5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
78 #ifdef PERL_IN_XSUB_RE
84 #define RF_tainted 1 /* tainted information used? e.g. locale */
85 #define RF_warned 2 /* warned about big count? */
87 #define RF_utf8 8 /* Pattern contains multibyte chars? */
89 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
91 #define RS_init 1 /* eval environment created */
92 #define RS_set 2 /* replsv value is set */
98 /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
99 * call if there are no complications: i.e., if everything matchable is
100 * straight forward in the bitmap */
101 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
102 : ANYOF_BITMAP_TEST(p,*(c)))
108 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
109 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
111 #define HOPc(pos,off) \
112 (char *)(PL_reg_match_utf8 \
113 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
115 #define HOPBACKc(pos, off) \
116 (char*)(PL_reg_match_utf8\
117 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
118 : (pos - off >= PL_bostr) \
122 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
123 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
125 /* these are unrolled below in the CCC_TRY_XXX defined */
126 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
127 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
129 /* Doesn't do an assert to verify that is correct */
130 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
131 if (!CAT2(PL_utf8_,class)) { bool throw_away; ENTER; save_re_context(); throw_away = CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END
133 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
134 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
135 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
137 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
138 LOAD_UTF8_CHARCLASS(X_begin, " "); \
139 LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
140 /* These are utf8 constants, and not utf-ebcdic constants, so the \
141 * assert should likely and hopefully fail on an EBCDIC machine */ \
142 LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
144 /* No asserts are done for these, in case called on an early \
145 * Unicode version in which they map to nothing */ \
146 LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
147 LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
148 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
149 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
150 LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
151 LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
152 LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
154 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
156 /* The actual code for CCC_TRY, which uses several variables from the routine
157 * it's callable from. It is designed to be the bulk of a case statement.
158 * FUNC is the macro or function to call on non-utf8 targets that indicate if
159 * nextchr matches the class.
160 * UTF8_TEST is the whole test string to use for utf8 targets
161 * LOAD is what to use to test, and if not present to load in the swash for the
163 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
165 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
166 * utf8 and a variant, load the swash if necessary and test using the utf8
167 * test. Advance to the next character if test is ok, otherwise fail; If not
168 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
169 * fails, or advance to the next character */
171 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
172 if (locinput >= PL_regeol) { \
175 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
176 LOAD_UTF8_CHARCLASS(CLASS, STR); \
177 if (POS_OR_NEG (UTF8_TEST)) { \
180 locinput += PL_utf8skip[nextchr]; \
181 nextchr = UCHARAT(locinput); \
184 if (POS_OR_NEG (FUNC(nextchr))) { \
187 nextchr = UCHARAT(++locinput); \
190 /* Handle the non-locale cases for a character class and its complement. It
191 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
192 * This is because that code fails when the test succeeds, so we want to have
193 * the test fail so that the code succeeds. The swash is stored in a
194 * predictable PL_ place */
195 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
198 _CCC_TRY_CODE( !, FUNC, \
199 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
200 (U8*)locinput, TRUE)), \
203 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
204 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
205 (U8*)locinput, TRUE)), \
208 /* Generate the case statements for both locale and non-locale character
209 * classes in regmatch for classes that don't have special unicode semantics.
210 * Locales don't use an immediate swash, but an intermediary special locale
211 * function that is called on the pointer to the current place in the input
212 * string. That function will resolve to needing the same swash. One might
213 * think that because we don't know what the locale will match, we shouldn't
214 * check with the swash loading function that it loaded properly; ie, that we
215 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
216 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
218 #define CCC_TRY(NAME, NNAME, FUNC, \
219 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
220 NAMEA, NNAMEA, FUNCA, \
223 PL_reg_flags |= RF_tainted; \
224 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
226 PL_reg_flags |= RF_tainted; \
227 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
230 if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
233 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
234 nextchr = UCHARAT(++locinput); \
237 if (locinput >= PL_regeol || FUNCA(nextchr)) { \
241 locinput += PL_utf8skip[nextchr]; \
242 nextchr = UCHARAT(locinput); \
245 nextchr = UCHARAT(++locinput); \
248 /* Generate the non-locale cases */ \
249 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
251 /* This is like CCC_TRY, but has an extra set of parameters for generating case
252 * statements to handle separate Unicode semantics nodes */
253 #define CCC_TRY_U(NAME, NNAME, FUNC, \
254 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
255 NAMEU, NNAMEU, FUNCU, \
256 NAMEA, NNAMEA, FUNCA, \
258 CCC_TRY(NAME, NNAME, FUNC, \
259 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
260 NAMEA, NNAMEA, FUNCA, \
262 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
264 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
266 /* for use after a quantifier and before an EXACT-like node -- japhy */
267 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
269 * NOTE that *nothing* that affects backtracking should be in here, specifically
270 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
271 * node that is in between two EXACT like nodes when ascertaining what the required
272 * "follow" character is. This should probably be moved to regex compile time
273 * although it may be done at run time beause of the REF possibility - more
274 * investigation required. -- demerphq
276 #define JUMPABLE(rn) ( \
278 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
280 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
281 OP(rn) == PLUS || OP(rn) == MINMOD || \
283 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
285 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
287 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
290 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
291 we don't need this definition. */
292 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
293 #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF )
294 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
297 /* ... so we use this as its faster. */
298 #define IS_TEXT(rn) ( OP(rn)==EXACT )
299 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA)
300 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
301 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
306 Search for mandatory following text node; for lookahead, the text must
307 follow but for lookbehind (rn->flags != 0) we skip to the next step.
309 #define FIND_NEXT_IMPT(rn) STMT_START { \
310 while (JUMPABLE(rn)) { \
311 const OPCODE type = OP(rn); \
312 if (type == SUSPEND || PL_regkind[type] == CURLY) \
313 rn = NEXTOPER(NEXTOPER(rn)); \
314 else if (type == PLUS) \
316 else if (type == IFMATCH) \
317 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
318 else rn += NEXT_OFF(rn); \
323 static void restore_pos(pTHX_ void *arg);
325 #define REGCP_PAREN_ELEMS 4
326 #define REGCP_OTHER_ELEMS 5
327 #define REGCP_FRAME_ELEMS 1
328 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
329 * are needed for the regexp context stack bookkeeping. */
332 S_regcppush(pTHX_ I32 parenfloor)
335 const int retval = PL_savestack_ix;
336 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
337 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
338 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
340 GET_RE_DEBUG_FLAGS_DECL;
342 if (paren_elems_to_push < 0)
343 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
345 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
346 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
347 " out of range (%lu-%ld)",
348 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
350 SSGROW(total_elems + REGCP_FRAME_ELEMS);
352 for (p = PL_regsize; p > parenfloor; p--) {
353 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
354 SSPUSHINT(PL_regoffs[p].end);
355 SSPUSHINT(PL_regoffs[p].start);
356 SSPUSHPTR(PL_reg_start_tmp[p]);
358 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
359 " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
360 (UV)p, (IV)PL_regoffs[p].start,
361 (IV)(PL_reg_start_tmp[p] - PL_bostr),
362 (IV)PL_regoffs[p].end
365 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
366 SSPUSHPTR(PL_regoffs);
367 SSPUSHINT(PL_regsize);
368 SSPUSHINT(*PL_reglastparen);
369 SSPUSHINT(*PL_reglastcloseparen);
370 SSPUSHPTR(PL_reginput);
371 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
376 /* These are needed since we do not localize EVAL nodes: */
377 #define REGCP_SET(cp) \
379 PerlIO_printf(Perl_debug_log, \
380 " Setting an EVAL scope, savestack=%"IVdf"\n", \
381 (IV)PL_savestack_ix)); \
384 #define REGCP_UNWIND(cp) \
386 if (cp != PL_savestack_ix) \
387 PerlIO_printf(Perl_debug_log, \
388 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
389 (IV)(cp), (IV)PL_savestack_ix)); \
393 S_regcppop(pTHX_ const regexp *rex)
398 GET_RE_DEBUG_FLAGS_DECL;
400 PERL_ARGS_ASSERT_REGCPPOP;
402 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
404 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
405 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
406 input = (char *) SSPOPPTR;
407 *PL_reglastcloseparen = SSPOPINT;
408 *PL_reglastparen = SSPOPINT;
409 PL_regsize = SSPOPINT;
410 PL_regoffs=(regexp_paren_pair *) SSPOPPTR;
412 i -= REGCP_OTHER_ELEMS;
413 /* Now restore the parentheses context. */
414 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
416 U32 paren = (U32)SSPOPINT;
417 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
418 PL_regoffs[paren].start = SSPOPINT;
420 if (paren <= *PL_reglastparen)
421 PL_regoffs[paren].end = tmps;
423 PerlIO_printf(Perl_debug_log,
424 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
425 (UV)paren, (IV)PL_regoffs[paren].start,
426 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
427 (IV)PL_regoffs[paren].end,
428 (paren > *PL_reglastparen ? "(no)" : ""));
432 if (*PL_reglastparen + 1 <= rex->nparens) {
433 PerlIO_printf(Perl_debug_log,
434 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
435 (IV)(*PL_reglastparen + 1), (IV)rex->nparens);
439 /* It would seem that the similar code in regtry()
440 * already takes care of this, and in fact it is in
441 * a better location to since this code can #if 0-ed out
442 * but the code in regtry() is needed or otherwise tests
443 * requiring null fields (pat.t#187 and split.t#{13,14}
444 * (as of patchlevel 7877) will fail. Then again,
445 * this code seems to be necessary or otherwise
446 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
447 * --jhi updated by dapm */
448 for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
450 PL_regoffs[i].start = -1;
451 PL_regoffs[i].end = -1;
457 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
460 * pregexec and friends
463 #ifndef PERL_IN_XSUB_RE
465 - pregexec - match a regexp against a string
468 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
469 char *strbeg, I32 minend, SV *screamer, U32 nosave)
470 /* strend: pointer to null at end of string */
471 /* strbeg: real beginning of string */
472 /* minend: end of match must be >=minend after stringarg. */
473 /* nosave: For optimizations. */
475 PERL_ARGS_ASSERT_PREGEXEC;
478 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
479 nosave ? 0 : REXEC_COPY_STR);
484 * Need to implement the following flags for reg_anch:
486 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
488 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
489 * INTUIT_AUTORITATIVE_ML
490 * INTUIT_ONCE_NOML - Intuit can match in one location only.
493 * Another flag for this function: SECOND_TIME (so that float substrs
494 * with giant delta may be not rechecked).
497 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
499 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
500 Otherwise, only SvCUR(sv) is used to get strbeg. */
502 /* XXXX We assume that strpos is strbeg unless sv. */
504 /* XXXX Some places assume that there is a fixed substring.
505 An update may be needed if optimizer marks as "INTUITable"
506 RExen without fixed substrings. Similarly, it is assumed that
507 lengths of all the strings are no more than minlen, thus they
508 cannot come from lookahead.
509 (Or minlen should take into account lookahead.)
510 NOTE: Some of this comment is not correct. minlen does now take account
511 of lookahead/behind. Further research is required. -- demerphq
515 /* A failure to find a constant substring means that there is no need to make
516 an expensive call to REx engine, thus we celebrate a failure. Similarly,
517 finding a substring too deep into the string means that less calls to
518 regtry() should be needed.
520 REx compiler's optimizer found 4 possible hints:
521 a) Anchored substring;
523 c) Whether we are anchored (beginning-of-line or \G);
524 d) First node (of those at offset 0) which may distinguish positions;
525 We use a)b)d) and multiline-part of c), and try to find a position in the
526 string which does not contradict any of them.
529 /* Most of decisions we do here should have been done at compile time.
530 The nodes of the REx which we used for the search should have been
531 deleted from the finite automaton. */
534 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
535 char *strend, const U32 flags, re_scream_pos_data *data)
538 struct regexp *const prog = (struct regexp *)SvANY(rx);
539 register I32 start_shift = 0;
540 /* Should be nonnegative! */
541 register I32 end_shift = 0;
546 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
548 register char *other_last = NULL; /* other substr checked before this */
549 char *check_at = NULL; /* check substr found at this pos */
550 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
551 RXi_GET_DECL(prog,progi);
553 const char * const i_strpos = strpos;
555 GET_RE_DEBUG_FLAGS_DECL;
557 PERL_ARGS_ASSERT_RE_INTUIT_START;
559 RX_MATCH_UTF8_set(rx,utf8_target);
562 PL_reg_flags |= RF_utf8;
565 debug_start_match(rx, utf8_target, strpos, strend,
566 sv ? "Guessing start of match in sv for"
567 : "Guessing start of match in string for");
570 /* CHR_DIST() would be more correct here but it makes things slow. */
571 if (prog->minlen > strend - strpos) {
572 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
573 "String too short... [re_intuit_start]\n"));
577 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
580 if (!prog->check_utf8 && prog->check_substr)
581 to_utf8_substr(prog);
582 check = prog->check_utf8;
584 if (!prog->check_substr && prog->check_utf8)
585 to_byte_substr(prog);
586 check = prog->check_substr;
588 if (check == &PL_sv_undef) {
589 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
590 "Non-utf8 string cannot match utf8 check string\n"));
593 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
594 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
595 || ( (prog->extflags & RXf_ANCH_BOL)
596 && !multiline ) ); /* Check after \n? */
599 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
600 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
601 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
603 && (strpos != strbeg)) {
604 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
607 if (prog->check_offset_min == prog->check_offset_max &&
608 !(prog->extflags & RXf_CANY_SEEN)) {
609 /* Substring at constant offset from beg-of-str... */
612 s = HOP3c(strpos, prog->check_offset_min, strend);
615 slen = SvCUR(check); /* >= 1 */
617 if ( strend - s > slen || strend - s < slen - 1
618 || (strend - s == slen && strend[-1] != '\n')) {
619 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
622 /* Now should match s[0..slen-2] */
624 if (slen && (*SvPVX_const(check) != *s
626 && memNE(SvPVX_const(check), s, slen)))) {
628 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
632 else if (*SvPVX_const(check) != *s
633 || ((slen = SvCUR(check)) > 1
634 && memNE(SvPVX_const(check), s, slen)))
637 goto success_at_start;
640 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
642 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
643 end_shift = prog->check_end_shift;
646 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
647 - (SvTAIL(check) != 0);
648 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
650 if (end_shift < eshift)
654 else { /* Can match at random position */
657 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
658 end_shift = prog->check_end_shift;
660 /* end shift should be non negative here */
663 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
665 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
666 (IV)end_shift, RX_PRECOMP(prog));
670 /* Find a possible match in the region s..strend by looking for
671 the "check" substring in the region corrected by start/end_shift. */
674 I32 srch_start_shift = start_shift;
675 I32 srch_end_shift = end_shift;
676 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
677 srch_end_shift -= ((strbeg - s) - srch_start_shift);
678 srch_start_shift = strbeg - s;
680 DEBUG_OPTIMISE_MORE_r({
681 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
682 (IV)prog->check_offset_min,
683 (IV)srch_start_shift,
685 (IV)prog->check_end_shift);
688 if (flags & REXEC_SCREAM) {
689 I32 p = -1; /* Internal iterator of scream. */
690 I32 * const pp = data ? data->scream_pos : &p;
692 if (PL_screamfirst[BmRARE(check)] >= 0
693 || ( BmRARE(check) == '\n'
694 && (BmPREVIOUS(check) == SvCUR(check) - 1)
696 s = screaminstr(sv, check,
697 srch_start_shift + (s - strbeg), srch_end_shift, pp, 0);
700 /* we may be pointing at the wrong string */
701 if (s && RXp_MATCH_COPIED(prog))
702 s = strbeg + (s - SvPVX_const(sv));
704 *data->scream_olds = s;
709 if (prog->extflags & RXf_CANY_SEEN) {
710 start_point= (U8*)(s + srch_start_shift);
711 end_point= (U8*)(strend - srch_end_shift);
713 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
714 end_point= HOP3(strend, -srch_end_shift, strbeg);
716 DEBUG_OPTIMISE_MORE_r({
717 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
718 (int)(end_point - start_point),
719 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
723 s = fbm_instr( start_point, end_point,
724 check, multiline ? FBMrf_MULTILINE : 0);
727 /* Update the count-of-usability, remove useless subpatterns,
731 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
732 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
733 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
734 (s ? "Found" : "Did not find"),
735 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
736 ? "anchored" : "floating"),
739 (s ? " at offset " : "...\n") );
744 /* Finish the diagnostic message */
745 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
747 /* XXX dmq: first branch is for positive lookbehind...
748 Our check string is offset from the beginning of the pattern.
749 So we need to do any stclass tests offset forward from that
758 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
759 Start with the other substr.
760 XXXX no SCREAM optimization yet - and a very coarse implementation
761 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
762 *always* match. Probably should be marked during compile...
763 Probably it is right to do no SCREAM here...
766 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
767 : (prog->float_substr && prog->anchored_substr))
769 /* Take into account the "other" substring. */
770 /* XXXX May be hopelessly wrong for UTF... */
773 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
776 char * const last = HOP3c(s, -start_shift, strbeg);
778 char * const saved_s = s;
781 t = s - prog->check_offset_max;
782 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
784 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
789 t = HOP3c(t, prog->anchored_offset, strend);
790 if (t < other_last) /* These positions already checked */
792 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
795 /* XXXX It is not documented what units *_offsets are in.
796 We assume bytes, but this is clearly wrong.
797 Meaning this code needs to be carefully reviewed for errors.
801 /* On end-of-str: see comment below. */
802 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
803 if (must == &PL_sv_undef) {
805 DEBUG_r(must = prog->anchored_utf8); /* for debug */
810 HOP3(HOP3(last1, prog->anchored_offset, strend)
811 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
813 multiline ? FBMrf_MULTILINE : 0
816 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
817 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
818 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
819 (s ? "Found" : "Contradicts"),
820 quoted, RE_SV_TAIL(must));
825 if (last1 >= last2) {
826 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
827 ", giving up...\n"));
830 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
831 ", trying floating at offset %ld...\n",
832 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
833 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
834 s = HOP3c(last, 1, strend);
838 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
839 (long)(s - i_strpos)));
840 t = HOP3c(s, -prog->anchored_offset, strbeg);
841 other_last = HOP3c(s, 1, strend);
849 else { /* Take into account the floating substring. */
851 char * const saved_s = s;
854 t = HOP3c(s, -start_shift, strbeg);
856 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
857 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
858 last = HOP3c(t, prog->float_max_offset, strend);
859 s = HOP3c(t, prog->float_min_offset, strend);
862 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
863 must = utf8_target ? prog->float_utf8 : prog->float_substr;
864 /* fbm_instr() takes into account exact value of end-of-str
865 if the check is SvTAIL(ed). Since false positives are OK,
866 and end-of-str is not later than strend we are OK. */
867 if (must == &PL_sv_undef) {
869 DEBUG_r(must = prog->float_utf8); /* for debug message */
872 s = fbm_instr((unsigned char*)s,
873 (unsigned char*)last + SvCUR(must)
875 must, multiline ? FBMrf_MULTILINE : 0);
877 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
878 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
879 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
880 (s ? "Found" : "Contradicts"),
881 quoted, RE_SV_TAIL(must));
885 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
886 ", giving up...\n"));
889 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
890 ", trying anchored starting at offset %ld...\n",
891 (long)(saved_s + 1 - i_strpos)));
893 s = HOP3c(t, 1, strend);
897 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
898 (long)(s - i_strpos)));
899 other_last = s; /* Fix this later. --Hugo */
909 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
911 DEBUG_OPTIMISE_MORE_r(
912 PerlIO_printf(Perl_debug_log,
913 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
914 (IV)prog->check_offset_min,
915 (IV)prog->check_offset_max,
923 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
925 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
928 /* Fixed substring is found far enough so that the match
929 cannot start at strpos. */
931 if (ml_anch && t[-1] != '\n') {
932 /* Eventually fbm_*() should handle this, but often
933 anchored_offset is not 0, so this check will not be wasted. */
934 /* XXXX In the code below we prefer to look for "^" even in
935 presence of anchored substrings. And we search even
936 beyond the found float position. These pessimizations
937 are historical artefacts only. */
939 while (t < strend - prog->minlen) {
941 if (t < check_at - prog->check_offset_min) {
942 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
943 /* Since we moved from the found position,
944 we definitely contradict the found anchored
945 substr. Due to the above check we do not
946 contradict "check" substr.
947 Thus we can arrive here only if check substr
948 is float. Redo checking for "other"=="fixed".
951 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
952 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
953 goto do_other_anchored;
955 /* We don't contradict the found floating substring. */
956 /* XXXX Why not check for STCLASS? */
958 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
959 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
962 /* Position contradicts check-string */
963 /* XXXX probably better to look for check-string
964 than for "\n", so one should lower the limit for t? */
965 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
966 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
967 other_last = strpos = s = t + 1;
972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
973 PL_colors[0], PL_colors[1]));
977 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
978 PL_colors[0], PL_colors[1]));
982 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
985 /* The found string does not prohibit matching at strpos,
986 - no optimization of calling REx engine can be performed,
987 unless it was an MBOL and we are not after MBOL,
988 or a future STCLASS check will fail this. */
990 /* Even in this situation we may use MBOL flag if strpos is offset
991 wrt the start of the string. */
992 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
993 && (strpos != strbeg) && strpos[-1] != '\n'
994 /* May be due to an implicit anchor of m{.*foo} */
995 && !(prog->intflags & PREGf_IMPLICIT))
1000 DEBUG_EXECUTE_r( if (ml_anch)
1001 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1002 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1005 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1007 prog->check_utf8 /* Could be deleted already */
1008 && --BmUSEFUL(prog->check_utf8) < 0
1009 && (prog->check_utf8 == prog->float_utf8)
1011 prog->check_substr /* Could be deleted already */
1012 && --BmUSEFUL(prog->check_substr) < 0
1013 && (prog->check_substr == prog->float_substr)
1016 /* If flags & SOMETHING - do not do it many times on the same match */
1017 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1018 /* XXX Does the destruction order has to change with utf8_target? */
1019 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1020 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1021 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1022 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1023 check = NULL; /* abort */
1025 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1026 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1027 if (prog->intflags & PREGf_IMPLICIT)
1028 prog->extflags &= ~RXf_ANCH_MBOL;
1029 /* XXXX This is a remnant of the old implementation. It
1030 looks wasteful, since now INTUIT can use many
1031 other heuristics. */
1032 prog->extflags &= ~RXf_USE_INTUIT;
1033 /* XXXX What other flags might need to be cleared in this branch? */
1039 /* Last resort... */
1040 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1041 /* trie stclasses are too expensive to use here, we are better off to
1042 leave it to regmatch itself */
1043 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1044 /* minlen == 0 is possible if regstclass is \b or \B,
1045 and the fixed substr is ''$.
1046 Since minlen is already taken into account, s+1 is before strend;
1047 accidentally, minlen >= 1 guaranties no false positives at s + 1
1048 even for \b or \B. But (minlen? 1 : 0) below assumes that
1049 regstclass does not come from lookahead... */
1050 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1051 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1052 const U8* const str = (U8*)STRING(progi->regstclass);
1053 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1054 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1057 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1058 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1059 else if (prog->float_substr || prog->float_utf8)
1060 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1064 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n",
1065 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg)));
1068 s = find_byclass(prog, progi->regstclass, s, endpos, NULL);
1071 const char *what = NULL;
1073 if (endpos == strend) {
1074 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1075 "Could not match STCLASS...\n") );
1078 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1079 "This position contradicts STCLASS...\n") );
1080 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1082 /* Contradict one of substrings */
1083 if (prog->anchored_substr || prog->anchored_utf8) {
1084 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1085 DEBUG_EXECUTE_r( what = "anchored" );
1087 s = HOP3c(t, 1, strend);
1088 if (s + start_shift + end_shift > strend) {
1089 /* XXXX Should be taken into account earlier? */
1090 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1091 "Could not match STCLASS...\n") );
1096 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1097 "Looking for %s substr starting at offset %ld...\n",
1098 what, (long)(s + start_shift - i_strpos)) );
1101 /* Have both, check_string is floating */
1102 if (t + start_shift >= check_at) /* Contradicts floating=check */
1103 goto retry_floating_check;
1104 /* Recheck anchored substring, but not floating... */
1108 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1109 "Looking for anchored substr starting at offset %ld...\n",
1110 (long)(other_last - i_strpos)) );
1111 goto do_other_anchored;
1113 /* Another way we could have checked stclass at the
1114 current position only: */
1119 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1120 "Looking for /%s^%s/m starting at offset %ld...\n",
1121 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1124 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1126 /* Check is floating substring. */
1127 retry_floating_check:
1128 t = check_at - start_shift;
1129 DEBUG_EXECUTE_r( what = "floating" );
1130 goto hop_and_restart;
1133 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1134 "By STCLASS: moving %ld --> %ld\n",
1135 (long)(t - i_strpos), (long)(s - i_strpos))
1139 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1140 "Does not contradict STCLASS...\n");
1145 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1146 PL_colors[4], (check ? "Guessed" : "Giving up"),
1147 PL_colors[5], (long)(s - i_strpos)) );
1150 fail_finish: /* Substring not found */
1151 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1152 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1154 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1155 PL_colors[4], PL_colors[5]));
1159 #define DECL_TRIE_TYPE(scan) \
1160 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1161 trie_type = (scan->flags != EXACT) \
1162 ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \
1163 : (utf8_target ? trie_utf8 : trie_plain)
1165 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1166 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1167 switch (trie_type) { \
1168 case trie_utf8_fold: \
1169 if ( foldlen>0 ) { \
1170 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1175 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1176 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1177 foldlen -= UNISKIP( uvc ); \
1178 uscan = foldbuf + UNISKIP( uvc ); \
1181 case trie_latin_utf8_fold: \
1182 if ( foldlen>0 ) { \
1183 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
1189 uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
1190 foldlen -= UNISKIP( uvc ); \
1191 uscan = foldbuf + UNISKIP( uvc ); \
1195 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
1202 charid = trie->charmap[ uvc ]; \
1206 if (widecharmap) { \
1207 SV** const svpp = hv_fetch(widecharmap, \
1208 (char*)&uvc, sizeof(UV), 0); \
1210 charid = (U16)SvIV(*svpp); \
1215 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1219 && (ln == 1 || folder(s, pat_string, ln)) \
1220 && (!reginfo || regtry(reginfo, &s)) ) \
1226 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1228 while (s + (uskip = UTF8SKIP(s)) <= strend) { \
1234 #define REXEC_FBC_SCAN(CoDe) \
1236 while (s < strend) { \
1242 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1243 REXEC_FBC_UTF8_SCAN( \
1245 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1254 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1257 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1266 #define REXEC_FBC_TRYIT \
1267 if ((!reginfo || regtry(reginfo, &s))) \
1270 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1271 if (utf8_target) { \
1272 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1275 REXEC_FBC_CLASS_SCAN(CoNd); \
1278 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1279 if (utf8_target) { \
1281 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1284 REXEC_FBC_CLASS_SCAN(CoNd); \
1287 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1288 PL_reg_flags |= RF_tainted; \
1289 if (utf8_target) { \
1290 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1293 REXEC_FBC_CLASS_SCAN(CoNd); \
1296 #define DUMP_EXEC_POS(li,s,doutf8) \
1297 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1300 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1301 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1302 tmp = TEST_NON_UTF8(tmp); \
1303 REXEC_FBC_UTF8_SCAN( \
1304 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1313 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1314 if (s == PL_bostr) { \
1318 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1319 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1322 LOAD_UTF8_CHARCLASS_ALNUM(); \
1323 REXEC_FBC_UTF8_SCAN( \
1324 if (tmp == ! (TeSt2_UtF8)) { \
1333 /* The only difference between the BOUND and NBOUND cases is that
1334 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1335 * NBOUND. This is accomplished by passing it in either the if or else clause,
1336 * with the other one being empty */
1337 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1338 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1340 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1341 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1343 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1344 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1346 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1347 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1350 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1351 * be passed in completely with the variable name being tested, which isn't
1352 * such a clean interface, but this is easier to read than it was before. We
1353 * are looking for the boundary (or non-boundary between a word and non-word
1354 * character. The utf8 and non-utf8 cases have the same logic, but the details
1355 * must be different. Find the "wordness" of the character just prior to this
1356 * one, and compare it with the wordness of this one. If they differ, we have
1357 * a boundary. At the beginning of the string, pretend that the previous
1358 * character was a new-line */
1359 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1360 if (utf8_target) { \
1363 else { /* Not utf8 */ \
1364 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1365 tmp = TEST_NON_UTF8(tmp); \
1367 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1376 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1379 /* We know what class REx starts with. Try to find this position... */
1380 /* if reginfo is NULL, its a dryrun */
1381 /* annoyingly all the vars in this routine have different names from their counterparts
1382 in regmatch. /grrr */
1385 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1386 const char *strend, regmatch_info *reginfo)
1389 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1390 char *pat_string; /* The pattern's exactish string */
1391 char *pat_end; /* ptr to end char of pat_string */
1392 re_fold_t folder; /* Function for computing non-utf8 folds */
1393 const U8 *fold_array; /* array for folding ords < 256 */
1396 register STRLEN uskip;
1400 register I32 tmp = 1; /* Scratch variable? */
1401 register const bool utf8_target = PL_reg_match_utf8;
1402 UV utf8_fold_flags = 0;
1403 RXi_GET_DECL(prog,progi);
1405 PERL_ARGS_ASSERT_FIND_BYCLASS;
1407 /* We know what class it must start with. */
1411 if (utf8_target || OP(c) == ANYOFV) {
1412 STRLEN inclasslen = strend - s;
1413 REXEC_FBC_UTF8_CLASS_SCAN(
1414 reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
1417 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1422 if (tmp && (!reginfo || regtry(reginfo, &s)))
1430 if (UTF_PATTERN || utf8_target) {
1431 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1432 goto do_exactf_utf8;
1434 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1435 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1436 goto do_exactf_non_utf8; /* isn't dealt with by these */
1439 if (UTF_PATTERN || utf8_target) {
1440 utf8_fold_flags = 0;
1441 goto do_exactf_utf8;
1443 fold_array = PL_fold_latin1;
1444 folder = foldEQ_latin1;
1445 /* XXX This uses the full utf8 fold because if the pattern contains
1446 * 'ss' it could match LATIN_SMALL_LETTER SHARP_S in the string.
1447 * There could be a new node type, say EXACTFU_SS, which is
1448 * generated by regcomp only if there is an 'ss', and then every
1449 * other case could goto do_exactf_non_utf8;*/
1450 goto do_exactf_utf8;
1453 if (UTF_PATTERN || utf8_target) {
1454 utf8_fold_flags = 0;
1455 goto do_exactf_utf8;
1457 fold_array = PL_fold;
1459 goto do_exactf_non_utf8;
1462 if (UTF_PATTERN || utf8_target) {
1463 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1464 goto do_exactf_utf8;
1466 fold_array = PL_fold_locale;
1467 folder = foldEQ_locale;
1471 do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */
1473 /* The idea in the non-utf8 EXACTF* cases is to first find the
1474 * first character of the EXACTF* node and then, if necessary,
1475 * case-insensitively compare the full text of the node. c1 is the
1476 * first character. c2 is its fold. This logic will not work for
1477 * Unicode semantics and the german sharp ss, which hence should
1478 * not be compiled into a node that gets here. */
1479 pat_string = STRING(c);
1480 ln = STR_LEN(c); /* length to match in octets/bytes */
1482 e = HOP3c(strend, -((I32)ln), s);
1484 if (!reginfo && e < s) {
1485 e = s; /* Due to minlen logic of intuit() */
1489 c2 = fold_array[c1];
1490 if (c1 == c2) { /* If char and fold are the same */
1491 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1494 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1500 /* If one of the operands is in utf8, we can't use the simpler
1501 * folding above, due to the fact that many different characters
1502 * can have the same fold, or portion of a fold, or different-
1504 pat_string = STRING(c);
1505 ln = STR_LEN(c); /* length to match in octets/bytes */
1506 pat_end = pat_string + ln;
1507 lnc = (UTF_PATTERN) /* length to match in characters */
1508 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1511 e = HOP3c(strend, -((I32)lnc), s);
1513 if (!reginfo && e < s) {
1514 e = s; /* Due to minlen logic of intuit() */
1518 char *my_strend= (char *)strend;
1519 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1520 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1521 && (!reginfo || regtry(reginfo, &s)) )
1529 PL_reg_flags |= RF_tainted;
1530 FBC_BOUND(isALNUM_LC,
1531 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1532 isALNUM_LC_utf8((U8*)s));
1535 PL_reg_flags |= RF_tainted;
1536 FBC_NBOUND(isALNUM_LC,
1537 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1538 isALNUM_LC_utf8((U8*)s));
1541 FBC_BOUND(isWORDCHAR,
1543 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1546 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1548 isWORDCHAR_A((U8*)s));
1551 FBC_NBOUND(isWORDCHAR,
1553 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1556 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1558 isWORDCHAR_A((U8*)s));
1561 FBC_BOUND(isWORDCHAR_L1,
1563 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1566 FBC_NBOUND(isWORDCHAR_L1,
1568 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1571 REXEC_FBC_CSCAN_TAINT(
1572 isALNUM_LC_utf8((U8*)s),
1577 REXEC_FBC_CSCAN_PRELOAD(
1578 LOAD_UTF8_CHARCLASS_ALNUM(),
1579 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1580 isWORDCHAR_L1((U8) *s)
1584 REXEC_FBC_CSCAN_PRELOAD(
1585 LOAD_UTF8_CHARCLASS_ALNUM(),
1586 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1591 /* Don't need to worry about utf8, as it can match only a single
1592 * byte invariant character */
1593 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1596 REXEC_FBC_CSCAN_PRELOAD(
1597 LOAD_UTF8_CHARCLASS_ALNUM(),
1598 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1599 ! isWORDCHAR_L1((U8) *s)
1603 REXEC_FBC_CSCAN_PRELOAD(
1604 LOAD_UTF8_CHARCLASS_ALNUM(),
1605 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1616 REXEC_FBC_CSCAN_TAINT(
1617 !isALNUM_LC_utf8((U8*)s),
1622 REXEC_FBC_CSCAN_PRELOAD(
1623 LOAD_UTF8_CHARCLASS_SPACE(),
1624 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1629 REXEC_FBC_CSCAN_PRELOAD(
1630 LOAD_UTF8_CHARCLASS_SPACE(),
1631 *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
1636 /* Don't need to worry about utf8, as it can match only a single
1637 * byte invariant character */
1638 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1641 REXEC_FBC_CSCAN_TAINT(
1642 isSPACE_LC_utf8((U8*)s),
1647 REXEC_FBC_CSCAN_PRELOAD(
1648 LOAD_UTF8_CHARCLASS_SPACE(),
1649 !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1650 ! isSPACE_L1((U8) *s)
1654 REXEC_FBC_CSCAN_PRELOAD(
1655 LOAD_UTF8_CHARCLASS_SPACE(),
1656 !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
1667 REXEC_FBC_CSCAN_TAINT(
1668 !isSPACE_LC_utf8((U8*)s),
1673 REXEC_FBC_CSCAN_PRELOAD(
1674 LOAD_UTF8_CHARCLASS_DIGIT(),
1675 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1680 /* Don't need to worry about utf8, as it can match only a single
1681 * byte invariant character */
1682 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1685 REXEC_FBC_CSCAN_TAINT(
1686 isDIGIT_LC_utf8((U8*)s),
1691 REXEC_FBC_CSCAN_PRELOAD(
1692 LOAD_UTF8_CHARCLASS_DIGIT(),
1693 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1704 REXEC_FBC_CSCAN_TAINT(
1705 !isDIGIT_LC_utf8((U8*)s),
1712 is_LNBREAK_latin1(s)
1724 !is_VERTWS_latin1(s)
1730 is_HORIZWS_latin1(s)
1735 !is_HORIZWS_utf8(s),
1736 !is_HORIZWS_latin1(s)
1743 /* what trie are we using right now */
1745 = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1747 = (reg_trie_data*)progi->data->data[ aho->trie ];
1748 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1750 const char *last_start = strend - trie->minlen;
1752 const char *real_start = s;
1754 STRLEN maxlen = trie->maxlen;
1756 U8 **points; /* map of where we were in the input string
1757 when reading a given char. For ASCII this
1758 is unnecessary overhead as the relationship
1759 is always 1:1, but for Unicode, especially
1760 case folded Unicode this is not true. */
1761 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1765 GET_RE_DEBUG_FLAGS_DECL;
1767 /* We can't just allocate points here. We need to wrap it in
1768 * an SV so it gets freed properly if there is a croak while
1769 * running the match */
1772 sv_points=newSV(maxlen * sizeof(U8 *));
1773 SvCUR_set(sv_points,
1774 maxlen * sizeof(U8 *));
1775 SvPOK_on(sv_points);
1776 sv_2mortal(sv_points);
1777 points=(U8**)SvPV_nolen(sv_points );
1778 if ( trie_type != trie_utf8_fold
1779 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1782 bitmap=(U8*)trie->bitmap;
1784 bitmap=(U8*)ANYOF_BITMAP(c);
1786 /* this is the Aho-Corasick algorithm modified a touch
1787 to include special handling for long "unknown char"
1788 sequences. The basic idea being that we use AC as long
1789 as we are dealing with a possible matching char, when
1790 we encounter an unknown char (and we have not encountered
1791 an accepting state) we scan forward until we find a legal
1793 AC matching is basically that of trie matching, except
1794 that when we encounter a failing transition, we fall back
1795 to the current states "fail state", and try the current char
1796 again, a process we repeat until we reach the root state,
1797 state 1, or a legal transition. If we fail on the root state
1798 then we can either terminate if we have reached an accepting
1799 state previously, or restart the entire process from the beginning
1803 while (s <= last_start) {
1804 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1812 U8 *uscan = (U8*)NULL;
1813 U8 *leftmost = NULL;
1815 U32 accepted_word= 0;
1819 while ( state && uc <= (U8*)strend ) {
1821 U32 word = aho->states[ state ].wordnum;
1825 DEBUG_TRIE_EXECUTE_r(
1826 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1827 dump_exec_pos( (char *)uc, c, strend, real_start,
1828 (char *)uc, utf8_target );
1829 PerlIO_printf( Perl_debug_log,
1830 " Scanning for legal start char...\n");
1834 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1838 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1844 if (uc >(U8*)last_start) break;
1848 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1849 if (!leftmost || lpos < leftmost) {
1850 DEBUG_r(accepted_word=word);
1856 points[pointpos++ % maxlen]= uc;
1857 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
1858 uscan, len, uvc, charid, foldlen,
1860 DEBUG_TRIE_EXECUTE_r({
1861 dump_exec_pos( (char *)uc, c, strend, real_start,
1863 PerlIO_printf(Perl_debug_log,
1864 " Charid:%3u CP:%4"UVxf" ",
1870 word = aho->states[ state ].wordnum;
1872 base = aho->states[ state ].trans.base;
1874 DEBUG_TRIE_EXECUTE_r({
1876 dump_exec_pos( (char *)uc, c, strend, real_start,
1878 PerlIO_printf( Perl_debug_log,
1879 "%sState: %4"UVxf", word=%"UVxf,
1880 failed ? " Fail transition to " : "",
1881 (UV)state, (UV)word);
1887 ( ((offset = base + charid
1888 - 1 - trie->uniquecharcount)) >= 0)
1889 && ((U32)offset < trie->lasttrans)
1890 && trie->trans[offset].check == state
1891 && (tmp=trie->trans[offset].next))
1893 DEBUG_TRIE_EXECUTE_r(
1894 PerlIO_printf( Perl_debug_log," - legal\n"));
1899 DEBUG_TRIE_EXECUTE_r(
1900 PerlIO_printf( Perl_debug_log," - fail\n"));
1902 state = aho->fail[state];
1906 /* we must be accepting here */
1907 DEBUG_TRIE_EXECUTE_r(
1908 PerlIO_printf( Perl_debug_log," - accepting\n"));
1917 if (!state) state = 1;
1920 if ( aho->states[ state ].wordnum ) {
1921 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
1922 if (!leftmost || lpos < leftmost) {
1923 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
1928 s = (char*)leftmost;
1929 DEBUG_TRIE_EXECUTE_r({
1931 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
1932 (UV)accepted_word, (IV)(s - real_start)
1935 if (!reginfo || regtry(reginfo, &s)) {
1941 DEBUG_TRIE_EXECUTE_r({
1942 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
1945 DEBUG_TRIE_EXECUTE_r(
1946 PerlIO_printf( Perl_debug_log,"No match.\n"));
1955 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1965 - regexec_flags - match a regexp against a string
1968 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
1969 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1970 /* strend: pointer to null at end of string */
1971 /* strbeg: real beginning of string */
1972 /* minend: end of match must be >=minend after stringarg. */
1973 /* data: May be used for some additional optimizations.
1974 Currently its only used, with a U32 cast, for transmitting
1975 the ganch offset when doing a /g match. This will change */
1976 /* nosave: For optimizations. */
1979 struct regexp *const prog = (struct regexp *)SvANY(rx);
1980 /*register*/ char *s;
1981 register regnode *c;
1982 /*register*/ char *startpos = stringarg;
1983 I32 minlen; /* must match at least this many chars */
1984 I32 dontbother = 0; /* how many characters not to try at end */
1985 I32 end_shift = 0; /* Same for the end. */ /* CC */
1986 I32 scream_pos = -1; /* Internal iterator of scream. */
1987 char *scream_olds = NULL;
1988 const bool utf8_target = cBOOL(DO_UTF8(sv));
1990 RXi_GET_DECL(prog,progi);
1991 regmatch_info reginfo; /* create some info to pass to regtry etc */
1992 regexp_paren_pair *swap = NULL;
1993 GET_RE_DEBUG_FLAGS_DECL;
1995 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
1996 PERL_UNUSED_ARG(data);
1998 /* Be paranoid... */
1999 if (prog == NULL || startpos == NULL) {
2000 Perl_croak(aTHX_ "NULL regexp parameter");
2004 multiline = prog->extflags & RXf_PMf_MULTILINE;
2005 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2007 RX_MATCH_UTF8_set(rx, utf8_target);
2009 debug_start_match(rx, utf8_target, startpos, strend,
2013 minlen = prog->minlen;
2015 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2016 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2017 "String too short [regexec_flags]...\n"));
2022 /* Check validity of program. */
2023 if (UCHARAT(progi->program) != REG_MAGIC) {
2024 Perl_croak(aTHX_ "corrupted regexp program");
2028 PL_reg_eval_set = 0;
2032 PL_reg_flags |= RF_utf8;
2034 /* Mark beginning of line for ^ and lookbehind. */
2035 reginfo.bol = startpos; /* XXX not used ??? */
2039 /* Mark end of line for $ (and such) */
2042 /* see how far we have to get to not match where we matched before */
2043 reginfo.till = startpos+minend;
2045 /* If there is a "must appear" string, look for it. */
2048 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2050 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2051 reginfo.ganch = startpos + prog->gofs;
2052 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2053 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2054 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2056 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2057 && mg->mg_len >= 0) {
2058 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2059 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2060 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2062 if (prog->extflags & RXf_ANCH_GPOS) {
2063 if (s > reginfo.ganch)
2065 s = reginfo.ganch - prog->gofs;
2066 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2067 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2073 reginfo.ganch = strbeg + PTR2UV(data);
2074 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2075 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2077 } else { /* pos() not defined */
2078 reginfo.ganch = strbeg;
2079 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2080 "GPOS: reginfo.ganch = strbeg\n"));
2083 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2084 /* We have to be careful. If the previous successful match
2085 was from this regex we don't want a subsequent partially
2086 successful match to clobber the old results.
2087 So when we detect this possibility we add a swap buffer
2088 to the re, and switch the buffer each match. If we fail
2089 we switch it back, otherwise we leave it swapped.
2092 /* do we need a save destructor here for eval dies? */
2093 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2095 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2096 re_scream_pos_data d;
2098 d.scream_olds = &scream_olds;
2099 d.scream_pos = &scream_pos;
2100 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2102 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2103 goto phooey; /* not present */
2109 /* Simplest case: anchored match need be tried only once. */
2110 /* [unless only anchor is BOL and multiline is set] */
2111 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2112 if (s == startpos && regtry(®info, &startpos))
2114 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2115 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2120 dontbother = minlen - 1;
2121 end = HOP3c(strend, -dontbother, strbeg) - 1;
2122 /* for multiline we only have to try after newlines */
2123 if (prog->check_substr || prog->check_utf8) {
2124 /* because of the goto we can not easily reuse the macros for bifurcating the
2125 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2128 goto after_try_utf8;
2130 if (regtry(®info, &s)) {
2137 if (prog->extflags & RXf_USE_INTUIT) {
2138 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2147 } /* end search for check string in unicode */
2149 if (s == startpos) {
2150 goto after_try_latin;
2153 if (regtry(®info, &s)) {
2160 if (prog->extflags & RXf_USE_INTUIT) {
2161 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2170 } /* end search for check string in latin*/
2171 } /* end search for check string */
2172 else { /* search for newline */
2174 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2177 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2179 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2180 if (regtry(®info, &s))
2184 } /* end search for newline */
2185 } /* end anchored/multiline check string search */
2187 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2189 /* the warning about reginfo.ganch being used without initialization
2190 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2191 and we only enter this block when the same bit is set. */
2192 char *tmp_s = reginfo.ganch - prog->gofs;
2194 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2199 /* Messy cases: unanchored match. */
2200 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2201 /* we have /x+whatever/ */
2202 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2207 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2208 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2209 ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
2214 DEBUG_EXECUTE_r( did_match = 1 );
2215 if (regtry(®info, &s)) goto got_it;
2217 while (s < strend && *s == ch)
2225 DEBUG_EXECUTE_r( did_match = 1 );
2226 if (regtry(®info, &s)) goto got_it;
2228 while (s < strend && *s == ch)
2233 DEBUG_EXECUTE_r(if (!did_match)
2234 PerlIO_printf(Perl_debug_log,
2235 "Did not find anchored character...\n")
2238 else if (prog->anchored_substr != NULL
2239 || prog->anchored_utf8 != NULL
2240 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2241 && prog->float_max_offset < strend - s)) {
2246 char *last1; /* Last position checked before */
2250 if (prog->anchored_substr || prog->anchored_utf8) {
2251 if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
2252 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2253 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
2254 back_max = back_min = prog->anchored_offset;
2256 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2257 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2258 must = utf8_target ? prog->float_utf8 : prog->float_substr;
2259 back_max = prog->float_max_offset;
2260 back_min = prog->float_min_offset;
2264 if (must == &PL_sv_undef)
2265 /* could not downgrade utf8 check substring, so must fail */
2271 last = HOP3c(strend, /* Cannot start after this */
2272 -(I32)(CHR_SVLEN(must)
2273 - (SvTAIL(must) != 0) + back_min), strbeg);
2276 last1 = HOPc(s, -1);
2278 last1 = s - 1; /* bogus */
2280 /* XXXX check_substr already used to find "s", can optimize if
2281 check_substr==must. */
2283 dontbother = end_shift;
2284 strend = HOPc(strend, -dontbother);
2285 while ( (s <= last) &&
2286 ((flags & REXEC_SCREAM)
2287 ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg,
2288 end_shift, &scream_pos, 0))
2289 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2290 (unsigned char*)strend, must,
2291 multiline ? FBMrf_MULTILINE : 0))) ) {
2292 /* we may be pointing at the wrong string */
2293 if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
2294 s = strbeg + (s - SvPVX_const(sv));
2295 DEBUG_EXECUTE_r( did_match = 1 );
2296 if (HOPc(s, -back_max) > last1) {
2297 last1 = HOPc(s, -back_min);
2298 s = HOPc(s, -back_max);
2301 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2303 last1 = HOPc(s, -back_min);
2307 while (s <= last1) {
2308 if (regtry(®info, &s))
2314 while (s <= last1) {
2315 if (regtry(®info, &s))
2321 DEBUG_EXECUTE_r(if (!did_match) {
2322 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2323 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2324 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2325 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2326 ? "anchored" : "floating"),
2327 quoted, RE_SV_TAIL(must));
2331 else if ( (c = progi->regstclass) ) {
2333 const OPCODE op = OP(progi->regstclass);
2334 /* don't bother with what can't match */
2335 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2336 strend = HOPc(strend, -(minlen - 1));
2339 SV * const prop = sv_newmortal();
2340 regprop(prog, prop, c);
2342 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2344 PerlIO_printf(Perl_debug_log,
2345 "Matching stclass %.*s against %s (%d bytes)\n",
2346 (int)SvCUR(prop), SvPVX_const(prop),
2347 quoted, (int)(strend - s));
2350 if (find_byclass(prog, c, s, strend, ®info))
2352 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2356 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2361 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
2362 utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
2363 float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
2365 if (flags & REXEC_SCREAM) {
2366 last = screaminstr(sv, float_real, s - strbeg,
2367 end_shift, &scream_pos, 1); /* last one */
2369 last = scream_olds; /* Only one occurrence. */
2370 /* we may be pointing at the wrong string */
2371 else if (RXp_MATCH_COPIED(prog))
2372 s = strbeg + (s - SvPVX_const(sv));
2376 const char * const little = SvPV_const(float_real, len);
2378 if (SvTAIL(float_real)) {
2379 if (memEQ(strend - len + 1, little, len - 1))
2380 last = strend - len + 1;
2381 else if (!multiline)
2382 last = memEQ(strend - len, little, len)
2383 ? strend - len : NULL;
2389 last = rninstr(s, strend, little, little + len);
2391 last = strend; /* matching "$" */
2396 PerlIO_printf(Perl_debug_log,
2397 "%sCan't trim the tail, match fails (should not happen)%s\n",
2398 PL_colors[4], PL_colors[5]));
2399 goto phooey; /* Should not happen! */
2401 dontbother = strend - last + prog->float_min_offset;
2403 if (minlen && (dontbother < minlen))
2404 dontbother = minlen - 1;
2405 strend -= dontbother; /* this one's always in bytes! */
2406 /* We don't know much -- general case. */
2409 if (regtry(®info, &s))
2418 if (regtry(®info, &s))
2420 } while (s++ < strend);
2429 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2431 if (PL_reg_eval_set)
2432 restore_pos(aTHX_ prog);
2433 if (RXp_PAREN_NAMES(prog))
2434 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2436 /* make sure $`, $&, $', and $digit will work later */
2437 if ( !(flags & REXEC_NOT_FIRST) ) {
2438 RX_MATCH_COPY_FREE(rx);
2439 if (flags & REXEC_COPY_STR) {
2440 const I32 i = PL_regeol - startpos + (stringarg - strbeg);
2441 #ifdef PERL_OLD_COPY_ON_WRITE
2443 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2445 PerlIO_printf(Perl_debug_log,
2446 "Copy on write: regexp capture, type %d\n",
2449 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2450 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2451 assert (SvPOKp(prog->saved_copy));
2455 RX_MATCH_COPIED_on(rx);
2456 s = savepvn(strbeg, i);
2462 prog->subbeg = strbeg;
2463 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2470 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2471 PL_colors[4], PL_colors[5]));
2472 if (PL_reg_eval_set)
2473 restore_pos(aTHX_ prog);
2475 /* we failed :-( roll it back */
2476 Safefree(prog->offs);
2485 - regtry - try match at specific point
2487 STATIC I32 /* 0 failure, 1 success */
2488 S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
2492 REGEXP *const rx = reginfo->prog;
2493 regexp *const prog = (struct regexp *)SvANY(rx);
2494 RXi_GET_DECL(prog,progi);
2495 GET_RE_DEBUG_FLAGS_DECL;
2497 PERL_ARGS_ASSERT_REGTRY;
2499 reginfo->cutpoint=NULL;
2501 if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
2504 PL_reg_eval_set = RS_init;
2505 DEBUG_EXECUTE_r(DEBUG_s(
2506 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2507 (IV)(PL_stack_sp - PL_stack_base));
2510 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2511 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2513 /* Apparently this is not needed, judging by wantarray. */
2514 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2515 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2518 /* Make $_ available to executed code. */
2519 if (reginfo->sv != DEFSV) {
2521 DEFSV_set(reginfo->sv);
2524 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2525 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2526 /* prepare for quick setting of pos */
2527 #ifdef PERL_OLD_COPY_ON_WRITE
2528 if (SvIsCOW(reginfo->sv))
2529 sv_force_normal_flags(reginfo->sv, 0);
2531 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2532 &PL_vtbl_mglob, NULL, 0);
2536 PL_reg_oldpos = mg->mg_len;
2537 SAVEDESTRUCTOR_X(restore_pos, prog);
2539 if (!PL_reg_curpm) {
2540 Newxz(PL_reg_curpm, 1, PMOP);
2543 SV* const repointer = &PL_sv_undef;
2544 /* this regexp is also owned by the new PL_reg_curpm, which
2545 will try to free it. */
2546 av_push(PL_regex_padav, repointer);
2547 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2548 PL_regex_pad = AvARRAY(PL_regex_padav);
2553 /* It seems that non-ithreads works both with and without this code.
2554 So for efficiency reasons it seems best not to have the code
2555 compiled when it is not needed. */
2556 /* This is safe against NULLs: */
2557 ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
2558 /* PM_reg_curpm owns a reference to this regexp. */
2559 (void)ReREFCNT_inc(rx);
2561 PM_SETRE(PL_reg_curpm, rx);
2562 PL_reg_oldcurpm = PL_curpm;
2563 PL_curpm = PL_reg_curpm;
2564 if (RXp_MATCH_COPIED(prog)) {
2565 /* Here is a serious problem: we cannot rewrite subbeg,
2566 since it may be needed if this match fails. Thus
2567 $` inside (?{}) could fail... */
2568 PL_reg_oldsaved = prog->subbeg;
2569 PL_reg_oldsavedlen = prog->sublen;
2570 #ifdef PERL_OLD_COPY_ON_WRITE
2571 PL_nrs = prog->saved_copy;
2573 RXp_MATCH_COPIED_off(prog);
2576 PL_reg_oldsaved = NULL;
2577 prog->subbeg = PL_bostr;
2578 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2580 DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
2581 prog->offs[0].start = *startpos - PL_bostr;
2582 PL_reginput = *startpos;
2583 PL_reglastparen = &prog->lastparen;
2584 PL_reglastcloseparen = &prog->lastcloseparen;
2585 prog->lastparen = 0;
2586 prog->lastcloseparen = 0;
2588 PL_regoffs = prog->offs;
2589 if (PL_reg_start_tmpl <= prog->nparens) {
2590 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2591 if(PL_reg_start_tmp)
2592 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2594 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2597 /* XXXX What this code is doing here?!!! There should be no need
2598 to do this again and again, PL_reglastparen should take care of
2601 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2602 * Actually, the code in regcppop() (which Ilya may be meaning by
2603 * PL_reglastparen), is not needed at all by the test suite
2604 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2605 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2606 * Meanwhile, this code *is* needed for the
2607 * above-mentioned test suite tests to succeed. The common theme
2608 * on those tests seems to be returning null fields from matches.
2609 * --jhi updated by dapm */
2611 if (prog->nparens) {
2612 regexp_paren_pair *pp = PL_regoffs;
2614 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2622 if (regmatch(reginfo, progi->program + 1)) {
2623 PL_regoffs[0].end = PL_reginput - PL_bostr;
2626 if (reginfo->cutpoint)
2627 *startpos= reginfo->cutpoint;
2628 REGCP_UNWIND(lastcp);
2633 #define sayYES goto yes
2634 #define sayNO goto no
2635 #define sayNO_SILENT goto no_silent
2637 /* we dont use STMT_START/END here because it leads to
2638 "unreachable code" warnings, which are bogus, but distracting. */
2639 #define CACHEsayNO \
2640 if (ST.cache_mask) \
2641 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2644 /* this is used to determine how far from the left messages like
2645 'failed...' are printed. It should be set such that messages
2646 are inline with the regop output that created them.
2648 #define REPORT_CODE_OFF 32
2651 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2652 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2654 #define SLAB_FIRST(s) (&(s)->states[0])
2655 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2657 /* grab a new slab and return the first slot in it */
2659 STATIC regmatch_state *
2662 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2665 regmatch_slab *s = PL_regmatch_slab->next;
2667 Newx(s, 1, regmatch_slab);
2668 s->prev = PL_regmatch_slab;
2670 PL_regmatch_slab->next = s;
2672 PL_regmatch_slab = s;
2673 return SLAB_FIRST(s);
2677 /* push a new state then goto it */
2679 #define PUSH_STATE_GOTO(state, node) \
2681 st->resume_state = state; \
2684 /* push a new state with success backtracking, then goto it */
2686 #define PUSH_YES_STATE_GOTO(state, node) \
2688 st->resume_state = state; \
2689 goto push_yes_state;
2695 regmatch() - main matching routine
2697 This is basically one big switch statement in a loop. We execute an op,
2698 set 'next' to point the next op, and continue. If we come to a point which
2699 we may need to backtrack to on failure such as (A|B|C), we push a
2700 backtrack state onto the backtrack stack. On failure, we pop the top
2701 state, and re-enter the loop at the state indicated. If there are no more
2702 states to pop, we return failure.
2704 Sometimes we also need to backtrack on success; for example /A+/, where
2705 after successfully matching one A, we need to go back and try to
2706 match another one; similarly for lookahead assertions: if the assertion
2707 completes successfully, we backtrack to the state just before the assertion
2708 and then carry on. In these cases, the pushed state is marked as
2709 'backtrack on success too'. This marking is in fact done by a chain of
2710 pointers, each pointing to the previous 'yes' state. On success, we pop to
2711 the nearest yes state, discarding any intermediate failure-only states.
2712 Sometimes a yes state is pushed just to force some cleanup code to be
2713 called at the end of a successful match or submatch; e.g. (??{$re}) uses
2714 it to free the inner regex.
2716 Note that failure backtracking rewinds the cursor position, while
2717 success backtracking leaves it alone.
2719 A pattern is complete when the END op is executed, while a subpattern
2720 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
2721 ops trigger the "pop to last yes state if any, otherwise return true"
2724 A common convention in this function is to use A and B to refer to the two
2725 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
2726 the subpattern to be matched possibly multiple times, while B is the entire
2727 rest of the pattern. Variable and state names reflect this convention.
2729 The states in the main switch are the union of ops and failure/success of
2730 substates associated with with that op. For example, IFMATCH is the op
2731 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
2732 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
2733 successfully matched A and IFMATCH_A_fail is a state saying that we have
2734 just failed to match A. Resume states always come in pairs. The backtrack
2735 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
2736 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
2737 on success or failure.
2739 The struct that holds a backtracking state is actually a big union, with
2740 one variant for each major type of op. The variable st points to the
2741 top-most backtrack struct. To make the code clearer, within each
2742 block of code we #define ST to alias the relevant union.
2744 Here's a concrete example of a (vastly oversimplified) IFMATCH
2750 #define ST st->u.ifmatch
2752 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2753 ST.foo = ...; // some state we wish to save
2755 // push a yes backtrack state with a resume value of
2756 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
2758 PUSH_YES_STATE_GOTO(IFMATCH_A, A);
2761 case IFMATCH_A: // we have successfully executed A; now continue with B
2763 bar = ST.foo; // do something with the preserved value
2766 case IFMATCH_A_fail: // A failed, so the assertion failed
2767 ...; // do some housekeeping, then ...
2768 sayNO; // propagate the failure
2775 For any old-timers reading this who are familiar with the old recursive
2776 approach, the code above is equivalent to:
2778 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
2787 ...; // do some housekeeping, then ...
2788 sayNO; // propagate the failure
2791 The topmost backtrack state, pointed to by st, is usually free. If you
2792 want to claim it, populate any ST.foo fields in it with values you wish to
2793 save, then do one of
2795 PUSH_STATE_GOTO(resume_state, node);
2796 PUSH_YES_STATE_GOTO(resume_state, node);
2798 which sets that backtrack state's resume value to 'resume_state', pushes a
2799 new free entry to the top of the backtrack stack, then goes to 'node'.
2800 On backtracking, the free slot is popped, and the saved state becomes the
2801 new free state. An ST.foo field in this new top state can be temporarily
2802 accessed to retrieve values, but once the main loop is re-entered, it
2803 becomes available for reuse.
2805 Note that the depth of the backtrack stack constantly increases during the
2806 left-to-right execution of the pattern, rather than going up and down with
2807 the pattern nesting. For example the stack is at its maximum at Z at the
2808 end of the pattern, rather than at X in the following:
2810 /(((X)+)+)+....(Y)+....Z/
2812 The only exceptions to this are lookahead/behind assertions and the cut,
2813 (?>A), which pop all the backtrack states associated with A before
2816 Backtrack state structs are allocated in slabs of about 4K in size.
2817 PL_regmatch_state and st always point to the currently active state,
2818 and PL_regmatch_slab points to the slab currently containing
2819 PL_regmatch_state. The first time regmatch() is called, the first slab is
2820 allocated, and is never freed until interpreter destruction. When the slab
2821 is full, a new one is allocated and chained to the end. At exit from
2822 regmatch(), slabs allocated since entry are freed.
2827 #define DEBUG_STATE_pp(pp) \
2829 DUMP_EXEC_POS(locinput, scan, utf8_target); \
2830 PerlIO_printf(Perl_debug_log, \
2831 " %*s"pp" %s%s%s%s%s\n", \
2833 PL_reg_name[st->resume_state], \
2834 ((st==yes_state||st==mark_state) ? "[" : ""), \
2835 ((st==yes_state) ? "Y" : ""), \
2836 ((st==mark_state) ? "M" : ""), \
2837 ((st==yes_state||st==mark_state) ? "]" : "") \
2842 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
2847 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
2848 const char *start, const char *end, const char *blurb)
2850 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
2852 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
2857 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
2858 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
2860 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
2861 start, end - start, 60);
2863 PerlIO_printf(Perl_debug_log,
2864 "%s%s REx%s %s against %s\n",
2865 PL_colors[4], blurb, PL_colors[5], s0, s1);
2867 if (utf8_target||utf8_pat)
2868 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
2869 utf8_pat ? "pattern" : "",
2870 utf8_pat && utf8_target ? " and " : "",
2871 utf8_target ? "string" : ""
2877 S_dump_exec_pos(pTHX_ const char *locinput,
2878 const regnode *scan,
2879 const char *loc_regeol,
2880 const char *loc_bostr,
2881 const char *loc_reg_starttry,
2882 const bool utf8_target)
2884 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
2885 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2886 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
2887 /* The part of the string before starttry has one color
2888 (pref0_len chars), between starttry and current
2889 position another one (pref_len - pref0_len chars),
2890 after the current position the third one.
2891 We assume that pref0_len <= pref_len, otherwise we
2892 decrease pref0_len. */
2893 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
2894 ? (5 + taill) - l : locinput - loc_bostr;
2897 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
2899 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2901 pref0_len = pref_len - (locinput - loc_reg_starttry);
2902 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
2903 l = ( loc_regeol - locinput > (5 + taill) - pref_len
2904 ? (5 + taill) - pref_len : loc_regeol - locinput);
2905 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2909 if (pref0_len > pref_len)
2910 pref0_len = pref_len;
2912 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
2914 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
2915 (locinput - pref_len),pref0_len, 60, 4, 5);
2917 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
2918 (locinput - pref_len + pref0_len),
2919 pref_len - pref0_len, 60, 2, 3);
2921 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
2922 locinput, loc_regeol - locinput, 10, 0, 1);
2924 const STRLEN tlen=len0+len1+len2;
2925 PerlIO_printf(Perl_debug_log,
2926 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
2927 (IV)(locinput - loc_bostr),
2930 (docolor ? "" : "> <"),
2932 (int)(tlen > 19 ? 0 : 19 - tlen),
2939 /* reg_check_named_buff_matched()
2940 * Checks to see if a named buffer has matched. The data array of
2941 * buffer numbers corresponding to the buffer is expected to reside
2942 * in the regexp->data->data array in the slot stored in the ARG() of
2943 * node involved. Note that this routine doesn't actually care about the
2944 * name, that information is not preserved from compilation to execution.
2945 * Returns the index of the leftmost defined buffer with the given name
2946 * or 0 if non of the buffers matched.
2949 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
2952 RXi_GET_DECL(rex,rexi);
2953 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
2954 I32 *nums=(I32*)SvPVX(sv_dat);
2956 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
2958 for ( n=0; n<SvIVX(sv_dat); n++ ) {
2959 if ((I32)*PL_reglastparen >= nums[n] &&
2960 PL_regoffs[nums[n]].end != -1)
2969 /* free all slabs above current one - called during LEAVE_SCOPE */
2972 S_clear_backtrack_stack(pTHX_ void *p)
2974 regmatch_slab *s = PL_regmatch_slab->next;
2979 PL_regmatch_slab->next = NULL;
2981 regmatch_slab * const osl = s;
2988 #define SETREX(Re1,Re2) \
2989 if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \
2992 STATIC I32 /* 0 failure, 1 success */
2993 S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
2995 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2999 register const bool utf8_target = PL_reg_match_utf8;
3000 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3001 REGEXP *rex_sv = reginfo->prog;
3002 regexp *rex = (struct regexp *)SvANY(rex_sv);
3003 RXi_GET_DECL(rex,rexi);
3005 /* the current state. This is a cached copy of PL_regmatch_state */
3006 register regmatch_state *st;
3007 /* cache heavy used fields of st in registers */
3008 register regnode *scan;
3009 register regnode *next;
3010 register U32 n = 0; /* general value; init to avoid compiler warning */
3011 register I32 ln = 0; /* len or last; init to avoid compiler warning */
3012 register char *locinput = PL_reginput;
3013 register I32 nextchr; /* is always set to UCHARAT(locinput) */
3015 bool result = 0; /* return value of S_regmatch */
3016 int depth = 0; /* depth of backtrack stack */
3017 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3018 const U32 max_nochange_depth =
3019 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3020 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3021 regmatch_state *yes_state = NULL; /* state to pop to on success of
3023 /* mark_state piggy backs on the yes_state logic so that when we unwind
3024 the stack on success we can update the mark_state as we go */
3025 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3026 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3027 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3029 bool no_final = 0; /* prevent failure from backtracking? */
3030 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3031 char *startpoint = PL_reginput;
3032 SV *popmark = NULL; /* are we looking for a mark? */
3033 SV *sv_commit = NULL; /* last mark name seen in failure */
3034 SV *sv_yes_mark = NULL; /* last mark name we have seen
3035 during a successful match */
3036 U32 lastopen = 0; /* last open we saw */
3037 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3038 SV* const oreplsv = GvSV(PL_replgv);
3039 /* these three flags are set by various ops to signal information to
3040 * the very next op. They have a useful lifetime of exactly one loop
3041 * iteration, and are not preserved or restored by state pushes/pops
3043 bool sw = 0; /* the condition value in (?(cond)a|b) */
3044 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3045 int logical = 0; /* the following EVAL is:
3049 or the following IFMATCH/UNLESSM is:
3050 false: plain (?=foo)
3051 true: used as a condition: (?(?=foo))
3054 GET_RE_DEBUG_FLAGS_DECL;
3057 PERL_ARGS_ASSERT_REGMATCH;
3059 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3060 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3062 /* on first ever call to regmatch, allocate first slab */
3063 if (!PL_regmatch_slab) {
3064 Newx(PL_regmatch_slab, 1, regmatch_slab);
3065 PL_regmatch_slab->prev = NULL;
3066 PL_regmatch_slab->next = NULL;
3067 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3070 oldsave = PL_savestack_ix;
3071 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3072 SAVEVPTR(PL_regmatch_slab);
3073 SAVEVPTR(PL_regmatch_state);
3075 /* grab next free state slot */
3076 st = ++PL_regmatch_state;
3077 if (st > SLAB_LAST(PL_regmatch_slab))
3078 st = PL_regmatch_state = S_push_slab(aTHX);
3080 /* Note that nextchr is a byte even in UTF */
3081 nextchr = UCHARAT(locinput);
3083 while (scan != NULL) {
3086 SV * const prop = sv_newmortal();
3087 regnode *rnext=regnext(scan);
3088 DUMP_EXEC_POS( locinput, scan, utf8_target );
3089 regprop(rex, prop, scan);
3091 PerlIO_printf(Perl_debug_log,
3092 "%3"IVdf":%*s%s(%"IVdf")\n",
3093 (IV)(scan - rexi->program), depth*2, "",
3095 (PL_regkind[OP(scan)] == END || !rnext) ?
3096 0 : (IV)(rnext - rexi->program));
3099 next = scan + NEXT_OFF(scan);
3102 state_num = OP(scan);
3104 REH_CALL_EXEC_NODE_HOOK(rex, scan, reginfo, st);
3107 assert(PL_reglastparen == &rex->lastparen);
3108 assert(PL_reglastcloseparen == &rex->lastcloseparen);
3109 assert(PL_regoffs == rex->offs);
3111 switch (state_num) {
3113 if (locinput == PL_bostr)
3115 /* reginfo->till = reginfo->bol; */
3120 if (locinput == PL_bostr ||
3121 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
3127 if (locinput == PL_bostr)
3131 if (locinput == reginfo->ganch)
3136 /* update the startpoint */
3137 st->u.keeper.val = PL_regoffs[0].start;
3138 PL_reginput = locinput;
3139 PL_regoffs[0].start = locinput - PL_bostr;
3140 PUSH_STATE_GOTO(KEEPS_next, next);
3142 case KEEPS_next_fail:
3143 /* rollback the start point change */
3144 PL_regoffs[0].start = st->u.keeper.val;
3150 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3155 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
3157 if (PL_regeol - locinput > 1)
3161 if (PL_regeol != locinput)
3165 if (!nextchr && locinput >= PL_regeol)
3168 locinput += PL_utf8skip[nextchr];
3169 if (locinput > PL_regeol)
3171 nextchr = UCHARAT(locinput);
3174 nextchr = UCHARAT(++locinput);
3177 if (!nextchr && locinput >= PL_regeol)
3179 nextchr = UCHARAT(++locinput);
3182 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
3185 locinput += PL_utf8skip[nextchr];
3186 if (locinput > PL_regeol)
3188 nextchr = UCHARAT(locinput);
3191 nextchr = UCHARAT(++locinput);
3195 #define ST st->u.trie
3197 /* In this case the charclass data is available inline so
3198 we can fail fast without a lot of extra overhead.
3200 if (scan->flags == EXACT || !utf8_target) {
3201 if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
3203 PerlIO_printf(Perl_debug_log,
3204 "%*s %sfailed to match trie start class...%s\n",
3205 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3213 /* the basic plan of execution of the trie is:
3214 * At the beginning, run though all the states, and
3215 * find the longest-matching word. Also remember the position
3216 * of the shortest matching word. For example, this pattern:
3219 * when matched against the string "abcde", will generate
3220 * accept states for all words except 3, with the longest
3221 * matching word being 4, and the shortest being 1 (with
3222 * the position being after char 1 of the string).
3224 * Then for each matching word, in word order (i.e. 1,2,4,5),
3225 * we run the remainder of the pattern; on each try setting
3226 * the current position to the character following the word,
3227 * returning to try the next word on failure.
3229 * We avoid having to build a list of words at runtime by
3230 * using a compile-time structure, wordinfo[].prev, which
3231 * gives, for each word, the previous accepting word (if any).
3232 * In the case above it would contain the mappings 1->2, 2->0,
3233 * 3->0, 4->5, 5->1. We can use this table to generate, from
3234 * the longest word (4 above), a list of all words, by
3235 * following the list of prev pointers; this gives us the
3236 * unordered list 4,5,1,2. Then given the current word we have
3237 * just tried, we can go through the list and find the
3238 * next-biggest word to try (so if we just failed on word 2,
3239 * the next in the list is 4).
3241 * Since at runtime we don't record the matching position in
3242 * the string for each word, we have to work that out for
3243 * each word we're about to process. The wordinfo table holds
3244 * the character length of each word; given that we recorded
3245 * at the start: the position of the shortest word and its
3246 * length in chars, we just need to move the pointer the
3247 * difference between the two char lengths. Depending on
3248 * Unicode status and folding, that's cheap or expensive.
3250 * This algorithm is optimised for the case where are only a
3251 * small number of accept states, i.e. 0,1, or maybe 2.
3252 * With lots of accepts states, and having to try all of them,
3253 * it becomes quadratic on number of accept states to find all
3258 /* what type of TRIE am I? (utf8 makes this contextual) */
3259 DECL_TRIE_TYPE(scan);
3261 /* what trie are we using right now */
3262 reg_trie_data * const trie
3263 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3264 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3265 U32 state = trie->startstate;
3267 if (trie->bitmap && trie_type != trie_utf8_fold &&
3268 !TRIE_BITMAP_TEST(trie,*locinput)
3270 if (trie->states[ state ].wordnum) {
3272 PerlIO_printf(Perl_debug_log,
3273 "%*s %smatched empty string...%s\n",
3274 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3280 PerlIO_printf(Perl_debug_log,
3281 "%*s %sfailed to match trie start class...%s\n",
3282 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3289 U8 *uc = ( U8* )locinput;
3293 U8 *uscan = (U8*)NULL;
3294 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3295 U32 charcount = 0; /* how many input chars we have matched */
3296 U32 accepted = 0; /* have we seen any accepting states? */
3299 ST.jump = trie->jump;
3302 ST.longfold = FALSE; /* char longer if folded => it's harder */
3305 /* fully traverse the TRIE; note the position of the
3306 shortest accept state and the wordnum of the longest
3309 while ( state && uc <= (U8*)PL_regeol ) {
3310 U32 base = trie->states[ state ].trans.base;
3314 wordnum = trie->states[ state ].wordnum;
3316 if (wordnum) { /* it's an accept state */
3319 /* record first match position */
3321 ST.firstpos = (U8*)locinput;
3326 ST.firstchars = charcount;
3329 if (!ST.nextword || wordnum < ST.nextword)
3330 ST.nextword = wordnum;
3331 ST.topword = wordnum;
3334 DEBUG_TRIE_EXECUTE_r({
3335 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3336 PerlIO_printf( Perl_debug_log,
3337 "%*s %sState: %4"UVxf" Accepted: %c ",
3338 2+depth * 2, "", PL_colors[4],
3339 (UV)state, (accepted ? 'Y' : 'N'));
3342 /* read a char and goto next state */
3345 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3346 uscan, len, uvc, charid, foldlen,
3353 base + charid - 1 - trie->uniquecharcount)) >= 0)
3355 && ((U32)offset < trie->lasttrans)
3356 && trie->trans[offset].check == state)
3358 state = trie->trans[offset].next;
3369 DEBUG_TRIE_EXECUTE_r(
3370 PerlIO_printf( Perl_debug_log,
3371 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
3372 charid, uvc, (UV)state, PL_colors[5] );
3378 /* calculate total number of accept states */
3383 w = trie->wordinfo[w].prev;
3386 ST.accepted = accepted;
3390 PerlIO_printf( Perl_debug_log,
3391 "%*s %sgot %"IVdf" possible matches%s\n",
3392 REPORT_CODE_OFF + depth * 2, "",
3393 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
3395 goto trie_first_try; /* jump into the fail handler */
3399 case TRIE_next_fail: /* we failed - try next alternative */
3401 REGCP_UNWIND(ST.cp);
3402 for (n = *PL_reglastparen; n > ST.lastparen; n--)
3403 PL_regoffs[n].end = -1;
3404 *PL_reglastparen = n;
3406 if (!--ST.accepted) {
3408 PerlIO_printf( Perl_debug_log,
3409 "%*s %sTRIE failed...%s\n",
3410 REPORT_CODE_OFF+depth*2, "",
3417 /* Find next-highest word to process. Note that this code
3418 * is O(N^2) per trie run (O(N) per branch), so keep tight */
3419 register U16 min = 0;
3421 register U16 const nextword = ST.nextword;
3422 register reg_trie_wordinfo * const wordinfo
3423 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
3424 for (word=ST.topword; word; word=wordinfo[word].prev) {
3425 if (word > nextword && (!min || word < min))
3438 ST.lastparen = *PL_reglastparen;
3442 /* find start char of end of current word */
3444 U32 chars; /* how many chars to skip */
3445 U8 *uc = ST.firstpos;
3446 reg_trie_data * const trie
3447 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
3449 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
3451 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
3455 /* the hard option - fold each char in turn and find
3456 * its folded length (which may be different */
3457 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
3465 uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
3473 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
3478 uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
3492 PL_reginput = (char *)uc;
3495 scan = (ST.jump && ST.jump[ST.nextword])
3496 ? ST.me + ST.jump[ST.nextword]
3500 PerlIO_printf( Perl_debug_log,
3501 "%*s %sTRIE matched word #%d, continuing%s\n",
3502 REPORT_CODE_OFF+depth*2, "",
3509 if (ST.accepted > 1 || has_cutgroup) {
3510 PUSH_STATE_GOTO(TRIE_next, scan);
3513 /* only one choice left - just continue */
3515 AV *const trie_words
3516 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
3517 SV ** const tmp = av_fetch( trie_words,
3519 SV *sv= tmp ? sv_newmortal() : NULL;
3521 PerlIO_printf( Perl_debug_log,
3522 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
3523 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
3525 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
3526 PL_colors[0], PL_colors[1],
3527 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
3529 : "not compiled under -Dr",
3533 locinput = PL_reginput;
3534 nextchr = UCHARAT(locinput);
3535 continue; /* execute rest of RE */
3540 char *s = STRING(scan);
3542 if (utf8_target != UTF_PATTERN) {
3543 /* The target and the pattern have differing utf8ness. */
3545 const char * const e = s + ln;
3548 /* The target is utf8, the pattern is not utf8. */
3553 if (NATIVE_TO_UNI(*(U8*)s) !=
3554 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
3562 /* The target is not utf8, the pattern is utf8. */
3567 if (NATIVE_TO_UNI(*((U8*)l)) !=
3568 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
3576 nextchr = UCHARAT(locinput);
3579 /* The target and the pattern have the same utf8ness. */
3580 /* Inline the first character, for speed. */
3581 if (UCHARAT(s) != nextchr)
3583 if (PL_regeol - locinput < ln)
3585 if (ln > 1 && memNE(s, locinput, ln))
3588 nextchr = UCHARAT(locinput);
3593 const U8 * fold_array;
3595 U32 fold_utf8_flags;
3597 PL_reg_flags |= RF_tainted;
3598 folder = foldEQ_locale;
3599 fold_array = PL_fold_locale;
3600 fold_utf8_flags = FOLDEQ_UTF8_LOCALE;
3604 folder = foldEQ_latin1;
3605 fold_array = PL_fold_latin1;
3606 fold_utf8_flags = 0;
3610 folder = foldEQ_latin1;
3611 fold_array = PL_fold_latin1;
3612 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
3617 fold_array = PL_fold;
3618 fold_utf8_flags = 0;
3624 if (utf8_target || UTF_PATTERN) {
3625 /* Either target or the pattern are utf8. */
3626 const char * const l = locinput;
3627 char *e = PL_regeol;
3629 if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN),
3630 l, &e, 0, utf8_target, fold_utf8_flags))
3635 nextchr = UCHARAT(locinput);
3639 /* Neither the target nor the pattern are utf8 */
3640 if (UCHARAT(s) != nextchr &&
3641 UCHARAT(s) != fold_array[nextchr])
3645 if (PL_regeol - locinput < ln)
3647 if (ln > 1 && ! folder(s, locinput, ln))
3650 nextchr = UCHARAT(locinput);
3654 /* XXX Could improve efficiency by separating these all out using a
3655 * macro or in-line function. At that point regcomp.c would no longer
3656 * have to set the FLAGS fields of these */
3659 PL_reg_flags |= RF_tainted;
3667 /* was last char in word? */
3669 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
3670 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
3672 if (locinput == PL_bostr)
3675 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
3677 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags);
3679 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
3680 ln = isALNUM_uni(ln);
3681 LOAD_UTF8_CHARCLASS_ALNUM();
3682 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target);
3685 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
3686 n = isALNUM_LC_utf8((U8*)locinput);
3691 /* Here the string isn't utf8, or is utf8 and only ascii
3692 * characters are to match \w. In the latter case looking at
3693 * the byte just prior to the current one may be just the final
3694 * byte of a multi-byte character. This is ok. There are two
3696 * 1) it is a single byte character, and then the test is doing
3697 * just what it's supposed to.
3698 * 2) it is a multi-byte character, in which case the final
3699 * byte is never mistakable for ASCII, and so the test
3700 * will say it is not a word character, which is the
3701 * correct answer. */
3702 ln = (locinput != PL_bostr) ?
3703 UCHARAT(locinput - 1) : '\n';
3704 switch (FLAGS(scan)) {
3705 case REGEX_UNICODE_CHARSET:
3706 ln = isWORDCHAR_L1(ln);
3707 n = isWORDCHAR_L1(nextchr);
3709 case REGEX_LOCALE_CHARSET:
3710 ln = isALNUM_LC(ln);
3711 n = isALNUM_LC(nextchr);
3713 case REGEX_DEPENDS_CHARSET:
3715 n = isALNUM(nextchr);
3717 case REGEX_ASCII_RESTRICTED_CHARSET:
3718 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
3719 ln = isWORDCHAR_A(ln);
3720 n = isWORDCHAR_A(nextchr);
3723 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
3727 /* Note requires that all BOUNDs be lower than all NBOUNDs in
3729 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
3734 if (utf8_target || state_num == ANYOFV) {
3735 STRLEN inclasslen = PL_regeol - locinput;
3736 if (locinput >= PL_regeol)
3739 if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
3741 locinput += inclasslen;
3742 nextchr = UCHARAT(locinput);
3747 nextchr = UCHARAT(locinput);
3748 if (!nextchr && locinput >= PL_regeol)
3750 if (!REGINCLASS(rex, scan, (U8*)locinput))
3752 nextchr = UCHARAT(++locinput);
3756 /* Special char classes - The defines start on line 129 or so */
3757 CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR,
3758 ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8,
3759 ALNUMU, NALNUMU, isWORDCHAR_L1,
3760 ALNUMA, NALNUMA, isWORDCHAR_A,
3763 CCC_TRY_U(SPACE, NSPACE, isSPACE,
3764 SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8,
3765 SPACEU, NSPACEU, isSPACE_L1,
3766 SPACEA, NSPACEA, isSPACE_A,
3769 CCC_TRY(DIGIT, NDIGIT, isDIGIT,
3770 DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8,
3771 DIGITA, NDIGITA, isDIGIT_A,
3774 case CLUMP: /* Match \X: logical Unicode character. This is defined as
3775 a Unicode extended Grapheme Cluster */
3776 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
3777 extended Grapheme Cluster is:
3780 | Prepend* Begin Extend*
3783 Begin is (Hangul-syllable | ! Control)
3784 Extend is (Grapheme_Extend | Spacing_Mark)
3785 Control is [ GCB_Control CR LF ]
3787 The discussion below shows how the code for CLUMP is derived
3788 from this regex. Note that most of these concepts are from
3789 property values of the Grapheme Cluster Boundary (GCB) property.
3790 No code point can have multiple property values for a given
3791 property. Thus a code point in Prepend can't be in Control, but
3792 it must be in !Control. This is why Control above includes
3793 GCB_Control plus CR plus LF. The latter two are used in the GCB
3794 property separately, and so can't be in GCB_Control, even though
3795 they logically are controls. Control is not the same as gc=cc,
3796 but includes format and other characters as well.
3798 The Unicode definition of Hangul-syllable is:
3800 | (L* ( ( V | LV ) V* | LVT ) T*)
3803 Each of these is a value for the GCB property, and hence must be
3804 disjoint, so the order they are tested is immaterial, so the
3805 above can safely be changed to
3808 | (L* ( LVT | ( V | LV ) V*) T*)
3810 The last two terms can be combined like this:
3812 | (( LVT | ( V | LV ) V*) T*))
3814 And refactored into this:
3815 L* (L | LVT T* | V V* T* | LV V* T*)
3817 That means that if we have seen any L's at all we can quit
3818 there, but if the next character is a LVT, a V or and LV we
3821 There is a subtlety with Prepend* which showed up in testing.
3822 Note that the Begin, and only the Begin is required in:
3823 | Prepend* Begin Extend*
3824 Also, Begin contains '! Control'. A Prepend must be a '!
3825 Control', which means it must be a Begin. What it comes down to
3826 is that if we match Prepend* and then find no suitable Begin
3827 afterwards, that if we backtrack the last Prepend, that one will
3828 be a suitable Begin.
3831 if (locinput >= PL_regeol)
3833 if (! utf8_target) {
3835 /* Match either CR LF or '.', as all the other possibilities
3837 locinput++; /* Match the . or CR */
3839 && locinput < PL_regeol
3840 && UCHARAT(locinput) == '\n') locinput++;
3844 /* Utf8: See if is ( CR LF ); already know that locinput <
3845 * PL_regeol, so locinput+1 is in bounds */
3846 if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') {
3850 /* In case have to backtrack to beginning, then match '.' */
3851 char *starting = locinput;
3853 /* In case have to backtrack the last prepend */
3854 char *previous_prepend = 0;
3856 LOAD_UTF8_CHARCLASS_GCB();
3858 /* Match (prepend)* */
3859 while (locinput < PL_regeol
3860 && swash_fetch(PL_utf8_X_prepend,
3861 (U8*)locinput, utf8_target))
3863 previous_prepend = locinput;
3864 locinput += UTF8SKIP(locinput);
3867 /* As noted above, if we matched a prepend character, but
3868 * the next thing won't match, back off the last prepend we
3869 * matched, as it is guaranteed to match the begin */
3870 if (previous_prepend
3871 && (locinput >= PL_regeol
3872 || ! swash_fetch(PL_utf8_X_begin,
3873 (U8*)locinput, utf8_target)))
3875 locinput = previous_prepend;
3878 /* Note that here we know PL_regeol > locinput, as we
3879 * tested that upon input to this switch case, and if we
3880 * moved locinput forward, we tested the result just above
3881 * and it either passed, or we backed off so that it will
3883 if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
3885 /* Here did not match the required 'Begin' in the
3886 * second term. So just match the very first
3887 * character, the '.' of the final term of the regex */
3888 locinput = starting + UTF8SKIP(starting);
3891 /* Here is the beginning of a character that can have
3892 * an extender. It is either a hangul syllable, or a
3894 if (swash_fetch(PL_utf8_X_non_hangul,
3895 (U8*)locinput, utf8_target))
3898 /* Here not a Hangul syllable, must be a
3899 * ('! * Control') */
3900 locinput += UTF8SKIP(locinput);
3903 /* Here is a Hangul syllable. It can be composed
3904 * of several individual characters. One
3905 * possibility is T+ */
3906 if (swash_fetch(PL_utf8_X_T,
3907 (U8*)locinput, utf8_target))
3909 while (locinput < PL_regeol
3910 && swash_fetch(PL_utf8_X_T,
3911 (U8*)locinput, utf8_target))
3913 locinput += UTF8SKIP(locinput);
3917 /* Here, not T+, but is a Hangul. That means
3918 * it is one of the others: L, LV, LVT or V,
3920 * L* (L | LVT T* | V V* T* | LV V* T*) */
3923 while (locinput < PL_regeol
3924 && swash_fetch(PL_utf8_X_L,
3925 (U8*)locinput, utf8_target))
3927 locinput += UTF8SKIP(locinput);
3930 /* Here, have exhausted L*. If the next
3931 * character is not an LV, LVT nor V, it means
3932 * we had to have at least one L, so matches L+
3933 * in the original equation, we have a complete
3934 * hangul syllable. Are done. */
3936 if (locinput < PL_regeol
3937 && swash_fetch(PL_utf8_X_LV_LVT_V,
3938 (U8*)locinput, utf8_target))
3941 /* Otherwise keep going. Must be LV, LVT
3942 * or V. See if LVT */
3943 if (swash_fetch(PL_utf8_X_LVT,
3944 (U8*)locinput, utf8_target))
3946 locinput += UTF8SKIP(locinput);
3949 /* Must be V or LV. Take it, then
3951 locinput += UTF8SKIP(locinput);
3952 while (locinput < PL_regeol
3953 && swash_fetch(PL_utf8_X_V,
3954 (U8*)locinput, utf8_target))
3956 locinput += UTF8SKIP(locinput);
3960 /* And any of LV, LVT, or V can be followed
3962 while (locinput < PL_regeol
3963 && swash_fetch(PL_utf8_X_T,
3967 locinput += UTF8SKIP(locinput);
3973 /* Match any extender */
3974 while (locinput < PL_regeol
3975 && swash_fetch(PL_utf8_X_extend,
3976 (U8*)locinput, utf8_target))
3978 locinput += UTF8SKIP(locinput);
3982 if (locinput > PL_regeol) sayNO;
3984 nextchr = UCHARAT(locinput);
3988 { /* The capture buffer cases. The ones beginning with N for the
3989 named buffers just convert to the equivalent numbered and
3990 pretend they were called as the corresponding numbered buffer
3992 /* don't initialize these in the declaration, it makes C++
3997 const U8 *fold_array;
4000 PL_reg_flags |= RF_tainted;
4001 folder = foldEQ_locale;
4002 fold_array = PL_fold_locale;
4004 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4008 folder = foldEQ_latin1;
4009 fold_array = PL_fold_latin1;
4011 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4015 folder = foldEQ_latin1;
4016 fold_array = PL_fold_latin1;
4018 utf8_fold_flags = 0;
4023 fold_array = PL_fold;
4025 utf8_fold_flags = 0;
4032 utf8_fold_flags = 0;
4035 /* For the named back references, find the corresponding buffer
4037 n = reg_check_named_buff_matched(rex,scan);
4042 goto do_nref_ref_common;
4045 PL_reg_flags |= RF_tainted;
4046 folder = foldEQ_locale;
4047 fold_array = PL_fold_locale;
4048 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
4052 folder = foldEQ_latin1;
4053 fold_array = PL_fold_latin1;
4054 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4058 folder = foldEQ_latin1;
4059 fold_array = PL_fold_latin1;
4060 utf8_fold_flags = 0;
4065 fold_array = PL_fold;
4066 utf8_fold_flags = 0;
4072 utf8_fold_flags = 0;
4076 n = ARG(scan); /* which paren pair */
4079 ln = PL_regoffs[n].start;
4080 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4081 if (*PL_reglastparen < n || ln == -1)
4082 sayNO; /* Do not match unless seen CLOSEn. */
4083 if (ln == PL_regoffs[n].end)
4087 if (type != REF /* REF can do byte comparison */
4088 && (utf8_target || type == REFFU))
4089 { /* XXX handle REFFL better */
4090 char * limit = PL_regeol;
4092 /* This call case insensitively compares the entire buffer
4093 * at s, with the current input starting at locinput, but
4094 * not going off the end given by PL_regeol, and returns in
4095 * limit upon success, how much of the current input was
4097 if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target,
4098 locinput, &limit, 0, utf8_target, utf8_fold_flags))
4103 nextchr = UCHARAT(locinput);
4107 /* Not utf8: Inline the first character, for speed. */
4108 if (UCHARAT(s) != nextchr &&
4110 UCHARAT(s) != fold_array[nextchr]))
4112 ln = PL_regoffs[n].end - ln;
4113 if (locinput + ln > PL_regeol)
4115 if (ln > 1 && (type == REF
4116 ? memNE(s, locinput, ln)
4117 : ! folder(s, locinput, ln)))
4120 nextchr = UCHARAT(locinput);
4130 #define ST st->u.eval
4135 regexp_internal *rei;
4136 regnode *startpoint;
4139 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
4140 if (cur_eval && cur_eval->locinput==locinput) {
4141 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
4142 Perl_croak(aTHX_ "Infinite recursion in regex");
4143 if ( ++nochange_depth > max_nochange_depth )
4145 "Pattern subroutine nesting without pos change"
4146 " exceeded limit in regex");
4153 (void)ReREFCNT_inc(rex_sv);
4154 if (OP(scan)==GOSUB) {
4155 startpoint = scan + ARG2L(scan);
4156 ST.close_paren = ARG(scan);
4158 startpoint = rei->program+1;
4161 goto eval_recurse_doit;
4163 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
4164 if (cur_eval && cur_eval->locinput==locinput) {
4165 if ( ++nochange_depth > max_nochange_depth )
4166 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
4171 /* execute the code in the {...} */
4173 SV ** const before = SP;
4174 OP_4tree * const oop = PL_op;
4175 COP * const ocurcop = PL_curcop;
4177 char *saved_regeol = PL_regeol;
4178 struct re_save_state saved_state;
4180 /* To not corrupt the existing regex state while executing the
4181 * eval we would normally put it on the save stack, like with
4182 * save_re_context. However, re-evals have a weird scoping so we
4183 * can't just add ENTER/LEAVE here. With that, things like
4185 * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a})
4187 * would break, as they expect the localisation to be unwound
4188 * only when the re-engine backtracks through the bit that
4191 * What we do instead is just saving the state in a local c
4194 Copy(&PL_reg_state, &saved_state, 1, struct re_save_state);
4197 PL_op = (OP_4tree*)rexi->data->data[n];
4198 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
4199 " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
4200 PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]);
4201 PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
4204 SV *sv_mrk = get_sv("REGMARK", 1);
4205 sv_setsv(sv_mrk, sv_yes_mark);
4208 CALLRUNOPS(aTHX); /* Scalar context. */
4211 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
4217 Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
4220 PAD_RESTORE_LOCAL(old_comppad);
4221 PL_curcop = ocurcop;
4222 PL_regeol = saved_regeol;
4225 sv_setsv(save_scalar(PL_replgv), ret);
4229 if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
4232 /* extract RE object from returned value; compiling if
4238 SV *const sv = SvRV(ret);
4240 if (SvTYPE(sv) == SVt_REGEXP) {
4242 } else if (SvSMAGICAL(sv)) {
4243 mg = mg_find(sv, PERL_MAGIC_qr);
4246 } else if (SvTYPE(ret) == SVt_REGEXP) {
4248 } else if (SvSMAGICAL(ret)) {
4249 if (SvGMAGICAL(ret)) {
4250 /* I don't believe that there is ever qr magic
4252 assert(!mg_find(ret, PERL_MAGIC_qr));
4253 sv_unmagic(ret, PERL_MAGIC_qr);
4256 mg = mg_find(ret, PERL_MAGIC_qr);
4257 /* testing suggests mg only ends up non-NULL for
4258 scalars who were upgraded and compiled in the
4259 else block below. In turn, this is only
4260 triggered in the "postponed utf8 string" tests
4266 rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
4270 rx = reg_temp_copy(NULL, rx);
4274 const I32 osize = PL_regsize;
4277 assert (SvUTF8(ret));
4278 } else if (SvUTF8(ret)) {
4279 /* Not doing UTF-8, despite what the SV says. Is
4280 this only if we're trapped in use 'bytes'? */
4281 /* Make a copy of the octet sequence, but without
4282 the flag on, as the compiler now honours the
4283 SvUTF8 flag on ret. */
4285 const char *const p = SvPV(ret, len);
4286 ret = newSVpvn_flags(p, len, SVs_TEMP);
4288 rx = CALLREGCOMP(ret, pm_flags);
4290 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
4292 /* This isn't a first class regexp. Instead, it's
4293 caching a regexp onto an existing, Perl visible
4295 sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
4300 re = (struct regexp *)SvANY(rx);
4302 RXp_MATCH_COPIED_off(re);
4303 re->subbeg = rex->subbeg;
4304 re->sublen = rex->sublen;
4307 debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
4308 "Matching embedded");
4310 startpoint = rei->program + 1;
4311 ST.close_paren = 0; /* only used for GOSUB */
4312 /* borrowed from regtry */
4313 if (PL_reg_start_tmpl <= re->nparens) {
4314 PL_reg_start_tmpl = re->nparens*3/2 + 3;
4315 if(PL_reg_start_tmp)
4316 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4318 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
4321 eval_recurse_doit: /* Share code with GOSUB below this line */
4322 /* run the pattern returned from (??{...}) */
4323 ST.cp = regcppush(0); /* Save *all* the positions. */
4324 REGCP_SET(ST.lastcp);
4326 PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
4328 /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
4329 PL_reglastparen = &re->lastparen;
4330 PL_reglastcloseparen = &re->lastcloseparen;
4332 re->lastcloseparen = 0;
4334 PL_reginput = locinput;
4337 /* XXXX This is too dramatic a measure... */
4340 ST.toggle_reg_flags = PL_reg_flags;
4342 PL_reg_flags |= RF_utf8;
4344 PL_reg_flags &= ~RF_utf8;
4345 ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
4347 ST.prev_rex = rex_sv;
4348 ST.prev_curlyx = cur_curlyx;
4349 SETREX(rex_sv,re_sv);
4354 ST.prev_eval = cur_eval;
4356 /* now continue from first node in postoned RE */
4357 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
4360 /* logical is 1, /(?(?{...})X|Y)/ */
4361 sw = cBOOL(SvTRUE(ret));
4366 case EVAL_AB: /* cleanup after a successful (??{A})B */
4367 /* note: this is called twice; first after popping B, then A */
4368 PL_reg_flags ^= ST.toggle_reg_flags;
4369 ReREFCNT_dec(rex_sv);
4370 SETREX(rex_sv,ST.prev_rex);
4371 rex = (struct regexp *)SvANY(rex_sv);
4372 rexi = RXi_GET(rex);
4374 cur_eval = ST.prev_eval;
4375 cur_curlyx = ST.prev_curlyx;
4377 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4378 PL_reglastparen = &rex->lastparen;
4379 PL_reglastcloseparen = &rex->lastcloseparen;
4380 /* also update PL_regoffs */
4381 PL_regoffs = rex->offs;
4383 /* XXXX This is too dramatic a measure... */
4385 if ( nochange_depth )
4390 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
4391 /* note: this is called twice; first after popping B, then A */
4392 PL_reg_flags ^= ST.toggle_reg_flags;
4393 ReREFCNT_dec(rex_sv);
4394 SETREX(rex_sv,ST.prev_rex);
4395 rex = (struct regexp *)SvANY(rex_sv);
4396 rexi = RXi_GET(rex);
4397 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
4398 PL_reglastparen = &rex->lastparen;
4399 PL_reglastcloseparen = &rex->lastcloseparen;
4401 PL_reginput = locinput;
4402 REGCP_UNWIND(ST.lastcp);
4404 cur_eval = ST.prev_eval;
4405 cur_curlyx = ST.prev_curlyx;
4406 /* XXXX This is too dramatic a measure... */
4408 if ( nochange_depth )
4414 n = ARG(scan); /* which paren pair */
4415 PL_reg_start_tmp[n] = locinput;
4421 n = ARG(scan); /* which paren pair */
4422 PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr;
4423 PL_regoffs[n].end = locinput - PL_bostr;
4424 /*if (n > PL_regsize)
4426 if (n > *PL_reglastparen)
4427 *PL_reglastparen = n;
4428 *PL_reglastcloseparen = n;
4429 if (cur_eval && cur_eval->u.eval.close_paren == n) {
4437 cursor && OP(cursor)!=END;
4438 cursor=regnext(cursor))
4440 if ( OP(cursor)==CLOSE ){
4442 if ( n <= lastopen ) {
4444 = PL_reg_start_tmp[n] - PL_bostr;
4445 PL_regoffs[n].end = locinput - PL_bostr;
4446 /*if (n > PL_regsize)
4448 if (n > *PL_reglastparen)
4449 *PL_reglastparen = n;
4450 *PL_reglastcloseparen = n;
4451 if ( n == ARG(scan) || (cur_eval &&
4452 cur_eval->u.eval.close_paren == n))
4461 n = ARG(scan); /* which paren pair */
4462 sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1);
4465 /* reg_check_named_buff_matched returns 0 for no match */
4466 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
4470 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
4476 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
4478 next = NEXTOPER(NEXTOPER(scan));
4480 next = scan + ARG(scan);
4481 if (OP(next) == IFTHEN) /* Fake one. */
4482 next = NEXTOPER(NEXTOPER(next));
4486 logical = scan->flags;
4489 /*******************************************************************
4491 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
4492 pattern, where A and B are subpatterns. (For simple A, CURLYM or
4493 STAR/PLUS/CURLY/CURLYN are used instead.)
4495 A*B is compiled as <CURLYX><A><WHILEM><B>
4497 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
4498 state, which contains the current count, initialised to -1. It also sets
4499 cur_curlyx to point to this state, with any previous value saved in the
4502 CURLYX then jumps straight to the WHILEM op, rather than executing A,
4503 since the pattern may possibly match zero times (i.e. it's a while {} loop
4504 rather than a do {} while loop).
4506 Each entry to WHILEM represents a successful match of A. The count in the
4507 CURLYX block is incremented, another WHILEM state is pushed, and execution
4508 passes to A or B depending on greediness and the current count.
4510 For example, if matching against the string a1a2a3b (where the aN are
4511 substrings that match /A/), then the match progresses as follows: (the
4512 pushed states are interspersed with the bits of strings matched so far):
4515 <CURLYX cnt=0><WHILEM>
4516 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
4517 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
4518 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
4519 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
4521 (Contrast this with something like CURLYM, which maintains only a single
4525 a1 <CURLYM cnt=1> a2
4526 a1 a2 <CURLYM cnt=2> a3
4527 a1 a2 a3 <CURLYM cnt=3> b
4530 Each WHILEM state block marks a point to backtrack to upon partial failure
4531 of A or B, and also contains some minor state data related to that
4532 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
4533 overall state, such as the count, and pointers to the A and B ops.
4535 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
4536 must always point to the *current* CURLYX block, the rules are:
4538 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
4539 and set cur_curlyx to point the new block.
4541 When popping the CURLYX block after a successful or unsuccessful match,
4542 restore the previous cur_curlyx.
4544 When WHILEM is about to execute B, save the current cur_curlyx, and set it
4545 to the outer one saved in the CURLYX block.
4547 When popping the WHILEM block after a successful or unsuccessful B match,
4548 restore the previous cur_curlyx.
4550 Here's an example for the pattern (AI* BI)*BO
4551 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
4554 curlyx backtrack stack
4555 ------ ---------------
4557 CO <CO prev=NULL> <WO>
4558 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4559 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4560 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
4562 At this point the pattern succeeds, and we work back down the stack to
4563 clean up, restoring as we go:
4565 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
4566 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
4567 CO <CO prev=NULL> <WO>
4570 *******************************************************************/
4572 #define ST st->u.curlyx
4574 case CURLYX: /* start of /A*B/ (for complex A) */
4576 /* No need to save/restore up to this paren */
4577 I32 parenfloor = scan->flags;
4579 assert(next); /* keep Coverity happy */
4580 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
4583 /* XXXX Probably it is better to teach regpush to support
4584 parenfloor > PL_regsize... */
4585 if (parenfloor > (I32)*PL_reglastparen)
4586 parenfloor = *PL_reglastparen; /* Pessimization... */
4588 ST.prev_curlyx= cur_curlyx;
4590 ST.cp = PL_savestack_ix;
4592 /* these fields contain the state of the current curly.
4593 * they are accessed by subsequent WHILEMs */
4594 ST.parenfloor = parenfloor;
4599 ST.count = -1; /* this will be updated by WHILEM */
4600 ST.lastloc = NULL; /* this will be updated by WHILEM */
4602 PL_reginput = locinput;
4603 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
4607 case CURLYX_end: /* just finished matching all of A*B */
4608 cur_curlyx = ST.prev_curlyx;
4612 case CURLYX_end_fail: /* just failed to match all of A*B */
4614 cur_curlyx = ST.prev_curlyx;
4620 #define ST st->u.whilem
4622 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
4624 /* see the discussion above about CURLYX/WHILEM */
4626 int min = ARG1(cur_curlyx->u.curlyx.me);
4627 int max = ARG2(cur_curlyx->u.curlyx.me);
4628 regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
4630 assert(cur_curlyx); /* keep Coverity happy */
4631 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
4632 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
4633 ST.cache_offset = 0;
4636 PL_reginput = locinput;
4638 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4639 "%*s whilem: matched %ld out of %d..%d\n",
4640 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
4643 /* First just match a string of min A's. */
4646 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4647 cur_curlyx->u.curlyx.lastloc = locinput;
4648 REGCP_SET(ST.lastcp);
4650 PUSH_STATE_GOTO(WHILEM_A_pre, A);
4654 /* If degenerate A matches "", assume A done. */
4656 if (locinput == cur_curlyx->u.curlyx.lastloc) {
4657 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4658 "%*s whilem: empty match detected, trying continuation...\n",
4659 REPORT_CODE_OFF+depth*2, "")
4661 goto do_whilem_B_max;
4664 /* super-linear cache processing */
4668 if (!PL_reg_maxiter) {
4669 /* start the countdown: Postpone detection until we
4670 * know the match is not *that* much linear. */
4671 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
4672 /* possible overflow for long strings and many CURLYX's */
4673 if (PL_reg_maxiter < 0)
4674 PL_reg_maxiter = I32_MAX;
4675 PL_reg_leftiter = PL_reg_maxiter;
4678 if (PL_reg_leftiter-- == 0) {
4679 /* initialise cache */
4680 const I32 size = (PL_reg_maxiter + 7)/8;
4681 if (PL_reg_poscache) {
4682 if ((I32)PL_reg_poscache_size < size) {
4683 Renew(PL_reg_poscache, size, char);
4684 PL_reg_poscache_size = size;
4686 Zero(PL_reg_poscache, size, char);
4689 PL_reg_poscache_size = size;
4690 Newxz(PL_reg_poscache, size, char);
4692 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4693 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
4694 PL_colors[4], PL_colors[5])
4698 if (PL_reg_leftiter < 0) {
4699 /* have we already failed at this position? */
4701 offset = (scan->flags & 0xf) - 1
4702 + (locinput - PL_bostr) * (scan->flags>>4);
4703 mask = 1 << (offset % 8);
4705 if (PL_reg_poscache[offset] & mask) {
4706 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
4707 "%*s whilem: (cache) already tried at this position...\n",
4708 REPORT_CODE_OFF+depth*2, "")
4710 sayNO; /* cache records failure */
4712 ST.cache_offset = offset;
4713 ST.cache_mask = mask;
4717 /* Prefer B over A for minimal matching. */
4719 if (cur_curlyx->u.curlyx.minmod) {
4720 ST.save_curlyx = cur_curlyx;
4721 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4722 ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor);
4723 REGCP_SET(ST.lastcp);
4724 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
4728 /* Prefer A over B for maximal matching. */
4730 if (n < max) { /* More greed allowed? */
4731 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4732 cur_curlyx->u.curlyx.lastloc = locinput;
4733 REGCP_SET(ST.lastcp);
4734 PUSH_STATE_GOTO(WHILEM_A_max, A);
4737 goto do_whilem_B_max;
4741 case WHILEM_B_min: /* just matched B in a minimal match */
4742 case WHILEM_B_max: /* just matched B in a maximal match */
4743 cur_curlyx = ST.save_curlyx;
4747 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
4748 cur_curlyx = ST.save_curlyx;
4749 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4750 cur_curlyx->u.curlyx.count--;
4754 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
4756 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
4757 REGCP_UNWIND(ST.lastcp);
4759 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
4760 cur_curlyx->u.curlyx.count--;
4764 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
4765 REGCP_UNWIND(ST.lastcp);
4766 regcppop(rex); /* Restore some previous $<digit>s? */
4767 PL_reginput = locinput;
4768 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4769 "%*s whilem: failed, trying continuation...\n",
4770 REPORT_CODE_OFF+depth*2, "")
4773 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4774 && ckWARN(WARN_REGEXP)
4775 && !(PL_reg_flags & RF_warned))
4777 PL_reg_flags |= RF_warned;
4778 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
4779 "Complex regular subexpression recursion",
4784 ST.save_curlyx = cur_curlyx;
4785 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4786 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
4789 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
4790 cur_curlyx = ST.save_curlyx;
4791 REGCP_UNWIND(ST.lastcp);
4794 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
4795 /* Maximum greed exceeded */
4796 if (cur_curlyx->u.curlyx.count >= REG_INFTY
4797 && ckWARN(WARN_REGEXP)
4798 && !(PL_reg_flags & RF_warned))
4800 PL_reg_flags |= RF_warned;
4801 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
4802 "%s limit (%d) exceeded",
4803 "Complex regular subexpression recursion",
4806 cur_curlyx->u.curlyx.count--;
4810 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4811 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
4813 /* Try grabbing another A and see if it helps. */
4814 PL_reginput = locinput;
4815 cur_curlyx->u.curlyx.lastloc = locinput;
4816 ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor);
4817 REGCP_SET(ST.lastcp);
4818 PUSH_STATE_GOTO(WHILEM_A_min,
4819 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
4823 #define ST st->u.branch
4825 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
4826 next = scan + ARG(scan);
4829 scan = NEXTOPER(scan);
4832 case BRANCH: /* /(...|A|...)/ */
4833 scan = NEXTOPER(scan); /* scan now points to inner node */
4834 ST.lastparen = *PL_reglastparen;
4835 ST.next_branch = next;
4837 PL_reginput = locinput;
4839 /* Now go into the branch */
4841 PUSH_YES_STATE_GOTO(BRANCH_next, scan);
4843 PUSH_STATE_GOTO(BRANCH_next, scan);
4847 PL_reginput = locinput;
4848 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
4849 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4850 PUSH_STATE_GOTO(CUTGROUP_next,next);
4852 case CUTGROUP_next_fail:
4855 if (st->u.mark.mark_name)
4856 sv_commit = st->u.mark.mark_name;
4862 case BRANCH_next_fail: /* that branch failed; try the next, if any */
4867 REGCP_UNWIND(ST.cp);
4868 for (n = *PL_reglastparen; n > ST.lastparen; n--)
4869 PL_regoffs[n].end = -1;
4870 *PL_reglastparen = n;
4871 /*dmq: *PL_reglastcloseparen = n; */
4872 scan = ST.next_branch;
4873 /* no more branches? */
4874 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
4876 PerlIO_printf( Perl_debug_log,
4877 "%*s %sBRANCH failed...%s\n",
4878 REPORT_CODE_OFF+depth*2, "",
4884 continue; /* execute next BRANCH[J] op */
4892 #define ST st->u.curlym
4894 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
4896 /* This is an optimisation of CURLYX that enables us to push
4897 * only a single backtracking state, no matter how many matches
4898 * there are in {m,n}. It relies on the pattern being constant
4899 * length, with no parens to influence future backrefs
4903 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
4905 /* if paren positive, emulate an OPEN/CLOSE around A */
4907 U32 paren = ST.me->flags;
4908 if (paren > PL_regsize)
4910 if (paren > *PL_reglastparen)
4911 *PL_reglastparen = paren;
4912 scan += NEXT_OFF(scan); /* Skip former OPEN. */
4920 ST.c1 = CHRTEST_UNINIT;
4923 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
4926 curlym_do_A: /* execute the A in /A{m,n}B/ */
4927 PL_reginput = locinput;
4928 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
4931 case CURLYM_A: /* we've just matched an A */
4932 locinput = st->locinput;
4933 nextchr = UCHARAT(locinput);
4936 /* after first match, determine A's length: u.curlym.alen */
4937 if (ST.count == 1) {
4938 if (PL_reg_match_utf8) {
4940 while (s < PL_reginput) {
4946 ST.alen = PL_reginput - locinput;
4949 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
4952 PerlIO_printf(Perl_debug_log,
4953 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
4954 (int)(REPORT_CODE_OFF+(depth*2)), "",
4955 (IV) ST.count, (IV)ST.alen)
4958 locinput = PL_reginput;
4960 if (cur_eval && cur_eval->u.eval.close_paren &&
4961 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
4965 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
4966 if ( max == REG_INFTY || ST.count < max )
4967 goto curlym_do_A; /* try to match another A */
4969 goto curlym_do_B; /* try to match B */
4971 case CURLYM_A_fail: /* just failed to match an A */
4972 REGCP_UNWIND(ST.cp);
4974 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
4975 || (cur_eval && cur_eval->u.eval.close_paren &&
4976 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
4979 curlym_do_B: /* execute the B in /A{m,n}B/ */
4980 PL_reginput = locinput;
4981 if (ST.c1 == CHRTEST_UNINIT) {
4982 /* calculate c1 and c2 for possible match of 1st char
4983 * following curly */
4984 ST.c1 = ST.c2 = CHRTEST_VOID;
4985 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
4986 regnode *text_node = ST.B;
4987 if (! HAS_TEXT(text_node))
4988 FIND_NEXT_IMPT(text_node);
4991 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
4993 But the former is redundant in light of the latter.
4995 if this changes back then the macro for
4996 IS_TEXT and friends need to change.
4998 if (PL_regkind[OP(text_node)] == EXACT)
5001 ST.c1 = (U8)*STRING(text_node);
5002 switch (OP(text_node)) {
5003 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5005 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5006 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5007 default: ST.c2 = ST.c1;
5014 PerlIO_printf(Perl_debug_log,
5015 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5016 (int)(REPORT_CODE_OFF+(depth*2)),
5019 if (ST.c1 != CHRTEST_VOID
5020 && UCHARAT(PL_reginput) != ST.c1
5021 && UCHARAT(PL_reginput) != ST.c2)
5023 /* simulate B failing */
5025 PerlIO_printf(Perl_debug_log,
5026 "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n",
5027 (int)(REPORT_CODE_OFF+(depth*2)),"",
5030 state_num = CURLYM_B_fail;
5031 goto reenter_switch;
5035 /* mark current A as captured */
5036 I32 paren = ST.me->flags;
5038 PL_regoffs[paren].start
5039 = HOPc(PL_reginput, -ST.alen) - PL_bostr;
5040 PL_regoffs[paren].end = PL_reginput - PL_bostr;
5041 /*dmq: *PL_reglastcloseparen = paren; */
5044 PL_regoffs[paren].end = -1;
5045 if (cur_eval && cur_eval->u.eval.close_paren &&
5046 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
5055 PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
5058 case CURLYM_B_fail: /* just failed to match a B */
5059 REGCP_UNWIND(ST.cp);
5061 I32 max = ARG2(ST.me);
5062 if (max != REG_INFTY && ST.count == max)
5064 goto curlym_do_A; /* try to match a further A */
5066 /* backtrack one A */
5067 if (ST.count == ARG1(ST.me) /* min */)
5070 locinput = HOPc(locinput, -ST.alen);
5071 goto curlym_do_B; /* try to match B */
5074 #define ST st->u.curly
5076 #define CURLY_SETPAREN(paren, success) \
5079 PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \
5080 PL_regoffs[paren].end = locinput - PL_bostr; \
5081 *PL_reglastcloseparen = paren; \
5084 PL_regoffs[paren].end = -1; \
5087 case STAR: /* /A*B/ where A is width 1 */
5091 scan = NEXTOPER(scan);
5093 case PLUS: /* /A+B/ where A is width 1 */
5097 scan = NEXTOPER(scan);
5099 case CURLYN: /* /(A){m,n}B/ where A is width 1 */
5100 ST.paren = scan->flags; /* Which paren to set */
5101 if (ST.paren > PL_regsize)
5102 PL_regsize = ST.paren;
5103 if (ST.paren > *PL_reglastparen)
5104 *PL_reglastparen = ST.paren;
5105 ST.min = ARG1(scan); /* min to match */
5106 ST.max = ARG2(scan); /* max to match */
5107 if (cur_eval && cur_eval->u.eval.close_paren &&
5108 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5112 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
5114 case CURLY: /* /A{m,n}B/ where A is width 1 */
5116 ST.min = ARG1(scan); /* min to match */
5117 ST.max = ARG2(scan); /* max to match */
5118 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
5121 * Lookahead to avoid useless match attempts
5122 * when we know what character comes next.
5124 * Used to only do .*x and .*?x, but now it allows
5125 * for )'s, ('s and (?{ ... })'s to be in the way
5126 * of the quantifier and the EXACT-like node. -- japhy
5129 if (ST.min > ST.max) /* XXX make this a compile-time check? */
5131 if (HAS_TEXT(next) || JUMPABLE(next)) {
5133 regnode *text_node = next;
5135 if (! HAS_TEXT(text_node))
5136 FIND_NEXT_IMPT(text_node);
5138 if (! HAS_TEXT(text_node))
5139 ST.c1 = ST.c2 = CHRTEST_VOID;
5141 if ( PL_regkind[OP(text_node)] != EXACT ) {
5142 ST.c1 = ST.c2 = CHRTEST_VOID;
5143 goto assume_ok_easy;
5146 s = (U8*)STRING(text_node);
5148 /* Currently we only get here when
5150 PL_rekind[OP(text_node)] == EXACT
5152 if this changes back then the macro for IS_TEXT and
5153 friends need to change. */
5156 switch (OP(text_node)) {
5157 case EXACTF: ST.c2 = PL_fold[ST.c1]; break;
5159 case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break;
5160 case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break;
5161 default: ST.c2 = ST.c1; break;
5164 else { /* UTF_PATTERN */
5165 if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) {
5166 STRLEN ulen1, ulen2;
5167 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
5168 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
5170 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
5171 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
5173 ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0,
5175 0 : UTF8_ALLOW_ANY);
5176 ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0,
5178 0 : UTF8_ALLOW_ANY);
5180 ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
5182 ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
5187 ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
5194 ST.c1 = ST.c2 = CHRTEST_VOID;
5199 PL_reginput = locinput;
5202 if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
5205 locinput = PL_reginput;
5207 if (ST.c1 == CHRTEST_VOID)
5208 goto curly_try_B_min;
5210 ST.oldloc = locinput;
5212 /* set ST.maxpos to the furthest point along the
5213 * string that could possibly match */
5214 if (ST.max == REG_INFTY) {
5215 ST.maxpos = PL_regeol - 1;
5217 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
5220 else if (utf8_target) {
5221 int m = ST.max - ST.min;
5222 for (ST.maxpos = locinput;
5223 m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--)
5224 ST.maxpos += UTF8SKIP(ST.maxpos);
5227 ST.maxpos = locinput + ST.max - ST.min;
5228 if (ST.maxpos >= PL_regeol)
5229 ST.maxpos = PL_regeol - 1;
5231 goto curly_try_B_min_known;
5235 ST.count = regrepeat(rex, ST.A, ST.max, depth);
5236 locinput = PL_reginput;
5237 if (ST.count < ST.min)
5239 if ((ST.count > ST.min)
5240 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
5242 /* A{m,n} must come at the end of the string, there's
5243 * no point in backing off ... */
5245 /* ...except that $ and \Z can match before *and* after
5246 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
5247 We may back off by one in this case. */
5248 if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
5252 goto curly_try_B_max;
5257 case CURLY_B_min_known_fail:
5258 /* failed to find B in a non-greedy match where c1,c2 valid */
5259 if (ST.paren && ST.count)
5260 PL_regoffs[ST.paren].end = -1;
5262 PL_reginput = locinput; /* Could be reset... */
5263 REGCP_UNWIND(ST.cp);
5264 /* Couldn't or didn't -- move forward. */
5265 ST.oldloc = locinput;
5267 locinput += UTF8SKIP(locinput);
5271 curly_try_B_min_known:
5272 /* find the next place where 'B' could work, then call B */
5276 n = (ST.oldloc == locinput) ? 0 : 1;
5277 if (ST.c1 == ST.c2) {
5279 /* set n to utf8_distance(oldloc, locinput) */
5280 while (locinput <= ST.maxpos &&
5281 utf8n_to_uvchr((U8*)locinput,
5282 UTF8_MAXBYTES, &len,
5283 uniflags) != (UV)ST.c1) {
5289 /* set n to utf8_distance(oldloc, locinput) */
5290 while (locinput <= ST.maxpos) {
5292 const UV c = utf8n_to_uvchr((U8*)locinput,
5293 UTF8_MAXBYTES, &len,
5295 if (c == (UV)ST.c1 || c == (UV)ST.c2)
5303 if (ST.c1 == ST.c2) {
5304 while (locinput <= ST.maxpos &&
5305 UCHARAT(locinput) != ST.c1)
5309 while (locinput <= ST.maxpos
5310 && UCHARAT(locinput) != ST.c1
5311 && UCHARAT(locinput) != ST.c2)
5314 n = locinput - ST.oldloc;
5316 if (locinput > ST.maxpos)
5318 /* PL_reginput == oldloc now */
5321 if (regrepeat(rex, ST.A, n, depth) < n)
5324 PL_reginput = locinput;
5325 CURLY_SETPAREN(ST.paren, ST.count);
5326 if (cur_eval && cur_eval->u.eval.close_paren &&
5327 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5330 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
5335 case CURLY_B_min_fail:
5336 /* failed to find B in a non-greedy match where c1,c2 invalid */
5337 if (ST.paren && ST.count)
5338 PL_regoffs[ST.paren].end = -1;
5340 REGCP_UNWIND(ST.cp);
5341 /* failed -- move forward one */
5342 PL_reginput = locinput;
5343 if (regrepeat(rex, ST.A, 1, depth)) {
5345 locinput = PL_reginput;
5346 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
5347 ST.count > 0)) /* count overflow ? */
5350 CURLY_SETPAREN(ST.paren, ST.count);
5351 if (cur_eval && cur_eval->u.eval.close_paren &&
5352 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5355 PUSH_STATE_GOTO(CURLY_B_min, ST.B);
5363 /* a successful greedy match: now try to match B */
5364 if (cur_eval && cur_eval->u.eval.close_paren &&
5365 cur_eval->u.eval.close_paren == (U32)ST.paren) {
5370 if (ST.c1 != CHRTEST_VOID)
5371 c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
5372 UTF8_MAXBYTES, 0, uniflags)
5373 : (UV) UCHARAT(PL_reginput);
5374 /* If it could work, try it. */
5375 if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
5376 CURLY_SETPAREN(ST.paren, ST.count);
5377 PUSH_STATE_GOTO(CURLY_B_max, ST.B);
5382 case CURLY_B_max_fail:
5383 /* failed to find B in a greedy match */
5384 if (ST.paren && ST.count)
5385 PL_regoffs[ST.paren].end = -1;
5387 REGCP_UNWIND(ST.cp);
5389 if (--ST.count < ST.min)
5391 PL_reginput = locinput = HOPc(locinput, -1);
5392 goto curly_try_B_max;
5399 /* we've just finished A in /(??{A})B/; now continue with B */
5401 st->u.eval.toggle_reg_flags
5402 = cur_eval->u.eval.toggle_reg_flags;
5403 PL_reg_flags ^= st->u.eval.toggle_reg_flags;
5405 st->u.eval.prev_rex = rex_sv; /* inner */
5406 SETREX(rex_sv,cur_eval->u.eval.prev_rex);
5407 rex = (struct regexp *)SvANY(rex_sv);
5408 rexi = RXi_GET(rex);
5409 cur_curlyx = cur_eval->u.eval.prev_curlyx;
5410 (void)ReREFCNT_inc(rex_sv);
5411 st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
5413 /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
5414 PL_reglastparen = &rex->lastparen;
5415 PL_reglastcloseparen = &rex->lastcloseparen;
5417 REGCP_SET(st->u.eval.lastcp);
5418 PL_reginput = locinput;
5420 /* Restore parens of the outer rex without popping the
5422 tmpix = PL_savestack_ix;
5423 PL_savestack_ix = cur_eval->u.eval.lastcp;
5425 PL_savestack_ix = tmpix;
5427 st->u.eval.prev_eval = cur_eval;
5428 cur_eval = cur_eval->u.eval.prev_eval;
5430 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
5431 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
5432 if ( nochange_depth )
5435 PUSH_YES_STATE_GOTO(EVAL_AB,
5436 st->u.eval.prev_eval->u.eval.B); /* match B */
5439 if (locinput < reginfo->till) {
5440 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5441 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
5443 (long)(locinput - PL_reg_starttry),
5444 (long)(reginfo->till - PL_reg_starttry),
5447 sayNO_SILENT; /* Cannot match: too short. */
5449 PL_reginput = locinput; /* put where regtry can find it */
5450 sayYES; /* Success! */
5452 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
5454 PerlIO_printf(Perl_debug_log,
5455 "%*s %ssubpattern success...%s\n",
5456 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
5457 PL_reginput = locinput; /* put where regtry can find it */
5458 sayYES; /* Success! */
5461 #define ST st->u.ifmatch
5463 case SUSPEND: /* (?>A) */
5465 PL_reginput = locinput;
5468 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
5470 goto ifmatch_trivial_fail_test;
5472 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
5474 ifmatch_trivial_fail_test:
5476 char * const s = HOPBACKc(locinput, scan->flags);
5481 sw = 1 - cBOOL(ST.wanted);
5485 next = scan + ARG(scan);
5493 PL_reginput = locinput;
5497 ST.logical = logical;
5498 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
5500 /* execute body of (?...A) */
5501 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
5504 case IFMATCH_A_fail: /* body of (?...A) failed */
5505 ST.wanted = !ST.wanted;
5508 case IFMATCH_A: /* body of (?...A) succeeded */
5510 sw = cBOOL(ST.wanted);
5512 else if (!ST.wanted)
5515 if (OP(ST.me) == SUSPEND)
5516 locinput = PL_reginput;
5518 locinput = PL_reginput = st->locinput;
5519 nextchr = UCHARAT(locinput);
5521 scan = ST.me + ARG(ST.me);
5524 continue; /* execute B */
5529 next = scan + ARG(scan);
5534 reginfo->cutpoint = PL_regeol;
5537 PL_reginput = locinput;
5539 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5540 PUSH_STATE_GOTO(COMMIT_next,next);
5542 case COMMIT_next_fail:
5549 #define ST st->u.mark
5551 ST.prev_mark = mark_state;
5552 ST.mark_name = sv_commit = sv_yes_mark
5553 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5555 ST.mark_loc = PL_reginput = locinput;
5556 PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
5558 case MARKPOINT_next:
5559 mark_state = ST.prev_mark;
5562 case MARKPOINT_next_fail:
5563 if (popmark && sv_eq(ST.mark_name,popmark))
5565 if (ST.mark_loc > startpoint)
5566 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5567 popmark = NULL; /* we found our mark */
5568 sv_commit = ST.mark_name;
5571 PerlIO_printf(Perl_debug_log,
5572 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
5573 REPORT_CODE_OFF+depth*2, "",
5574 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
5577 mark_state = ST.prev_mark;
5578 sv_yes_mark = mark_state ?
5579 mark_state->u.mark.mark_name : NULL;
5583 PL_reginput = locinput;
5585 /* (*SKIP) : if we fail we cut here*/
5586 ST.mark_name = NULL;
5587 ST.mark_loc = locinput;
5588 PUSH_STATE_GOTO(SKIP_next,next);
5590 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5591 otherwise do nothing. Meaning we need to scan
5593 regmatch_state *cur = mark_state;
5594 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5597 if ( sv_eq( cur->u.mark.mark_name,
5600 ST.mark_name = find;
5601 PUSH_STATE_GOTO( SKIP_next, next );
5603 cur = cur->u.mark.prev_mark;
5606 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5608 case SKIP_next_fail:
5610 /* (*CUT:NAME) - Set up to search for the name as we
5611 collapse the stack*/
5612 popmark = ST.mark_name;
5614 /* (*CUT) - No name, we cut here.*/
5615 if (ST.mark_loc > startpoint)
5616 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5617 /* but we set sv_commit to latest mark_name if there
5618 is one so they can test to see how things lead to this
5621 sv_commit=mark_state->u.mark.mark_name;
5629 if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) {
5631 } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) {
5634 U8 folded[UTF8_MAXBYTES_CASE+1];
5636 const char * const l = locinput;
5637 char *e = PL_regeol;
5638 to_uni_fold(n, folded, &foldlen);
5640 if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1,
5641 l, &e, 0, utf8_target)) {
5646 nextchr = UCHARAT(locinput);
5649 if ((n=is_LNBREAK(locinput,utf8_target))) {
5651 nextchr = UCHARAT(locinput);
5656 #define CASE_CLASS(nAmE) \
5658 if ((n=is_##nAmE(locinput,utf8_target))) { \
5660 nextchr = UCHARAT(locinput); \
5665 if ((n=is_##nAmE(locinput,utf8_target))) { \
5668 locinput += UTF8SKIP(locinput); \
5669 nextchr = UCHARAT(locinput); \
5674 CASE_CLASS(HORIZWS);
5678 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
5679 PTR2UV(scan), OP(scan));
5680 Perl_croak(aTHX_ "regexp memory corruption");
5684 /* switch break jumps here */
5685 scan = next; /* prepare to execute the next op and ... */
5686 continue; /* ... jump back to the top, reusing st */
5690 /* push a state that backtracks on success */
5691 st->u.yes.prev_yes_state = yes_state;
5695 /* push a new regex state, then continue at scan */
5697 regmatch_state *newst;
5700 regmatch_state *cur = st;
5701 regmatch_state *curyes = yes_state;
5703 regmatch_slab *slab = PL_regmatch_slab;
5704 for (;curd > -1;cur--,curd--) {
5705 if (cur < SLAB_FIRST(slab)) {
5707 cur = SLAB_LAST(slab);
5709 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
5710 REPORT_CODE_OFF + 2 + depth * 2,"",
5711 curd, PL_reg_name[cur->resume_state],
5712 (curyes == cur) ? "yes" : ""
5715 curyes = cur->u.yes.prev_yes_state;
5718 DEBUG_STATE_pp("push")
5721 st->locinput = locinput;
5723 if (newst > SLAB_LAST(PL_regmatch_slab))
5724 newst = S_push_slab(aTHX);
5725 PL_regmatch_state = newst;
5727 locinput = PL_reginput;
5728 nextchr = UCHARAT(locinput);
5736 * We get here only if there's trouble -- normally "case END" is
5737 * the terminating point.
5739 Perl_croak(aTHX_ "corrupted regexp pointers");
5745 /* we have successfully completed a subexpression, but we must now
5746 * pop to the state marked by yes_state and continue from there */
5747 assert(st != yes_state);
5749 while (st != yes_state) {
5751 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5752 PL_regmatch_slab = PL_regmatch_slab->prev;
5753 st = SLAB_LAST(PL_regmatch_slab);
5757 DEBUG_STATE_pp("pop (no final)");
5759 DEBUG_STATE_pp("pop (yes)");
5765 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
5766 || yes_state > SLAB_LAST(PL_regmatch_slab))
5768 /* not in this slab, pop slab */
5769 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
5770 PL_regmatch_slab = PL_regmatch_slab->prev;
5771 st = SLAB_LAST(PL_regmatch_slab);
5773 depth -= (st - yes_state);
5776 yes_state = st->u.yes.prev_yes_state;
5777 PL_regmatch_state = st;
5780 locinput= st->locinput;
5781 nextchr = UCHARAT(locinput);
5783 state_num = st->resume_state + no_final;
5784 goto reenter_switch;
5787 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
5788 PL_colors[4], PL_colors[5]));
5790 if (PL_reg_eval_set) {
5791 /* each successfully executed (?{...}) block does the equivalent of
5792 * local $^R = do {...}
5793 * When popping the save stack, all these locals would be undone;
5794 * bypass this by setting the outermost saved $^R to the latest
5796 if (oreplsv != GvSV(PL_replgv))
5797 sv_setsv(oreplsv, GvSV(PL_replgv));
5804 PerlIO_printf(Perl_debug_log,
5805 "%*s %sfailed...%s\n",
5806 REPORT_CODE_OFF+depth*2, "",
5807 PL_colors[4], PL_colors[5])
5819 /* there's a previous state to backtrack to */
5821 if (st < SLAB_FIRST(PL_regmatch_slab)) {
5822 PL_regmatch_slab = PL_regmatch_slab->prev;
5823 st = SLAB_LAST(PL_regmatch_slab);
5825 PL_regmatch_state = st;
5826 locinput= st->locinput;
5827 nextchr = UCHARAT(locinput);
5829 DEBUG_STATE_pp("pop");
5831 if (yes_state == st)
5832 yes_state = st->u.yes.prev_yes_state;
5834 state_num = st->resume_state + 1; /* failure = success + 1 */
5835 goto reenter_switch;
5840 if (rex->intflags & PREGf_VERBARG_SEEN) {
5841 SV *sv_err = get_sv("REGERROR", 1);
5842 SV *sv_mrk = get_sv("REGMARK", 1);
5844 sv_commit = &PL_sv_no;
5846 sv_yes_mark = &PL_sv_yes;
5849 sv_commit = &PL_sv_yes;
5850 sv_yes_mark = &PL_sv_no;
5852 sv_setsv(sv_err, sv_commit);
5853 sv_setsv(sv_mrk, sv_yes_mark);
5856 /* clean up; in particular, free all slabs above current one */
5857 LEAVE_SCOPE(oldsave);
5863 - regrepeat - repeatedly match something simple, report how many
5866 * [This routine now assumes that it will only match on things of length 1.
5867 * That was true before, but now we assume scan - reginput is the count,
5868 * rather than incrementing count on every character. [Er, except utf8.]]
5871 S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
5874 register char *scan;
5876 register char *loceol = PL_regeol;
5877 register I32 hardcount = 0;
5878 register bool utf8_target = PL_reg_match_utf8;
5881 PERL_UNUSED_ARG(depth);
5884 PERL_ARGS_ASSERT_REGREPEAT;
5887 if (max == REG_INFTY)
5889 else if (max < loceol - scan)
5890 loceol = scan + max;
5895 while (scan < loceol && hardcount < max && *scan != '\n') {
5896 scan += UTF8SKIP(scan);
5900 while (scan < loceol && *scan != '\n')
5907 while (scan < loceol && hardcount < max) {
5908 scan += UTF8SKIP(scan);
5919 /* To get here, EXACTish nodes must have *byte* length == 1. That
5920 * means they match only characters in the string that can be expressed
5921 * as a single byte. For non-utf8 strings, that means a simple match.
5922 * For utf8 strings, the character matched must be an invariant, or
5923 * downgradable to a single byte. The pattern's utf8ness is
5924 * irrelevant, as since it's a single byte, it either isn't utf8, or if
5925 * it is, it's an invariant */
5928 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5930 if (! utf8_target || UNI_IS_INVARIANT(c)) {
5931 while (scan < loceol && UCHARAT(scan) == c) {
5937 /* Here, the string is utf8, and the pattern char is different
5938 * in utf8 than not, so can't compare them directly. Outside the
5939 * loop, find find the two utf8 bytes that represent c, and then
5940 * look for those in sequence in the utf8 string */
5941 U8 high = UTF8_TWO_BYTE_HI(c);
5942 U8 low = UTF8_TWO_BYTE_LO(c);
5945 while (hardcount < max
5946 && scan + 1 < loceol
5947 && UCHARAT(scan) == high
5948 && UCHARAT(scan + 1) == low)
5956 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5960 PL_reg_flags |= RF_tainted;
5961 utf8_flags = FOLDEQ_UTF8_LOCALE;
5968 /* The comments for the EXACT case above apply as well to these fold
5973 assert(! UTF_PATTERN || UNI_IS_INVARIANT(c));
5975 if (utf8_target) { /* Use full Unicode fold matching */
5976 char *tmpeol = loceol;
5977 while (hardcount < max
5978 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
5979 STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags))
5986 /* XXX Note that the above handles properly the German sharp s in
5987 * the pattern matching ss in the string. But it doesn't handle
5988 * properly cases where the string contains say 'LIGATURE ff' and
5989 * the pattern is 'f+'. This would require, say, a new function or
5990 * revised interface to foldEQ_utf8(), in which the maximum number
5991 * of characters to match could be passed and it would return how
5992 * many actually did. This is just one of many cases where
5993 * multi-char folds don't work properly, and so the fix is being
5999 /* Here, the string isn't utf8 and c is a single byte; and either
6000 * the pattern isn't utf8 or c is an invariant, so its utf8ness
6001 * doesn't affect c. Can just do simple comparisons for exact or
6004 case EXACTF: folded = PL_fold[c]; break;
6006 case EXACTFU: folded = PL_fold_latin1[c]; break;
6007 case EXACTFL: folded = PL_fold_locale[c]; break;
6008 default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p));
6010 while (scan < loceol &&
6011 (UCHARAT(scan) == c || UCHARAT(scan) == folded))
6019 if (utf8_target || OP(p) == ANYOFV) {
6022 inclasslen = loceol - scan;
6023 while (hardcount < max
6024 && ((inclasslen = loceol - scan) > 0)
6025 && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target))
6031 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
6039 LOAD_UTF8_CHARCLASS_ALNUM();
6040 while (hardcount < max && scan < loceol &&
6041 swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6043 scan += UTF8SKIP(scan);
6047 while (scan < loceol && isWORDCHAR_L1((U8) *scan)) {
6055 while (scan < loceol && isALNUM((U8) *scan)) {
6060 while (scan < loceol && isWORDCHAR_A((U8) *scan)) {
6065 PL_reg_flags |= RF_tainted;
6068 while (hardcount < max && scan < loceol &&
6069 isALNUM_LC_utf8((U8*)scan)) {
6070 scan += UTF8SKIP(scan);
6074 while (scan < loceol && isALNUM_LC(*scan))
6084 LOAD_UTF8_CHARCLASS_ALNUM();
6085 while (hardcount < max && scan < loceol &&
6086 ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target))
6088 scan += UTF8SKIP(scan);
6092 while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) {
6099 goto utf8_Nwordchar;
6100 while (scan < loceol && ! isALNUM((U8) *scan)) {
6106 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6107 scan += UTF8SKIP(scan);
6111 while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
6117 PL_reg_flags |= RF_tainted;
6120 while (hardcount < max && scan < loceol &&
6121 !isALNUM_LC_utf8((U8*)scan)) {
6122 scan += UTF8SKIP(scan);
6126 while (scan < loceol && !isALNUM_LC(*scan))
6136 LOAD_UTF8_CHARCLASS_SPACE();
6137 while (hardcount < max && scan < loceol &&
6139 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6141 scan += UTF8SKIP(scan);
6147 while (scan < loceol && isSPACE_L1((U8) *scan)) {
6156 while (scan < loceol && isSPACE((U8) *scan)) {
6161 while (scan < loceol && isSPACE_A((U8) *scan)) {
6166 PL_reg_flags |= RF_tainted;
6169 while (hardcount < max && scan < loceol &&
6170 isSPACE_LC_utf8((U8*)scan)) {
6171 scan += UTF8SKIP(scan);
6175 while (scan < loceol && isSPACE_LC(*scan))
6185 LOAD_UTF8_CHARCLASS_SPACE();
6186 while (hardcount < max && scan < loceol &&
6188 swash_fetch(PL_utf8_space,(U8*)scan, utf8_target)))
6190 scan += UTF8SKIP(scan);
6196 while (scan < loceol && ! isSPACE_L1((U8) *scan)) {
6205 while (scan < loceol && ! isSPACE((U8) *scan)) {
6211 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6212 scan += UTF8SKIP(scan);
6216 while (scan < loceol && ! isSPACE_A((U8) *scan)) {
6222 PL_reg_flags |= RF_tainted;
6225 while (hardcount < max && scan < loceol &&
6226 !isSPACE_LC_utf8((U8*)scan)) {
6227 scan += UTF8SKIP(scan);
6231 while (scan < loceol && !isSPACE_LC(*scan))
6238 LOAD_UTF8_CHARCLASS_DIGIT();
6239 while (hardcount < max && scan < loceol &&
6240 swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6241 scan += UTF8SKIP(scan);
6245 while (scan < loceol && isDIGIT(*scan))
6250 while (scan < loceol && isDIGIT_A((U8) *scan)) {
6255 PL_reg_flags |= RF_tainted;
6258 while (hardcount < max && scan < loceol &&
6259 isDIGIT_LC_utf8((U8*)scan)) {
6260 scan += UTF8SKIP(scan);
6264 while (scan < loceol && isDIGIT_LC(*scan))
6271 LOAD_UTF8_CHARCLASS_DIGIT();
6272 while (hardcount < max && scan < loceol &&
6273 !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) {
6274 scan += UTF8SKIP(scan);
6278 while (scan < loceol && !isDIGIT(*scan))
6284 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6285 scan += UTF8SKIP(scan);
6289 while (scan < loceol && ! isDIGIT_A((U8) *scan)) {
6295 PL_reg_flags |= RF_tainted;
6298 while (hardcount < max && scan < loceol &&
6299 !isDIGIT_LC_utf8((U8*)scan)) {
6300 scan += UTF8SKIP(scan);
6304 while (scan < loceol && !isDIGIT_LC(*scan))
6311 while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) {
6317 LNBREAK can match two latin chars, which is ok,
6318 because we have a null terminated string, but we
6319 have to use hardcount in this situation
6321 while (scan < loceol && (c=is_LNBREAK_latin1(scan))) {
6330 while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) {
6335 while (scan < loceol && is_HORIZWS_latin1(scan))
6342 while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) {
6343 scan += UTF8SKIP(scan);
6347 while (scan < loceol && !is_HORIZWS_latin1(scan))
6355 while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) {
6360 while (scan < loceol && is_VERTWS_latin1(scan))
6368 while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) {
6369 scan += UTF8SKIP(scan);
6373 while (scan < loceol && !is_VERTWS_latin1(scan))
6379 default: /* Called on something of 0 width. */
6380 break; /* So match right here or not at all. */
6386 c = scan - PL_reginput;
6390 GET_RE_DEBUG_FLAGS_DECL;
6392 SV * const prop = sv_newmortal();
6393 regprop(prog, prop, p);
6394 PerlIO_printf(Perl_debug_log,
6395 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
6396 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
6404 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
6406 - regclass_swash - prepare the utf8 swash
6410 Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
6416 RXi_GET_DECL(prog,progi);
6417 const struct reg_data * const data = prog ? progi->data : NULL;
6419 PERL_ARGS_ASSERT_REGCLASS_SWASH;
6421 assert(ANYOF_NONBITMAP(node));
6423 if (data && data->count) {
6424 const U32 n = ARG(node);
6426 if (data->what[n] == 's') {
6427 SV * const rv = MUTABLE_SV(data->data[n]);
6428 AV * const av = MUTABLE_AV(SvRV(rv));
6429 SV **const ary = AvARRAY(av);
6432 /* See the end of regcomp.c:S_regclass() for
6433 * documentation of these array elements. */
6436 a = SvROK(ary[1]) ? &ary[1] : NULL;
6437 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
6441 else if (si && doinit) {
6442 sw = swash_init("utf8", "", si, 1, 0);
6443 (void)av_store(av, 1, sw);
6460 - reginclass - determine if a character falls into a character class
6462 n is the ANYOF regnode
6463 p is the target string
6464 lenp is pointer to the maximum number of bytes of how far to go in p
6465 (This is assumed wthout checking to always be at least the current
6467 utf8_target tells whether p is in UTF-8.
6469 Returns true if matched; false otherwise. If lenp is not NULL, on return
6470 from a successful match, the value it points to will be updated to how many
6471 bytes in p were matched. If there was no match, the value is undefined,
6472 possibly changed from the input.
6474 Note that this can be a synthetic start class, a combination of various
6475 nodes, so things you think might be mutually exclusive, such as locale,
6476 aren't. It can match both locale and non-locale
6481 S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target)
6484 const char flags = ANYOF_FLAGS(n);
6490 PERL_ARGS_ASSERT_REGINCLASS;
6492 /* If c is not already the code point, get it */
6493 if (utf8_target && !UTF8_IS_INVARIANT(c)) {
6494 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len,
6495 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
6496 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
6497 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
6498 * UTF8_ALLOW_FFFF */
6499 if (c_len == (STRLEN)-1)
6500 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
6506 /* Use passed in max length, or one character if none passed in or less
6507 * than one character. And assume will match just one character. This is
6508 * overwritten later if matched more. */
6510 maxlen = (*lenp > c_len) ? *lenp : c_len;
6518 /* If this character is potentially in the bitmap, check it */
6520 if (ANYOF_BITMAP_TEST(n, c))
6522 else if (flags & ANYOF_NON_UTF8_LATIN1_ALL
6529 else if (flags & ANYOF_LOCALE) {
6530 PL_reg_flags |= RF_tainted;
6532 if ((flags & ANYOF_LOC_NONBITMAP_FOLD)
6533 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
6537 else if (ANYOF_CLASS_TEST_ANY_SET(n) &&
6538 ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
6539 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
6540 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
6541 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
6542 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
6543 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
6544 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
6545 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
6546 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
6547 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
6548 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
6549 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
6550 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
6551 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
6552 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
6553 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
6554 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
6555 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
6556 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
6557 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
6558 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
6559 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
6560 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
6561 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
6562 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
6563 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
6564 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
6565 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
6566 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
6567 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
6568 ) /* How's that for a conditional? */
6575 /* If the bitmap didn't (or couldn't) match, and something outside the
6576 * bitmap could match, try that. Locale nodes specifiy completely the
6577 * behavior of code points in the bit map (otherwise, a utf8 target would
6578 * cause them to be treated as Unicode and not locale), except in
6579 * the very unlikely event when this node is a synthetic start class, which
6580 * could be a combination of locale and non-locale nodes. So allow locale
6581 * to match for the synthetic start class, which will give a false
6582 * positive that will be resolved when the match is done again as not part
6583 * of the synthetic start class */
6585 if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) {
6586 match = TRUE; /* Everything above 255 matches */
6588 else if (ANYOF_NONBITMAP(n)
6589 && ((flags & ANYOF_NONBITMAP_NON_UTF8)
6592 || (! (flags & ANYOF_LOCALE))
6593 || (flags & ANYOF_IS_SYNTHETIC)))))
6596 SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
6604 /* Not utf8. Convert as much of the string as available up
6605 * to the limit of how far the (single) character in the
6606 * pattern can possibly match (no need to go further). If
6607 * the node is a straight ANYOF or not folding, it can't
6608 * match more than one. Otherwise, It can match up to how
6609 * far a single char can fold to. Since not utf8, each
6610 * character is a single byte, so the max it can be in
6611 * bytes is the same as the max it can be in characters */
6612 STRLEN len = (OP(n) == ANYOF
6613 || ! (flags & ANYOF_LOC_NONBITMAP_FOLD))
6615 : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND)
6617 : UTF8_MAX_FOLD_CHAR_EXPAND;
6618 utf8_p = bytes_to_utf8(p, &len);
6621 if (swash_fetch(sw, utf8_p, TRUE))
6623 else if (flags & ANYOF_LOC_NONBITMAP_FOLD) {
6625 /* Here, we need to test if the fold of the target string
6626 * matches. The non-multi char folds have all been moved to
6627 * the compilation phase, and the multi-char folds have
6628 * been stored by regcomp into 'av'; we linearly check to
6629 * see if any match the target string (folded). We know
6630 * that the originals were each one character, but we don't
6631 * currently know how many characters/bytes each folded to,
6632 * except we do know that there are small limits imposed by
6633 * Unicode. XXX A performance enhancement would be to have
6634 * regcomp.c store the max number of chars/bytes that are
6635 * in an av entry, as, say the 0th element. Even better
6636 * would be to have a hash of the few characters that can
6637 * start a multi-char fold to the max number of chars of
6640 * If there is a match, we will need to advance (if lenp is
6641 * specified) the match pointer in the target string. But
6642 * what we are comparing here isn't that string directly,
6643 * but its fold, whose length may differ from the original.
6644 * As we go along in constructing the fold, therefore, we
6645 * create a map so that we know how many bytes in the
6646 * source to advance given that we have matched a certain
6647 * number of bytes in the fold. This map is stored in
6648 * 'map_fold_len_back'. Let n mean the number of bytes in
6649 * the fold of the first character that we are folding.
6650 * Then map_fold_len_back[n] is set to the number of bytes
6651 * in that first character. Similarly let m be the
6652 * corresponding number for the second character to be
6653 * folded. Then map_fold_len_back[n+m] is set to the
6654 * number of bytes occupied by the first two source
6655 * characters. ... */
6656 U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 };
6657 U8 folded[UTF8_MAXBYTES_CASE+1];
6658 STRLEN foldlen = 0; /* num bytes in fold of 1st char */
6659 STRLEN total_foldlen = 0; /* num bytes in fold of all
6662 if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) {
6664 /* Here, only need to fold the first char of the target
6665 * string. It the source wasn't utf8, is 1 byte long */
6666 to_utf8_fold(utf8_p, folded, &foldlen);
6667 total_foldlen = foldlen;
6668 map_fold_len_back[foldlen] = (utf8_target)
6674 /* Here, need to fold more than the first char. Do so
6675 * up to the limits */
6676 U8* source_ptr = utf8_p; /* The source for the fold
6679 U8* folded_ptr = folded;
6680 U8* e = utf8_p + maxlen; /* Can't go beyond last
6681 available byte in the
6685 i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e;
6689 /* Fold the next character */
6690 U8 this_char_folded[UTF8_MAXBYTES_CASE+1];
6691 STRLEN this_char_foldlen;
6692 to_utf8_fold(source_ptr,
6694 &this_char_foldlen);
6696 /* Bail if it would exceed the byte limit for
6697 * folding a single char. */
6698 if (this_char_foldlen + folded_ptr - folded >
6704 /* Add the fold of this character */
6705 Copy(this_char_folded,
6709 source_ptr += UTF8SKIP(source_ptr);
6710 folded_ptr += this_char_foldlen;
6711 total_foldlen = folded_ptr - folded;
6713 /* Create map from the number of bytes in the fold
6714 * back to the number of bytes in the source. If
6715 * the source isn't utf8, the byte count is just
6716 * the number of characters so far */
6717 map_fold_len_back[total_foldlen]
6719 ? source_ptr - utf8_p
6726 /* Do the linear search to see if the fold is in the list
6727 * of multi-char folds. */
6730 for (i = 0; i <= av_len(av); i++) {
6731 SV* const sv = *av_fetch(av, i, FALSE);
6733 const char * const s = SvPV_const(sv, len);
6735 if (len <= total_foldlen
6736 && memEQ(s, (char*)folded, len)
6738 /* If 0, means matched a partial char. See
6740 && map_fold_len_back[len])
6743 /* Advance the target string ptr to account for
6744 * this fold, but have to translate from the
6745 * folded length to the corresponding source
6748 *lenp = map_fold_len_back[len];
6757 /* If we allocated a string above, free it */
6758 if (! utf8_target) Safefree(utf8_p);
6763 return (flags & ANYOF_INVERT) ? !match : match;
6767 S_reghop3(U8 *s, I32 off, const U8* lim)
6771 PERL_ARGS_ASSERT_REGHOP3;
6774 while (off-- && s < lim) {
6775 /* XXX could check well-formedness here */
6780 while (off++ && s > lim) {
6782 if (UTF8_IS_CONTINUED(*s)) {
6783 while (s > lim && UTF8_IS_CONTINUATION(*s))
6786 /* XXX could check well-formedness here */
6793 /* there are a bunch of places where we use two reghop3's that should
6794 be replaced with this routine. but since thats not done yet
6795 we ifdef it out - dmq
6798 S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
6802 PERL_ARGS_ASSERT_REGHOP4;
6805 while (off-- && s < rlim) {
6806 /* XXX could check well-formedness here */
6811 while (off++ && s > llim) {
6813 if (UTF8_IS_CONTINUED(*s)) {
6814 while (s > llim && UTF8_IS_CONTINUATION(*s))
6817 /* XXX could check well-formedness here */
6825 S_reghopmaybe3(U8* s, I32 off, const U8* lim)
6829 PERL_ARGS_ASSERT_REGHOPMAYBE3;
6832 while (off-- && s < lim) {
6833 /* XXX could check well-formedness here */
6840 while (off++ && s > lim) {
6842 if (UTF8_IS_CONTINUED(*s)) {
6843 while (s > lim && UTF8_IS_CONTINUATION(*s))
6846 /* XXX could check well-formedness here */
6855 restore_pos(pTHX_ void *arg)
6858 regexp * const rex = (regexp *)arg;
6859 if (PL_reg_eval_set) {
6860 if (PL_reg_oldsaved) {
6861 rex->subbeg = PL_reg_oldsaved;
6862 rex->sublen = PL_reg_oldsavedlen;
6863 #ifdef PERL_OLD_COPY_ON_WRITE
6864 rex->saved_copy = PL_nrs;
6866 RXp_MATCH_COPIED_on(rex);
6868 PL_reg_magic->mg_len = PL_reg_oldpos;
6869 PL_reg_eval_set = 0;
6870 PL_curpm = PL_reg_oldcurpm;
6875 S_to_utf8_substr(pTHX_ register regexp *prog)
6879 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
6882 if (prog->substrs->data[i].substr
6883 && !prog->substrs->data[i].utf8_substr) {
6884 SV* const sv = newSVsv(prog->substrs->data[i].substr);
6885 prog->substrs->data[i].utf8_substr = sv;
6886 sv_utf8_upgrade(sv);
6887 if (SvVALID(prog->substrs->data[i].substr)) {
6888 const U8 flags = BmFLAGS(prog->substrs->data[i].substr);
6889 if (flags & FBMcf_TAIL) {
6890 /* Trim the trailing \n that fbm_compile added last
6892 SvCUR_set(sv, SvCUR(sv) - 1);
6893 /* Whilst this makes the SV technically "invalid" (as its
6894 buffer is no longer followed by "\0") when fbm_compile()
6895 adds the "\n" back, a "\0" is restored. */
6897 fbm_compile(sv, flags);
6899 if (prog->substrs->data[i].substr == prog->check_substr)
6900 prog->check_utf8 = sv;
6906 S_to_byte_substr(pTHX_ register regexp *prog)
6911 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
6914 if (prog->substrs->data[i].utf8_substr
6915 && !prog->substrs->data[i].substr) {
6916 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
6917 if (sv_utf8_downgrade(sv, TRUE)) {
6918 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
6920 = BmFLAGS(prog->substrs->data[i].utf8_substr);
6921 if (flags & FBMcf_TAIL) {
6922 /* Trim the trailing \n that fbm_compile added last
6924 SvCUR_set(sv, SvCUR(sv) - 1);
6926 fbm_compile(sv, flags);
6932 prog->substrs->data[i].substr = sv;
6933 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
6934 prog->check_substr = sv;
6941 * c-indentation-style: bsd
6943 * indent-tabs-mode: t
6946 * ex: set ts=8 sts=4 sw=4 noet: