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")
168 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
169 /* No asserts are done for some of these, in case called on a */ \
170 /* Unicode version in which they map to nothing */ \
171 LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
172 LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
174 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
176 /* The actual code for CCC_TRY, which uses several variables from the routine
177 * it's callable from. It is designed to be the bulk of a case statement.
178 * FUNC is the macro or function to call on non-utf8 targets that indicate if
179 * nextchr matches the class.
180 * UTF8_TEST is the whole test string to use for utf8 targets
181 * LOAD is what to use to test, and if not present to load in the swash for the
183 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
185 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
186 * utf8 and a variant, load the swash if necessary and test using the utf8
187 * test. Advance to the next character if test is ok, otherwise fail; If not
188 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
189 * fails, or advance to the next character */
191 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
192 if (NEXTCHR_IS_EOS) { \
195 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
196 LOAD_UTF8_CHARCLASS(CLASS, STR); \
197 if (POS_OR_NEG (UTF8_TEST)) { \
201 else if (POS_OR_NEG (FUNC(nextchr))) { \
204 goto increment_locinput;
206 /* Handle the non-locale cases for a character class and its complement. It
207 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
208 * This is because that code fails when the test succeeds, so we want to have
209 * the test fail so that the code succeeds. The swash is stored in a
210 * predictable PL_ place */
211 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
214 _CCC_TRY_CODE( !, FUNC, \
215 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
216 (U8*)locinput, TRUE)), \
219 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
220 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
221 (U8*)locinput, TRUE)), \
224 /* Generate the case statements for both locale and non-locale character
225 * classes in regmatch for classes that don't have special unicode semantics.
226 * Locales don't use an immediate swash, but an intermediary special locale
227 * function that is called on the pointer to the current place in the input
228 * string. That function will resolve to needing the same swash. One might
229 * think that because we don't know what the locale will match, we shouldn't
230 * check with the swash loading function that it loaded properly; ie, that we
231 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
232 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
234 #define CCC_TRY(NAME, NNAME, FUNC, \
235 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
236 NAMEA, NNAMEA, FUNCA, \
239 PL_reg_flags |= RF_tainted; \
240 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
242 PL_reg_flags |= RF_tainted; \
243 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
246 if (NEXTCHR_IS_EOS || ! FUNCA(nextchr)) { \
249 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
253 if (NEXTCHR_IS_EOS || FUNCA(nextchr)) { \
256 goto increment_locinput; \
257 /* Generate the non-locale cases */ \
258 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
260 /* This is like CCC_TRY, but has an extra set of parameters for generating case
261 * statements to handle separate Unicode semantics nodes */
262 #define CCC_TRY_U(NAME, NNAME, FUNC, \
263 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
264 NAMEU, NNAMEU, FUNCU, \
265 NAMEA, NNAMEA, FUNCA, \
267 CCC_TRY(NAME, NNAME, FUNC, \
268 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
269 NAMEA, NNAMEA, FUNCA, \
271 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
273 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
275 /* for use after a quantifier and before an EXACT-like node -- japhy */
276 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
278 * NOTE that *nothing* that affects backtracking should be in here, specifically
279 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
280 * node that is in between two EXACT like nodes when ascertaining what the required
281 * "follow" character is. This should probably be moved to regex compile time
282 * although it may be done at run time beause of the REF possibility - more
283 * investigation required. -- demerphq
285 #define JUMPABLE(rn) ( \
287 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
289 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
290 OP(rn) == PLUS || OP(rn) == MINMOD || \
292 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
294 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
296 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
299 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
300 we don't need this definition. */
301 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
302 #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 )
303 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
306 /* ... so we use this as its faster. */
307 #define IS_TEXT(rn) ( OP(rn)==EXACT )
308 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
309 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
310 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
315 Search for mandatory following text node; for lookahead, the text must
316 follow but for lookbehind (rn->flags != 0) we skip to the next step.
318 #define FIND_NEXT_IMPT(rn) STMT_START { \
319 while (JUMPABLE(rn)) { \
320 const OPCODE type = OP(rn); \
321 if (type == SUSPEND || PL_regkind[type] == CURLY) \
322 rn = NEXTOPER(NEXTOPER(rn)); \
323 else if (type == PLUS) \
325 else if (type == IFMATCH) \
326 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
327 else rn += NEXT_OFF(rn); \
332 static void restore_pos(pTHX_ void *arg);
334 #define REGCP_PAREN_ELEMS 3
335 #define REGCP_OTHER_ELEMS 3
336 #define REGCP_FRAME_ELEMS 1
337 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
338 * are needed for the regexp context stack bookkeeping. */
341 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
344 const int retval = PL_savestack_ix;
345 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
346 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
347 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
349 GET_RE_DEBUG_FLAGS_DECL;
351 PERL_ARGS_ASSERT_REGCPPUSH;
353 if (paren_elems_to_push < 0)
354 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
355 paren_elems_to_push);
357 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
358 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
359 " out of range (%lu-%ld)",
360 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
362 SSGROW(total_elems + REGCP_FRAME_ELEMS);
365 if ((int)PL_regsize > (int)parenfloor)
366 PerlIO_printf(Perl_debug_log,
367 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
372 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
373 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
374 SSPUSHINT(rex->offs[p].end);
375 SSPUSHINT(rex->offs[p].start);
376 SSPUSHINT(rex->offs[p].start_tmp);
377 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
378 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
380 (IV)rex->offs[p].start,
381 (IV)rex->offs[p].start_tmp,
385 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
386 SSPUSHINT(PL_regsize);
387 SSPUSHINT(rex->lastparen);
388 SSPUSHINT(rex->lastcloseparen);
389 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
394 /* These are needed since we do not localize EVAL nodes: */
395 #define REGCP_SET(cp) \
397 PerlIO_printf(Perl_debug_log, \
398 " Setting an EVAL scope, savestack=%"IVdf"\n", \
399 (IV)PL_savestack_ix)); \
402 #define REGCP_UNWIND(cp) \
404 if (cp != PL_savestack_ix) \
405 PerlIO_printf(Perl_debug_log, \
406 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
407 (IV)(cp), (IV)PL_savestack_ix)); \
410 #define UNWIND_PAREN(lp, lcp) \
411 for (n = rex->lastparen; n > lp; n--) \
412 rex->offs[n].end = -1; \
413 rex->lastparen = n; \
414 rex->lastcloseparen = lcp;
418 S_regcppop(pTHX_ regexp *rex)
423 GET_RE_DEBUG_FLAGS_DECL;
425 PERL_ARGS_ASSERT_REGCPPOP;
427 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
429 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
430 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
431 rex->lastcloseparen = SSPOPINT;
432 rex->lastparen = SSPOPINT;
433 PL_regsize = SSPOPINT;
435 i -= REGCP_OTHER_ELEMS;
436 /* Now restore the parentheses context. */
438 if (i || rex->lastparen + 1 <= rex->nparens)
439 PerlIO_printf(Perl_debug_log,
440 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
446 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
448 rex->offs[paren].start_tmp = SSPOPINT;
449 rex->offs[paren].start = SSPOPINT;
451 if (paren <= rex->lastparen)
452 rex->offs[paren].end = tmps;
453 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
454 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
456 (IV)rex->offs[paren].start,
457 (IV)rex->offs[paren].start_tmp,
458 (IV)rex->offs[paren].end,
459 (paren > rex->lastparen ? "(skipped)" : ""));
464 /* It would seem that the similar code in regtry()
465 * already takes care of this, and in fact it is in
466 * a better location to since this code can #if 0-ed out
467 * but the code in regtry() is needed or otherwise tests
468 * requiring null fields (pat.t#187 and split.t#{13,14}
469 * (as of patchlevel 7877) will fail. Then again,
470 * this code seems to be necessary or otherwise
471 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
472 * --jhi updated by dapm */
473 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
475 rex->offs[i].start = -1;
476 rex->offs[i].end = -1;
477 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
478 " \\%"UVuf": %s ..-1 undeffing\n",
480 (i > PL_regsize) ? "-1" : " "
486 /* restore the parens and associated vars at savestack position ix,
487 * but without popping the stack */
490 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
492 I32 tmpix = PL_savestack_ix;
493 PL_savestack_ix = ix;
495 PL_savestack_ix = tmpix;
498 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
501 * pregexec and friends
504 #ifndef PERL_IN_XSUB_RE
506 - pregexec - match a regexp against a string
509 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
510 char *strbeg, I32 minend, SV *screamer, U32 nosave)
511 /* stringarg: the point in the string at which to begin matching */
512 /* strend: pointer to null at end of string */
513 /* strbeg: real beginning of string */
514 /* minend: end of match must be >= minend bytes after stringarg. */
515 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
516 * itself is accessed via the pointers above */
517 /* nosave: For optimizations. */
519 PERL_ARGS_ASSERT_PREGEXEC;
522 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
523 nosave ? 0 : REXEC_COPY_STR);
528 * Need to implement the following flags for reg_anch:
530 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
532 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
533 * INTUIT_AUTORITATIVE_ML
534 * INTUIT_ONCE_NOML - Intuit can match in one location only.
537 * Another flag for this function: SECOND_TIME (so that float substrs
538 * with giant delta may be not rechecked).
541 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
543 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
544 Otherwise, only SvCUR(sv) is used to get strbeg. */
546 /* XXXX We assume that strpos is strbeg unless sv. */
548 /* XXXX Some places assume that there is a fixed substring.
549 An update may be needed if optimizer marks as "INTUITable"
550 RExen without fixed substrings. Similarly, it is assumed that
551 lengths of all the strings are no more than minlen, thus they
552 cannot come from lookahead.
553 (Or minlen should take into account lookahead.)
554 NOTE: Some of this comment is not correct. minlen does now take account
555 of lookahead/behind. Further research is required. -- demerphq
559 /* A failure to find a constant substring means that there is no need to make
560 an expensive call to REx engine, thus we celebrate a failure. Similarly,
561 finding a substring too deep into the string means that less calls to
562 regtry() should be needed.
564 REx compiler's optimizer found 4 possible hints:
565 a) Anchored substring;
567 c) Whether we are anchored (beginning-of-line or \G);
568 d) First node (of those at offset 0) which may distinguish positions;
569 We use a)b)d) and multiline-part of c), and try to find a position in the
570 string which does not contradict any of them.
573 /* Most of decisions we do here should have been done at compile time.
574 The nodes of the REx which we used for the search should have been
575 deleted from the finite automaton. */
578 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
579 char *strend, const U32 flags, re_scream_pos_data *data)
582 struct regexp *const prog = ReANY(rx);
584 /* Should be nonnegative! */
590 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
592 char *other_last = NULL; /* other substr checked before this */
593 char *check_at = NULL; /* check substr found at this pos */
594 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
595 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
596 RXi_GET_DECL(prog,progi);
598 const char * const i_strpos = strpos;
600 GET_RE_DEBUG_FLAGS_DECL;
602 PERL_ARGS_ASSERT_RE_INTUIT_START;
603 PERL_UNUSED_ARG(flags);
604 PERL_UNUSED_ARG(data);
606 RX_MATCH_UTF8_set(rx,utf8_target);
609 PL_reg_flags |= RF_utf8;
612 debug_start_match(rx, utf8_target, strpos, strend,
613 sv ? "Guessing start of match in sv for"
614 : "Guessing start of match in string for");
617 /* CHR_DIST() would be more correct here but it makes things slow. */
618 if (prog->minlen > strend - strpos) {
619 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
620 "String too short... [re_intuit_start]\n"));
624 /* XXX we need to pass strbeg as a separate arg: the following is
625 * guesswork and can be wrong... */
626 if (sv && SvPOK(sv)) {
627 char * p = SvPVX(sv);
628 STRLEN cur = SvCUR(sv);
629 if (p <= strpos && strpos < p + cur) {
631 assert(p <= strend && strend <= p + cur);
634 strbeg = strend - cur;
641 if (!prog->check_utf8 && prog->check_substr)
642 to_utf8_substr(prog);
643 check = prog->check_utf8;
645 if (!prog->check_substr && prog->check_utf8) {
646 if (! to_byte_substr(prog)) {
647 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
650 check = prog->check_substr;
652 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
653 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
654 || ( (prog->extflags & RXf_ANCH_BOL)
655 && !multiline ) ); /* Check after \n? */
658 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
659 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
660 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
662 && (strpos != strbeg)) {
663 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
666 if (prog->check_offset_min == prog->check_offset_max
667 && !(prog->extflags & RXf_CANY_SEEN)
668 && ! multiline) /* /m can cause \n's to match that aren't
669 accounted for in the string max length.
670 See [perl #115242] */
672 /* Substring at constant offset from beg-of-str... */
675 s = HOP3c(strpos, prog->check_offset_min, strend);
678 slen = SvCUR(check); /* >= 1 */
680 if ( strend - s > slen || strend - s < slen - 1
681 || (strend - s == slen && strend[-1] != '\n')) {
682 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
685 /* Now should match s[0..slen-2] */
687 if (slen && (*SvPVX_const(check) != *s
689 && memNE(SvPVX_const(check), s, slen)))) {
691 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
695 else if (*SvPVX_const(check) != *s
696 || ((slen = SvCUR(check)) > 1
697 && memNE(SvPVX_const(check), s, slen)))
700 goto success_at_start;
703 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
705 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
706 end_shift = prog->check_end_shift;
709 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
710 - (SvTAIL(check) != 0);
711 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
713 if (end_shift < eshift)
717 else { /* Can match at random position */
720 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
721 end_shift = prog->check_end_shift;
723 /* end shift should be non negative here */
726 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
728 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
729 (IV)end_shift, RX_PRECOMP(prog));
733 /* Find a possible match in the region s..strend by looking for
734 the "check" substring in the region corrected by start/end_shift. */
737 I32 srch_start_shift = start_shift;
738 I32 srch_end_shift = end_shift;
741 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
742 srch_end_shift -= ((strbeg - s) - srch_start_shift);
743 srch_start_shift = strbeg - s;
745 DEBUG_OPTIMISE_MORE_r({
746 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
747 (IV)prog->check_offset_min,
748 (IV)srch_start_shift,
750 (IV)prog->check_end_shift);
753 if (prog->extflags & RXf_CANY_SEEN) {
754 start_point= (U8*)(s + srch_start_shift);
755 end_point= (U8*)(strend - srch_end_shift);
757 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
758 end_point= HOP3(strend, -srch_end_shift, strbeg);
760 DEBUG_OPTIMISE_MORE_r({
761 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
762 (int)(end_point - start_point),
763 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
767 s = fbm_instr( start_point, end_point,
768 check, multiline ? FBMrf_MULTILINE : 0);
770 /* Update the count-of-usability, remove useless subpatterns,
774 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
775 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
776 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
777 (s ? "Found" : "Did not find"),
778 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
779 ? "anchored" : "floating"),
782 (s ? " at offset " : "...\n") );
787 /* Finish the diagnostic message */
788 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
790 /* XXX dmq: first branch is for positive lookbehind...
791 Our check string is offset from the beginning of the pattern.
792 So we need to do any stclass tests offset forward from that
801 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
802 Start with the other substr.
803 XXXX no SCREAM optimization yet - and a very coarse implementation
804 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
805 *always* match. Probably should be marked during compile...
806 Probably it is right to do no SCREAM here...
809 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
810 : (prog->float_substr && prog->anchored_substr))
812 /* Take into account the "other" substring. */
813 /* XXXX May be hopelessly wrong for UTF... */
816 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
819 char * const last = HOP3c(s, -start_shift, strbeg);
821 char * const saved_s = s;
824 t = s - prog->check_offset_max;
825 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
827 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
832 t = HOP3c(t, prog->anchored_offset, strend);
833 if (t < other_last) /* These positions already checked */
835 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
838 /* XXXX It is not documented what units *_offsets are in.
839 We assume bytes, but this is clearly wrong.
840 Meaning this code needs to be carefully reviewed for errors.
844 /* On end-of-str: see comment below. */
845 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
846 if (must == &PL_sv_undef) {
848 DEBUG_r(must = prog->anchored_utf8); /* for debug */
853 HOP3(HOP3(last1, prog->anchored_offset, strend)
854 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
856 multiline ? FBMrf_MULTILINE : 0
859 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
860 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
861 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
862 (s ? "Found" : "Contradicts"),
863 quoted, RE_SV_TAIL(must));
868 if (last1 >= last2) {
869 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
870 ", giving up...\n"));
873 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
874 ", trying floating at offset %ld...\n",
875 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
876 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
877 s = HOP3c(last, 1, strend);
881 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
882 (long)(s - i_strpos)));
883 t = HOP3c(s, -prog->anchored_offset, strbeg);
884 other_last = HOP3c(s, 1, strend);
892 else { /* Take into account the floating substring. */
894 char * const saved_s = s;
897 t = HOP3c(s, -start_shift, strbeg);
899 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
900 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
901 last = HOP3c(t, prog->float_max_offset, strend);
902 s = HOP3c(t, prog->float_min_offset, strend);
905 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
906 must = utf8_target ? prog->float_utf8 : prog->float_substr;
907 /* fbm_instr() takes into account exact value of end-of-str
908 if the check is SvTAIL(ed). Since false positives are OK,
909 and end-of-str is not later than strend we are OK. */
910 if (must == &PL_sv_undef) {
912 DEBUG_r(must = prog->float_utf8); /* for debug message */
915 s = fbm_instr((unsigned char*)s,
916 (unsigned char*)last + SvCUR(must)
918 must, multiline ? FBMrf_MULTILINE : 0);
920 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
921 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
922 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
923 (s ? "Found" : "Contradicts"),
924 quoted, RE_SV_TAIL(must));
928 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
929 ", giving up...\n"));
932 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
933 ", trying anchored starting at offset %ld...\n",
934 (long)(saved_s + 1 - i_strpos)));
936 s = HOP3c(t, 1, strend);
940 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
941 (long)(s - i_strpos)));
942 other_last = s; /* Fix this later. --Hugo */
952 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
954 DEBUG_OPTIMISE_MORE_r(
955 PerlIO_printf(Perl_debug_log,
956 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
957 (IV)prog->check_offset_min,
958 (IV)prog->check_offset_max,
966 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
968 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
971 /* Fixed substring is found far enough so that the match
972 cannot start at strpos. */
974 if (ml_anch && t[-1] != '\n') {
975 /* Eventually fbm_*() should handle this, but often
976 anchored_offset is not 0, so this check will not be wasted. */
977 /* XXXX In the code below we prefer to look for "^" even in
978 presence of anchored substrings. And we search even
979 beyond the found float position. These pessimizations
980 are historical artefacts only. */
982 while (t < strend - prog->minlen) {
984 if (t < check_at - prog->check_offset_min) {
985 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
986 /* Since we moved from the found position,
987 we definitely contradict the found anchored
988 substr. Due to the above check we do not
989 contradict "check" substr.
990 Thus we can arrive here only if check substr
991 is float. Redo checking for "other"=="fixed".
994 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
995 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
996 goto do_other_anchored;
998 /* We don't contradict the found floating substring. */
999 /* XXXX Why not check for STCLASS? */
1001 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1002 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1005 /* Position contradicts check-string */
1006 /* XXXX probably better to look for check-string
1007 than for "\n", so one should lower the limit for t? */
1008 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1009 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1010 other_last = strpos = s = t + 1;
1015 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1016 PL_colors[0], PL_colors[1]));
1020 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1021 PL_colors[0], PL_colors[1]));
1025 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1028 /* The found string does not prohibit matching at strpos,
1029 - no optimization of calling REx engine can be performed,
1030 unless it was an MBOL and we are not after MBOL,
1031 or a future STCLASS check will fail this. */
1033 /* Even in this situation we may use MBOL flag if strpos is offset
1034 wrt the start of the string. */
1035 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1036 && (strpos != strbeg) && strpos[-1] != '\n'
1037 /* May be due to an implicit anchor of m{.*foo} */
1038 && !(prog->intflags & PREGf_IMPLICIT))
1043 DEBUG_EXECUTE_r( if (ml_anch)
1044 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1045 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1048 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1050 prog->check_utf8 /* Could be deleted already */
1051 && --BmUSEFUL(prog->check_utf8) < 0
1052 && (prog->check_utf8 == prog->float_utf8)
1054 prog->check_substr /* Could be deleted already */
1055 && --BmUSEFUL(prog->check_substr) < 0
1056 && (prog->check_substr == prog->float_substr)
1059 /* If flags & SOMETHING - do not do it many times on the same match */
1060 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1061 /* XXX Does the destruction order has to change with utf8_target? */
1062 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1063 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1064 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1065 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1066 check = NULL; /* abort */
1068 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1069 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1070 if (prog->intflags & PREGf_IMPLICIT)
1071 prog->extflags &= ~RXf_ANCH_MBOL;
1072 /* XXXX This is a remnant of the old implementation. It
1073 looks wasteful, since now INTUIT can use many
1074 other heuristics. */
1075 prog->extflags &= ~RXf_USE_INTUIT;
1076 /* XXXX What other flags might need to be cleared in this branch? */
1082 /* Last resort... */
1083 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1084 /* trie stclasses are too expensive to use here, we are better off to
1085 leave it to regmatch itself */
1086 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1087 /* minlen == 0 is possible if regstclass is \b or \B,
1088 and the fixed substr is ''$.
