8*)strend ) {
1929 U32 word = aho->states[ state ].wordnum;
1933 DEBUG_TRIE_EXECUTE_r(
1934 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1935 dump_exec_pos( (char *)uc, c, strend, real_start,
1936 (char *)uc, utf8_target );
1937 PerlIO_printf( Perl_debug_log,
1938 " Scanning for legal start char...\n");
1942 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1946 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1952 if (uc >(U8*)last_start) break;
1956 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1957 if (!leftmost || lpos < leftmost) {
1958 DEBUG_r(accepted_word=word);
1964 points[pointpos++ % maxlen]= uc;
1965 if (foldlen || uc < (U8*)strend) {
1966 REXEC_TRIE_READ_CHAR(trie_type, trie,
1968 uscan, len, uvc, charid, foldlen,
1970 DEBUG_TRIE_EXECUTE_r({
1971 dump_exec_pos( (char *)uc, c, strend,
1972 real_start, s, utf8_target);
1973 PerlIO_printf(Perl_debug_log,
1974 " Charid:%3u CP:%4"UVxf" ",
1986 word = aho->states[ state ].wordnum;
1988 base = aho->states[ state ].trans.base;
1990 DEBUG_TRIE_EXECUTE_r({
1992 dump_exec_pos( (char *)uc, c, strend, real_start,
1994 PerlIO_printf( Perl_debug_log,
1995 "%sState: %4"UVxf", word=%"UVxf,
1996 failed ? " Fail transition to " : "",
1997 (UV)state, (UV)word);
2003 ( ((offset = base + charid
2004 - 1 - trie->uniquecharcount)) >= 0)
2005 && ((U32)offset < trie->lasttrans)
2006 && trie->trans[offset].check == state
2007 && (tmp=trie->trans[offset].next))
2009 DEBUG_TRIE_EXECUTE_r(
2010 PerlIO_printf( Perl_debug_log," - legal\n"));
2015 DEBUG_TRIE_EXECUTE_r(
2016 PerlIO_printf( Perl_debug_log," - fail\n"));
2018 state = aho->fail[state];
2022 /* we must be accepting here */
2023 DEBUG_TRIE_EXECUTE_r(
2024 PerlIO_printf( Perl_debug_log," - accepting\n"));
2033 if (!state) state = 1;
2036 if ( aho->states[ state ].wordnum ) {
2037 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2038 if (!leftmost || lpos < leftmost) {
2039 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2044 s = (char*)leftmost;
2045 DEBUG_TRIE_EXECUTE_r({
2047 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2048 (UV)accepted_word, (IV)(s - real_start)
2051 if (!reginfo || regtry(reginfo, &s)) {
2057 DEBUG_TRIE_EXECUTE_r({
2058 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2061 DEBUG_TRIE_EXECUTE_r(
2062 PerlIO_printf( Perl_debug_log,"No match.\n"));
2071 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2081 - regexec_flags - match a regexp against a string
2084 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2085 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2086 /* stringarg: the point in the string at which to begin matching */
2087 /* strend: pointer to null at end of string */
2088 /* strbeg: real beginning of string */
2089 /* minend: end of match must be >= minend bytes after stringarg. */
2090 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2091 * itself is accessed via the pointers above */
2092 /* data: May be used for some additional optimizations.
2093 Currently its only used, with a U32 cast, for transmitting
2094 the ganch offset when doing a /g match. This will change */
2095 /* nosave: For optimizations. */
2099 struct regexp *const prog = ReANY(rx);
2100 /*register*/ char *s;
2102 /*register*/ char *startpos = stringarg;
2103 I32 minlen; /* must match at least this many chars */
2104 I32 dontbother = 0; /* how many characters not to try at end */
2105 I32 end_shift = 0; /* Same for the end. */ /* CC */
2106 I32 scream_pos = -1; /* Internal iterator of scream. */
2107 char *scream_olds = NULL;
2108 const bool utf8_target = cBOOL(DO_UTF8(sv));
2110 RXi_GET_DECL(prog,progi);
2111 regmatch_info reginfo; /* create some info to pass to regtry etc */
2112 regexp_paren_pair *swap = NULL;
2113 GET_RE_DEBUG_FLAGS_DECL;
2115 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2116 PERL_UNUSED_ARG(data);
2118 /* Be paranoid... */
2119 if (prog == NULL || startpos == NULL) {
2120 Perl_croak(aTHX_ "NULL regexp parameter");
2124 multiline = prog->extflags & RXf_PMf_MULTILINE;
2125 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2127 RX_MATCH_UTF8_set(rx, utf8_target);
2129 debug_start_match(rx, utf8_target, startpos, strend,
2133 minlen = prog->minlen;
2135 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2136 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2137 "String too short [regexec_flags]...\n"));
2142 /* Check validity of program. */
2143 if (UCHARAT(progi->program) != REG_MAGIC) {
2144 Perl_croak(aTHX_ "corrupted regexp program");
2148 PL_reg_state.re_state_eval_setup_done = FALSE;
2152 PL_reg_flags |= RF_utf8;
2154 /* Mark beginning of line for ^ and lookbehind. */
2155 reginfo.bol = startpos; /* XXX not used ??? */
2159 /* Mark end of line for $ (and such) */
2162 /* see how far we have to get to not match where we matched before */
2163 reginfo.till = startpos+minend;
2165 /* If there is a "must appear" string, look for it. */
2168 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2170 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2171 reginfo.ganch = startpos + prog->gofs;
2172 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2173 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2174 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2176 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2177 && mg->mg_len >= 0) {
2178 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2179 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2180 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2182 if (prog->extflags & RXf_ANCH_GPOS) {
2183 if (s > reginfo.ganch)
2185 s = reginfo.ganch - prog->gofs;
2186 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2187 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2193 reginfo.ganch = strbeg + PTR2UV(data);
2194 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2195 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2197 } else { /* pos() not defined */
2198 reginfo.ganch = strbeg;
2199 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2200 "GPOS: reginfo.ganch = strbeg\n"));
2203 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2204 /* We have to be careful. If the previous successful match
2205 was from this regex we don't want a subsequent partially
2206 successful match to clobber the old results.
2207 So when we detect this possibility we add a swap buffer
2208 to the re, and switch the buffer each match. If we fail
2209 we switch it back, otherwise we leave it swapped.
2212 /* do we need a save destructor here for eval dies? */
2213 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2214 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2215 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2221 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2222 re_scream_pos_data d;
2224 d.scream_olds = &scream_olds;
2225 d.scream_pos = &scream_pos;
2226 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2228 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2229 goto phooey; /* not present */
2235 /* Simplest case: anchored match need be tried only once. */
2236 /* [unless only anchor is BOL and multiline is set] */
2237 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2238 if (s == startpos && regtry(®info, &startpos))
2240 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2241 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2246 dontbother = minlen - 1;
2247 end = HOP3c(strend, -dontbother, strbeg) - 1;
2248 /* for multiline we only have to try after newlines */
2249 if (prog->check_substr || prog->check_utf8) {
2250 /* because of the goto we can not easily reuse the macros for bifurcating the
2251 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2254 goto after_try_utf8;
2256 if (regtry(®info, &s)) {
2263 if (prog->extflags & RXf_USE_INTUIT) {
2264 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2273 } /* end search for check string in unicode */
2275 if (s == startpos) {
2276 goto after_try_latin;
2279 if (regtry(®info, &s)) {
2286 if (prog->extflags & RXf_USE_INTUIT) {
2287 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2296 } /* end search for check string in latin*/
2297 } /* end search for check string */
2298 else { /* search for newline */
2300 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2303 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2304 while (s <= end) { /* note it could be possible to match at the end of the string */
2305 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2306 if (regtry(®info, &s))
2310 } /* end search for newline */
2311 } /* end anchored/multiline check string search */
2313 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2315 /* the warning about reginfo.ganch being used without initialization
2316 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2317 and we only enter this block when the same bit is set. */
2318 char *tmp_s = reginfo.ganch - prog->gofs;
2320 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2325 /* Messy cases: unanchored match. */
2326 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2327 /* we have /x+whatever/ */
2328 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2334 if (! prog->anchored_utf8) {
2335 to_utf8_substr(prog);
2337 ch = SvPVX_const(prog->anchored_utf8)[0];
2340 DEBUG_EXECUTE_r( did_match = 1 );
2341 if (regtry(®info, &s)) goto got_it;
2343 while (s < strend && *s == ch)
2350 if (! prog->anchored_substr) {
2351 if (! to_byte_substr(prog)) {
2352 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2355 ch = SvPVX_const(prog->anchored_substr)[0];
2358 DEBUG_EXECUTE_r( did_match = 1 );
2359 if (regtry(®info, &s)) goto got_it;
2361 while (s < strend && *s == ch)
2366 DEBUG_EXECUTE_r(if (!did_match)
2367 PerlIO_printf(Perl_debug_log,
2368 "Did not find anchored character...\n")
2371 else if (prog->anchored_substr != NULL
2372 || prog->anchored_utf8 != NULL
2373 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2374 && prog->float_max_offset < strend - s)) {
2379 char *last1; /* Last position checked before */
2383 if (prog->anchored_substr || prog->anchored_utf8) {
2385 if (! prog->anchored_utf8) {
2386 to_utf8_substr(prog);
2388 must = prog->anchored_utf8;
2391 if (! prog->anchored_substr) {
2392 if (! to_byte_substr(prog)) {
2393 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2396 must = prog->anchored_substr;
2398 back_max = back_min = prog->anchored_offset;
2401 if (! prog->float_utf8) {
2402 to_utf8_substr(prog);
2404 must = prog->float_utf8;
2407 if (! prog->float_substr) {
2408 if (! to_byte_substr(prog)) {
2409 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2412 must = prog->float_substr;
2414 back_max = prog->float_max_offset;
2415 back_min = prog->float_min_offset;
2421 last = HOP3c(strend, /* Cannot start after this */
2422 -(I32)(CHR_SVLEN(must)
2423 - (SvTAIL(must) != 0) + back_min), strbeg);
2426 last1 = HOPc(s, -1);
2428 last1 = s - 1; /* bogus */
2430 /* XXXX check_substr already used to find "s", can optimize if
2431 check_substr==must. */
2433 dontbother = end_shift;
2434 strend = HOPc(strend, -dontbother);
2435 while ( (s <= last) &&
2436 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2437 (unsigned char*)strend, must,
2438 multiline ? FBMrf_MULTILINE : 0)) ) {
2439 DEBUG_EXECUTE_r( did_match = 1 );
2440 if (HOPc(s, -back_max) > last1) {
2441 last1 = HOPc(s, -back_min);
2442 s = HOPc(s, -back_max);
2445 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2447 last1 = HOPc(s, -back_min);
2451 while (s <= last1) {
2452 if (regtry(®info, &s))
2455 s++; /* to break out of outer loop */
2462 while (s <= last1) {
2463 if (regtry(®info, &s))
2469 DEBUG_EXECUTE_r(if (!did_match) {
2470 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2471 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2472 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2473 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2474 ? "anchored" : "floating"),
2475 quoted, RE_SV_TAIL(must));
2479 else if ( (c = progi->regstclass) ) {
2481 const OPCODE op = OP(progi->regstclass);
2482 /* don't bother with what can't match */
2483 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2484 strend = HOPc(strend, -(minlen - 1));
2487 SV * const prop = sv_newmortal();
2488 regprop(prog, prop, c);
2490 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2492 PerlIO_printf(Perl_debug_log,
2493 "Matching stclass %.*s against %s (%d bytes)\n",
2494 (int)SvCUR(prop), SvPVX_const(prop),
2495 quoted, (int)(strend - s));
2498 if (find_byclass(prog, c, s, strend, ®info))
2500 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2504 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2512 if (! prog->float_utf8) {
2513 to_utf8_substr(prog);
2515 float_real = prog->float_utf8;
2518 if (! prog->float_substr) {
2519 if (! to_byte_substr(prog)) {
2520 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2523 float_real = prog->float_substr;
2526 little = SvPV_const(float_real, len);
2527 if (SvTAIL(float_real)) {
2528 /* This means that float_real contains an artificial \n on
2529 * the end due to the presence of something like this:
2530 * /foo$/ where we can match both "foo" and "foo\n" at the
2531 * end of the string. So we have to compare the end of the
2532 * string first against the float_real without the \n and
2533 * then against the full float_real with the string. We
2534 * have to watch out for cases where the string might be
2535 * smaller than the float_real or the float_real without
2537 char *checkpos= strend - len;
2539 PerlIO_printf(Perl_debug_log,
2540 "%sChecking for float_real.%s\n",
2541 PL_colors[4], PL_colors[5]));
2542 if (checkpos + 1 < strbeg) {
2543 /* can't match, even if we remove the trailing \n
2544 * string is too short to match */
2546 PerlIO_printf(Perl_debug_log,
2547 "%sString shorter than required trailing substring, cannot match.%s\n",
2548 PL_colors[4], PL_colors[5]));
2550 } else if (memEQ(checkpos + 1, little, len - 1)) {
2551 /* can match, the end of the string matches without the
2553 last = checkpos + 1;
2554 } else if (checkpos < strbeg) {
2555 /* cant match, string is too short when the "\n" is
2558 PerlIO_printf(Perl_debug_log,
2559 "%sString does not contain required trailing substring, cannot match.%s\n",
2560 PL_colors[4], PL_colors[5]));
2562 } else if (!multiline) {
2563 /* non multiline match, so compare with the "\n" at the
2564 * end of the string */
2565 if (memEQ(checkpos, little, len)) {
2569 PerlIO_printf(Perl_debug_log,
2570 "%sString does not contain required trailing substring, cannot match.%s\n",
2571 PL_colors[4], PL_colors[5]));
2575 /* multiline match, so we have to search for a place
2576 * where the full string is located */
2582 last = rninstr(s, strend, little, little + len);
2584 last = strend; /* matching "$" */
2587 /* at one point this block contained a comment which was
2588 * probably incorrect, which said that this was a "should not
2589 * happen" case. Even if it was true when it was written I am
2590 * pretty sure it is not anymore, so I have removed the comment
2591 * and replaced it with this one. Yves */
2593 PerlIO_printf(Perl_debug_log,
2594 "String does not contain required substring, cannot match.\n"
2598 dontbother = strend - last + prog->float_min_offset;
2600 if (minlen && (dontbother < minlen))
2601 dontbother = minlen - 1;
2602 strend -= dontbother; /* this one's always in bytes! */
2603 /* We don't know much -- general case. */
2606 if (regtry(®info, &s))
2615 if (regtry(®info, &s))
2617 } while (s++ < strend);
2627 PerlIO_printf(Perl_debug_log,
2628 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2634 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2636 if (PL_reg_state.re_state_eval_setup_done)
2637 restore_pos(aTHX_ prog);
2638 if (RXp_PAREN_NAMES(prog))
2639 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2641 /* make sure $`, $&, $', and $digit will work later */
2642 if ( !(flags & REXEC_NOT_FIRST) ) {
2643 if (flags & REXEC_COPY_STR) {
2644 #ifdef PERL_OLD_COPY_ON_WRITE
2646 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2648 PerlIO_printf(Perl_debug_log,
2649 "Copy on write: regexp capture, type %d\n",
2652 RX_MATCH_COPY_FREE(rx);
2653 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2654 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2655 assert (SvPOKp(prog->saved_copy));
2656 prog->sublen = PL_regeol - strbeg;
2657 prog->suboffset = 0;
2658 prog->subcoffset = 0;
2663 I32 max = PL_regeol - strbeg;
2666 if ( (flags & REXEC_COPY_SKIP_POST)
2667 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2668 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2669 ) { /* don't copy $' part of string */
2672 /* calculate the right-most part of the string covered
2673 * by a capture. Due to look-ahead, this may be to
2674 * the right of $&, so we have to scan all captures */
2675 while (n <= prog->lastparen) {
2676 if (prog->offs[n].end > max)
2677 max = prog->offs[n].end;
2681 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2682 ? prog->offs[0].start
2684 assert(max >= 0 && max <= PL_regeol - strbeg);
2687 if ( (flags & REXEC_COPY_SKIP_PRE)
2688 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2689 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2690 ) { /* don't copy $` part of string */
2693 /* calculate the left-most part of the string covered
2694 * by a capture. Due to look-behind, this may be to
2695 * the left of $&, so we have to scan all captures */
2696 while (min && n <= prog->lastparen) {
2697 if ( prog->offs[n].start != -1
2698 && prog->offs[n].start < min)
2700 min = prog->offs[n].start;
2704 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2705 && min > prog->offs[0].end
2707 min = prog->offs[0].end;
2711 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2714 if (RX_MATCH_COPIED(rx)) {
2715 if (sublen > prog->sublen)
2717 (char*)saferealloc(prog->subbeg, sublen+1);
2720 prog->subbeg = (char*)safemalloc(sublen+1);
2721 Copy(strbeg + min, prog->subbeg, sublen, char);
2722 prog->subbeg[sublen] = '\0';
2723 prog->suboffset = min;
2724 prog->sublen = sublen;
2725 RX_MATCH_COPIED_on(rx);
2727 prog->subcoffset = prog->suboffset;
2728 if (prog->suboffset && utf8_target) {
2729 /* Convert byte offset to chars.
2730 * XXX ideally should only compute this if @-/@+
2731 * has been seen, a la PL_sawampersand ??? */
2733 /* If there's a direct correspondence between the
2734 * string which we're matching and the original SV,
2735 * then we can use the utf8 len cache associated with
2736 * the SV. In particular, it means that under //g,
2737 * sv_pos_b2u() will use the previously cached
2738 * position to speed up working out the new length of
2739 * subcoffset, rather than counting from the start of
2740 * the string each time. This stops
2741 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2742 * from going quadratic */
2743 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2744 sv_pos_b2u(sv, &(prog->subcoffset));
2746 prog->subcoffset = utf8_length((U8*)strbeg,
2747 (U8*)(strbeg+prog->suboffset));
2751 RX_MATCH_COPY_FREE(rx);
2752 prog->subbeg = strbeg;
2753 prog->suboffset = 0;
2754 prog->subcoffset = 0;
2755 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2762 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2763 PL_colors[4], PL_colors[5]));
2764 if (PL_reg_state.re_state_eval_setup_done)
2765 restore_pos(aTHX_ prog);
2767 /* we failed :-( roll it back */
2768 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2769 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2774 Safefree(prog->offs);
2781 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2782 * Do inc before dec, in case old and new rex are the same */
2783 #define SET_reg_curpm(Re2) \
2784 if (PL_reg_state.re_state_eval_setup_done) { \
2785 (void)ReREFCNT_inc(Re2); \
2786 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2787 PM_SETRE((PL_reg_curpm), (Re2)); \
2792 - regtry - try match at specific point
2794 STATIC I32 /* 0 failure, 1 success */
2795 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2799 REGEXP *const rx = reginfo->prog;
2800 regexp *const prog = ReANY(rx);
2802 RXi_GET_DECL(prog,progi);
2803 GET_RE_DEBUG_FLAGS_DECL;
2805 PERL_ARGS_ASSERT_REGTRY;
2807 reginfo->cutpoint=NULL;
2809 if ((prog->extflags & RXf_EVAL_SEEN)
2810 && !PL_reg_state.re_state_eval_setup_done)
2814 PL_reg_state.re_state_eval_setup_done = TRUE;
2816 /* Make $_ available to executed code. */
2817 if (reginfo->sv != DEFSV) {
2819 DEFSV_set(reginfo->sv);
2822 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2823 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2824 /* prepare for quick setting of pos */
2825 #ifdef PERL_OLD_COPY_ON_WRITE
2826 if (SvIsCOW(reginfo->sv))
2827 sv_force_normal_flags(reginfo->sv, 0);
2829 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2830 &PL_vtbl_mglob, NULL, 0);
2834 PL_reg_oldpos = mg->mg_len;
2835 SAVEDESTRUCTOR_X(restore_pos, prog);
2837 if (!PL_reg_curpm) {
2838 Newxz(PL_reg_curpm, 1, PMOP);
2841 SV* const repointer = &PL_sv_undef;
2842 /* this regexp is also owned by the new PL_reg_curpm, which
2843 will try to free it. */
2844 av_push(PL_regex_padav, repointer);
2845 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2846 PL_regex_pad = AvARRAY(PL_regex_padav);
2851 PL_reg_oldcurpm = PL_curpm;
2852 PL_curpm = PL_reg_curpm;
2853 if (RXp_MATCH_COPIED(prog)) {
2854 /* Here is a serious problem: we cannot rewrite subbeg,
2855 since it may be needed if this match fails. Thus
2856 $` inside (?{}) could fail... */
2857 PL_reg_oldsaved = prog->subbeg;
2858 PL_reg_oldsavedlen = prog->sublen;
2859 PL_reg_oldsavedoffset = prog->suboffset;
2860 PL_reg_oldsavedcoffset = prog->suboffset;
2861 #ifdef PERL_OLD_COPY_ON_WRITE
2862 PL_nrs = prog->saved_copy;
2864 RXp_MATCH_COPIED_off(prog);
2867 PL_reg_oldsaved = NULL;
2868 prog->subbeg = PL_bostr;
2869 prog->suboffset = 0;
2870 prog->subcoffset = 0;
2871 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2874 PL_reg_starttry = *startposp;
2876 prog->offs[0].start = *startposp - PL_bostr;
2877 prog->lastparen = 0;
2878 prog->lastcloseparen = 0;
2881 /* XXXX What this code is doing here?!!! There should be no need
2882 to do this again and again, prog->lastparen should take care of
2885 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2886 * Actually, the code in regcppop() (which Ilya may be meaning by
2887 * prog->lastparen), is not needed at all by the test suite
2888 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2889 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2890 * Meanwhile, this code *is* needed for the
2891 * above-mentioned test suite tests to succeed. The common theme
2892 * on those tests seems to be returning null fields from matches.
2893 * --jhi updated by dapm */
2895 if (prog->nparens) {
2896 regexp_paren_pair *pp = prog->offs;
2898 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2906 result = regmatch(reginfo, *startposp, progi->program + 1);
2908 prog->offs[0].end = result;
2911 if (reginfo->cutpoint)
2912 *startposp= reginfo->cutpoint;
2913 REGCP_UNWIND(lastcp);
2918 #define sayYES goto yes
2919 #define sayNO goto no
2920 #define sayNO_SILENT goto no_silent
2922 /* we dont use STMT_START/END here because it leads to
2923 "unreachable code" warnings, which are bogus, but distracting. */
2924 #define CACHEsayNO \
2925 if (ST.cache_mask) \
2926 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2929 /* this is used to determine how far from the left messages like
2930 'failed...' are printed. It should be set such that messages
2931 are inline with the regop output that created them.
2933 #define REPORT_CODE_OFF 32
2936 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2937 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2938 #define CHRTEST_NOT_A_CP_1 -999
2939 #define CHRTEST_NOT_A_CP_2 -998
2941 #define SLAB_FIRST(s) (&(s)->states[0])
2942 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2944 /* grab a new slab and return the first slot in it */
2946 STATIC regmatch_state *
2949 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2952 regmatch_slab *s = PL_regmatch_slab->next;
2954 Newx(s, 1, regmatch_slab);
2955 s->prev = PL_regmatch_slab;
2957 PL_regmatch_slab->next = s;
2959 PL_regmatch_slab = s;
2960 return SLAB_FIRST(s);
2964 /* push a new state then goto it */
2966 #define PUSH_STATE_GOTO(state, node, input) \
2967 pushinput = input; \
2969 st->resume_state = state; \
2972 /* push a new state with success backtracking, then goto it */
2974 #define PUSH_YES_STATE_GOTO(state, node, input) \
2975 pushinput = input; \
2977 st->resume_state = state; \
2978 goto push_yes_state;
2985 regmatch() - main matching routine
2987 This is basically one big switch statement in a loop. We execute an op,
2988 set 'next' to point the next op, and continue. If we come to a point which
2989 we may need to backtrack to on failure such as (A|B|C), we push a
2990 backtrack state onto the backtrack stack. On failure, we pop the top
2991 state, and re-enter the loop at the state indicated. If there are no more
2992 states to pop, we return failure.
2994 Sometimes we also need to backtrack on success; for example /A+/, where
2995 after successfully matching one A, we need to go back and try to
2996 match another one; similarly for lookahead assertions: if the assertion
2997 completes successfully, we backtrack to the state just before the assertion
2998 and then carry on. In these cases, the pushed state is marked as
2999 'backtrack on success too'. This marking is in fact done by a chain of
3000 pointers, each pointing to the previous 'yes' state. On success, we pop to
3001 the nearest yes state, discarding any intermediate failure-only states.
3002 Sometimes a yes state is pushed just to force some cleanup code to be
3003 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3004 it to free the inner regex.
3006 Note that failure backtracking rewinds the cursor position, while
3007 success backtracking leaves it alone.
3009 A pattern is complete when the END op is executed, while a subpattern
3010 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3011 ops trigger the "pop to last yes state if any, otherwise return true"
3014 A common convention in this function is to use A and B to refer to the two
3015 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3016 the subpattern to be matched possibly multiple times, while B is the entire
3017 rest of the pattern. Variable and state names reflect this convention.
3019 The states in the main switch are the union of ops and failure/success of
3020 substates associated with with that op. For example, IFMATCH is the op
3021 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3022 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3023 successfully matched A and IFMATCH_A_fail is a state saying that we have
3024 just failed to match A. Resume states always come in pairs. The backtrack
3025 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3026 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3027 on success or failure.
3029 The struct that holds a backtracking state is actually a big union, with
3030 one variant for each major type of op. The variable st points to the
3031 top-most backtrack struct. To make the code clearer, within each
3032 block of code we #define ST to alias the relevant union.
3034 Here's a concrete example of a (vastly oversimplified) IFMATCH
3040 #define ST st->u.ifmatch
3042 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3043 ST.foo = ...; // some state we wish to save
3045 // push a yes backtrack state with a resume value of
3046 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3048 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3051 case IFMATCH_A: // we have successfully executed A; now continue with B
3053 bar = ST.foo; // do something with the preserved value
3056 case IFMATCH_A_fail: // A failed, so the assertion failed
3057 ...; // do some housekeeping, then ...
3058 sayNO; // propagate the failure
3065 For any old-timers reading this who are familiar with the old recursive
3066 approach, the code above is equivalent to:
3068 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3077 ...; // do some housekeeping, then ...
3078 sayNO; // propagate the failure
3081 The topmost backtrack state, pointed to by st, is usually free. If you
3082 want to claim it, populate any ST.foo fields in it with values you wish to
3083 save, then do one of
3085 PUSH_STATE_GOTO(resume_state, node, newinput);
3086 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3088 which sets that backtrack state's resume value to 'resume_state', pushes a
3089 new free entry to the top of the backtrack stack, then goes to 'node'.
3090 On backtracking, the free slot is popped, and the saved state becomes the
3091 new free state. An ST.foo field in this new top state can be temporarily
3092 accessed to retrieve values, but once the main loop is re-entered, it
3093 becomes available for reuse.
3095 Note that the depth of the backtrack stack constantly increases during the
3096 left-to-right execution of the pattern, rather than going up and down with
3097 the pattern nesting. For example the stack is at its maximum at Z at the
3098 end of the pattern, rather than at X in the following:
3100 /(((X)+)+)+....(Y)+....Z/
3102 The only exceptions to this are lookahead/behind assertions and the cut,
3103 (?>A), which pop all the backtrack states associated with A before
3106 Backtrack state structs are allocated in slabs of about 4K in size.
3107 PL_regmatch_state and st always point to the currently active state,
3108 and PL_regmatch_slab points to the slab currently containing
3109 PL_regmatch_state. The first time regmatch() is called, the first slab is
3110 allocated, and is never freed until interpreter destruction. When the slab
3111 is full, a new one is allocated and chained to the end. At exit from
3112 regmatch(), slabs allocated since entry are freed.
3117 #define DEBUG_STATE_pp(pp) \
3119 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3120 PerlIO_printf(Perl_debug_log, \
3121 " %*s"pp" %s%s%s%s%s\n", \
3123 PL_reg_name[st->resume_state], \
3124 ((st==yes_state||st==mark_state) ? "[" : ""), \
3125 ((st==yes_state) ? "Y" : ""), \
3126 ((st==mark_state) ? "M" : ""), \
3127 ((st==yes_state||st==mark_state) ? "]" : "") \
3132 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3137 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3138 const char *start, const char *end, const char *blurb)
3140 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3142 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3147 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3148 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3150 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3151 start, end - start, 60);
3153 PerlIO_printf(Perl_debug_log,
3154 "%s%s REx%s %s against %s\n",
3155 PL_colors[4], blurb, PL_colors[5], s0, s1);
3157 if (utf8_target||utf8_pat)
3158 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3159 utf8_pat ? "pattern" : "",
3160 utf8_pat && utf8_target ? " and " : "",
3161 utf8_target ? "string" : ""
3167 S_dump_exec_pos(pTHX_ const char *locinput,
3168 const regnode *scan,
3169 const char *loc_regeol,
3170 const char *loc_bostr,
3171 const char *loc_reg_starttry,
3172 const bool utf8_target)
3174 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3175 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3176 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3177 /* The part of the string before starttry has one color
3178 (pref0_len chars), between starttry and current
3179 position another one (pref_len - pref0_len chars),
3180 after the current position the third one.
3181 We assume that pref0_len <= pref_len, otherwise we
3182 decrease pref0_len. */
3183 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3184 ? (5 + taill) - l : locinput - loc_bostr;
3187 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3189 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3191 pref0_len = pref_len - (locinput - loc_reg_starttry);
3192 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3193 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3194 ? (5 + taill) - pref_len : loc_regeol - locinput);
3195 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3199 if (pref0_len > pref_len)
3200 pref0_len = pref_len;
3202 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3204 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3205 (locinput - pref_len),pref0_len, 60, 4, 5);
3207 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3208 (locinput - pref_len + pref0_len),
3209 pref_len - pref0_len, 60, 2, 3);
3211 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3212 locinput, loc_regeol - locinput, 10, 0, 1);
3214 const STRLEN tlen=len0+len1+len2;
3215 PerlIO_printf(Perl_debug_log,
3216 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3217 (IV)(locinput - loc_bostr),
3220 (docolor ? "" : "> <"),
3222 (int)(tlen > 19 ? 0 : 19 - tlen),
3229 /* reg_check_named_buff_matched()
3230 * Checks to see if a named buffer has matched. The data array of
3231 * buffer numbers corresponding to the buffer is expected to reside
3232 * in the regexp->data->data array in the slot stored in the ARG() of
3233 * node involved. Note that this routine doesn't actually care about the
3234 * name, that information is not preserved from compilation to execution.
3235 * Returns the index of the leftmost defined buffer with the given name
3236 * or 0 if non of the buffers matched.
3239 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3242 RXi_GET_DECL(rex,rexi);
3243 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3244 I32 *nums=(I32*)SvPVX(sv_dat);
3246 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3248 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3249 if ((I32)rex->lastparen >= nums[n] &&
3250 rex->offs[nums[n]].end != -1)
3259 /* free all slabs above current one - called during LEAVE_SCOPE */
3262 S_clear_backtrack_stack(pTHX_ void *p)
3264 regmatch_slab *s = PL_regmatch_slab->next;
3269 PL_regmatch_slab->next = NULL;
3271 regmatch_slab * const osl = s;
3277 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, U8* c1_utf8, int *c2p, U8* c2_utf8)
3279 /* This function determines if there are one or two characters that match
3280 * the first character of the passed-in EXACTish node <text_node>, and if
3281 * so, returns them in the passed-in pointers.
3283 * If it determines that no possible character in the target string can
3284 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3285 * the first character in <text_node> requires UTF-8 to represent, and the
3286 * target string isn't in UTF-8.)
3288 * If there are more than two characters that could match the beginning of
3289 * <text_node>, or if more context is required to determine a match or not,
3290 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3292 * The motiviation behind this function is to allow the caller to set up
3293 * tight loops for matching. If <text_node> is of type EXACT, there is
3294 * only one possible character that can match its first character, and so
3295 * the situation is quite simple. But things get much more complicated if
3296 * folding is involved. It may be that the first character of an EXACTFish
3297 * node doesn't participate in any possible fold, e.g., punctuation, so it
3298 * can be matched only by itself. The vast majority of characters that are
3299 * in folds match just two things, their lower and upper-case equivalents.
3300 * But not all are like that; some have multiple possible matches, or match
3301 * sequences of more than one character. This function sorts all that out.
3303 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3304 * loop of trying to match A*, we know we can't exit where the thing
3305 * following it isn't a B. And something can't be a B unless it is the
3306 * beginning of B. By putting a quick test for that beginning in a tight
3307 * loop, we can rule out things that can't possibly be B without having to
3308 * break out of the loop, thus avoiding work. Similarly, if A is a single
3309 * character, we can make a tight loop matching A*, using the outputs of
3312 * If the target string to match isn't in UTF-8, and there aren't
3313 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3314 * the one or two possible octets (which are characters in this situation)
3315 * that can match. In all cases, if there is only one character that can
3316 * match, *<c1p> and *<c2p> will be identical.
3318 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3319 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3320 * can match the beginning of <text_node>. They should be declared with at
3321 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3322 * undefined what these contain.) If one or both of the buffers are
3323 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3324 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3325 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3326 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3327 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3329 const bool utf8_target = PL_reg_match_utf8;
3331 UV c1 = CHRTEST_NOT_A_CP_1;
3332 UV c2 = CHRTEST_NOT_A_CP_2;
3333 bool use_chrtest_void = FALSE;
3335 /* Used when we have both utf8 input and utf8 output, to avoid converting
3336 * to/from code points */
3337 bool utf8_has_been_setup = FALSE;
3341 U8 *pat = (U8*)STRING(text_node);
3343 if (OP(text_node) == EXACT) {
3345 /* In an exact node, only one thing can be matched, that first
3346 * character. If both the pat and the target are UTF-8, we can just
3347 * copy the input to the output, avoiding finding the code point of
3349 if (! UTF_PATTERN) {
3352 else if (utf8_target) {
3353 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3354 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3355 utf8_has_been_setup = TRUE;
3358 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3361 else /* an EXACTFish node */
3363 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3364 pat + STR_LEN(text_node)))
3366 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3367 pat + STR_LEN(text_node))))
3369 /* Multi-character folds require more context to sort out. Also
3370 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3371 * handled outside this routine */
3372 use_chrtest_void = TRUE;
3374 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3375 c1 = (UTF_PATTERN) ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3377 /* Load the folds hash, if not already done */
3379 if (! PL_utf8_foldclosures) {
3380 if (! PL_utf8_tofold) {
3381 U8 dummy[UTF8_MAXBYTES+1];
3383 /* Force loading this by folding an above-Latin1 char */
3384 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3385 assert(PL_utf8_tofold); /* Verify that worked */
3387 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3390 /* The fold closures data structure is a hash with the keys being
3391 * the UTF-8 of every character that is folded to, like 'k', and
3392 * the values each an array of all code points that fold to its
3393 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3395 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3400 /* Not found in the hash, therefore there are no folds
3401 * containing it, so there is only a single character that
3405 else { /* Does participate in folds */
3406 AV* list = (AV*) *listp;
3407 if (av_len(list) != 1) {
3409 /* If there aren't exactly two folds to this, it is outside
3410 * the scope of this function */
3411 use_chrtest_void = TRUE;
3413 else { /* There are two. Get them */
3414 SV** c_p = av_fetch(list, 0, FALSE);
3416 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3420 c_p = av_fetch(list, 1, FALSE);
3422 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3426 /* Folds that cross the 255/256 boundary are forbidden if
3427 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3428 * pattern character is above 256, and its only other match
3429 * is below 256, the only legal match will be to itself.
3430 * We have thrown away the original, so have to compute
3431 * which is the one above 255 */
3432 if ((c1 < 256) != (c2 < 256)) {
3433 if (OP(text_node) == EXACTFL
3434 || (OP(text_node) == EXACTFA
3435 && (isASCII(c1) || isASCII(c2))))
3448 else /* Here, c1 is < 255 */
3450 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3451 && OP(text_node) != EXACTFL
3452 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3454 /* Here, there could be something above Latin1 in the target which
3455 * folds to this character in the pattern. All such cases except
3456 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3457 * involved in their folds, so are outside the scope of this
3459 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3460 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3463 use_chrtest_void = TRUE;
3466 else { /* Here nothing above Latin1 can fold to the pattern character */
3467 switch (OP(text_node)) {
3469 case EXACTFL: /* /l rules */
3470 c2 = PL_fold_locale[c1];
3474 if (! utf8_target) { /* /d rules */
3479 /* /u rules for all these. This happens to work for
3480 * EXACTFA as nothing in Latin1 folds to ASCII */
3482 case EXACTFU_TRICKYFOLD:
3485 c2 = PL_fold_latin1[c1];
3489 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3490 assert(0); /* NOTREACHED */
3495 /* Here have figured things out. Set up the returns */
3496 if (use_chrtest_void) {
3497 *c2p = *c1p = CHRTEST_VOID;
3499 else if (utf8_target) {
3500 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3501 uvchr_to_utf8(c1_utf8, c1);
3502 uvchr_to_utf8(c2_utf8, c2);
3505 /* Invariants are stored in both the utf8 and byte outputs; Use
3506 * negative numbers otherwise for the byte ones. Make sure that the
3507 * byte ones are the same iff the utf8 ones are the same */
3508 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3509 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3512 ? CHRTEST_NOT_A_CP_1
3513 : CHRTEST_NOT_A_CP_2;
3515 else if (c1 > 255) {
3516 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3521 *c1p = *c2p = c2; /* c2 is the only representable value */
3523 else { /* c1 is representable; see about c2 */
3525 *c2p = (c2 < 256) ? c2 : c1;
3531 /* returns -1 on failure, $+[0] on success */
3533 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3535 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3539 const bool utf8_target = PL_reg_match_utf8;
3540 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3541 REGEXP *rex_sv = reginfo->prog;
3542 regexp *rex = ReANY(rex_sv);
3543 RXi_GET_DECL(rex,rexi);
3545 /* the current state. This is a cached copy of PL_regmatch_state */
3547 /* cache heavy used fields of st in registers */
3550 U32 n = 0; /* general value; init to avoid compiler warning */
3551 I32 ln = 0; /* len or last; init to avoid compiler warning */
3552 char *locinput = startpos;
3553 char *pushinput; /* where to continue after a PUSH */
3554 I32 nextchr; /* is always set to UCHARAT(locinput) */
3556 bool result = 0; /* return value of S_regmatch */
3557 int depth = 0; /* depth of backtrack stack */
3558 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3559 const U32 max_nochange_depth =
3560 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3561 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3562 regmatch_state *yes_state = NULL; /* state to pop to on success of
3564 /* mark_state piggy backs on the yes_state logic so that when we unwind
3565 the stack on success we can update the mark_state as we go */
3566 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3567 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3568 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3570 bool no_final = 0; /* prevent failure from backtracking? */
3571 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3572 char *startpoint = locinput;
3573 SV *popmark = NULL; /* are we looking for a mark? */
3574 SV *sv_commit = NULL; /* last mark name seen in failure */
3575 SV *sv_yes_mark = NULL; /* last mark name we have seen
3576 during a successful match */
3577 U32 lastopen = 0; /* last open we saw */
3578 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3579 SV* const oreplsv = GvSV(PL_replgv);
3580 /* these three flags are set by various ops to signal information to
3581 * the very next op. They have a useful lifetime of exactly one loop
3582 * iteration, and are not preserved or restored by state pushes/pops
3584 bool sw = 0; /* the condition value in (?(cond)a|b) */
3585 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3586 int logical = 0; /* the following EVAL is:
3590 or the following IFMATCH/UNLESSM is:
3591 false: plain (?=foo)
3592 true: used as a condition: (?(?=foo))
3594 PAD* last_pad = NULL;
3596 I32 gimme = G_SCALAR;
3597 CV *caller_cv = NULL; /* who called us */
3598 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3599 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3602 GET_RE_DEBUG_FLAGS_DECL;
3605 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3606 multicall_oldcatch = 0;
3607 multicall_cv = NULL;
3609 PERL_UNUSED_VAR(multicall_cop);
3610 PERL_UNUSED_VAR(newsp);
3613 PERL_ARGS_ASSERT_REGMATCH;
3615 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3616 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3618 /* on first ever call to regmatch, allocate first slab */
3619 if (!PL_regmatch_slab) {
3620 Newx(PL_regmatch_slab, 1, regmatch_slab);
3621 PL_regmatch_slab->prev = NULL;
3622 PL_regmatch_slab->next = NULL;
3623 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3626 oldsave = PL_savestack_ix;
3627 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3628 SAVEVPTR(PL_regmatch_slab);
3629 SAVEVPTR(PL_regmatch_state);
3631 /* grab next free state slot */
3632 st = ++PL_regmatch_state;
3633 if (st > SLAB_LAST(PL_regmatch_slab))
3634 st = PL_regmatch_state = S_push_slab(aTHX);
3636 /* Note that nextchr is a byte even in UTF */
3639 while (scan != NULL) {
3642 SV * const prop = sv_newmortal();
3643 regnode *rnext=regnext(scan);
3644 DUMP_EXEC_POS( locinput, scan, utf8_target );
3645 regprop(rex, prop, scan);
3647 PerlIO_printf(Perl_debug_log,
3648 "%3"IVdf":%*s%s(%"IVdf")\n",
3649 (IV)(scan - rexi->program), depth*2, "",
3651 (PL_regkind[OP(scan)] == END || !rnext) ?
3652 0 : (IV)(rnext - rexi->program));
3655 next = scan + NEXT_OFF(scan);
3658 state_num = OP(scan);
3663 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3665 switch (state_num) {
3666 case BOL: /* /^../ */
3667 if (locinput == PL_bostr)
3669 /* reginfo->till = reginfo->bol; */
3674 case MBOL: /* /^../m */
3675 if (locinput == PL_bostr ||
3676 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3682 case SBOL: /* /^../s */
3683 if (locinput == PL_bostr)
3688 if (locinput == reginfo->ganch)
3692 case KEEPS: /* \K */
3693 /* update the startpoint */
3694 st->u.keeper.val = rex->offs[0].start;
3695 rex->offs[0].start = locinput - PL_bostr;
3696 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3697 assert(0); /*NOTREACHED*/
3698 case KEEPS_next_fail:
3699 /* rollback the start point change */
3700 rex->offs[0].start = st->u.keeper.val;
3702 assert(0); /*NOTREACHED*/
3704 case EOL: /* /..$/ */
3707 case MEOL: /* /..$/m */
3708 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3712 case SEOL: /* /..$/s */
3714 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3716 if (PL_regeol - locinput > 1)
3721 if (!NEXTCHR_IS_EOS)
3725 case SANY: /* /./s */
3728 goto increment_locinput;
3736 case REG_ANY: /* /./ */
3737 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3739 goto increment_locinput;
3743 #define ST st->u.trie
3744 case TRIEC: /* (ab|cd) with known charclass */
3745 /* In this case the charclass data is available inline so
3746 we can fail fast without a lot of extra overhead.
3748 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3750 PerlIO_printf(Perl_debug_log,
3751 "%*s %sfailed to match trie start class...%s\n",
3752 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3755 assert(0); /* NOTREACHED */
3758 case TRIE: /* (ab|cd) */
3759 /* the basic plan of execution of the trie is:
3760 * At the beginning, run though all the states, and
3761 * find the longest-matching word. Also remember the position
3762 * of the shortest matching word. For example, this pattern:
3765 * when matched against the string "abcde", will generate
3766 * accept states for all words except 3, with the longest
3767 * matching word being 4, and the shortest being 2 (with
3768 * the position being after char 1 of the string).
3770 * Then for each matching word, in word order (i.e. 1,2,4,5),
3771 * we run the remainder of the pattern; on each try setting
3772 * the current position to the character following the word,
3773 * returning to try the next word on failure.
3775 * We avoid having to build a list of words at runtime by
3776 * using a compile-time structure, wordinfo[].prev, which
3777 * gives, for each word, the previous accepting word (if any).
3778 * In the case above it would contain the mappings 1->2, 2->0,
3779 * 3->0, 4->5, 5->1. We can use this table to generate, from
3780 * the longest word (4 above), a list of all words, by
3781 * following the list of prev pointers; this gives us the
3782 * unordered list 4,5,1,2. Then given the current word we have
3783 * just tried, we can go through the list and find the
3784 * next-biggest word to try (so if we just failed on word 2,
3785 * the next in the list is 4).
3787 * Since at runtime we don't record the matching position in
3788 * the string for each word, we have to work that out for
3789 * each word we're about to process. The wordinfo table holds
3790 * the character length of each word; given that we recorded
3791 * at the start: the position of the shortest word and its
3792 * length in chars, we just need to move the pointer the
3793 * difference between the two char lengths. Depending on
3794 * Unicode status and folding, that's cheap or expensive.
3796 * This algorithm is optimised for the case where are only a
3797 * small number of accept states, i.e. 0,1, or maybe 2.
3798 * With lots of accepts states, and having to try all of them,
3799 * it becomes quadratic on number of accept states to find all
3804 /* what type of TRIE am I? (utf8 makes this contextual) */
3805 DECL_TRIE_TYPE(scan);
3807 /* what trie are we using right now */
3808 reg_trie_data * const trie
3809 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3810 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3811 U32 state = trie->startstate;
3814 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3816 if (trie->states[ state ].wordnum) {
3818 PerlIO_printf(Perl_debug_log,
3819 "%*s %smatched empty string...%s\n",
3820 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3826 PerlIO_printf(Perl_debug_log,
3827 "%*s %sfailed to match trie start class...%s\n",
3828 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3835 U8 *uc = ( U8* )locinput;
3839 U8 *uscan = (U8*)NULL;
3840 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3841 U32 charcount = 0; /* how many input chars we have matched */
3842 U32 accepted = 0; /* have we seen any accepting states? */
3844 ST.jump = trie->jump;
3847 ST.longfold = FALSE; /* char longer if folded => it's harder */
3850 /* fully traverse the TRIE; note the position of the
3851 shortest accept state and the wordnum of the longest
3854 while ( state && uc <= (U8*)PL_regeol ) {
3855 U32 base = trie->states[ state ].trans.base;
3859 wordnum = trie->states[ state ].wordnum;
3861 if (wordnum) { /* it's an accept state */
3864 /* record first match position */