1089 Since minlen is already taken into account, s+1 is before strend;
1090 accidentally, minlen >= 1 guaranties no false positives at s + 1
1091 even for \b or \B. But (minlen? 1 : 0) below assumes that
1092 regstclass does not come from lookahead... */
1093 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1094 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1095 const U8* const str = (U8*)STRING(progi->regstclass);
1096 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1097 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1100 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1101 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1102 else if (prog->float_substr || prog->float_utf8)
1103 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1107 if (checked_upto < s)
1109 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1110 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1113 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1118 const char *what = NULL;
1120 if (endpos == strend) {
1121 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1122 "Could not match STCLASS...\n") );
1125 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1126 "This position contradicts STCLASS...\n") );
1127 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1129 checked_upto = HOPBACKc(endpos, start_shift);
1130 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1131 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1132 /* Contradict one of substrings */
1133 if (prog->anchored_substr || prog->anchored_utf8) {
1134 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1135 DEBUG_EXECUTE_r( what = "anchored" );
1137 s = HOP3c(t, 1, strend);
1138 if (s + start_shift + end_shift > strend) {
1139 /* XXXX Should be taken into account earlier? */
1140 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1141 "Could not match STCLASS...\n") );
1146 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1147 "Looking for %s substr starting at offset %ld...\n",
1148 what, (long)(s + start_shift - i_strpos)) );
1151 /* Have both, check_string is floating */
1152 if (t + start_shift >= check_at) /* Contradicts floating=check */
1153 goto retry_floating_check;
1154 /* Recheck anchored substring, but not floating... */
1158 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1159 "Looking for anchored substr starting at offset %ld...\n",
1160 (long)(other_last - i_strpos)) );
1161 goto do_other_anchored;
1163 /* Another way we could have checked stclass at the
1164 current position only: */
1169 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1170 "Looking for /%s^%s/m starting at offset %ld...\n",
1171 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1174 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1176 /* Check is floating substring. */
1177 retry_floating_check:
1178 t = check_at - start_shift;
1179 DEBUG_EXECUTE_r( what = "floating" );
1180 goto hop_and_restart;
1183 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1184 "By STCLASS: moving %ld --> %ld\n",
1185 (long)(t - i_strpos), (long)(s - i_strpos))
1189 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1190 "Does not contradict STCLASS...\n");
1195 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1196 PL_colors[4], (check ? "Guessed" : "Giving up"),
1197 PL_colors[5], (long)(s - i_strpos)) );
1200 fail_finish: /* Substring not found */
1201 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1202 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1204 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1205 PL_colors[4], PL_colors[5]));
1209 #define DECL_TRIE_TYPE(scan) \
1210 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1211 trie_type = ((scan->flags == EXACT) \
1212 ? (utf8_target ? trie_utf8 : trie_plain) \
1213 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1215 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1216 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1218 switch (trie_type) { \
1219 case trie_utf8_fold: \
1220 if ( foldlen>0 ) { \
1221 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1226 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1227 len = UTF8SKIP(uc); \
1228 skiplen = UNISKIP( uvc ); \
1229 foldlen -= skiplen; \
1230 uscan = foldbuf + skiplen; \
1233 case trie_latin_utf8_fold: \
1234 if ( foldlen>0 ) { \
1235 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1241 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1242 skiplen = UNISKIP( uvc ); \
1243 foldlen -= skiplen; \
1244 uscan = foldbuf + skiplen; \
1248 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1255 charid = trie->charmap[ uvc ]; \
1259 if (widecharmap) { \
1260 SV** const svpp = hv_fetch(widecharmap, \
1261 (char*)&uvc, sizeof(UV), 0); \
1263 charid = (U16)SvIV(*svpp); \
1268 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1272 && (ln == 1 || folder(s, pat_string, ln)) \
1273 && (!reginfo || regtry(reginfo, &s)) ) \
1279 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1281 while (s < strend && s + (uskip = UTF8SKIP(s)) <= strend) { \
1287 #define REXEC_FBC_SCAN(CoDe) \
1289 while (s < strend) { \
1295 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1296 REXEC_FBC_UTF8_SCAN( \
1298 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1307 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1310 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1319 #define REXEC_FBC_TRYIT \
1320 if ((!reginfo || regtry(reginfo, &s))) \
1323 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1324 if (utf8_target) { \
1325 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1328 REXEC_FBC_CLASS_SCAN(CoNd); \
1331 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1332 if (utf8_target) { \
1334 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1337 REXEC_FBC_CLASS_SCAN(CoNd); \
1340 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1341 PL_reg_flags |= RF_tainted; \
1342 if (utf8_target) { \
1343 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1346 REXEC_FBC_CLASS_SCAN(CoNd); \
1349 #define DUMP_EXEC_POS(li,s,doutf8) \
1350 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1353 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1354 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1355 tmp = TEST_NON_UTF8(tmp); \
1356 REXEC_FBC_UTF8_SCAN( \
1357 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1366 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1367 if (s == PL_bostr) { \
1371 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1372 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1375 LOAD_UTF8_CHARCLASS_ALNUM(); \
1376 REXEC_FBC_UTF8_SCAN( \
1377 if (tmp == ! (TeSt2_UtF8)) { \
1386 /* The only difference between the BOUND and NBOUND cases is that
1387 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1388 * NBOUND. This is accomplished by passing it in either the if or else clause,
1389 * with the other one being empty */
1390 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1391 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1393 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1394 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1396 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1397 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1399 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1400 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1403 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1404 * be passed in completely with the variable name being tested, which isn't
1405 * such a clean interface, but this is easier to read than it was before. We
1406 * are looking for the boundary (or non-boundary between a word and non-word
1407 * character. The utf8 and non-utf8 cases have the same logic, but the details
1408 * must be different. Find the "wordness" of the character just prior to this
1409 * one, and compare it with the wordness of this one. If they differ, we have
1410 * a boundary. At the beginning of the string, pretend that the previous
1411 * character was a new-line */
1412 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1413 if (utf8_target) { \
1416 else { /* Not utf8 */ \
1417 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1418 tmp = TEST_NON_UTF8(tmp); \
1420 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1429 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1432 /* We know what class REx starts with. Try to find this position... */
1433 /* if reginfo is NULL, its a dryrun */
1434 /* annoyingly all the vars in this routine have different names from their counterparts
1435 in regmatch. /grrr */
1438 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1439 const char *strend, regmatch_info *reginfo)
1442 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1443 char *pat_string; /* The pattern's exactish string */
1444 char *pat_end; /* ptr to end char of pat_string */
1445 re_fold_t folder; /* Function for computing non-utf8 folds */
1446 const U8 *fold_array; /* array for folding ords < 256 */
1453 I32 tmp = 1; /* Scratch variable? */
1454 const bool utf8_target = PL_reg_match_utf8;
1455 UV utf8_fold_flags = 0;
1456 RXi_GET_DECL(prog,progi);
1458 PERL_ARGS_ASSERT_FIND_BYCLASS;
1460 /* We know what class it must start with. */
1464 REXEC_FBC_UTF8_CLASS_SCAN(
1465 reginclass(prog, c, (U8*)s, utf8_target));
1468 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1473 if (tmp && (!reginfo || regtry(reginfo, &s)))
1481 if (UTF_PATTERN || utf8_target) {
1482 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1483 goto do_exactf_utf8;
1485 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1486 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1487 goto do_exactf_non_utf8; /* isn't dealt with by these */
1492 /* regcomp.c already folded this if pattern is in UTF-8 */
1493 utf8_fold_flags = 0;
1494 goto do_exactf_utf8;
1496 fold_array = PL_fold;
1498 goto do_exactf_non_utf8;
1501 if (UTF_PATTERN || utf8_target) {
1502 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1503 goto do_exactf_utf8;
1505 fold_array = PL_fold_locale;
1506 folder = foldEQ_locale;
1507 goto do_exactf_non_utf8;
1511 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1513 goto do_exactf_utf8;
1515 case EXACTFU_TRICKYFOLD:
1517 if (UTF_PATTERN || utf8_target) {
1518 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1519 goto do_exactf_utf8;
1522 /* Any 'ss' in the pattern should have been replaced by regcomp,
1523 * so we don't have to worry here about this single special case
1524 * in the Latin1 range */
1525 fold_array = PL_fold_latin1;
1526 folder = foldEQ_latin1;
1530 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1531 are no glitches with fold-length differences
1532 between the target string and pattern */
1534 /* The idea in the non-utf8 EXACTF* cases is to first find the
1535 * first character of the EXACTF* node and then, if necessary,
1536 * case-insensitively compare the full text of the node. c1 is the
1537 * first character. c2 is its fold. This logic will not work for
1538 * Unicode semantics and the german sharp ss, which hence should
1539 * not be compiled into a node that gets here. */
1540 pat_string = STRING(c);
1541 ln = STR_LEN(c); /* length to match in octets/bytes */
1543 /* We know that we have to match at least 'ln' bytes (which is the
1544 * same as characters, since not utf8). If we have to match 3
1545 * characters, and there are only 2 availabe, we know without
1546 * trying that it will fail; so don't start a match past the
1547 * required minimum number from the far end */
1548 e = HOP3c(strend, -((I32)ln), s);
1550 if (!reginfo && e < s) {
1551 e = s; /* Due to minlen logic of intuit() */
1555 c2 = fold_array[c1];
1556 if (c1 == c2) { /* If char and fold are the same */
1557 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1560 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1569 /* If one of the operands is in utf8, we can't use the simpler
1570 * folding above, due to the fact that many different characters
1571 * can have the same fold, or portion of a fold, or different-
1573 pat_string = STRING(c);
1574 ln = STR_LEN(c); /* length to match in octets/bytes */
1575 pat_end = pat_string + ln;
1576 lnc = (UTF_PATTERN) /* length to match in characters */
1577 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1580 /* We have 'lnc' characters to match in the pattern, but because of
1581 * multi-character folding, each character in the target can match
1582 * up to 3 characters (Unicode guarantees it will never exceed
1583 * this) if it is utf8-encoded; and up to 2 if not (based on the
1584 * fact that the Latin 1 folds are already determined, and the
1585 * only multi-char fold in that range is the sharp-s folding to
1586 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1587 * string character. Adjust lnc accordingly, rounding up, so that
1588 * if we need to match at least 4+1/3 chars, that really is 5. */
1589 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1590 lnc = (lnc + expansion - 1) / expansion;
1592 /* As in the non-UTF8 case, if we have to match 3 characters, and
1593 * only 2 are left, it's guaranteed to fail, so don't start a
1594 * match that would require us to go beyond the end of the string
1596 e = HOP3c(strend, -((I32)lnc), s);
1598 if (!reginfo && e < s) {
1599 e = s; /* Due to minlen logic of intuit() */
1602 /* XXX Note that we could recalculate e to stop the loop earlier,
1603 * as the worst case expansion above will rarely be met, and as we
1604 * go along we would usually find that e moves further to the left.
1605 * This would happen only after we reached the point in the loop
1606 * where if there were no expansion we should fail. Unclear if
1607 * worth the expense */
1610 char *my_strend= (char *)strend;
1611 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1612 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1613 && (!reginfo || regtry(reginfo, &s)) )
1617 s += (utf8_target) ? UTF8SKIP(s) : 1;
1622 PL_reg_flags |= RF_tainted;
1623 FBC_BOUND(isALNUM_LC,
1624 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1625 isALNUM_LC_utf8((U8*)s));
1628 PL_reg_flags |= RF_tainted;
1629 FBC_NBOUND(isALNUM_LC,
1630 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1631 isALNUM_LC_utf8((U8*)s));
1634 FBC_BOUND(isWORDCHAR,
1636 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1639 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1641 isWORDCHAR_A((U8*)s));
1644 FBC_NBOUND(isWORDCHAR,
1646 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1649 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1651 isWORDCHAR_A((U8*)s));
1654 FBC_BOUND(isWORDCHAR_L1,
1656 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1659 FBC_NBOUND(isWORDCHAR_L1,
1661 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1664 REXEC_FBC_CSCAN_TAINT(
1665 isALNUM_LC_utf8((U8*)s),
1670 REXEC_FBC_CSCAN_PRELOAD(
1671 LOAD_UTF8_CHARCLASS_ALNUM(),
1672 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1673 isWORDCHAR_L1((U8) *s)
1677 REXEC_FBC_CSCAN_PRELOAD(
1678 LOAD_UTF8_CHARCLASS_ALNUM(),
1679 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1684 /* Don't need to worry about utf8, as it can match only a single
1685 * byte invariant character */
1686 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1689 REXEC_FBC_CSCAN_PRELOAD(
1690 LOAD_UTF8_CHARCLASS_ALNUM(),
1691 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1692 ! isWORDCHAR_L1((U8) *s)
1696 REXEC_FBC_CSCAN_PRELOAD(
1697 LOAD_UTF8_CHARCLASS_ALNUM(),
1698 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1709 REXEC_FBC_CSCAN_TAINT(
1710 !isALNUM_LC_utf8((U8*)s),
1716 is_XPERLSPACE_utf8(s),
1722 is_XPERLSPACE_utf8(s),
1727 /* Don't need to worry about utf8, as it can match only a single
1728 * byte invariant character */
1729 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1732 REXEC_FBC_CSCAN_TAINT(
1733 isSPACE_LC_utf8((U8*)s),
1739 ! is_XPERLSPACE_utf8(s),
1740 ! isSPACE_L1((U8) *s)
1745 ! is_XPERLSPACE_utf8(s),
1756 REXEC_FBC_CSCAN_TAINT(
1757 !isSPACE_LC_utf8((U8*)s),
1762 REXEC_FBC_CSCAN_PRELOAD(
1763 LOAD_UTF8_CHARCLASS_DIGIT(),
1764 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1769 /* Don't need to worry about utf8, as it can match only a single
1770 * byte invariant character */
1771 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1774 REXEC_FBC_CSCAN_TAINT(
1775 isDIGIT_LC_utf8((U8*)s),
1780 REXEC_FBC_CSCAN_PRELOAD(
1781 LOAD_UTF8_CHARCLASS_DIGIT(),
1782 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1793 REXEC_FBC_CSCAN_TAINT(
1794 !isDIGIT_LC_utf8((U8*)s),
1800 is_LNBREAK_utf8_safe(s, strend),
1801 is_LNBREAK_latin1_safe(s, strend)
1806 is_VERTWS_utf8_safe(s, strend),
1807 is_VERTWS_latin1_safe(s, strend)
1812 !is_VERTWS_utf8_safe(s, strend),
1813 !is_VERTWS_latin1_safe(s, strend)
1818 is_HORIZWS_utf8_safe(s, strend),
1819 is_HORIZWS_latin1_safe(s, strend)
1824 !is_HORIZWS_utf8_safe(s, strend),
1825 !is_HORIZWS_latin1_safe(s, strend)
1829 /* Don't need to worry about utf8, as it can match only a single
1830 * byte invariant character. The flag in this node type is the
1831 * class number to pass to _generic_isCC() to build a mask for
1832 * searching in PL_charclass[] */
1833 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1837 !_generic_isCC_A(*s, FLAGS(c)),
1838 !_generic_isCC_A(*s, FLAGS(c))
1846 /* what trie are we using right now */
1848 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1850 = (reg_trie_data*)progi->data->data[ aho->trie ];
1851 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1853 const char *last_start = strend - trie->minlen;
1855 const char *real_start = s;
1857 STRLEN maxlen = trie->maxlen;
1859 U8 **points; /* map of where we were in the input string
1860 when reading a given char. For ASCII this
1861 is unnecessary overhead as the relationship
1862 is always 1:1, but for Unicode, especially
1863 case folded Unicode this is not true. */
1864 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1868 GET_RE_DEBUG_FLAGS_DECL;
1870 /* We can't just allocate points here. We need to wrap it in
1871 * an SV so it gets freed properly if there is a croak while
1872 * running the match */
1875 sv_points=newSV(maxlen * sizeof(U8 *));
1876 SvCUR_set(sv_points,
1877 maxlen * sizeof(U8 *));
1878 SvPOK_on(sv_points);
1879 sv_2mortal(sv_points);
1880 points=(U8**)SvPV_nolen(sv_points );
1881 if ( trie_type != trie_utf8_fold
1882 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1885 bitmap=(U8*)trie->bitmap;
1887 bitmap=(U8*)ANYOF_BITMAP(c);
1889 /* this is the Aho-Corasick algorithm modified a touch
1890 to include special handling for long "unknown char"
1891 sequences. The basic idea being that we use AC as long
1892 as we are dealing with a possible matching char, when
1893 we encounter an unknown char (and we have not encountered
1894 an accepting state) we scan forward until we find a legal
1896 AC matching is basically that of trie matching, except
1897 that when we encounter a failing transition, we fall back
1898 to the current states "fail state", and try the current char
1899 again, a process we repeat until we reach the root state,
1900 state 1, or a legal transition. If we fail on the root state
1901 then we can either terminate if we have reached an accepting
1902 state previously, or restart the entire process from the beginning
1906 while (s <= last_start) {
1907 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1915 U8 *uscan = (U8*)NULL;
1916 U8 *leftmost = NULL;
1918 U32 accepted_word= 0;
1922 while ( state && uc <= (U8*)strend ) {
1924 U32 word = aho->states[ state ].wordnum;
1928 DEBUG_TRIE_EXECUTE_r(
1929 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1930 dump_exec_pos( (char *)uc, c, strend, real_start,
1931 (char *)uc, utf8_target );
1932 PerlIO_printf( Perl_debug_log,
1933 " Scanning for legal start char...\n");
1937 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1941 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1947 if (uc >(U8*)last_start) break;
1951 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1952 if (!leftmost || lpos < leftmost) {
1953 DEBUG_r(accepted_word=word);
1959 points[pointpos++ % maxlen]= uc;
1960 if (foldlen || uc < (U8*)strend) {
1961 REXEC_TRIE_READ_CHAR(trie_type, trie,
1963 uscan, len, uvc, charid, foldlen,
1965 DEBUG_TRIE_EXECUTE_r({
1966 dump_exec_pos( (char *)uc, c, strend,
1967 real_start, s, utf8_target);
1968 PerlIO_printf(Perl_debug_log,
1969 " Charid:%3u CP:%4"UVxf" ",
1981 word = aho->states[ state ].wordnum;
1983 base = aho->states[ state ].trans.base;
1985 DEBUG_TRIE_EXECUTE_r({
1987 dump_exec_pos( (char *)uc, c, strend, real_start,
1989 PerlIO_printf( Perl_debug_log,
1990 "%sState: %4"UVxf", word=%"UVxf,
1991 failed ? " Fail transition to " : "",
1992 (UV)state, (UV)word);
1998 ( ((offset = base + charid
1999 - 1 - trie->uniquecharcount)) >= 0)
2000 && ((U32)offset < trie->lasttrans)
2001 && trie->trans[offset].check == state
2002 && (tmp=trie->trans[offset].next))
2004 DEBUG_TRIE_EXECUTE_r(
2005 PerlIO_printf( Perl_debug_log," - legal\n"));
2010 DEBUG_TRIE_EXECUTE_r(
2011 PerlIO_printf( Perl_debug_log," - fail\n"));
2013 state = aho->fail[state];
2017 /* we must be accepting here */
2018 DEBUG_TRIE_EXECUTE_r(
2019 PerlIO_printf( Perl_debug_log," - accepting\n"));
2028 if (!state) state = 1;
2031 if ( aho->states[ state ].wordnum ) {
2032 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2033 if (!leftmost || lpos < leftmost) {
2034 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2039 s = (char*)leftmost;
2040 DEBUG_TRIE_EXECUTE_r({
2042 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2043 (UV)accepted_word, (IV)(s - real_start)
2046 if (!reginfo || regtry(reginfo, &s)) {
2052 DEBUG_TRIE_EXECUTE_r({
2053 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2056 DEBUG_TRIE_EXECUTE_r(
2057 PerlIO_printf( Perl_debug_log,"No match.\n"));
2066 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2076 - regexec_flags - match a regexp against a string
2079 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
2080 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2081 /* stringarg: the point in the string at which to begin matching */
2082 /* strend: pointer to null at end of string */
2083 /* strbeg: real beginning of string */
2084 /* minend: end of match must be >= minend bytes after stringarg. */
2085 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2086 * itself is accessed via the pointers above */
2087 /* data: May be used for some additional optimizations.
2088 Currently its only used, with a U32 cast, for transmitting
2089 the ganch offset when doing a /g match. This will change */
2090 /* nosave: For optimizations. */
2094 struct regexp *const prog = ReANY(rx);
2095 /*register*/ char *s;
2097 /*register*/ char *startpos = stringarg;
2098 I32 minlen; /* must match at least this many chars */
2099 I32 dontbother = 0; /* how many characters not to try at end */
2100 I32 end_shift = 0; /* Same for the end. */ /* CC */
2101 I32 scream_pos = -1; /* Internal iterator of scream. */
2102 char *scream_olds = NULL;
2103 const bool utf8_target = cBOOL(DO_UTF8(sv));
2105 RXi_GET_DECL(prog,progi);
2106 regmatch_info reginfo; /* create some info to pass to regtry etc */
2107 regexp_paren_pair *swap = NULL;
2108 GET_RE_DEBUG_FLAGS_DECL;
2110 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2111 PERL_UNUSED_ARG(data);
2113 /* Be paranoid... */
2114 if (prog == NULL || startpos == NULL) {
2115 Perl_croak(aTHX_ "NULL regexp parameter");
2119 multiline = prog->extflags & RXf_PMf_MULTILINE;
2120 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2122 RX_MATCH_UTF8_set(rx, utf8_target);
2124 debug_start_match(rx, utf8_target, startpos, strend,
2128 minlen = prog->minlen;
2130 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2131 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2132 "String too short [regexec_flags]...\n"));
2137 /* Check validity of program. */
2138 if (UCHARAT(progi->program) != REG_MAGIC) {
2139 Perl_croak(aTHX_ "corrupted regexp program");
2143 PL_reg_state.re_state_eval_setup_done = FALSE;
2147 PL_reg_flags |= RF_utf8;
2149 /* Mark beginning of line for ^ and lookbehind. */
2150 reginfo.bol = startpos; /* XXX not used ??? */
2154 /* Mark end of line for $ (and such) */
2157 /* see how far we have to get to not match where we matched before */
2158 reginfo.till = startpos+minend;
2160 /* If there is a "must appear" string, look for it. */
2163 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2165 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2166 reginfo.ganch = startpos + prog->gofs;
2167 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2168 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2169 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2171 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2172 && mg->mg_len >= 0) {
2173 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2174 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2175 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2177 if (prog->extflags & RXf_ANCH_GPOS) {
2178 if (s > reginfo.ganch)
2180 s = reginfo.ganch - prog->gofs;
2181 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2182 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2188 reginfo.ganch = strbeg + PTR2UV(data);
2189 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2190 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2192 } else { /* pos() not defined */
2193 reginfo.ganch = strbeg;
2194 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2195 "GPOS: reginfo.ganch = strbeg\n"));
2198 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2199 /* We have to be careful. If the previous successful match
2200 was from this regex we don't want a subsequent partially
2201 successful match to clobber the old results.
2202 So when we detect this possibility we add a swap buffer
2203 to the re, and switch the buffer each match. If we fail
2204 we switch it back, otherwise we leave it swapped.
2207 /* do we need a save destructor here for eval dies? */
2208 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2209 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2210 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2216 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2217 re_scream_pos_data d;
2219 d.scream_olds = &scream_olds;
2220 d.scream_pos = &scream_pos;
2221 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2223 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2224 goto phooey; /* not present */
2230 /* Simplest case: anchored match need be tried only once. */
2231 /* [unless only anchor is BOL and multiline is set] */
2232 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2233 if (s == startpos && regtry(®info, &startpos))
2235 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2236 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2241 dontbother = minlen - 1;
2242 end = HOP3c(strend, -dontbother, strbeg) - 1;
2243 /* for multiline we only have to try after newlines */
2244 if (prog->check_substr || prog->check_utf8) {
2245 /* because of the goto we can not easily reuse the macros for bifurcating the
2246 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2249 goto after_try_utf8;
2251 if (regtry(®info, &s)) {
2258 if (prog->extflags & RXf_USE_INTUIT) {
2259 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2268 } /* end search for check string in unicode */
2270 if (s == startpos) {
2271 goto after_try_latin;
2274 if (regtry(®info, &s)) {
2281 if (prog->extflags & RXf_USE_INTUIT) {
2282 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2291 } /* end search for check string in latin*/
2292 } /* end search for check string */
2293 else { /* search for newline */
2295 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2298 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2299 while (s <= end) { /* note it could be possible to match at the end of the string */
2300 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2301 if (regtry(®info, &s))
2305 } /* end search for newline */
2306 } /* end anchored/multiline check string search */
2308 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2310 /* the warning about reginfo.ganch being used without initialization
2311 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2312 and we only enter this block when the same bit is set. */
2313 char *tmp_s = reginfo.ganch - prog->gofs;
2315 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2320 /* Messy cases: unanchored match. */
2321 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2322 /* we have /x+whatever/ */
2323 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2329 if (! prog->anchored_utf8) {
2330 to_utf8_substr(prog);
2332 ch = SvPVX_const(prog->anchored_utf8)[0];
2335 DEBUG_EXECUTE_r( did_match = 1 );
2336 if (regtry(®info, &s)) goto got_it;
2338 while (s < strend && *s == ch)
2345 if (! prog->anchored_substr) {
2346 if (! to_byte_substr(prog)) {
2347 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2350 ch = SvPVX_const(prog->anchored_substr)[0];
2353 DEBUG_EXECUTE_r( did_match = 1 );
2354 if (regtry(®info, &s)) goto got_it;
2356 while (s < strend && *s == ch)
2361 DEBUG_EXECUTE_r(if (!did_match)
2362 PerlIO_printf(Perl_debug_log,
2363 "Did not find anchored character...\n")
2366 else if (prog->anchored_substr != NULL
2367 || prog->anchored_utf8 != NULL
2368 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2369 && prog->float_max_offset < strend - s)) {
2374 char *last1; /* Last position checked before */
2378 if (prog->anchored_substr || prog->anchored_utf8) {
2380 if (! prog->anchored_utf8) {
2381 to_utf8_substr(prog);
2383 must = prog->anchored_utf8;
2386 if (! prog->anchored_substr) {
2387 if (! to_byte_substr(prog)) {
2388 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2391 must = prog->anchored_substr;
2393 back_max = back_min = prog->anchored_offset;
2396 if (! prog->float_utf8) {
2397 to_utf8_substr(prog);
2399 must = prog->float_utf8;
2402 if (! prog->float_substr) {
2403 if (! to_byte_substr(prog)) {
2404 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2407 must = prog->float_substr;
2409 back_max = prog->float_max_offset;
2410 back_min = prog->float_min_offset;
2416 last = HOP3c(strend, /* Cannot start after this */
2417 -(I32)(CHR_SVLEN(must)
2418 - (SvTAIL(must) != 0) + back_min), strbeg);
2421 last1 = HOPc(s, -1);
2423 last1 = s - 1; /* bogus */
2425 /* XXXX check_substr already used to find "s", can optimize if
2426 check_substr==must. */
2428 dontbother = end_shift;
2429 strend = HOPc(strend, -dontbother);
2430 while ( (s <= last) &&
2431 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2432 (unsigned char*)strend, must,
2433 multiline ? FBMrf_MULTILINE : 0)) ) {
2434 DEBUG_EXECUTE_r( did_match = 1 );
2435 if (HOPc(s, -back_max) > last1) {
2436 last1 = HOPc(s, -back_min);
2437 s = HOPc(s, -back_max);
2440 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2442 last1 = HOPc(s, -back_min);
2446 while (s <= last1) {
2447 if (regtry(®info, &s))
2450 s++; /* to break out of outer loop */
2457 while (s <= last1) {
2458 if (regtry(®info, &s))
2464 DEBUG_EXECUTE_r(if (!did_match) {
2465 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2466 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2467 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2468 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2469 ? "anchored" : "floating"),
2470 quoted, RE_SV_TAIL(must));
2474 else if ( (c = progi->regstclass) ) {
2476 const OPCODE op = OP(progi->regstclass);
2477 /* don't bother with what can't match */
2478 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2479 strend = HOPc(strend, -(minlen - 1));
2482 SV * const prop = sv_newmortal();
2483 regprop(prog, prop, c);
2485 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2487 PerlIO_printf(Perl_debug_log,
2488 "Matching stclass %.*s against %s (%d bytes)\n",
2489 (int)SvCUR(prop), SvPVX_const(prop),
2490 quoted, (int)(strend - s));
2493 if (find_byclass(prog, c, s, strend, ®info))
2495 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2499 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2507 if (! prog->float_utf8) {
2508 to_utf8_substr(prog);
2510 float_real = prog->float_utf8;
2513 if (! prog->float_substr) {
2514 if (! to_byte_substr(prog)) {
2515 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2518 float_real = prog->float_substr;
2521 little = SvPV_const(float_real, len);
2522 if (SvTAIL(float_real)) {
2523 /* This means that float_real contains an artificial \n on
2524 * the end due to the presence of something like this:
2525 * /foo$/ where we can match both "foo" and "foo\n" at the
2526 * end of the string. So we have to compare the end of the
2527 * string first against the float_real without the \n and
2528 * then against the full float_real with the string. We
2529 * have to watch out for cases where the string might be
2530 * smaller than the float_real or the float_real without
2532 char *checkpos= strend - len;
2534 PerlIO_printf(Perl_debug_log,
2535 "%sChecking for float_real.%s\n",
2536 PL_colors[4], PL_colors[5]));
2537 if (checkpos + 1 < strbeg) {
2538 /* can't match, even if we remove the trailing \n
2539 * string is too short to match */
2541 PerlIO_printf(Perl_debug_log,
2542 "%sString shorter than required trailing substring, cannot match.%s\n",
2543 PL_colors[4], PL_colors[5]));
2545 } else if (memEQ(checkpos + 1, little, len - 1)) {
2546 /* can match, the end of the string matches without the
2548 last = checkpos + 1;
2549 } else if (checkpos < strbeg) {
2550 /* cant match, string is too short when the "\n" is
2553 PerlIO_printf(Perl_debug_log,
2554 "%sString does not contain required trailing substring, cannot match.%s\n",
2555 PL_colors[4], PL_colors[5]));
2557 } else if (!multiline) {
2558 /* non multiline match, so compare with the "\n" at the
2559 * end of the string */
2560 if (memEQ(checkpos, little, len)) {
2564 PerlIO_printf(Perl_debug_log,
2565 "%sString does not contain required trailing substring, cannot match.%s\n",
2566 PL_colors[4], PL_colors[5]));
2570 /* multiline match, so we have to search for a place
2571 * where the full string is located */
2577 last = rninstr(s, strend, little, little + len);
2579 last = strend; /* matching "$" */
2582 /* at one point this block contained a comment which was
2583 * probably incorrect, which said that this was a "should not
2584 * happen" case. Even if it was true when it was written I am
2585 * pretty sure it is not anymore, so I have removed the comment
2586 * and replaced it with this one. Yves */
2588 PerlIO_printf(Perl_debug_log,
2589 "String does not contain required substring, cannot match.\n"
2593 dontbother = strend - last + prog->float_min_offset;
2595 if (minlen && (dontbother < minlen))
2596 dontbother = minlen - 1;
2597 strend -= dontbother; /* this one's always in bytes! */
2598 /* We don't know much -- general case. */
2601 if (regtry(®info, &s))
2610 if (regtry(®info, &s))
2612 } while (s++ < strend);
2622 PerlIO_printf(Perl_debug_log,
2623 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2629 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2631 if (PL_reg_state.re_state_eval_setup_done)
2632 restore_pos(aTHX_ prog);
2633 if (RXp_PAREN_NAMES(prog))
2634 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2636 /* make sure $`, $&, $', and $digit will work later */
2637 if ( !(flags & REXEC_NOT_FIRST) ) {
2638 if (flags & REXEC_COPY_STR) {
2639 #ifdef PERL_OLD_COPY_ON_WRITE
2641 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2643 PerlIO_printf(Perl_debug_log,
2644 "Copy on write: regexp capture, type %d\n",
2647 RX_MATCH_COPY_FREE(rx);
2648 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2649 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2650 assert (SvPOKp(prog->saved_copy));
2651 prog->sublen = PL_regeol - strbeg;
2652 prog->suboffset = 0;
2653 prog->subcoffset = 0;
2658 I32 max = PL_regeol - strbeg;
2661 if ( (flags & REXEC_COPY_SKIP_POST)
2662 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2663 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2664 ) { /* don't copy $' part of string */
2667 /* calculate the right-most part of the string covered
2668 * by a capture. Due to look-ahead, this may be to
2669 * the right of $&, so we have to scan all captures */
2670 while (n <= prog->lastparen) {
2671 if (prog->offs[n].end > max)
2672 max = prog->offs[n].end;
2676 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2677 ? prog->offs[0].start
2679 assert(max >= 0 && max <= PL_regeol - strbeg);
2682 if ( (flags & REXEC_COPY_SKIP_PRE)
2683 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2684 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2685 ) { /* don't copy $` part of string */
2688 /* calculate the left-most part of the string covered
2689 * by a capture. Due to look-behind, this may be to
2690 * the left of $&, so we have to scan all captures */
2691 while (min && n <= prog->lastparen) {
2692 if ( prog->offs[n].start != -1
2693 && prog->offs[n].start < min)
2695 min = prog->offs[n].start;
2699 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2700 && min > prog->offs[0].end
2702 min = prog->offs[0].end;
2706 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2709 if (RX_MATCH_COPIED(rx)) {
2710 if (sublen > prog->sublen)
2712 (char*)saferealloc(prog->subbeg, sublen+1);
2715 prog->subbeg = (char*)safemalloc(sublen+1);
2716 Copy(strbeg + min, prog->subbeg, sublen, char);
2717 prog->subbeg[sublen] = '\0';
2718 prog->suboffset = min;
2719 prog->sublen = sublen;
2720 RX_MATCH_COPIED_on(rx);
2722 prog->subcoffset = prog->suboffset;
2723 if (prog->suboffset && utf8_target) {
2724 /* Convert byte offset to chars.
2725 * XXX ideally should only compute this if @-/@+
2726 * has been seen, a la PL_sawampersand ??? */
2728 /* If there's a direct correspondence between the
2729 * string which we're matching and the original SV,
2730 * then we can use the utf8 len cache associated with
2731 * the SV. In particular, it means that under //g,
2732 * sv_pos_b2u() will use the previously cached
2733 * position to speed up working out the new length of
2734 * subcoffset, rather than counting from the start of
2735 * the string each time. This stops
2736 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2737 * from going quadratic */
2738 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2739 sv_pos_b2u(sv, &(prog->subcoffset));
2741 prog->subcoffset = utf8_length((U8*)strbeg,
2742 (U8*)(strbeg+prog->suboffset));
2746 RX_MATCH_COPY_FREE(rx);
2747 prog->subbeg = strbeg;
2748 prog->suboffset = 0;
2749 prog->subcoffset = 0;
2750 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2757 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2758 PL_colors[4], PL_colors[5]));
2759 if (PL_reg_state.re_state_eval_setup_done)
2760 restore_pos(aTHX_ prog);
2762 /* we failed :-( roll it back */
2763 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2764 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2769 Safefree(prog->offs);
2776 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2777 * Do inc before dec, in case old and new rex are the same */
2778 #define SET_reg_curpm(Re2) \
2779 if (PL_reg_state.re_state_eval_setup_done) { \
2780 (void)ReREFCNT_inc(Re2); \
2781 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2782 PM_SETRE((PL_reg_curpm), (Re2)); \
2787 - regtry - try match at specific point
2789 STATIC I32 /* 0 failure, 1 success */
2790 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2794 REGEXP *const rx = reginfo->prog;
2795 regexp *const prog = ReANY(rx);
2797 RXi_GET_DECL(prog,progi);
2798 GET_RE_DEBUG_FLAGS_DECL;
2800 PERL_ARGS_ASSERT_REGTRY;
2802 reginfo->cutpoint=NULL;
2804 if ((prog->extflags & RXf_EVAL_SEEN)
2805 && !PL_reg_state.re_state_eval_setup_done)
2809 PL_reg_state.re_state_eval_setup_done = TRUE;
2811 /* Make $_ available to executed code. */
2812 if (reginfo->sv != DEFSV) {
2814 DEFSV_set(reginfo->sv);
2817 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2818 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2819 /* prepare for quick setting of pos */
2820 #ifdef PERL_OLD_COPY_ON_WRITE
2821 if (SvIsCOW(reginfo->sv))
2822 sv_force_normal_flags(reginfo->sv, 0);
2824 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2825 &PL_vtbl_mglob, NULL, 0);
2829 PL_reg_oldpos = mg->mg_len;
2830 SAVEDESTRUCTOR_X(restore_pos, prog);
2832 if (!PL_reg_curpm) {
2833 Newxz(PL_reg_curpm, 1, PMOP);
2836 SV* const repointer = &PL_sv_undef;
2837 /* this regexp is also owned by the new PL_reg_curpm, which
2838 will try to free it. */
2839 av_push(PL_regex_padav, repointer);
2840 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2841 PL_regex_pad = AvARRAY(PL_regex_padav);
2846 PL_reg_oldcurpm = PL_curpm;
2847 PL_curpm = PL_reg_curpm;
2848 if (RXp_MATCH_COPIED(prog)) {
2849 /* Here is a serious problem: we cannot rewrite subbeg,
2850 since it may be needed if this match fails. Thus
2851 $` inside (?{}) could fail... */
2852 PL_reg_oldsaved = prog->subbeg;
2853 PL_reg_oldsavedlen = prog->sublen;
2854 PL_reg_oldsavedoffset = prog->suboffset;
2855 PL_reg_oldsavedcoffset = prog->suboffset;
2856 #ifdef PERL_OLD_COPY_ON_WRITE
2857 PL_nrs = prog->saved_copy;
2859 RXp_MATCH_COPIED_off(prog);
2862 PL_reg_oldsaved = NULL;
2863 prog->subbeg = PL_bostr;
2864 prog->suboffset = 0;
2865 prog->subcoffset = 0;
2866 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2869 PL_reg_starttry = *startposp;
2871 prog->offs[0].start = *startposp - PL_bostr;
2872 prog->lastparen = 0;
2873 prog->lastcloseparen = 0;
2876 /* XXXX What this code is doing here?!!! There should be no need
2877 to do this again and again, prog->lastparen should take care of
2880 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2881 * Actually, the code in regcppop() (which Ilya may be meaning by
2882 * prog->lastparen), is not needed at all by the test suite
2883 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2884 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2885 * Meanwhile, this code *is* needed for the
2886 * above-mentioned test suite tests to succeed. The common theme
2887 * on those tests seems to be returning null fields from matches.
2888 * --jhi updated by dapm */
2890 if (prog->nparens) {
2891 regexp_paren_pair *pp = prog->offs;
2893 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2901 result = regmatch(reginfo, *startposp, progi->program + 1);
2903 prog->offs[0].end = result;
2906 if (reginfo->cutpoint)
2907 *startposp= reginfo->cutpoint;
2908 REGCP_UNWIND(lastcp);
2913 #define sayYES goto yes
2914 #define sayNO goto no
2915 #define sayNO_SILENT goto no_silent
2917 /* we dont use STMT_START/END here because it leads to
2918 "unreachable code" warnings, which are bogus, but distracting. */
2919 #define CACHEsayNO \
2920 if (ST.cache_mask) \
2921 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2924 /* this is used to determine how far from the left messages like
2925 'failed...' are printed. It should be set such that messages
2926 are inline with the regop output that created them.
2928 #define REPORT_CODE_OFF 32
2931 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2932 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2933 #define CHRTEST_NOT_A_CP_1 -999
2934 #define CHRTEST_NOT_A_CP_2 -998
2936 #define SLAB_FIRST(s) (&(s)->states[0])
2937 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2939 /* grab a new slab and return the first slot in it */
2941 STATIC regmatch_state *
2944 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2947 regmatch_slab *s = PL_regmatch_slab->next;
2949 Newx(s, 1, regmatch_slab);
2950 s->prev = PL_regmatch_slab;
2952 PL_regmatch_slab->next = s;
2954 PL_regmatch_slab = s;
2955 return SLAB_FIRST(s);
2959 /* push a new state then goto it */
2961 #define PUSH_STATE_GOTO(state, node, input) \
2962 pushinput = input; \
2964 st->resume_state = state; \
2967 /* push a new state with success backtracking, then goto it */
2969 #define PUSH_YES_STATE_GOTO(state, node, input) \
2970 pushinput = input; \
2972 st->resume_state = state; \
2973 goto push_yes_state;
2980 regmatch() - main matching routine
2982 This is basically one big switch statement in a loop. We execute an op,
2983 set 'next' to point the next op, and continue. If we come to a point which
2984 we may need to backtrack to on failure such as (A|B|C), we push a
2985 backtrack state onto the backtrack stack. On failure, we pop the top
2986 state, and re-enter the loop at the state indicated. If there are no more
2987 states to pop, we return failure.
2989 Sometimes we also need to backtrack on success; for example /A+/, where
2990 after successfully matching one A, we need to go back and try to
2991 match another one; similarly for lookahead assertions: if the assertion
2992 completes successfully, we backtrack to the state just before the assertion
2993 and then carry on. In these cases, the pushed state is marked as
2994 'backtrack on success too'. This marking is in fact done by a chain of
2995 pointers, each pointing to the previous 'yes' state. On success, we pop to
2996 the nearest yes state, discarding any intermediate failure-only states.
2997 Sometimes a yes state is pushed just to force some cleanup code to be
2998 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2999 it to free the inner regex.
3001 Note that failure backtracking rewinds the cursor position, while
3002 success backtracking leaves it alone.
3004 A pattern is complete when the END op is executed, while a subpattern
3005 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3006 ops trigger the "pop to last yes state if any, otherwise return true"
3009 A common convention in this function is to use A and B to refer to the two
3010 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3011 the subpattern to be matched possibly multiple times, while B is the entire
3012 rest of the pattern. Variable and state names reflect this convention.
3014 The states in the main switch are the union of ops and failure/success of
3015 substates associated with with that op. For example, IFMATCH is the op
3016 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3017 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3018 successfully matched A and IFMATCH_A_fail is a state saying that we have
3019 just failed to match A. Resume states always come in pairs. The backtrack
3020 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3021 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3022 on success or failure.
3024 The struct that holds a backtracking state is actually a big union, with
3025 one variant for each major type of op. The variable st points to the
3026 top-most backtrack struct. To make the code clearer, within each
3027 block of code we #define ST to alias the relevant union.
3029 Here's a concrete example of a (vastly oversimplified) IFMATCH
3035 #define ST st->u.ifmatch
3037 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3038 ST.foo = ...; // some state we wish to save
3040 // push a yes backtrack state with a resume value of
3041 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3043 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3046 case IFMATCH_A: // we have successfully executed A; now continue with B
3048 bar = ST.foo; // do something with the preserved value
3051 case IFMATCH_A_fail: // A failed, so the assertion failed
3052 ...; // do some housekeeping, then ...
3053 sayNO; // propagate the failure
3060 For any old-timers reading this who are familiar with the old recursive
3061 approach, the code above is equivalent to:
3063 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3072 ...; // do some housekeeping, then ...
3073 sayNO; // propagate the failure
3076 The topmost backtrack state, pointed to by st, is usually free. If you
3077 want to claim it, populate any ST.foo fields in it with values you wish to
3078 save, then do one of
3080 PUSH_STATE_GOTO(resume_state, node, newinput);
3081 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3083 which sets that backtrack state's resume value to 'resume_state', pushes a
3084 new free entry to the top of the backtrack stack, then goes to 'node'.
3085 On backtracking, the free slot is popped, and the saved state becomes the
3086 new free state. An ST.foo field in this new top state can be temporarily
3087 accessed to retrieve values, but once the main loop is re-entered, it
3088 becomes available for reuse.
3090 Note that the depth of the backtrack stack constantly increases during the
3091 left-to-right execution of the pattern, rather than going up and down with
3092 the pattern nesting. For example the stack is at its maximum at Z at the
3093 end of the pattern, rather than at X in the following:
3095 /(((X)+)+)+....(Y)+....Z/
3097 The only exceptions to this are lookahead/behind assertions and the cut,
3098 (?>A), which pop all the backtrack states associated with A before
3101 Backtrack state structs are allocated in slabs of about 4K in size.
3102 PL_regmatch_state and st always point to the currently active state,
3103 and PL_regmatch_slab points to the slab currently containing
3104 PL_regmatch_state. The first time regmatch() is called, the first slab is
3105 allocated, and is never freed until interpreter destruction. When the slab
3106 is full, a new one is allocated and chained to the end. At exit from
3107 regmatch(), slabs allocated since entry are freed.
3112 #define DEBUG_STATE_pp(pp) \
3114 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3115 PerlIO_printf(Perl_debug_log, \
3116 " %*s"pp" %s%s%s%s%s\n", \
3118 PL_reg_name[st->resume_state], \
3119 ((st==yes_state||st==mark_state) ? "[" : ""), \
3120 ((st==yes_state) ? "Y" : ""), \
3121 ((st==mark_state) ? "M" : ""), \
3122 ((st==yes_state||st==mark_state) ? "]" : "") \
3127 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3132 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3133 const char *start, const char *end, const char *blurb)
3135 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3137 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3142 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3143 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3145 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3146 start, end - start, 60);
3148 PerlIO_printf(Perl_debug_log,
3149 "%s%s REx%s %s against %s\n",
3150 PL_colors[4], blurb, PL_colors[5], s0, s1);
3152 if (utf8_target||utf8_pat)
3153 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3154 utf8_pat ? "pattern" : "",
3155 utf8_pat && utf8_target ? " and " : "",
3156 utf8_target ? "string" : ""
3162 S_dump_exec_pos(pTHX_ const char *locinput,
3163 const regnode *scan,
3164 const char *loc_regeol,
3165 const char *loc_bostr,
3166 const char *loc_reg_starttry,
3167 const bool utf8_target)
3169 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3170 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3171 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3172 /* The part of the string before starttry has one color
3173 (pref0_len chars), between starttry and current
3174 position another one (pref_len - pref0_len chars),
3175 after the current position the third one.
3176 We assume that pref0_len <= pref_len, otherwise we
3177 decrease pref0_len. */
3178 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3179 ? (5 + taill) - l : locinput - loc_bostr;
3182 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3184 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3186 pref0_len = pref_len - (locinput - loc_reg_starttry);
3187 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3188 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3189 ? (5 + taill) - pref_len : loc_regeol - locinput);
3190 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3194 if (pref0_len > pref_len)
3195 pref0_len = pref_len;
3197 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3199 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3200 (locinput - pref_len),pref0_len, 60, 4, 5);
3202 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3203 (locinput - pref_len + pref0_len),
3204 pref_len - pref0_len, 60, 2, 3);
3206 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3207 locinput, loc_regeol - locinput, 10, 0, 1);
3209 const STRLEN tlen=len0+len1+len2;
3210 PerlIO_printf(Perl_debug_log,
3211 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3212 (IV)(locinput - loc_bostr),
3215 (docolor ? "" : "> <"),
3217 (int)(tlen > 19 ? 0 : 19 - tlen),
3224 /* reg_check_named_buff_matched()
3225 * Checks to see if a named buffer has matched. The data array of
3226 * buffer numbers corresponding to the buffer is expected to reside
3227 * in the regexp->data->data array in the slot stored in the ARG() of
3228 * node involved. Note that this routine doesn't actually care about the
3229 * name, that information is not preserved from compilation to execution.
3230 * Returns the index of the leftmost defined buffer with the given name
3231 * or 0 if non of the buffers matched.
3234 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3237 RXi_GET_DECL(rex,rexi);
3238 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3239 I32 *nums=(I32*)SvPVX(sv_dat);
3241 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3243 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3244 if ((I32)rex->lastparen >= nums[n] &&
3245 rex->offs[nums[n]].end != -1)
3254 /* free all slabs above current one - called during LEAVE_SCOPE */
3257 S_clear_backtrack_stack(pTHX_ void *p)
3259 regmatch_slab *s = PL_regmatch_slab->next;
3264 PL_regmatch_slab->next = NULL;
3266 regmatch_slab * const osl = s;
3272 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, U8* c1_utf8, int *c2p, U8* c2_utf8)
3274 /* This function determines if there are one or two characters that match
3275 * the first character of the passed-in EXACTish node <text_node>, and if
3276 * so, returns them in the passed-in pointers.
3278 * If it determines that no possible character in the target string can
3279 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3280 * the first character in <text_node> requires UTF-8 to represent, and the
3281 * target string isn't in UTF-8.)
3283 * If there are more than two characters that could match the beginning of
3284 * <text_node>, or if more context is required to determine a match or not,
3285 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3287 * The motiviation behind this function is to allow the caller to set up
3288 * tight loops for matching. If <text_node> is of type EXACT, there is
3289 * only one possible character that can match its first character, and so
3290 * the situation is quite simple. But things get much more complicated if
3291 * folding is involved. It may be that the first character of an EXACTFish
3292 * node doesn't participate in any possible fold, e.g., punctuation, so it
3293 * can be matched only by itself. The vast majority of characters that are
3294 * in folds match just two things, their lower and upper-case equivalents.
3295 * But not all are like that; some have multiple possible matches, or match
3296 * sequences of more than one character. This function sorts all that out.
3298 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3299 * loop of trying to match A*, we know we can't exit where the thing
3300 * following it isn't a B. And something can't be a B unless it is the
3301 * beginning of B. By putting a quick test for that beginning in a tight
3302 * loop, we can rule out things that can't possibly be B without having to
3303 * break out of the loop, thus avoiding work. Similarly, if A is a single
3304 * character, we can make a tight loop matching A*, using the outputs of
3307 * If the target string to match isn't in UTF-8, and there aren't
3308 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3309 * the one or two possible octets (which are characters in this situation)
3310 * that can match. In all cases, if there is only one character that can
3311 * match, *<c1p> and *<c2p> will be identical.
3313 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3314 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3315 * can match the beginning of <text_node>. They should be declared with at
3316 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3317 * undefined what these contain.) If one or both of the buffers are
3318 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3319 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3320 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3321 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3322 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3324 const bool utf8_target = PL_reg_match_utf8;
3326 UV c1 = CHRTEST_NOT_A_CP_1;
3327 UV c2 = CHRTEST_NOT_A_CP_2;
3328 bool use_chrtest_void = FALSE;
3330 /* Used when we have both utf8 input and utf8 output, to avoid converting
3331 * to/from code points */
3332 bool utf8_has_been_setup = FALSE;
3336 U8 *pat = (U8*)STRING(text_node);
3338 if (OP(text_node) == EXACT) {
3340 /* In an exact node, only one thing can be matched, that first
3341 * character. If both the pat and the target are UTF-8, we can just
3342 * copy the input to the output, avoiding finding the code point of
3344 if (! UTF_PATTERN) {
3347 else if (utf8_target) {
3348 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3349 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3350 utf8_has_been_setup = TRUE;
3353 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3356 else /* an EXACTFish node */
3358 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3359 pat + STR_LEN(text_node)))
3361 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3362 pat + STR_LEN(text_node))))
3364 /* Multi-character folds require more context to sort out. Also
3365 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3366 * handled outside this routine */
3367 use_chrtest_void = TRUE;
3369 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3370 c1 = (UTF_PATTERN) ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3372 /* Load the folds hash, if not already done */
3374 if (! PL_utf8_foldclosures) {
3375 if (! PL_utf8_tofold) {
3376 U8 dummy[UTF8_MAXBYTES+1];
3378 /* Force loading this by folding an above-Latin1 char */
3379 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3380 assert(PL_utf8_tofold); /* Verify that worked */
3382 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3385 /* The fold closures data structure is a hash with the keys being
3386 * the UTF-8 of every character that is folded to, like 'k', and
3387 * the values each an array of all code points that fold to its
3388 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3390 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3395 /* Not found in the hash, therefore there are no folds
3396 * containing it, so there is only a single character that
3400 else { /* Does participate in folds */
3401 AV* list = (AV*) *listp;
3402 if (av_len(list) != 1) {
3404 /* If there aren't exactly two folds to this, it is outside
3405 * the scope of this function */
3406 use_chrtest_void = TRUE;
3408 else { /* There are two. Get them */
3409 SV** c_p = av_fetch(list, 0, FALSE);
3411 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3415 c_p = av_fetch(list, 1, FALSE);
3417 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3421 /* Folds that cross the 255/256 boundary are forbidden if
3422 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3423 * pattern character is above 256, and its only other match
3424 * is below 256, the only legal match will be to itself.
3425 * We have thrown away the original, so have to compute
3426 * which is the one above 255 */
3427 if ((c1 < 256) != (c2 < 256)) {
3428 if (OP(text_node) == EXACTFL
3429 || (OP(text_node) == EXACTFA
3430 && (isASCII(c1) || isASCII(c2))))
3443 else /* Here, c1 is < 255 */
3445 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3446 && OP(text_node) != EXACTFL
3447 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3449 /* Here, there could be something above Latin1 in the target which
3450 * folds to this character in the pattern. All such cases except
3451 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3452 * involved in their folds, so are outside the scope of this
3454 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3455 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3458 use_chrtest_void = TRUE;
3461 else { /* Here nothing above Latin1 can fold to the pattern character */
3462 switch (OP(text_node)) {
3464 case EXACTFL: /* /l rules */
3465 c2 = PL_fold_locale[c1];
3469 if (! utf8_target) { /* /d rules */
3474 /* /u rules for all these. This happens to work for
3475 * EXACTFA as nothing in Latin1 folds to ASCII */
3477 case EXACTFU_TRICKYFOLD:
3480 c2 = PL_fold_latin1[c1];
3484 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3485 assert(0); /* NOTREACHED */
3490 /* Here have figured things out. Set up the returns */
3491 if (use_chrtest_void) {
3492 *c2p = *c1p = CHRTEST_VOID;
3494 else if (utf8_target) {
3495 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3496 uvchr_to_utf8(c1_utf8, c1);
3497 uvchr_to_utf8(c2_utf8, c2);
3500 /* Invariants are stored in both the utf8 and byte outputs; Use
3501 * negative numbers otherwise for the byte ones. Make sure that the
3502 * byte ones are the same iff the utf8 ones are the same */
3503 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3504 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3507 ? CHRTEST_NOT_A_CP_1
3508 : CHRTEST_NOT_A_CP_2;
3510 else if (c1 > 255) {
3511 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3516 *c1p = *c2p = c2; /* c2 is the only representable value */
3518 else { /* c1 is representable; see about c2 */
3520 *c2p = (c2 < 256) ? c2 : c1;
3526 /* returns -1 on failure, $+[0] on success */
3528 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3530 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3534 const bool utf8_target = PL_reg_match_utf8;
3535 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3536 REGEXP *rex_sv = reginfo->prog;
3537 regexp *rex = ReANY(rex_sv);
3538 RXi_GET_DECL(rex,rexi);
3540 /* the current state. This is a cached copy of PL_regmatch_state */
3542 /* cache heavy used fields of st in registers */
3545 U32 n = 0; /* general value; init to avoid compiler warning */
3546 I32 ln = 0; /* len or last; init to avoid compiler warning */
3547 char *locinput = startpos;
3548 char *pushinput; /* where to continue after a PUSH */
3549 I32 nextchr; /* is always set to UCHARAT(locinput) */
3551 bool result = 0; /* return value of S_regmatch */
3552 int depth = 0; /* depth of backtrack stack */
3553 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3554 const U32 max_nochange_depth =
3555 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3556 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3557 regmatch_state *yes_state = NULL; /* state to pop to on success of
3559 /* mark_state piggy backs on the yes_state logic so that when we unwind
3560 the stack on success we can update the mark_state as we go */
3561 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3562 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3563 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3565 bool no_final = 0; /* prevent failure from backtracking? */
3566 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3567 char *startpoint = locinput;
3568 SV *popmark = NULL; /* are we looking for a mark? */
3569 SV *sv_commit = NULL; /* last mark name seen in failure */
3570 SV *sv_yes_mark = NULL; /* last mark name we have seen
3571 during a successful match */
3572 U32 lastopen = 0; /* last open we saw */
3573 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3574 SV* const oreplsv = GvSV(PL_replgv);
3575 /* these three flags are set by various ops to signal information to
3576 * the very next op. They have a useful lifetime of exactly one loop
3577 * iteration, and are not preserved or restored by state pushes/pops
3579 bool sw = 0; /* the condition value in (?(cond)a|b) */
3580 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3581 int logical = 0; /* the following EVAL is:
3585 or the following IFMATCH/UNLESSM is:
3586 false: plain (?=foo)
3587 true: used as a condition: (?(?=foo))
3589 PAD* last_pad = NULL;
3591 I32 gimme = G_SCALAR;
3592 CV *caller_cv = NULL; /* who called us */
3593 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3594 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3597 GET_RE_DEBUG_FLAGS_DECL;
3600 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3601 multicall_oldcatch = 0;
3602 multicall_cv = NULL;
3604 PERL_UNUSED_VAR(multicall_cop);
3605 PERL_UNUSED_VAR(newsp);
3608 PERL_ARGS_ASSERT_REGMATCH;
3610 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3611 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3613 /* on first ever call to regmatch, allocate first slab */
3614 if (!PL_regmatch_slab) {
3615 Newx(PL_regmatch_slab, 1, regmatch_slab);
3616 PL_regmatch_slab->prev = NULL;
3617 PL_regmatch_slab->next = NULL;
3618 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3621 oldsave = PL_savestack_ix;
3622 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3623 SAVEVPTR(PL_regmatch_slab);
3624 SAVEVPTR(PL_regmatch_state);
3626 /* grab next free state slot */
3627 st = ++PL_regmatch_state;
3628 if (st > SLAB_LAST(PL_regmatch_slab))
3629 st = PL_regmatch_state = S_push_slab(aTHX);
3631 /* Note that nextchr is a byte even in UTF */
3634 while (scan != NULL) {
3637 SV * const prop = sv_newmortal();
3638 regnode *rnext=regnext(scan);
3639 DUMP_EXEC_POS( locinput, scan, utf8_target );
3640 regprop(rex, prop, scan);
3642 PerlIO_printf(Perl_debug_log,
3643 "%3"IVdf":%*s%s(%"IVdf")\n",
3644 (IV)(scan - rexi->program), depth*2, "",
3646 (PL_regkind[OP(scan)] == END || !rnext) ?
3647 0 : (IV)(rnext - rexi->program));
3650 next = scan + NEXT_OFF(scan);
3653 state_num = OP(scan);
3658 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3660 switch (state_num) {
3661 case BOL: /* /^../ */
3662 if (locinput == PL_bostr)
3664 /* reginfo->till = reginfo->bol; */
3669 case MBOL: /* /^../m */
3670 if (locinput == PL_bostr ||
3671 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3677 case SBOL: /* /^../s */
3678 if (locinput == PL_bostr)
3683 if (locinput == reginfo->ganch)
3687 case KEEPS: /* \K */
3688 /* update the startpoint */
3689 st->u.keeper.val = rex->offs[0].start;
3690 rex->offs[0].start = locinput - PL_bostr;
3691 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3692 assert(0); /*NOTREACHED*/
3693 case KEEPS_next_fail:
3694 /* rollback the start point change */
3695 rex->offs[0].start = st->u.keeper.val;
3697 assert(0); /*NOTREACHED*/
3699 case EOL: /* /..$/ */
3702 case MEOL: /* /..$/m */
3703 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3707 case SEOL: /* /..$/s */
3709 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3711 if (PL_regeol - locinput > 1)
3716 if (!NEXTCHR_IS_EOS)
3720 case SANY: /* /./s */
3723 goto increment_locinput;
3731 case REG_ANY: /* /./ */
3732 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3734 goto increment_locinput;
3738 #define ST st->u.trie
3739 case TRIEC: /* (ab|cd) with known charclass */
3740 /* In this case the charclass data is available inline so
3741 we can fail fast without a lot of extra overhead.
3743 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3745 PerlIO_printf(Perl_debug_log,
3746 "%*s %sfailed to match trie start class...%s\n",
3747 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3750 assert(0); /* NOTREACHED */
3753 case TRIE: /* (ab|cd) */
3754 /* the basic plan of execution of the trie is:
3755 * At the beginning, run though all the states, and
3756 * find the longest-matching word. Also remember the position
3757 * of the shortest matching word. For example, this pattern:
3760 * when matched against the string "abcde", will generate
3761 * accept states for all words except 3, with the longest
3762 * matching word being 4, and the shortest being 2 (with
3763 * the position being after char 1 of the string).
3765 * Then for each matching word, in word order (i.e. 1,2,4,5),
3766 * we run the remainder of the pattern; on each try setting
3767 * the current position to the character following the word,
3768 * returning to try the next word on failure.
3770 * We avoid having to build a list of words at runtime by
3771 * using a compile-time structure, wordinfo[].prev, which
3772 * gives, for each word, the previous accepting word (if any).
3773 * In the case above it would contain the mappings 1->2, 2->0,
3774 * 3->0, 4->5, 5->1. We can use this table to generate, from
3775 * the longest word (4 above), a list of all words, by
3776 * following the list of prev pointers; this gives us the
3777 * unordered list 4,5,1,2. Then given the current word we have
3778 * just tried, we can go through the list and find the
3779 * next-biggest word to try (so if we just failed on word 2,
3780 * the next in the list is 4).
3782 * Since at runtime we don't record the matching position in
3783 * the string for each word, we have to work that out for
3784 * each word we're about to process. The wordinfo table holds
3785 * the character length of each word; given that we recorded
3786 * at the start: the position of the shortest word and its
3787 * length in chars, we just need to move the pointer the
3788 * difference between the two char lengths. Depending on
3789 * Unicode status and folding, that's cheap or expensive.
3791 * This algorithm is optimised for the case where are only a
3792 * small number of accept states, i.e. 0,1, or maybe 2.
3793 * With lots of accepts states, and having to try all of them,
3794 * it becomes quadratic on number of accept states to find all
3799 /* what type of TRIE am I? (utf8 makes this contextual) */
3800 DECL_TRIE_TYPE(scan);
3802 /* what trie are we using right now */
3803 reg_trie_data * const trie
3804 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3805 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3806 U32 state = trie->startstate;
3809 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3811 if (trie->states[ state ].wordnum) {
3813 PerlIO_printf(Perl_debug_log,
3814 "%*s %smatched empty string...%s\n",
3815 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3821 PerlIO_printf(Perl_debug_log,
3822 "%*s %sfailed to match trie start class...%s\n",
3823 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3830 U8 *uc = ( U8* )locinput;
3834 U8 *uscan = (U8*)NULL;
3835 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3836 U32 charcount = 0; /* how many input chars we have matched */
3837 U32 accepted = 0; /* have we seen any accepting states? */
3839 ST.jump = trie->jump;
3842 ST.longfold = FALSE; /* char longer if folded => it's harder */
3845 /* fully traverse the TRIE; note the position of the
3846 shortest accept state and the wordnum of the longest
3849 while ( state && uc <= (U8*)PL_regeol ) {
3850 U32 base = trie->states[ state ].trans.base;
3854 wordnum = trie->states[ state ].wordnum;
3856 if (wordnum) { /* it's an accept state */
3859 /* record first match position */
3861 ST.firstpos = (U8*)locinput;
3866 ST.firstchars = charcount;
3869 if (!ST.nextword || wordnum < ST.nextword)
3870 ST.nextword = wordnum;
3871 ST.topword = wordnum;
3874 DEBUG_TRIE_EXECUTE_r({
3875 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3876 PerlIO_printf( Perl_debug_log,
3877 "%*s %sState: %4"UVxf" Accepted: %c ",
3878 2+depth * 2, "", PL_colors[4],
3879 (UV)state, (accepted ? 'Y' : 'N'));
3882 /* read a char and goto next state */
3883 if ( base && (foldlen || uc < (U8*)PL_regeol)) {
3885 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3886 uscan, len, uvc, charid, foldlen,
3893 base + charid - 1 - trie->uniquecharcount)) >= 0)
3895 && ((U32)offset < trie->lasttrans)
3896 && trie->trans[offset].check == state)
3898 state = trie->trans[offset].next;
3909 DEBUG_TRIE_EXECUTE_r(
3910 PerlIO_printf( Perl_debug_log,
3911 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3912 charid, uvc, (UV)state, PL_colors[5] );
3918 /* calculate total number of accept states */
3923 w = trie->wordinfo[w].prev;
3926 ST.accepted = accepted;
3930 PerlIO_printf( Perl_debug_log,
3931 "%*s %sgot %"IVdf" possible matches%s\n",
3932 REPORT_CODE_OFF + depth * 2, "",
3933 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3935 goto trie_first_try; /* jump into the fail handler */
3937 assert(0); /* NOTREACHED */
3939 case TRIE_next_fail: /* we failed - try next alternative */
3943 REGCP_UNWIND(ST.cp);
3944 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
3946 if (!--ST.accepted) {
3948 PerlIO_printf( Perl_debug_log,
3949 "%*s %sTRIE failed...%s\n",
3950 REPORT_CODE_OFF+depth*2, "",
3957 /* Find next-highest word to process. Note that this code
3958 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3961 U16 const nextword = ST.nextword;
3962 reg_trie_wordinfo * const wordinfo
3963 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3964 for (word=ST.topword; word; word=wordinfo[word].prev) {
3965 if (word > nextword && (!min || word < min))
3978 ST.lastparen = rex->lastparen;
3979 ST.lastcloseparen = rex->lastcloseparen;
3983 /* find start char of end of current word */
3985 U32 chars; /* how many chars to skip */
3986 reg_trie_data * const trie
3987 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3989 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3991 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3996 /* the hard option - fold each char in turn and find
3997 * its folded length (which may be different */
3998 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4006 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
4014 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4019 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
4035 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4036 ? ST.jump[ST.nextword]
4040 PerlIO_printf( Perl_debug_log,
4041 "%*s %sTRIE matched word #%d, continuing%s\n",
4042 REPORT_CODE_OFF+depth*2, "",
4049 if (ST.accepted > 1 || has_cutgroup) {
4050 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4051 assert(0); /* NOTREACHED */
4053 /* only one choice left - just continue */
4055 AV *const trie_words
4056 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4057 SV ** const tmp = av_fetch( trie_words,
4059 SV *sv= tmp ? sv_newmortal() : NULL;
4061 PerlIO_printf( Perl_debug_log,
4062 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4063 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4065 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4066 PL_colors[0], PL_colors[1],
4067 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4069 : "not compiled under -Dr",
4073 locinput = (char*)uc;
4074 continue; /* execute rest of RE */
4075 assert(0); /* NOTREACHED */
4079 case EXACT: { /* /abc/ */
4080 char *s = STRING(scan);
4082 if (utf8_target != UTF_PATTERN) {
4083 /* The target and the pattern have differing utf8ness. */
4085 const char * const e = s + ln;
4088 /* The target is utf8, the pattern is not utf8.
4089 * Above-Latin1 code points can't match the pattern;
4090 * invariants match exactly, and the other Latin1 ones need
4091 * to be downgraded to a single byte in order to do the
4092 * comparison. (If we could be confident that the target
4093 * is not malformed, this could be refactored to have fewer
4094 * tests by just assuming that if the first bytes match, it
4095 * is an invariant, but there are tests in the test suite
4096 * dealing with (??{...}) which violate this) */
4100 if (UTF8_IS_ABOVE_LATIN1(* (U8*) l)) {
4103 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4110 if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
4119 /* The target is not utf8, the pattern is utf8. */
4121 if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4125 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4132 if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
4143 /* The target and the pattern have the same utf8ness. */
4144 /* Inline the first character, for speed. */
4145 if (UCHARAT(s) != nextchr)
4147 if (PL_regeol - locinput < ln)
4149 if (ln > 1 && memNE(s, locinput, ln))
4155 case EXACTFL: { /* /abc/il */
4157 const U8 * fold_array;
4159 U32 fold_utf8_flags;
4161 PL_reg_flags |= RF_tainted;
4162 folder = foldEQ_locale;
4163 fold_array = PL_fold_locale;
4164 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
4167 case EXACTFU_SS: /* /\x{df}/iu */
4168 case EXACTFU_TRICKYFOLD: /* /\x{390}/iu */
4169 case EXACTFU: /* /abc/iu */
4170 folder = foldEQ_latin1;
4171 fold_array = PL_fold_latin1;
4172 fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4175 case EXACTFA: /* /abc/iaa */
4176 folder = foldEQ_latin1;
4177 fold_array = PL_fold_latin1;
4178 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4181 case EXACTF: /* /abc/i */
4183 fold_array = PL_fold;
4184 fold_utf8_flags = 0;
4190 if (utf8_target || UTF_PATTERN || state_num == EXACTFU_SS) {
4191 /* Either target or the pattern are utf8, or has the issue where
4192 * the fold lengths may differ. */
4193 const char * const l = locinput;
4194 char *e = PL_regeol;
4196 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
4197 l, &e, 0, utf8_target, fold_utf8_flags))
4205 /* Neither the target nor the pattern are utf8 */
4206 if (UCHARAT(s) != nextchr
4208 && UCHARAT(s) != fold_array[nextchr])
4212 if (PL_regeol - locinput < ln)
4214 if (ln > 1 && ! folder(s, locinput, ln))
4220 /* XXX Could improve efficiency by separating these all out using a
4221 * macro or in-line function. At that point regcomp.c would no longer
4222 * have to set the FLAGS fields of these */
4223 case BOUNDL: /* /\b/l */
4224 case NBOUNDL: /* /\B/l */
4225 PL_reg_flags |= RF_tainted;
4227 case BOUND: /* /\b/ */
4228 case BOUNDU: /* /\b/u */
4229 case BOUNDA: /* /\b/a */
4230 case NBOUND: /* /\B/ */
4231 case NBOUNDU: /* /\B/u */
4232 case NBOUNDA: /* /\B/a */
4233 /* was last char in word? */
4235 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4236 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4238 if (locinput == PL_bostr)
4241 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
4243 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
4245 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4246 ln = isALNUM_uni(ln);
4250 LOAD_UTF8_CHARCLASS_ALNUM();
4251 n = swash_fetch(PL_utf8_alnum, (U8*)locinput,
4256 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
4257 n = NEXTCHR_IS_EOS ? 0 : isALNUM_LC_utf8((U8*)locinput);
4262 /* Here the string isn't utf8, or is utf8 and only ascii
4263 * characters are to match \w. In the latter case looking at
4264 * the byte just prior to the current one may be just the final
4265 * byte of a multi-byte character. This is ok. There are two
4267 * 1) it is a single byte character, and then the test is doing
4268 * just what it's supposed to.
4269 * 2) it is a multi-byte character, in which case the final
4270 * byte is never mistakable for ASCII, and so the test
4271 * will say it is not a word character, which is the
4272 * correct answer. */
4273 ln = (locinput != PL_bostr) ?
4274 UCHARAT(locinput - 1) : '\n';
4275 switch (FLAGS(scan)) {
4276 case REGEX_UNICODE_CHARSET:
4277 ln = isWORDCHAR_L1(ln);
4278 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4280 case REGEX_LOCALE_CHARSET:
4281 ln = isALNUM_LC(ln);
4282 n = NEXTCHR_IS_EOS ? 0 : isALNUM_LC(nextchr);
4284 case REGEX_DEPENDS_CHARSET:
4286 n = NEXTCHR_IS_EOS ? 0 : isALNUM(nextchr);
4288 case REGEX_ASCII_RESTRICTED_CHARSET:
4289 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4290 ln = isWORDCHAR_A(ln);
4291 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4294 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4298 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4300 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4304 case ANYOF: /* /[abc]/ */
4308 if (!reginclass(rex, scan, (U8*)locinput, utf8_target))
4310 locinput += UTF8SKIP(locinput);
4314 if (!REGINCLASS(rex, scan, (U8*)locinput))
4321 /* Special char classes: \d, \w etc.
4322 * The defines start on line 166 or so */
4323 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
4324 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
4325 ALNUMU, NALNUMU, isWORDCHAR_L1,
4326 ALNUMA, NALNUMA, isWORDCHAR_A,
4330 PL_reg_flags |= RF_tainted;
4331 if (NEXTCHR_IS_EOS) {
4334 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) {
4335 if (! isSPACE_LC_utf8((U8 *) locinput)) {
4339 else if (! isSPACE_LC((U8) nextchr)) {
4342 goto increment_locinput;
4345 PL_reg_flags |= RF_tainted;
4346 if (NEXTCHR_IS_EOS) {
4349 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) {
4350 if (isSPACE_LC_utf8((U8 *) locinput)) {
4354 else if (isSPACE_LC(nextchr)) {
4357 goto increment_locinput;
4365 if (NEXTCHR_IS_EOS || ! isSPACE_A(nextchr)) {
4368 /* Matched a utf8-invariant, so don't have to worry about utf8 */
4378 if (NEXTCHR_IS_EOS || isSPACE_A(nextchr)) {
4381 goto increment_locinput;
4385 if (NEXTCHR_IS_EOS || ! is_XPERLSPACE(locinput, utf8_target)) {
4388 goto increment_locinput;
4392 if (NEXTCHR_IS_EOS || is_XPERLSPACE(locinput, utf8_target)) {
4395 goto increment_locinput;
4397 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
4398 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
4399 DIGITA, NDIGITA, isDIGIT_A,
4402 case POSIXA: /* /[[:ascii:]]/ etc */
4403 if (NEXTCHR_IS_EOS || ! _generic_isCC_A(nextchr, FLAGS(scan))) {
4406 /* Matched a utf8-invariant, so don't have to worry about utf8 */
4410 case NPOSIXA: /* /[^[:ascii:]]/ etc */
4411 if (NEXTCHR_IS_EOS || _generic_isCC_A(nextchr, FLAGS(scan))) {
4414 goto increment_locinput;
4416 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4417 a Unicode extended Grapheme Cluster */
4418 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4419 extended Grapheme Cluster is:
4422 | Prepend* Begin Extend*
4425 Begin is: ( Special_Begin | ! Control )
4426 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4427 Extend is: ( Grapheme_Extend | Spacing_Mark )
4428 Control is: [ GCB_Control CR LF ]
4429 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4431 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4434 Begin is ( Regular_Begin + Special Begin )
4436 It turns out that 98.4% of all Unicode code points match
4437 Regular_Begin. Doing it this way eliminates a table match in
4438 the previous implementation for almost all Unicode code points.
4440 There is a subtlety with Prepend* which showed up in testing.
4441 Note that the Begin, and only the Begin is required in:
4442 | Prepend* Begin Extend*
4443 Also, Begin contains '! Control'. A Prepend must be a
4444 '! Control', which means it must also be a Begin. What it
4445 comes down to is that if we match Prepend* and then find no
4446 suitable Begin afterwards, that if we backtrack the last
4447 Prepend, that one will be a suitable Begin.
4452 if (! utf8_target) {
4454 /* Match either CR LF or '.', as all the other possibilities
4456 locinput++; /* Match the . or CR */
4457 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4459 && locinput < PL_regeol
4460 && UCHARAT(locinput) == '\n') locinput++;
4464 /* Utf8: See if is ( CR LF ); already know that locinput <
4465 * PL_regeol, so locinput+1 is in bounds */
4466 if ( nextchr == '\r' && locinput+1 < PL_regeol
4467 && UCHARAT(locinput + 1) == '\n')
4474 /* In case have to backtrack to beginning, then match '.' */
4475 char *starting = locinput;
4477 /* In case have to backtrack the last prepend */
4478 char *previous_prepend = 0;
4480 LOAD_UTF8_CHARCLASS_GCB();
4482 /* Match (prepend)* */
4483 while (locinput < PL_regeol
4484 && (len = is_GCB_Prepend_utf8(locinput)))
4486 previous_prepend = locinput;
4490 /* As noted above, if we matched a prepend character, but
4491 * the next thing won't match, back off the last prepend we
4492 * matched, as it is guaranteed to match the begin */
4493 if (previous_prepend
4494 && (locinput >= PL_regeol
4495 || (! swash_fetch(PL_utf8_X_regular_begin,
4496 (U8*)locinput, utf8_target)
4497 && ! is_GCB_SPECIAL_BEGIN_utf8(locinput)))
4500 locinput = previous_prepend;
4503 /* Note that here we know PL_regeol > locinput, as we
4504 * tested that upon input to this switch case, and if we
4505 * moved locinput forward, we tested the result just above
4506 * and it either passed, or we backed off so that it will
4508 if (swash_fetch(PL_utf8_X_regular_begin,
4509 (U8*)locinput, utf8_target)) {
4510 locinput += UTF8SKIP(locinput);
4512 else if (! is_GCB_SPECIAL_BEGIN_utf8(locinput)) {
4514 /* Here did not match the required 'Begin' in the
4515 * second term. So just match the very first
4516 * character, the '.' of the final term of the regex */
4517 locinput = starting + UTF8SKIP(starting);
4521 /* Here is a special begin. It can be composed of
4522 * several individual characters. One possibility is
4524 if ((len = is_GCB_RI_utf8(locinput))) {
4526 while (locinput < PL_regeol
4527 && (len = is_GCB_RI_utf8(locinput)))
4531 } else if ((len = is_GCB_T_utf8(locinput))) {
4532 /* Another possibility is T+ */
4534 while (locinput < PL_regeol
4535 && (len = is_GCB_T_utf8(locinput)))
4541 /* Here, neither RI+ nor T+; must be some other
4542 * Hangul. That means it is one of the others: L,
4543 * LV, LVT or V, and matches:
4544 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4547 while (locinput < PL_regeol
4548 && (len = is_GCB_L_utf8(locinput)))
4553 /* Here, have exhausted L*. If the next character
4554 * is not an LV, LVT nor V, it means we had to have
4555 * at least one L, so matches L+ in the original
4556 * equation, we have a complete hangul syllable.
4559 if (locinput < PL_regeol
4560 && is_GCB_LV_LVT_V_utf8(locinput))
4563 /* Otherwise keep going. Must be LV, LVT or V.
4565 if (is_utf8_X_LVT((U8*)locinput)) {
4566 locinput += UTF8SKIP(locinput);
4569 /* Must be V or LV. Take it, then match
4571 locinput += UTF8SKIP(locinput);
4572 while (locinput < PL_regeol
4573 && (len = is_GCB_V_utf8(locinput)))
4579 /* And any of LV, LVT, or V can be followed
4581 while (locinput < PL_regeol
4582 && (len = is_GCB_T_utf8(locinput)))
4590 /* Match any extender */
4591 while (locinput < PL_regeol
4592 && swash_fetch(PL_utf8_X_extend,
4593 (U8*)locinput, utf8_target))
4595 locinput += UTF8SKIP(locinput);
4599 if (locinput > PL_regeol) sayNO;
4603 case NREFFL: /* /\g{name}/il */
4604 { /* The capture buffer cases. The ones beginning with N for the
4605 named buffers just convert to the equivalent numbered and
4606 pretend they were called as the corresponding numbered buffer
4608 /* don't initialize these in the declaration, it makes C++
4613 const U8 *fold_array;
4616 PL_reg_flags |= RF_tainted;
4617 folder = foldEQ_locale;
4618 fold_array = PL_fold_locale;
4620 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4623 case NREFFA: /* /\g{name}/iaa */
4624 folder = foldEQ_latin1;
4625 fold_array = PL_fold_latin1;
4627 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4630 case NREFFU: /* /\g{name}/iu */
4631 folder = foldEQ_latin1;
4632 fold_array = PL_fold_latin1;
4634 utf8_fold_flags = 0;
4637 case NREFF: /* /\g{name}/i */
4639 fold_array = PL_fold;
4641 utf8_fold_flags = 0;
4644 case NREF: /* /\g{name}/ */
4648 utf8_fold_flags = 0;
4651 /* For the named back references, find the corresponding buffer
4653 n = reg_check_named_buff_matched(rex,scan);
4658 goto do_nref_ref_common;
4660 case REFFL: /* /\1/il */
4661 PL_reg_flags |= RF_tainted;
4662 folder = foldEQ_locale;
4663 fold_array = PL_fold_locale;
4664 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4667 case REFFA: /* /\1/iaa */
4668 folder = foldEQ_latin1;
4669 fold_array = PL_fold_latin1;
4670 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4673 case REFFU: /* /\1/iu */
4674 folder = foldEQ_latin1;
4675 fold_array = PL_fold_latin1;
4676 utf8_fold_flags = 0;
4679 case REFF: /* /\1/i */
4681 fold_array = PL_fold;
4682 utf8_fold_flags = 0;
4685 case REF: /* /\1/ */
4688 utf8_fold_flags = 0;
4692 n = ARG(scan); /* which paren pair */
4695 ln = rex->offs[n].start;
4696 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4697 if (rex->lastparen < n || ln == -1)
4698 sayNO; /* Do not match unless seen CLOSEn. */
4699 if (ln == rex->offs[n].end)
4703 if (type != REF /* REF can do byte comparison */
4704 && (utf8_target || type == REFFU))
4705 { /* XXX handle REFFL better */
4706 char * limit = PL_regeol;
4708 /* This call case insensitively compares the entire buffer
4709 * at s, with the current input starting at locinput, but
4710 * not going off the end given by PL_regeol, and returns in
4711 * <limit> upon success, how much of the current input was
4713 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
4714 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4722 /* Not utf8: Inline the first character, for speed. */
4723 if (!NEXTCHR_IS_EOS &&
4724 UCHARAT(s) != nextchr &&
4726 UCHARAT(s) != fold_array[nextchr]))
4728 ln = rex->offs[n].end - ln;
4729 if (locinput + ln > PL_regeol)
4731 if (ln > 1 && (type == REF
4732 ? memNE(s, locinput, ln)
4733 : ! folder(s, locinput, ln)))
4739 case NOTHING: /* null op; e.g. the 'nothing' following
4740 * the '*' in m{(a+|b)*}' */
4742 case TAIL: /* placeholder while compiling (A|B|C) */
4745 case BACK: /* ??? doesn't appear to be used ??? */
4749 #define ST st->u.eval
4754 regexp_internal *rei;
4755 regnode *startpoint;
4757 case GOSTART: /* (?R) */
4758 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4759 if (cur_eval && cur_eval->locinput==locinput) {
4760 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4761 Perl_croak(aTHX_ "Infinite recursion in regex");
4762 if ( ++nochange_depth > max_nochange_depth )
4764 "Pattern subroutine nesting without pos change"
4765 " exceeded limit in regex");
4772 if (OP(scan)==GOSUB) {
4773 startpoint = scan + ARG2L(scan);
4774 ST.close_paren = ARG(scan);
4776 startpoint = rei->program+1;
4779 goto eval_recurse_doit;
4780 assert(0); /* NOTREACHED */
4782 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4783 if (cur_eval && cur_eval->locinput==locinput) {
4784 if ( ++nochange_depth > max_nochange_depth )
4785 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4790 /* execute the code in the {...} */
4794 OP * const oop = PL_op;
4795 COP * const ocurcop = PL_curcop;
4797 char *saved_regeol = PL_regeol;
4798 struct re_save_state saved_state;
4801 /* save *all* paren positions */
4803 REGCP_SET(runops_cp);
4805 /* To not corrupt the existing regex state while executing the
4806 * eval we would normally put it on the save stack, like with
4807 * save_re_context. However, re-evals have a weird scoping so we
4808 * can't just add ENTER/LEAVE here. With that, things like
4810 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4812 * would break, as they expect the localisation to be unwound
4813 * only when the re-engine backtracks through the bit that
4816 * What we do instead is just saving the state in a local c
4819 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4821 PL_reg_state.re_reparsing = FALSE;
4824 caller_cv = find_runcv(NULL);
4828 if (rexi->data->what[n] == 'r') { /* code from an external qr */
4830 (REGEXP*)(rexi->data->data[n])
4833 nop = (OP*)rexi->data->data[n+1];
4835 else if (rexi->data->what[n] == 'l') { /* literal code */
4837 nop = (OP*)rexi->data->data[n];
4838 assert(CvDEPTH(newcv));
4841 /* literal with own CV */
4842 assert(rexi->data->what[n] == 'L');
4843 newcv = rex->qr_anoncv;
4844 nop = (OP*)rexi->data->data[n];
4847 /* normally if we're about to execute code from the same
4848 * CV that we used previously, we just use the existing
4849 * CX stack entry. However, its possible that in the
4850 * meantime we may have backtracked, popped from the save
4851 * stack, and undone the SAVECOMPPAD(s) associated with
4852 * PUSH_MULTICALL; in which case PL_comppad no longer
4853 * points to newcv's pad. */
4854 if (newcv != last_pushed_cv || PL_comppad != last_pad)
4856 I32 depth = (newcv == caller_cv) ? 0 : 1;
4857 if (last_pushed_cv) {
4858 CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
4861 PUSH_MULTICALL_WITHDEPTH(newcv, depth);
4863 last_pushed_cv = newcv;
4865 last_pad = PL_comppad;
4867 /* the initial nextstate you would normally execute
4868 * at the start of an eval (which would cause error
4869 * messages to come from the eval), may be optimised
4870 * away from the execution path in the regex code blocks;
4871 * so manually set PL_curcop to it initially */
4873 OP *o = cUNOPx(nop)->op_first;
4874 assert(o->op_type == OP_NULL);
4875 if (o->op_targ == OP_SCOPE) {
4876 o = cUNOPo->op_first;
4879 assert(o->op_targ == OP_LEAVE);
4880 o = cUNOPo->op_first;
4881 assert(o->op_type == OP_ENTER);
4885 if (o->op_type != OP_STUB) {
4886 assert( o->op_type == OP_NEXTSTATE
4887 || o->op_type == OP_DBSTATE
4888 || (o->op_type == OP_NULL
4889 && ( o->op_targ == OP_NEXTSTATE
4890 || o->op_targ == OP_DBSTATE
4894 PL_curcop = (COP*)o;
4899 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4900 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
4902 rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4905 SV *sv_mrk = get_sv("REGMARK", 1);
4906 sv_setsv(sv_mrk, sv_yes_mark);
4909 /* we don't use MULTICALL here as we want to call the
4910 * first op of the block of interest, rather than the
4911 * first op of the sub */
4912 before = SP-PL_stack_base;
4914 CALLRUNOPS(aTHX); /* Scalar context. */
4916 if (SP-PL_stack_base == before)
4917 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4923 /* before restoring everything, evaluate the returned
4924 * value, so that 'uninit' warnings don't use the wrong
4925 * PL_op or pad. Also need to process any magic vars
4926 * (e.g. $1) *before* parentheses are restored */
4931 if (logical == 0) /* (?{})/ */
4932 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
4933 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
4934 sw = cBOOL(SvTRUE(ret));
4937 else { /* /(??{}) */
4938 /* if its overloaded, let the regex compiler handle
4939 * it; otherwise extract regex, or stringify */
4940 if (!SvAMAGIC(ret)) {
4944 if (SvTYPE(sv) == SVt_REGEXP)
4945 re_sv = (REGEXP*) sv;
4946 else if (SvSMAGICAL(sv)) {
4947 MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
4949 re_sv = (REGEXP *) mg->mg_obj;
4952 /* force any magic, undef warnings here */
4954 ret = sv_mortalcopy(ret);
4955 (void) SvPV_force_nolen(ret);
4961 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4963 /* *** Note that at this point we don't restore
4964 * PL_comppad, (or pop the CxSUB) on the assumption it may
4965 * be used again soon. This is safe as long as nothing
4966 * in the regexp code uses the pad ! */
4968 PL_curcop = ocurcop;
4969 PL_regeol = saved_regeol;
4970 S_regcp_restore(aTHX_ rex, runops_cp);
4976 /* only /(??{})/ from now on */
4979 /* extract RE object from returned value; compiling if
4983 re_sv = reg_temp_copy(NULL, re_sv);
4987 const I32 osize = PL_regsize;
4989 if (SvUTF8(ret) && IN_BYTES) {
4990 /* In use 'bytes': make a copy of the octet
4991 * sequence, but without the flag on */
4993 const char *const p = SvPV(ret, len);
4994 ret = newSVpvn_flags(p, len, SVs_TEMP);
4996 if (rex->intflags & PREGf_USE_RE_EVAL)
4997 pm_flags |= PMf_USE_RE_EVAL;
4999 /* if we got here, it should be an engine which
5000 * supports compiling code blocks and stuff */
5001 assert(rex->engine && rex->engine->op_comp);
5002 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5003 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5004 rex->engine, NULL, NULL,
5005 /* copy /msix etc to inner pattern */
5010 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
5012 /* This isn't a first class regexp. Instead, it's
5013 caching a regexp onto an existing, Perl visible
5015 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5018 /* safe to do now that any $1 etc has been
5019 * interpolated into the new pattern string and
5021 S_regcp_restore(aTHX_ rex, runops_cp);
5026 RXp_MATCH_COPIED_off(re);
5027 re->subbeg = rex->subbeg;
5028 re->sublen = rex->sublen;
5029 re->suboffset = rex->suboffset;
5030 re->subcoffset = rex->subcoffset;
5033 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
5034 "Matching embedded");
5036 startpoint = rei->program + 1;
5037 ST.close_paren = 0; /* only used for GOSUB */
5039 eval_recurse_doit: /* Share code with GOSUB below this line */
5040 /* run the pattern returned from (??{...}) */
5041 ST.cp = regcppush(rex, 0); /* Save *all* the positions. */
5042 REGCP_SET(ST.lastcp);
5045 re->lastcloseparen = 0;
5049 /* XXXX This is too dramatic a measure... */
5052 ST.toggle_reg_flags = PL_reg_flags;
5054 PL_reg_flags |= RF_utf8;
5056 PL_reg_flags &= ~RF_utf8;
5057 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
5059 ST.prev_rex = rex_sv;
5060 ST.prev_curlyx = cur_curlyx;
5062 SET_reg_curpm(rex_sv);
5067 ST.prev_eval = cur_eval;
5069 /* now continue from first node in postoned RE */
5070 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5071 assert(0); /* NOTREACHED */
5074 case EVAL_AB: /* cleanup after a successful (??{A})B */
5075 /* note: this is called twice; first after popping B, then A */
5076 PL_reg_flags ^= ST.toggle_reg_flags;
5077 rex_sv = ST.prev_rex;
5078 SET_reg_curpm(rex_sv);
5079 rex = ReANY(rex_sv);
5080 rexi = RXi_GET(rex);
5082 cur_eval = ST.prev_eval;
5083 cur_curlyx = ST.prev_curlyx;
5085 /* XXXX This is too dramatic a measure... */
5087 if ( nochange_depth )
5092 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5093 /* note: this is called twice; first after popping B, then A */
5094 PL_reg_flags ^= ST.toggle_reg_flags;
5095 rex_sv = ST.prev_rex;
5096 SET_reg_curpm(rex_sv);
5097 rex = ReANY(rex_sv);
5098 rexi = RXi_GET(rex);
5100 REGCP_UNWIND(ST.lastcp);
5102 cur_eval = ST.prev_eval;
5103 cur_curlyx = ST.prev_curlyx;
5104 /* XXXX This is too dramatic a measure... */
5106 if ( nochange_depth )
5112 n = ARG(scan); /* which paren pair */
5113 rex->offs[n].start_tmp = locinput - PL_bostr;
5116 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5117 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; regsize=%"UVuf"\n",
5121 (IV)rex->offs[n].start_tmp,
5127 /* XXX really need to log other places start/end are set too */
5128 #define CLOSE_CAPTURE \
5129 rex->offs[n].start = rex->offs[n].start_tmp; \
5130 rex->offs[n].end = locinput - PL_bostr; \
5131 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5132 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5134 PTR2UV(rex->offs), \
5136 (IV)rex->offs[n].start, \
5137 (IV)rex->offs[n].end \
5141 n = ARG(scan); /* which paren pair */
5143 /*if (n > PL_regsize)
5145 if (n > rex->lastparen)
5147 rex->lastcloseparen = n;
5148 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5153 case ACCEPT: /* (*ACCEPT) */
5157 cursor && OP(cursor)!=END;
5158 cursor=regnext(cursor))
5160 if ( OP(cursor)==CLOSE ){
5162 if ( n <= lastopen ) {
5164 /*if (n > PL_regsize)
5166 if (n > rex->lastparen)
5168 rex->lastcloseparen = n;
5169 if ( n == ARG(scan) || (cur_eval &&
5170 cur_eval->u.eval.close_paren == n))
5179 case GROUPP: /* (?(1)) */
5180 n = ARG(scan); /* which paren pair */
5181 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5184 case NGROUPP: /* (?(<name>)) */
5185 /* reg_check_named_buff_matched returns 0 for no match */
5186 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5189 case INSUBP: /* (?(R)) */
5191 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5194 case DEFINEP: /* (?(DEFINE)) */
5198 case IFTHEN: /* (?(cond)A|B) */
5199 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
5201 next = NEXTOPER(NEXTOPER(scan));
5203 next = scan + ARG(scan);
5204 if (OP(next) == IFTHEN) /* Fake one. */
5205 next = NEXTOPER(NEXTOPER(next));
5209 case LOGICAL: /* modifier for EVAL and IFMATCH */
5210 logical = scan->flags;
5213 /*******************************************************************
5215 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5216 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5217 STAR/PLUS/CURLY/CURLYN are used instead.)
5219 A*B is compiled as <CURLYX><A><WHILEM><B>
5221 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5222 state, which contains the current count, initialised to -1. It also sets
5223 cur_curlyx to point to this state, with any previous value saved in the
5226 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5227 since the pattern may possibly match zero times (i.e. it's a while {} loop
5228 rather than a do {} while loop).
5230 Each entry to WHILEM represents a successful match of A. The count in the
5231 CURLYX block is incremented, another WHILEM state is pushed, and execution
5232 passes to A or B depending on greediness and the current count.
5234 For example, if matching against the string a1a2a3b (where the aN are
5235 substrings that match /A/), then the match progresses as follows: (the
5236 pushed states are interspersed with the bits of strings matched so far):
5239 <CURLYX cnt=0><WHILEM>
5240 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5241 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5242 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5243 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5245 (Contrast this with something like CURLYM, which maintains only a single
5249 a1 <CURLYM cnt=1> a2
5250 a1 a2 <CURLYM cnt=2> a3
5251 a1 a2 a3 <CURLYM cnt=3> b
5254 Each WHILEM state block marks a point to backtrack to upon partial failure
5255 of A or B, and also contains some minor state data related to that
5256 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5257 overall state, such as the count, and pointers to the A and B ops.
5259 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5260 must always point to the *current* CURLYX block, the rules are:
5262 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5263 and set cur_curlyx to point the new block.
5265 When popping the CURLYX block after a successful or unsuccessful match,
5266 restore the previous cur_curlyx.
5268 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5269 to the outer one saved in the CURLYX block.
5271 When popping the WHILEM block after a successful or unsuccessful B match,
5272 restore the previous cur_curlyx.
5274 Here's an example for the pattern (AI* BI)*BO
5275 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5278 curlyx backtrack stack
5279 ------ ---------------
5281 CO <CO prev=NULL> <WO>
5282 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5283 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5284 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5286 At this point the pattern succeeds, and we work back down the stack to
5287 clean up, restoring as we go:
5289 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5290 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5291 CO <CO prev=NULL> <WO>
5294 *******************************************************************/
5296 #define ST st->u.curlyx
5298 case CURLYX: /* start of /A*B/ (for complex A) */
5300 /* No need to save/restore up to this paren */
5301 I32 parenfloor = scan->flags;
5303 assert(next); /* keep Coverity happy */
5304 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5307 /* XXXX Probably it is better to teach regpush to support
5308 parenfloor > PL_regsize... */
5309 if (parenfloor > (I32)rex->lastparen)
5310 parenfloor = rex->lastparen; /* Pessimization... */
5312 ST.prev_curlyx= cur_curlyx;
5314 ST.cp = PL_savestack_ix;
5316 /* these fields contain the state of the current curly.
5317 * they are accessed by subsequent WHILEMs */
5318 ST.parenfloor = parenfloor;
5323 ST.count = -1; /* this will be updated by WHILEM */
5324 ST.lastloc = NULL; /* this will be updated by WHILEM */
5326 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5327 assert(0); /* NOTREACHED */
5330 case CURLYX_end: /* just finished matching all of A*B */
5331 cur_curlyx = ST.prev_curlyx;
5333 assert(0); /* NOTREACHED */
5335 case CURLYX_end_fail: /* just failed to match all of A*B */
5337 cur_curlyx = ST.prev_curlyx;
5339 assert(0); /* NOTREACHED */
5343 #define ST st->u.whilem
5345 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5347 /* see the discussion above about CURLYX/WHILEM */
5349 int min = ARG1(cur_curlyx->u.curlyx.me);
5350 int max = ARG2(cur_curlyx->u.curlyx.me);
5351 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5353 assert(cur_curlyx); /* keep Coverity happy */
5354 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5355 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5356 ST.cache_offset = 0;
5360 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5361 "%*s whilem: matched %ld out of %d..%d\n",
5362 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5365 /* First just match a string of min A's. */
5368 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5369 cur_curlyx->u.curlyx.lastloc = locinput;
5370 REGCP_SET(ST.lastcp);
5372 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5373 assert(0); /* NOTREACHED */
5376 /* If degenerate A matches "", assume A done. */
5378 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5379 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5380 "%*s whilem: empty match detected, trying continuation...\n",
5381 REPORT_CODE_OFF+depth*2, "")
5383 goto do_whilem_B_max;
5386 /* super-linear cache processing */
5390 if (!PL_reg_maxiter) {
5391 /* start the countdown: Postpone detection until we
5392 * know the match is not *that* much linear. */
5393 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
5394 /* possible overflow for long strings and many CURLYX's */
5395 if (PL_reg_maxiter < 0)
5396 PL_reg_maxiter = I32_MAX;
5397 PL_reg_leftiter = PL_reg_maxiter;
5400 if (PL_reg_leftiter-- == 0) {
5401 /* initialise cache */
5402 const I32 size = (PL_reg_maxiter + 7)/8;
5403 if (PL_reg_poscache) {
5404 if ((I32)PL_reg_poscache_size < size) {
5405 Renew(PL_reg_poscache, size, char);
5406 PL_reg_poscache_size = size;
5408 Zero(PL_reg_poscache, size, char);
5411 PL_reg_poscache_size = size;
5412 Newxz(PL_reg_poscache, size, char);
5414 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5415 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5416 PL_colors[4], PL_colors[5])
5420 if (PL_reg_leftiter < 0) {
5421 /* have we already failed at this position? */
5423 offset = (scan->flags & 0xf) - 1
5424 + (locinput - PL_bostr) * (scan->flags>>4);
5425 mask = 1 << (offset % 8);
5427 if (PL_reg_poscache[offset] & mask) {
5428 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5429 "%*s whilem: (cache) already tried at this position...\n",
5430 REPORT_CODE_OFF+depth*2, "")
5432 sayNO; /* cache records failure */
5434 ST.cache_offset = offset;
5435 ST.cache_mask = mask;
5439 /* Prefer B over A for minimal matching. */
5441 if (cur_curlyx->u.curlyx.minmod) {
5442 ST.save_curlyx = cur_curlyx;
5443 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5444 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor);
5445 REGCP_SET(ST.lastcp);
5446 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5448 assert(0); /* NOTREACHED */
5451 /* Prefer A over B for maximal matching. */
5453 if (n < max) { /* More greed allowed? */
5454 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5455 cur_curlyx->u.curlyx.lastloc = locinput;
5456 REGCP_SET(ST.lastcp);
5457 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5458 assert(0); /* NOTREACHED */
5460 goto do_whilem_B_max;
5462 assert(0); /* NOTREACHED */
5464 case WHILEM_B_min: /* just matched B in a minimal match */
5465 case WHILEM_B_max: /* just matched B in a maximal match */
5466 cur_curlyx = ST.save_curlyx;
5468 assert(0); /* NOTREACHED */
5470 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5471 cur_curlyx = ST.save_curlyx;
5472 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5473 cur_curlyx->u.curlyx.count--;
5475 assert(0); /* NOTREACHED */
5477 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5479 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5480 REGCP_UNWIND(ST.lastcp);
5482 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5483 cur_curlyx->u.curlyx.count--;
5485 assert(0); /* NOTREACHED */
5487 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5488 REGCP_UNWIND(ST.lastcp);
5489 regcppop(rex); /* Restore some previous $<digit>s? */
5490 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5491 "%*s whilem: failed, trying continuation...\n",
5492 REPORT_CODE_OFF+depth*2, "")
5495 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5496 && ckWARN(WARN_REGEXP)
5497 && !(PL_reg_flags & RF_warned))
5499 PL_reg_flags |= RF_warned;
5500 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5501 "Complex regular subexpression recursion limit (%d) "
5507 ST.save_curlyx = cur_curlyx;
5508 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5509 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5511 assert(0); /* NOTREACHED */
5513 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
5514 cur_curlyx = ST.save_curlyx;
5515 REGCP_UNWIND(ST.lastcp);
5518 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
5519 /* Maximum greed exceeded */
5520 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5521 && ckWARN(WARN_REGEXP)
5522 && !(PL_reg_flags & RF_warned))
5524 PL_reg_flags |= RF_warned;
5525 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5526 "Complex regular subexpression recursion "
5527 "limit (%d) exceeded",
5530 cur_curlyx->u.curlyx.count--;
5534 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5535 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
5537 /* Try grabbing another A and see if it helps. */
5538 cur_curlyx->u.curlyx.lastloc = locinput;
5539 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
5540 REGCP_SET(ST.lastcp);
5541 PUSH_STATE_GOTO(WHILEM_A_min,
5542 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
5544 assert(0); /* NOTREACHED */
5547 #define ST st->u.branch
5549 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
5550 next = scan + ARG(scan);
5553 scan = NEXTOPER(scan);
5556 case BRANCH: /* /(...|A|...)/ */
5557 scan = NEXTOPER(scan); /* scan now points to inner node */
5558 ST.lastparen = rex->lastparen;
5559 ST.lastcloseparen = rex->lastcloseparen;
5560 ST.next_branch = next;
5563 /* Now go into the branch */
5565 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5567 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5569 assert(0); /* NOTREACHED */
5571 case CUTGROUP: /* /(*THEN)/ */
5572 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
5573 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5574 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
5575 assert(0); /* NOTREACHED */
5577 case CUTGROUP_next_fail:
5580 if (st->u.mark.mark_name)
5581 sv_commit = st->u.mark.mark_name;
5583 assert(0); /* NOTREACHED */
5587 assert(0); /* NOTREACHED */
5589 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5594 REGCP_UNWIND(ST.cp);
5595 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5596 scan = ST.next_branch;
5597 /* no more branches? */
5598 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
5600 PerlIO_printf( Perl_debug_log,
5601 "%*s %sBRANCH failed...%s\n",
5602 REPORT_CODE_OFF+depth*2, "",
5608 continue; /* execute next BRANCH[J] op */
5609 assert(0); /* NOTREACHED */
5611 case MINMOD: /* next op will be non-greedy, e.g. A*? */
5616 #define ST st->u.curlym
5618 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
5620 /* This is an optimisation of CURLYX that enables us to push
5621 * only a single backtracking state, no matter how many matches
5622 * there are in {m,n}. It relies on the pattern being constant
5623 * length, with no parens to influence future backrefs
5627 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5629 ST.lastparen = rex->lastparen;
5630 ST.lastcloseparen = rex->lastcloseparen;
5632 /* if paren positive, emulate an OPEN/CLOSE around A */
5634 U32 paren = ST.me->flags;
5635 if (paren > PL_regsize)
5637 scan += NEXT_OFF(scan); /* Skip former OPEN. */
5645 ST.c1 = CHRTEST_UNINIT;
5648 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
5651 curlym_do_A: /* execute the A in /A{m,n}B/ */
5652 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
5653 assert(0); /* NOTREACHED */
5655 case CURLYM_A: /* we've just matched an A */
5657 /* after first match, determine A's length: u.curlym.alen */
5658 if (ST.count == 1) {
5659 if (PL_reg_match_utf8) {
5660 char *s = st->locinput;
5661 while (s < locinput) {
5667 ST.alen = locinput - st->locinput;
5670 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
5673 PerlIO_printf(Perl_debug_log,
5674 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5675 (int)(REPORT_CODE_OFF+(depth*2)), "",
5676 (IV) ST.count, (IV)ST.alen)
5679 if (cur_eval && cur_eval->u.eval.close_paren &&
5680 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5684 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
5685 if ( max == REG_INFTY || ST.count < max )
5686 goto curlym_do_A; /* try to match another A */
5688 goto curlym_do_B; /* try to match B */
5690 case CURLYM_A_fail: /* just failed to match an A */
5691 REGCP_UNWIND(ST.cp);
5693 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
5694 || (cur_eval && cur_eval->u.eval.close_paren &&
5695 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
5698 curlym_do_B: /* execute the B in /A{m,n}B/ */
5699 if (ST.c1 == CHRTEST_UNINIT) {
5700 /* calculate c1 and c2 for possible match of 1st char
5701 * following curly */
5702 ST.c1 = ST.c2 = CHRTEST_VOID;
5703 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
5704 regnode *text_node = ST.B;
5705 if (! HAS_TEXT(text_node))
5706 FIND_NEXT_IMPT(text_node);
5709 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
5711 But the former is redundant in light of the latter.
5713 if this changes back then the macro for
5714 IS_TEXT and friends need to change.
5716 if (PL_regkind[OP(text_node)] == EXACT) {
5717 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5718 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8))
5727 PerlIO_printf(Perl_debug_log,
5728 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5729 (int)(REPORT_CODE_OFF+(depth*2)),
5732 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
5733 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
5734 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
5735 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
5737 /* simulate B failing */
5739 PerlIO_printf(Perl_debug_log,
5740 "%*s CURLYM Fast bail next target=U+%"UVXf" c1=U+%"UVXf" c2=U+%"UVXf"\n",
5741 (int)(REPORT_CODE_OFF+(depth*2)),"",
5742 valid_utf8_to_uvchr((U8 *) locinput, NULL),
5743 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
5744 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
5746 state_num = CURLYM_B_fail;
5747 goto reenter_switch;
5750 else if (nextchr != ST.c1 && nextchr != ST.c2) {
5751 /* simulate B failing */
5753 PerlIO_printf(Perl_debug_log,
5754 "%*s CURLYM Fast bail next target=U+%X c1=U+%X c2=U+%X\n",
5755 (int)(REPORT_CODE_OFF+(depth*2)),"",
5756 (int) nextchr, ST.c1, ST.c2)
5758 state_num = CURLYM_B_fail;
5759 goto reenter_switch;
5764 /* emulate CLOSE: mark current A as captured */
5765 I32 paren = ST.me->flags;
5767 rex->offs[paren].start
5768 = HOPc(locinput, -ST.alen) - PL_bostr;
5769 rex->offs[paren].end = locinput - PL_bostr;
5770 if ((U32)paren > rex->lastparen)
5771 rex->lastparen = paren;
5772 rex->lastcloseparen = paren;
5775 rex->offs[paren].end = -1;
5776 if (cur_eval && cur_eval->u.eval.close_paren &&
5777 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5786 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
5787 assert(0); /* NOTREACHED */
5789 case CURLYM_B_fail: /* just failed to match a B */
5790 REGCP_UNWIND(ST.cp);
5791 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5793 I32 max = ARG2(ST.me);
5794 if (max != REG_INFTY && ST.count == max)
5796 goto curlym_do_A; /* try to match a further A */
5798 /* backtrack one A */
5799 if (ST.count == ARG1(ST.me) /* min */)
5802 SET_locinput(HOPc(locinput, -ST.alen));
5803 goto curlym_do_B; /* try to match B */
5806 #define ST st->u.curly
5808 #define CURLY_SETPAREN(paren, success) \
5811 rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5812 rex->offs[paren].end = locinput - PL_bostr; \
5813 if (paren > rex->lastparen) \
5814 rex->lastparen = paren; \
5815 rex->lastcloseparen = paren; \
5818 rex->offs[paren].end = -1; \
5819 rex->lastparen = ST.lastparen; \
5820 rex->lastcloseparen = ST.lastcloseparen; \
5824 case STAR: /* /A*B/ where A is width 1 char */
5828 scan = NEXTOPER(scan);
5831 case PLUS: /* /A+B/ where A is width 1 char */
5835 scan = NEXTOPER(scan);
5838 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
5839 ST.paren = scan->flags; /* Which paren to set */
5840 ST.lastparen = rex->lastparen;
5841 ST.lastcloseparen = rex->lastcloseparen;
5842 if (ST.paren > PL_regsize)
5843 PL_regsize = ST.paren;
5844 ST.min = ARG1(scan); /* min to match */
5845 ST.max = ARG2(scan); /* max to match */
5846 if (cur_eval && cur_eval->u.eval.close_paren &&
5847 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5851 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5854 case CURLY: /* /A{m,n}B/ where A is width 1 char */
5856 ST.min = ARG1(scan); /* min to match */
5857 ST.max = ARG2(scan); /* max to match */
5858 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5861 * Lookahead to avoid useless match attempts
5862 * when we know what character comes next.
5864 * Used to only do .*x and .*?x, but now it allows
5865 * for )'s, ('s and (?{ ... })'s to be in the way
5866 * of the quantifier and the EXACT-like node. -- japhy
5869 assert(ST.min <= ST.max);
5870 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
5871 ST.c1 = ST.c2 = CHRTEST_VOID;
5874 regnode *text_node = next;
5876 if (! HAS_TEXT(text_node))
5877 FIND_NEXT_IMPT(text_node);
5879 if (! HAS_TEXT(text_node))
5880 ST.c1 = ST.c2 = CHRTEST_VOID;
5882 if ( PL_regkind[OP(text_node)] != EXACT ) {
5883 ST.c1 = ST.c2 = CHRTEST_VOID;
5887 /* Currently we only get here when
5889 PL_rekind[OP(text_node)] == EXACT
5891 if this changes back then the macro for IS_TEXT and
5892 friends need to change. */
5893 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
5894 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8))
5905 char *li = locinput;
5907 if (ST.min && regrepeat(rex, &li, ST.A, ST.min, depth) < ST.min)
5912 if (ST.c1 == CHRTEST_VOID)
5913 goto curly_try_B_min;
5915 ST.oldloc = locinput;
5917 /* set ST.maxpos to the furthest point along the
5918 * string that could possibly match */
5919 if (ST.max == REG_INFTY) {
5920 ST.maxpos = PL_regeol - 1;
5922 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5925 else if (utf8_target) {
5926 int m = ST.max - ST.min;
5927 for (ST.maxpos = locinput;
5928 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5929 ST.maxpos += UTF8SKIP(ST.maxpos);
5932 ST.maxpos = locinput + ST.max - ST.min;
5933 if (ST.maxpos >= PL_regeol)
5934 ST.maxpos = PL_regeol - 1;
5936 goto curly_try_B_min_known;
5940 /* avoid taking address of locinput, so it can remain
5942 char *li = locinput;
5943 ST.count = regrepeat(rex, &li, ST.A, ST.max, depth);
5944 if (ST.count < ST.min)
5947 if ((ST.count > ST.min)
5948 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5950 /* A{m,n} must come at the end of the string, there's
5951 * no point in backing off ... */
5953 /* ...except that $ and \Z can match before *and* after
5954 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5955 We may back off by one in this case. */
5956 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
5960 goto curly_try_B_max;
5962 assert(0); /* NOTREACHED */
5965 case CURLY_B_min_known_fail:
5966 /* failed to find B in a non-greedy match where c1,c2 valid */
5968 REGCP_UNWIND(ST.cp);
5970 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5972 /* Couldn't or didn't -- move forward. */
5973 ST.oldloc = locinput;
5975 locinput += UTF8SKIP(locinput);
5979 curly_try_B_min_known:
5980 /* find the next place where 'B' could work, then call B */
5984 n = (ST.oldloc == locinput) ? 0 : 1;
5985 if (ST.c1 == ST.c2) {
5986 /* set n to utf8_distance(oldloc, locinput) */
5987 while (locinput <= ST.maxpos
5988 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
5990 locinput += UTF8SKIP(locinput);
5995 /* set n to utf8_distance(oldloc, locinput) */
5996 while (locinput <= ST.maxpos
5997 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
5998 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6000 locinput += UTF8SKIP(locinput);
6005 else { /* Not utf8_target */
6006 if (ST.c1 == ST.c2) {
6007 while (locinput <= ST.maxpos &&
6008 UCHARAT(locinput) != ST.c1)
6012 while (locinput <= ST.maxpos
6013 && UCHARAT(locinput) != ST.c1
6014 && UCHARAT(locinput) != ST.c2)
6017 n = locinput - ST.oldloc;
6019 if (locinput > ST.maxpos)
6022 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6023 * at b; check that everything between oldloc and
6024 * locinput matches */
6025 char *li = ST.oldloc;
6027 if (regrepeat(rex, &li, ST.A, n, depth) < n)
6029 assert(n == REG_INFTY || locinput == li);
6031 CURLY_SETPAREN(ST.paren, ST.count);
6032 if (cur_eval && cur_eval->u.eval.close_paren &&
6033 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6036 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6038 assert(0); /* NOTREACHED */
6041 case CURLY_B_min_fail:
6042 /* failed to find B in a non-greedy match where c1,c2 invalid */
6044 REGCP_UNWIND(ST.cp);
6046 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6048 /* failed -- move forward one */
6050 char *li = locinput;
6051 if (!regrepeat(rex, &li, ST.A, 1, depth)) {
6058 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6059 ST.count > 0)) /* count overflow ? */
6062 CURLY_SETPAREN(ST.paren, ST.count);
6063 if (cur_eval && cur_eval->u.eval.close_paren &&
6064 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6067 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6071 assert(0); /* NOTREACHED */
6075 /* a successful greedy match: now try to match B */
6076 if (cur_eval && cur_eval->u.eval.close_paren &&
6077 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6081 bool could_match = locinput < PL_regeol;
6083 /* If it could work, try it. */
6084 if (ST.c1 != CHRTEST_VOID && could_match) {
6085 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6087 could_match = memEQ(locinput,
6092 UTF8SKIP(locinput));
6095 could_match = UCHARAT(locinput) == ST.c1
6096 || UCHARAT(locinput) == ST.c2;
6099 if (ST.c1 == CHRTEST_VOID || could_match) {
6100 CURLY_SETPAREN(ST.paren, ST.count);
6101 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6102 assert(0); /* NOTREACHED */
6107 case CURLY_B_max_fail:
6108 /* failed to find B in a greedy match */
6110 REGCP_UNWIND(ST.cp);
6112 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6115 if (--ST.count < ST.min)
6117 locinput = HOPc(locinput, -1);
6118 goto curly_try_B_max;
6122 case END: /* last op of main pattern */
6125 /* we've just finished A in /(??{A})B/; now continue with B */
6126 st->u.eval.toggle_reg_flags
6127 = cur_eval->u.eval.toggle_reg_flags;
6128 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
6130 st->u.eval.prev_rex = rex_sv; /* inner */
6131 st->u.eval.cp = regcppush(rex, 0); /* Save *all* the positions. */
6132 rex_sv = cur_eval->u.eval.prev_rex;
6133 SET_reg_curpm(rex_sv);
6134 rex = ReANY(rex_sv);
6135 rexi = RXi_GET(rex);
6136 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6138 REGCP_SET(st->u.eval.lastcp);
6140 /* Restore parens of the outer rex without popping the
6142 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp);
6144 st->u.eval.prev_eval = cur_eval;
6145 cur_eval = cur_eval->u.eval.prev_eval;
6147 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6148 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6149 if ( nochange_depth )
6152 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6153 locinput); /* match B */
6156 if (locinput < reginfo->till) {
6157 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6158 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6160 (long)(locinput - PL_reg_starttry),
6161 (long)(reginfo->till - PL_reg_starttry),
6164 sayNO_SILENT; /* Cannot match: too short. */
6166 sayYES; /* Success! */
6168 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6170 PerlIO_printf(Perl_debug_log,
6171 "%*s %ssubpattern success...%s\n",
6172 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6173 sayYES; /* Success! */
6176 #define ST st->u.ifmatch
6181 case SUSPEND: /* (?>A) */
6183 newstart = locinput;
6186 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6188 goto ifmatch_trivial_fail_test;
6190 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6192 ifmatch_trivial_fail_test:
6194 char * const s = HOPBACKc(locinput, scan->flags);
6199 sw = 1 - cBOOL(ST.wanted);
6203 next = scan + ARG(scan);
6211 newstart = locinput;
6215 ST.logical = logical;
6216 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6218 /* execute body of (?...A) */
6219 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6220 assert(0); /* NOTREACHED */
6223 case IFMATCH_A_fail: /* body of (?...A) failed */
6224 ST.wanted = !ST.wanted;
6227 case IFMATCH_A: /* body of (?...A) succeeded */
6229 sw = cBOOL(ST.wanted);
6231 else if (!ST.wanted)
6234 if (OP(ST.me) != SUSPEND) {
6235 /* restore old position except for (?>...) */
6236 locinput = st->locinput;
6238 scan = ST.me + ARG(ST.me);
6241 continue; /* execute B */
6245 case LONGJMP: /* alternative with many branches compiles to
6246 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6247 next = scan + ARG(scan);
6252 case COMMIT: /* (*COMMIT) */
6253 reginfo->cutpoint = PL_regeol;
6256 case PRUNE: /* (*PRUNE) */
6258 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6259 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6260 assert(0); /* NOTREACHED */
6262 case COMMIT_next_fail:
6266 case OPFAIL: /* (*FAIL) */
6268 assert(0); /* NOTREACHED */
6270 #define ST st->u.mark
6271 case MARKPOINT: /* (*MARK:foo) */
6272 ST.prev_mark = mark_state;
6273 ST.mark_name = sv_commit = sv_yes_mark
6274 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6276 ST.mark_loc = locinput;
6277 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6278 assert(0); /* NOTREACHED */
6280 case MARKPOINT_next:
6281 mark_state = ST.prev_mark;
6283 assert(0); /* NOTREACHED */
6285 case MARKPOINT_next_fail:
6286 if (popmark && sv_eq(ST.mark_name,popmark))
6288 if (ST.mark_loc > startpoint)
6289 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6290 popmark = NULL; /* we found our mark */
6291 sv_commit = ST.mark_name;
6294 PerlIO_printf(Perl_debug_log,
6295 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6296 REPORT_CODE_OFF+depth*2, "",
6297 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6300 mark_state = ST.prev_mark;
6301 sv_yes_mark = mark_state ?
6302 mark_state->u.mark.mark_name : NULL;
6304 assert(0); /* NOTREACHED */
6306 case SKIP: /* (*SKIP) */
6308 /* (*SKIP) : if we fail we cut here*/
6309 ST.mark_name = NULL;
6310 ST.mark_loc = locinput;
6311 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6313 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6314 otherwise do nothing. Meaning we need to scan
6316 regmatch_state *cur = mark_state;
6317 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6320 if ( sv_eq( cur->u.mark.mark_name,
6323 ST.mark_name = find;
6324 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6326 cur = cur->u.mark.prev_mark;
6329 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6332 case SKIP_next_fail:
6334 /* (*CUT:NAME) - Set up to search for the name as we
6335 collapse the stack*/
6336 popmark = ST.mark_name;
6338 /* (*CUT) - No name, we cut here.*/
6339 if (ST.mark_loc > startpoint)
6340 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6341 /* but we set sv_commit to latest mark_name if there
6342 is one so they can test to see how things lead to this
6345 sv_commit=mark_state->u.mark.mark_name;
6349 assert(0); /* NOTREACHED */
6352 case LNBREAK: /* \R */
6353 if ((n=is_LNBREAK_safe(locinput, PL_regeol, utf8_target))) {
6359 #define CASE_CLASS(nAmE) \
6361 if (NEXTCHR_IS_EOS) \
6363 if ((n=is_##nAmE(locinput,utf8_target))) { \
6369 if (NEXTCHR_IS_EOS) \
6371 if ((n=is_##nAmE(locinput,utf8_target))) { \
6374 locinput += UTF8SKIP(locinput); \
6378 CASE_CLASS(VERTWS); /* \v \V */
6379 CASE_CLASS(HORIZWS); /* \h \H */
6383 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6384 PTR2UV(scan), OP(scan));
6385 Perl_croak(aTHX_ "regexp memory corruption");
6387 /* this is a point to jump to in order to increment
6388 * locinput by one character */
6390 assert(!NEXTCHR_IS_EOS);
6392 locinput += PL_utf8skip[nextchr];
6393 /* locinput is allowed to go 1 char off the end, but not 2+ */
6394 if (locinput > PL_regeol)
6403 /* switch break jumps here */
6404 scan = next; /* prepare to execute the next op and ... */
6405 continue; /* ... jump back to the top, reusing st */
6406 assert(0); /* NOTREACHED */
6409 /* push a state that backtracks on success */
6410 st->u.yes.prev_yes_state = yes_state;
6414 /* push a new regex state, then continue at scan */
6416 regmatch_state *newst;
6419 regmatch_state *cur = st;
6420 regmatch_state *curyes = yes_state;
6422 regmatch_slab *slab = PL_regmatch_slab;
6423 for (;curd > -1;cur--,curd--) {
6424 if (cur < SLAB_FIRST(slab)) {
6426 cur = SLAB_LAST(slab);
6428 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6429 REPORT_CODE_OFF + 2 + depth * 2,"",
6430 curd, PL_reg_name[cur->resume_state],
6431 (curyes == cur) ? "yes" : ""
6434 curyes = cur->u.yes.prev_yes_state;
6437 DEBUG_STATE_pp("push")
6440 st->locinput = locinput;
6442 if (newst > SLAB_LAST(PL_regmatch_slab))
6443 newst = S_push_slab(aTHX);
6444 PL_regmatch_state = newst;
6446 locinput = pushinput;
6449 assert(0); /* NOTREACHED */
6454 * We get here only if there's trouble -- normally "case END" is
6455 * the terminating point.
6457 Perl_croak(aTHX_ "corrupted regexp pointers");
6463 /* we have successfully completed a subexpression, but we must now
6464 * pop to the state marked by yes_state and continue from there */
6465 assert(st != yes_state);
6467 while (st != yes_state) {
6469 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6470 PL_regmatch_slab = PL_regmatch_slab->prev;
6471 st = SLAB_LAST(PL_regmatch_slab);
6475 DEBUG_STATE_pp("pop (no final)");
6477 DEBUG_STATE_pp("pop (yes)");
6483 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6484 || yes_state > SLAB_LAST(PL_regmatch_slab))
6486 /* not in this slab, pop slab */
6487 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6488 PL_regmatch_slab = PL_regmatch_slab->prev;
6489 st = SLAB_LAST(PL_regmatch_slab);
6491 depth -= (st - yes_state);
6494 yes_state = st->u.yes.prev_yes_state;
6495 PL_regmatch_state = st;
6498 locinput= st->locinput;
6499 state_num = st->resume_state + no_final;
6500 goto reenter_switch;
6503 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6504 PL_colors[4], PL_colors[5]));
6506 if (PL_reg_state.re_state_eval_setup_done) {
6507 /* each successfully executed (?{...}) block does the equivalent of
6508 * local $^R = do {...}
6509 * When popping the save stack, all these locals would be undone;
6510 * bypass this by setting the outermost saved $^R to the latest
6512 if (oreplsv != GvSV(PL_replgv))
6513 sv_setsv(oreplsv, GvSV(PL_replgv));
6520 PerlIO_printf(Perl_debug_log,
6521 "%*s %sfailed...%s\n",
6522 REPORT_CODE_OFF+depth*2, "",
6523 PL_colors[4], PL_colors[5])
6535 /* there's a previous state to backtrack to */
6537 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6538 PL_regmatch_slab = PL_regmatch_slab->prev;
6539 st = SLAB_LAST(PL_regmatch_slab);
6541 PL_regmatch_state = st;
6542 locinput= st->locinput;
6544 DEBUG_STATE_pp("pop");
6546 if (yes_state == st)
6547 yes_state = st->u.yes.prev_yes_state;
6549 state_num = st->resume_state + 1; /* failure = success + 1 */
6550 goto reenter_switch;
6555 if (rex->intflags & PREGf_VERBARG_SEEN) {
6556 SV *sv_err = get_sv("REGERROR", 1);
6557 SV *sv_mrk = get_sv("REGMARK", 1);
6559 sv_commit = &PL_sv_no;
6561 sv_yes_mark = &PL_sv_yes;
6564 sv_commit = &PL_sv_yes;
6565 sv_yes_mark = &PL_sv_no;
6567 sv_setsv(sv_err, sv_commit);
6568 sv_setsv(sv_mrk, sv_yes_mark);
6572 if (last_pushed_cv) {
6575 PERL_UNUSED_VAR(SP);
6578 /* clean up; in particular, free all slabs above current one */
6579 LEAVE_SCOPE(oldsave);
6581 assert(!result || locinput - PL_bostr >= 0);
6582 return result ? locinput - PL_bostr : -1;
6586 - regrepeat - repeatedly match something simple, report how many
6588 * What 'simple' means is a node which can be the operand of a quantifier like
6591 * startposp - pointer a pointer to the start position. This is updated
6592 * to point to the byte following the highest successful
6594 * p - the regnode to be repeatedly matched against.
6595 * max - maximum number of things to match.
6596 * depth - (for debugging) backtracking depth.
6599 S_regrepeat(pTHX_ const regexp *prog, char **startposp, const regnode *p, I32 max, int depth)
6602 char *scan; /* Pointer to current position in target string */
6604 char *loceol = PL_regeol; /* local version */
6605 I32 hardcount = 0; /* How many matches so far */
6606 bool utf8_target = PL_reg_match_utf8;
6609 PERL_UNUSED_ARG(depth);
6612 PERL_ARGS_ASSERT_REGREPEAT;
6615 if (max == REG_INFTY)
6617 else if (! utf8_target && scan + max < loceol)
6618 loceol = scan + max;
6620 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
6621 * to the maximum of how far we should go in it (leaving it set to the real
6622 * end, if the maximum permissible would take us beyond that). This allows
6623 * us to make the loop exit condition that we haven't gone past <loceol> to
6624 * also mean that we haven't exceeded the max permissible count, saving a
6625 * test each time through the loop. But it assumes that the OP matches a
6626 * single byte, which is true for most of the OPs below when applied to a
6627 * non-UTF-8 target. Those relatively few OPs that don't have this
6628 * characteristic will have to compensate.
6630 * There is no adjustment for UTF-8 targets, as the number of bytes per
6631 * character varies. OPs will have to test both that the count is less
6632 * than the max permissible (using <hardcount> to keep track), and that we
6633 * are still within the bounds of the string (using <loceol>. A few OPs
6634 * match a single byte no matter what the encoding. They can omit the max
6635 * test if, for the UTF-8 case, they do the adjustment that was skipped
6638 * Thus, the code above sets things up for the common case; and exceptional
6639 * cases need extra work; the common case is to make sure <scan> doesn't
6640 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
6641 * count doesn't exceed the maximum permissible */
6646 while (scan < loceol && hardcount < max && *scan != '\n') {
6647 scan += UTF8SKIP(scan);
6651 while (scan < loceol && *scan != '\n')
6657 while (scan < loceol && hardcount < max) {
6658 scan += UTF8SKIP(scan);
6665 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
6666 if (utf8_target && scan + max < loceol) {
6668 /* <loceol> hadn't been adjusted in the UTF-8 case */
6676 assert(STR_LEN(p) == (UTF_PATTERN) ? UTF8SKIP(STRING(p)) : 1);
6680 /* Can use a simple loop if the pattern char to match on is invariant
6681 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
6682 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
6683 * true iff it doesn't matter if the argument is in UTF-8 or not */
6684 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! UTF_PATTERN)) {
6685 if (utf8_target && scan + max < loceol) {
6686 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6687 * since here, to match at all, 1 char == 1 byte */
6688 loceol = scan + max;
6690 while (scan < loceol && UCHARAT(scan) == c) {
6694 else if (UTF_PATTERN) {
6696 STRLEN scan_char_len;
6698 /* When both target and pattern are UTF-8, we have to do
6700 while (hardcount < max
6701 && scan + (scan_char_len = UTF8SKIP(scan)) <= loceol
6702 && scan_char_len <= STR_LEN(p)
6703 && memEQ(scan, STRING(p), scan_char_len))
6705 scan += scan_char_len;
6709 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
6711 /* Target isn't utf8; convert the character in the UTF-8
6712 * pattern to non-UTF8, and do a simple loop */
6713 c = TWO_BYTE_UTF8_TO_UNI(c, *(STRING(p) + 1));
6714 while (scan < loceol && UCHARAT(scan) == c) {
6717 } /* else pattern char is above Latin1, can't possibly match the
6722 /* Here, the string must be utf8; pattern isn't, and <c> is
6723 * different in utf8 than not, so can't compare them directly.
6724 * Outside the loop, find the two utf8 bytes that represent c, and
6725 * then look for those in sequence in the utf8 string */
6726 U8 high = UTF8_TWO_BYTE_HI(c);
6727 U8 low = UTF8_TWO_BYTE_LO(c);
6729 while (hardcount < max
6730 && scan + 1 < loceol
6731 && UCHARAT(scan) == high
6732 && UCHARAT(scan + 1) == low)
6741 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6745 PL_reg_flags |= RF_tainted;
6746 utf8_flags = FOLDEQ_UTF8_LOCALE;
6754 case EXACTFU_TRICKYFOLD:
6756 utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
6760 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
6762 assert(STR_LEN(p) == (UTF_PATTERN) ? UTF8SKIP(STRING(p)) : 1);
6764 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8)) {
6765 if (c1 == CHRTEST_VOID) {
6766 /* Use full Unicode fold matching */
6767 char *tmpeol = PL_regeol;
6768 STRLEN pat_len = (UTF_PATTERN) ? UTF8SKIP(STRING(p)) : 1;
6769 while (hardcount < max
6770 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
6771 STRING(p), NULL, pat_len,
6772 cBOOL(UTF_PATTERN), utf8_flags))
6779 else if (utf8_target) {
6781 while (scan < loceol
6783 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
6785 scan += UTF8SKIP(scan);
6790 while (scan < loceol
6792 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
6793 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
6795 scan += UTF8SKIP(scan);
6800 else if (c1 == c2) {
6801 while (scan < loceol && UCHARAT(scan) == c1) {
6806 while (scan < loceol &&
6807 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
6818 while (hardcount < max
6819 && scan + (inclasslen = UTF8SKIP(scan)) <= loceol
6820 && reginclass(prog, p, (U8*)scan, utf8_target))
6826 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6833 LOAD_UTF8_CHARCLASS_ALNUM();
6834 while (hardcount < max && scan < loceol &&
6835 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6837 scan += UTF8SKIP(scan);
6841 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6849 while (scan < loceol && isALNUM((U8) *scan)) {
6854 if (utf8_target && scan + max < loceol) {
6856 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6857 * since here, to match, 1 char == 1 byte */
6858 loceol = scan + max;
6860 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6865 PL_reg_flags |= RF_tainted;
6867 while (hardcount < max && scan < loceol &&
6868 isALNUM_LC_utf8((U8*)scan)) {
6869 scan += UTF8SKIP(scan);
6873 while (scan < loceol && isALNUM_LC(*scan))
6882 LOAD_UTF8_CHARCLASS_ALNUM();
6883 while (hardcount < max && scan < loceol &&
6884 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6886 scan += UTF8SKIP(scan);
6890 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6897 goto utf8_Nwordchar;
6898 while (scan < loceol && ! isALNUM((U8) *scan)) {
6904 if (utf8_target && scan + max < loceol) {
6906 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6907 * since here, to match, 1 char == 1 byte */
6908 loceol = scan + max;
6910 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
6916 while (scan < loceol && hardcount < max
6917 && ! _generic_isCC_A((U8) *scan, FLAGS(p)))
6919 scan += UTF8SKIP(scan);
6924 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
6931 while (scan < loceol && hardcount < max
6932 && ! isWORDCHAR_A((U8) *scan))
6934 scan += UTF8SKIP(scan);
6939 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6945 PL_reg_flags |= RF_tainted;
6947 while (hardcount < max && scan < loceol &&
6948 !isALNUM_LC_utf8((U8*)scan)) {
6949 scan += UTF8SKIP(scan);
6953 while (scan < loceol && !isALNUM_LC(*scan))
6962 while (hardcount < max && scan < loceol
6963 && is_XPERLSPACE_utf8((U8*)scan))
6965 scan += UTF8SKIP(scan);
6971 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6980 while (scan < loceol && isSPACE((U8) *scan)) {
6985 if (utf8_target && scan + max < loceol) {
6987 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
6988 * since here, to match, 1 char == 1 byte */
6989 loceol = scan + max;
6991 while (scan < loceol && isSPACE_A((U8) *scan)) {
6996 PL_reg_flags |= RF_tainted;
6998 while (hardcount < max && scan < loceol &&
6999 isSPACE_LC_utf8((U8*)scan)) {
7000 scan += UTF8SKIP(scan);
7004 while (scan < loceol && isSPACE_LC(*scan))
7013 while (hardcount < max && scan < loceol
7014 && ! is_XPERLSPACE_utf8((U8*)scan))
7016 scan += UTF8SKIP(scan);
7022 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
7031 while (scan < loceol && ! isSPACE((U8) *scan)) {
7037 while (hardcount < max && scan < loceol
7038 && ! isSPACE_A((U8) *scan))
7040 scan += UTF8SKIP(scan);
7045 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
7051 PL_reg_flags |= RF_tainted;
7053 while (hardcount < max && scan < loceol &&
7054 !isSPACE_LC_utf8((U8*)scan)) {
7055 scan += UTF8SKIP(scan);
7059 while (scan < loceol && !isSPACE_LC(*scan))
7065 LOAD_UTF8_CHARCLASS_DIGIT();
7066 while (hardcount < max && scan < loceol &&
7067 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
7068 scan += UTF8SKIP(scan);
7072 while (scan < loceol && isDIGIT(*scan))
7077 if (utf8_target && scan + max < loceol) {
7079 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
7080 * since here, to match, 1 char == 1 byte */
7081 loceol = scan + max;
7083 while (scan < loceol && isDIGIT_A((U8) *scan)) {
7088 PL_reg_flags |= RF_tainted;
7090 while (hardcount < max && scan < loceol &&
7091 isDIGIT_LC_utf8((U8*)scan)) {
7092 scan += UTF8SKIP(scan);
7096 while (scan < loceol && isDIGIT_LC(*scan))
7102 LOAD_UTF8_CHARCLASS_DIGIT();
7103 while (hardcount < max && scan < loceol &&
7104 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
7105 scan += UTF8SKIP(scan);
7109 while (scan < loceol && !isDIGIT(*scan))
7115 while (hardcount < max && scan < loceol
7116 && ! isDIGIT_A((U8) *scan)) {
7117 scan += UTF8SKIP(scan);
7122 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
7128 PL_reg_flags |= RF_tainted;
7130 while (hardcount < max && scan < loceol &&
7131 !isDIGIT_LC_utf8((U8*)scan)) {
7132 scan += UTF8SKIP(scan);
7136 while (scan < loceol && !isDIGIT_LC(*scan))
7142 while (hardcount < max && scan < loceol &&
7143 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7148 /* LNBREAK can match one or two latin chars, which is ok, but we
7149 * have to use hardcount in this situation, and throw away the
7150 * adjustment to <loceol> done before the switch statement */
7152 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7160 while (hardcount < max && scan < loceol &&
7161 (c=is_HORIZWS_utf8_safe(scan, loceol)))
7167 while (scan < loceol && is_HORIZWS_latin1_safe(scan, loceol))
7173 while (hardcount < max && scan < loceol &&
7174 !is_HORIZWS_utf8_safe(scan, loceol))
7176 scan += UTF8SKIP(scan);
7180 while (scan < loceol && !is_HORIZWS_latin1_safe(scan, loceol))
7187 while (hardcount < max && scan < loceol &&
7188 (c=is_VERTWS_utf8_safe(scan, loceol)))
7194 while (scan < loceol && is_VERTWS_latin1_safe(scan, loceol))
7201 while (hardcount < max && scan < loceol &&
7202 !is_VERTWS_utf8_safe(scan, loceol))
7204 scan += UTF8SKIP(scan);
7208 while (scan < loceol && !is_VERTWS_latin1_safe(scan, loceol))
7228 /* These are all 0 width, so match right here or not at all. */
7232 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7233 assert(0); /* NOTREACHED */
7240 c = scan - *startposp;
7244 GET_RE_DEBUG_FLAGS_DECL;
7246 SV * const prop = sv_newmortal();
7247 regprop(prog, prop, p);
7248 PerlIO_printf(Perl_debug_log,
7249 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7250 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7258 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7260 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7261 create a copy so that changes the caller makes won't change the shared one.
7262 If <altsvp> is non-null, will return NULL in it, for back-compat.
7265 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7267 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7273 return newSVsv(core_regclass_swash(prog, node, doinit, listsvp));
7278 S_core_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp)
7280 /* Returns the swash for the input 'node' in the regex 'prog'.
7281 * If <doinit> is true, will attempt to create the swash if not already
7283 * If <listsvp> is non-null, will return the swash initialization string in
7285 * Tied intimately to how regcomp.c sets up the data structure */
7292 RXi_GET_DECL(prog,progi);
7293 const struct reg_data * const data = prog ? progi->data : NULL;
7295 PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH;
7297 assert(ANYOF_NONBITMAP(node));
7299 if (data && data->count) {
7300 const U32 n = ARG(node);
7302 if (data->what[n] == 's') {
7303 SV * const rv = MUTABLE_SV(data->data[n]);
7304 AV * const av = MUTABLE_AV(SvRV(rv));
7305 SV **const ary = AvARRAY(av);
7306 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7308 si = *ary; /* ary[0] = the string to initialize the swash with */
7310 /* Elements 2 and 3 are either both present or both absent. [2] is
7311 * any inversion list generated at compile time; [3] indicates if
7312 * that inversion list has any user-defined properties in it. */
7313 if (av_len(av) >= 2) {
7316 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7323 /* Element [1] is reserved for the set-up swash. If already there,
7324 * return it; if not, create it and store it there */
7325 if (SvROK(ary[1])) {
7328 else if (si && doinit) {
7330 sw = _core_swash_init("utf8", /* the utf8 package */
7334 0, /* not from tr/// */
7337 (void)av_store(av, 1, sw);
7343 SV* matches_string = newSVpvn("", 0);
7345 /* Use the swash, if any, which has to have incorporated into it all
7347 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7348 && (si && si != &PL_sv_undef))
7351 /* If no swash, use the input initialization string, if available */
7352 sv_catsv(matches_string, si);
7355 /* Add the inversion list to whatever we have. This may have come from
7356 * the swash, or from an input parameter */
7358 sv_catsv(matches_string, _invlist_contents(invlist));
7360 *listsvp = matches_string;
7367 - reginclass - determine if a character falls into a character class
7369 n is the ANYOF regnode
7370 p is the target string
7371 utf8_target tells whether p is in UTF-8.
7373 Returns true if matched; false otherwise.
7375 Note that this can be a synthetic start class, a combination of various
7376 nodes, so things you think might be mutually exclusive, such as locale,
7377 aren't. It can match both locale and non-locale
7382 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, register const bool utf8_target)
7385 const char flags = ANYOF_FLAGS(n);
7389 PERL_ARGS_ASSERT_REGINCLASS;
7391 /* If c is not already the code point, get it. Note that
7392 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7393 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7395 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
7396 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7397 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7398 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7399 * UTF8_ALLOW_FFFF */
7400 if (c_len == (STRLEN)-1)
7401 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7404 /* If this character is potentially in the bitmap, check it */
7406 if (ANYOF_BITMAP_TEST(n, c))
7408 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
7414 else if (flags & ANYOF_LOCALE) {
7415 PL_reg_flags |= RF_tainted;
7417 if ((flags & ANYOF_LOC_FOLD)
7418 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7422 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
7423 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
7424 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
7425 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
7426 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
7427 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
7428 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
7429 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
7430 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
7431 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
7432 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
7433 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII_LC(c)) ||
7434 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII_LC(c)) ||
7435 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
7436 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
7437 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
7438 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
7439 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
7440 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
7441 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
7442 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
7443 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
7444 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
7445 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
7446 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
7447 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
7448 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
7449 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
7450 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
7451 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK_LC(c)) ||
7452 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK_LC(c))
7453 ) /* How's that for a conditional? */
7460 /* If the bitmap didn't (or couldn't) match, and something outside the
7461 * bitmap could match, try that. Locale nodes specify completely the
7462 * behavior of code points in the bit map (otherwise, a utf8 target would
7463 * cause them to be treated as Unicode and not locale), except in
7464 * the very unlikely event when this node is a synthetic start class, which
7465 * could be a combination of locale and non-locale nodes. So allow locale
7466 * to match for the synthetic start class, which will give a false
7467 * positive that will be resolved when the match is done again as not part
7468 * of the synthetic start class */
7470 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
7471 match = TRUE; /* Everything above 255 matches */
7473 else if (ANYOF_NONBITMAP(n)
7474 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
7477 || (! (flags & ANYOF_LOCALE))
7478 || (flags & ANYOF_IS_SYNTHETIC)))))
7480 SV * const sw = core_regclass_swash(prog, n, TRUE, 0);
7485 } else { /* Convert to utf8 */
7487 utf8_p = bytes_to_utf8(p, &len);
7490 if (swash_fetch(sw, utf8_p, TRUE)) {
7494 /* If we allocated a string above, free it */
7495 if (! utf8_target) Safefree(utf8_p);
7499 if (UNICODE_IS_SUPER(c)
7500 && (flags & ANYOF_WARN_SUPER)
7501 && ckWARN_d(WARN_NON_UNICODE))
7503 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7504 "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
7508 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7509 return cBOOL(flags & ANYOF_INVERT) ^ match;
7513 S_reghop3(U8 *s, I32 off, const U8* lim)
7515 /* return the position 'off' UTF-8 characters away from 's', forward if
7516 * 'off' >= 0, backwards if negative. But don't go outside of position
7517 * 'lim', which better be < s if off < 0 */
7521 PERL_ARGS_ASSERT_REGHOP3;
7524 while (off-- && s < lim) {
7525 /* XXX could check well-formedness here */
7530 while (off++ && s > lim) {
7532 if (UTF8_IS_CONTINUED(*s)) {
7533 while (s > lim && UTF8_IS_CONTINUATION(*s))
7536 /* XXX could check well-formedness here */
7543 /* there are a bunch of places where we use two reghop3's that should
7544 be replaced with this routine. but since thats not done yet
7545 we ifdef it out - dmq
7548 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
7552 PERL_ARGS_ASSERT_REGHOP4;
7555 while (off-- && s < rlim) {
7556 /* XXX could check well-formedness here */
7561 while (off++ && s > llim) {
7563 if (UTF8_IS_CONTINUED(*s)) {
7564 while (s > llim && UTF8_IS_CONTINUATION(*s))
7567 /* XXX could check well-formedness here */
7575 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
7579 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7582 while (off-- && s < lim) {
7583 /* XXX could check well-formedness here */
7590 while (off++ && s > lim) {
7592 if (UTF8_IS_CONTINUED(*s)) {
7593 while (s > lim && UTF8_IS_CONTINUATION(*s))
7596 /* XXX could check well-formedness here */
7605 restore_pos(pTHX_ void *arg)
7608 regexp * const rex = (regexp *)arg;
7609 if (PL_reg_state.re_state_eval_setup_done) {
7610 if (PL_reg_oldsaved) {
7611 rex->subbeg = PL_reg_oldsaved;
7612 rex->sublen = PL_reg_oldsavedlen;
7613 rex->suboffset = PL_reg_oldsavedoffset;
7614 rex->subcoffset = PL_reg_oldsavedcoffset;
7615 #ifdef PERL_OLD_COPY_ON_WRITE
7616 rex->saved_copy = PL_nrs;
7618 RXp_MATCH_COPIED_on(rex);
7620 PL_reg_magic->mg_len = PL_reg_oldpos;
7621 PL_reg_state.re_state_eval_setup_done = FALSE;
7622 PL_curpm = PL_reg_oldcurpm;
7627 S_to_utf8_substr(pTHX_ register regexp *prog)
7629 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
7630 * on the converted value */
7634 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
7637 if (prog->substrs->data[i].substr
7638 && !prog->substrs->data[i].utf8_substr) {
7639 SV* const sv = newSVsv(prog->substrs->data[i].substr);
7640 prog->substrs->data[i].utf8_substr = sv;
7641 sv_utf8_upgrade(sv);
7642 if (SvVALID(prog->substrs->data[i].substr)) {
7643 if (SvTAIL(prog->substrs->data[i].substr)) {
7644 /* Trim the trailing \n that fbm_compile added last
7646 SvCUR_set(sv, SvCUR(sv) - 1);
7647 /* Whilst this makes the SV technically "invalid" (as its
7648 buffer is no longer followed by "\0") when fbm_compile()
7649 adds the "\n" back, a "\0" is restored. */
7650 fbm_compile(sv, FBMcf_TAIL);
7654 if (prog->substrs->data[i].substr == prog->check_substr)
7655 prog->check_utf8 = sv;
7661 S_to_byte_substr(pTHX_ register regexp *prog)
7663 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
7664 * on the converted value; returns FALSE if can't be converted. */
7669 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
7672 if (prog->substrs->data[i].utf8_substr
7673 && !prog->substrs->data[i].substr) {
7674 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7675 if (! sv_utf8_downgrade(sv, TRUE)) {
7678 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
7679 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
7680 /* Trim the trailing \n that fbm_compile added last
7682 SvCUR_set(sv, SvCUR(sv) - 1);
7683 fbm_compile(sv, FBMcf_TAIL);
7687 prog->substrs->data[i].substr = sv;
7688 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
7689 prog->check_substr = sv;
7696 /* These constants are for finding GCB=LV and GCB=LVT. These are for the
7697 * pre-composed Hangul syllables, which are all in a contiguous block and
7698 * arranged there in such a way so as to facilitate alorithmic determination of
7699 * their characteristics. As such, they don't need a swash, but can be
7700 * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
7702 #define SBASE 0xAC00 /* Start of block */
7703 #define SCount 11172 /* Length of block */
7706 #if 0 /* This routine is not currently used */
7707 PERL_STATIC_INLINE bool
7708 S_is_utf8_X_LV(pTHX_ const U8 *p)
7710 /* Unlike most other similarly named routines here, this does not create a
7711 * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
7715 UV cp = valid_utf8_to_uvchr(p, NULL);
7717 PERL_ARGS_ASSERT_IS_UTF8_X_LV;
7719 /* The earliest Unicode releases did not have these precomposed Hangul
7720 * syllables. Set to point to undef in that case, so will return false on
7722 if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
7723 PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
7724 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
7725 SvREFCNT_dec(PL_utf8_X_LV);
7726 PL_utf8_X_LV = &PL_sv_undef;
7730 return (PL_utf8_X_LV != &PL_sv_undef
7731 && cp >= SBASE && cp < SBASE + SCount
7732 && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
7736 PERL_STATIC_INLINE bool
7737 S_is_utf8_X_LVT(pTHX_ const U8 *p)
7739 /* Unlike most other similarly named routines here, this does not create a
7740 * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
7744 UV cp = valid_utf8_to_uvchr(p, NULL);
7746 PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
7748 /* The earliest Unicode releases did not have these precomposed Hangul
7749 * syllables. Set to point to undef in that case, so will return false on
7751 if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
7752 PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
7753 if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
7754 SvREFCNT_dec(PL_utf8_X_LVT);
7755 PL_utf8_X_LVT = &PL_sv_undef;
7759 return (PL_utf8_X_LVT != &PL_sv_undef
7760 && cp >= SBASE && cp < SBASE + SCount
7761 && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
7766 * c-indentation-style: bsd
7768 * indent-tabs-mode: nil
7771 * ex: set ts=8 sts=4 sw=4 et: