5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
40 /* At least one required character in the target string is expressible only in
42 static const char* const non_utf8_target_but_utf8_required
43 = "Can't match, because target string needs to be in UTF-8\n";
45 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
46 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
51 * pregcomp and pregexec -- regsub and regerror are not used in perl
53 * Copyright (c) 1986 by University of Toronto.
54 * Written by Henry Spencer. Not derived from licensed software.
56 * Permission is granted to anyone to use this software for any
57 * purpose on any computer system, and to redistribute it freely,
58 * subject to the following restrictions:
60 * 1. The author is not responsible for the consequences of use of
61 * this software, no matter how awful, even if they arise
64 * 2. The origin of this software must not be misrepresented, either
65 * by explicit claim or by omission.
67 * 3. Altered versions must be plainly marked as such, and must not
68 * be misrepresented as being the original software.
70 **** Alterations to Henry's code are...
72 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
73 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
74 **** by Larry Wall and others
76 **** You may distribute under the terms of either the GNU General Public
77 **** License or the Artistic License, as specified in the README file.
79 * Beware that some of this code is subtly aware of the way operator
80 * precedence is structured in regular expressions. Serious changes in
81 * regular-expression syntax might require a total rethink.
84 #define PERL_IN_REGEXEC_C
87 #ifdef PERL_IN_XSUB_RE
93 #include "inline_invlist.c"
94 #include "unicode_constants.h"
96 #define RF_tainted 1 /* tainted information used? e.g. locale */
97 #define RF_warned 2 /* warned about big count? */
99 #define RF_utf8 8 /* Pattern contains multibyte chars? */
101 #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
103 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
106 #define STATIC static
109 /* Valid for non-utf8 strings: avoids the reginclass
110 * call if there are no complications: i.e., if everything matchable is
111 * straight forward in the bitmap */
112 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \
113 : ANYOF_BITMAP_TEST(p,*(c)))
119 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
120 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
122 #define HOPc(pos,off) \
123 (char *)(PL_reg_match_utf8 \
124 ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
126 #define HOPBACKc(pos, off) \
127 (char*)(PL_reg_match_utf8\
128 ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
129 : (pos - off >= PL_bostr) \
133 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
134 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
137 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
138 #define NEXTCHR_IS_EOS (nextchr < 0)
140 #define SET_nextchr \
141 nextchr = ((locinput < PL_regeol) ? UCHARAT(locinput) : NEXTCHR_EOS)
143 #define SET_locinput(p) \
148 /* these are unrolled below in the CCC_TRY_XXX defined */
149 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
150 if (!CAT2(PL_utf8_,class)) { \
152 ENTER; save_re_context(); \
153 ok=CAT2(is_utf8_,class)((const U8*)str); \
154 PERL_UNUSED_VAR(ok); \
155 assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
156 /* Doesn't do an assert to verify that is correct */
157 #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
158 if (!CAT2(PL_utf8_,class)) { \
159 bool throw_away PERL_UNUSED_DECL; \
160 ENTER; save_re_context(); \
161 throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
162 PERL_UNUSED_VAR(throw_away); \
165 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
166 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
168 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
169 /* No asserts are done for some of these, in case called on a */ \
170 /* Unicode version in which they map to nothing */ \
171 LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
172 LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
174 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
176 /* The actual code for CCC_TRY, which uses several variables from the routine
177 * it's callable from. It is designed to be the bulk of a case statement.
178 * FUNC is the macro or function to call on non-utf8 targets that indicate if
179 * nextchr matches the class.
180 * UTF8_TEST is the whole test string to use for utf8 targets
181 * LOAD is what to use to test, and if not present to load in the swash for the
183 * POS_OR_NEG is either empty or ! to complement the results of FUNC or
185 * The logic is: Fail if we're at the end-of-string; otherwise if the target is
186 * utf8 and a variant, load the swash if necessary and test using the utf8
187 * test. Advance to the next character if test is ok, otherwise fail; If not
188 * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
189 * fails, or advance to the next character */
191 #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
192 if (NEXTCHR_IS_EOS) { \
195 if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
196 LOAD_UTF8_CHARCLASS(CLASS, STR); \
197 if (POS_OR_NEG (UTF8_TEST)) { \
201 else if (POS_OR_NEG (FUNC(nextchr))) { \
204 goto increment_locinput;
206 /* Handle the non-locale cases for a character class and its complement. It
207 * calls _CCC_TRY_CODE with a ! to complement the test for the character class.
208 * This is because that code fails when the test succeeds, so we want to have
209 * the test fail so that the code succeeds. The swash is stored in a
210 * predictable PL_ place */
211 #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
214 _CCC_TRY_CODE( !, FUNC, \
215 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
216 (U8*)locinput, TRUE)), \
219 _CCC_TRY_CODE( PLACEHOLDER , FUNC, \
220 cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
221 (U8*)locinput, TRUE)), \
223 /* Generate the case statements for both locale and non-locale character
224 * classes in regmatch for classes that don't have special unicode semantics.
225 * Locales don't use an immediate swash, but an intermediary special locale
226 * function that is called on the pointer to the current place in the input
227 * string. That function will resolve to needing the same swash. One might
228 * think that because we don't know what the locale will match, we shouldn't
229 * check with the swash loading function that it loaded properly; ie, that we
230 * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
231 * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
233 #define CCC_TRY(NAME, NNAME, FUNC, \
234 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
235 NAMEA, NNAMEA, FUNCA, \
238 PL_reg_flags |= RF_tainted; \
239 _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
241 PL_reg_flags |= RF_tainted; \
242 _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
245 if (NEXTCHR_IS_EOS || ! FUNCA(nextchr)) { \
248 /* Matched a utf8-invariant, so don't have to worry about utf8 */ \
252 if (NEXTCHR_IS_EOS || FUNCA(nextchr)) { \
255 goto increment_locinput; \
256 /* Generate the non-locale cases */ \
257 _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
259 /* This is like CCC_TRY, but has an extra set of parameters for generating case
260 * statements to handle separate Unicode semantics nodes */
261 #define CCC_TRY_U(NAME, NNAME, FUNC, \
262 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
263 NAMEU, NNAMEU, FUNCU, \
264 NAMEA, NNAMEA, FUNCA, \
266 CCC_TRY(NAME, NNAME, FUNC, \
267 NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
268 NAMEA, NNAMEA, FUNCA, \
270 _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
272 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
274 /* for use after a quantifier and before an EXACT-like node -- japhy */
275 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
277 * NOTE that *nothing* that affects backtracking should be in here, specifically
278 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
279 * node that is in between two EXACT like nodes when ascertaining what the required
280 * "follow" character is. This should probably be moved to regex compile time
281 * although it may be done at run time beause of the REF possibility - more
282 * investigation required. -- demerphq
284 #define JUMPABLE(rn) ( \
286 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
288 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
289 OP(rn) == PLUS || OP(rn) == MINMOD || \
291 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
293 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
295 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
298 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
299 we don't need this definition. */
300 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
301 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
302 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
305 /* ... so we use this as its faster. */
306 #define IS_TEXT(rn) ( OP(rn)==EXACT )
307 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
308 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
309 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
314 Search for mandatory following text node; for lookahead, the text must
315 follow but for lookbehind (rn->flags != 0) we skip to the next step.
317 #define FIND_NEXT_IMPT(rn) STMT_START { \
318 while (JUMPABLE(rn)) { \
319 const OPCODE type = OP(rn); \
320 if (type == SUSPEND || PL_regkind[type] == CURLY) \
321 rn = NEXTOPER(NEXTOPER(rn)); \
322 else if (type == PLUS) \
324 else if (type == IFMATCH) \
325 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
326 else rn += NEXT_OFF(rn); \
330 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
331 * These are for the pre-composed Hangul syllables, which are all in a
332 * contiguous block and arranged there in such a way so as to facilitate
333 * alorithmic determination of their characteristics. As such, they don't need
334 * a swash, but can be determined by simple arithmetic. Almost all are
335 * GCB=LVT, but every 28th one is a GCB=LV */
336 #define SBASE 0xAC00 /* Start of block */
337 #define SCount 11172 /* Length of block */
340 static void restore_pos(pTHX_ void *arg);
342 #define REGCP_PAREN_ELEMS 3
343 #define REGCP_OTHER_ELEMS 3
344 #define REGCP_FRAME_ELEMS 1
345 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
346 * are needed for the regexp context stack bookkeeping. */
349 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
352 const int retval = PL_savestack_ix;
353 const int paren_elems_to_push =
354 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
355 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
356 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
358 GET_RE_DEBUG_FLAGS_DECL;
360 PERL_ARGS_ASSERT_REGCPPUSH;
362 if (paren_elems_to_push < 0)
363 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
364 paren_elems_to_push);
366 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
367 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
368 " out of range (%lu-%ld)",
370 (unsigned long)maxopenparen,
373 SSGROW(total_elems + REGCP_FRAME_ELEMS);
376 if ((int)maxopenparen > (int)parenfloor)
377 PerlIO_printf(Perl_debug_log,
378 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
383 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
384 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
385 SSPUSHINT(rex->offs[p].end);
386 SSPUSHINT(rex->offs[p].start);
387 SSPUSHINT(rex->offs[p].start_tmp);
388 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
389 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
391 (IV)rex->offs[p].start,
392 (IV)rex->offs[p].start_tmp,
396 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
397 SSPUSHINT(maxopenparen);
398 SSPUSHINT(rex->lastparen);
399 SSPUSHINT(rex->lastcloseparen);
400 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
405 /* These are needed since we do not localize EVAL nodes: */
406 #define REGCP_SET(cp) \
408 PerlIO_printf(Perl_debug_log, \
409 " Setting an EVAL scope, savestack=%"IVdf"\n", \
410 (IV)PL_savestack_ix)); \
413 #define REGCP_UNWIND(cp) \
415 if (cp != PL_savestack_ix) \
416 PerlIO_printf(Perl_debug_log, \
417 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
418 (IV)(cp), (IV)PL_savestack_ix)); \
421 #define UNWIND_PAREN(lp, lcp) \
422 for (n = rex->lastparen; n > lp; n--) \
423 rex->offs[n].end = -1; \
424 rex->lastparen = n; \
425 rex->lastcloseparen = lcp;
429 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
434 GET_RE_DEBUG_FLAGS_DECL;
436 PERL_ARGS_ASSERT_REGCPPOP;
438 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
440 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
441 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
442 rex->lastcloseparen = SSPOPINT;
443 rex->lastparen = SSPOPINT;
444 *maxopenparen_p = SSPOPINT;
446 i -= REGCP_OTHER_ELEMS;
447 /* Now restore the parentheses context. */
449 if (i || rex->lastparen + 1 <= rex->nparens)
450 PerlIO_printf(Perl_debug_log,
451 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
456 paren = *maxopenparen_p;
457 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
459 rex->offs[paren].start_tmp = SSPOPINT;
460 rex->offs[paren].start = SSPOPINT;
462 if (paren <= rex->lastparen)
463 rex->offs[paren].end = tmps;
464 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
465 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
467 (IV)rex->offs[paren].start,
468 (IV)rex->offs[paren].start_tmp,
469 (IV)rex->offs[paren].end,
470 (paren > rex->lastparen ? "(skipped)" : ""));
475 /* It would seem that the similar code in regtry()
476 * already takes care of this, and in fact it is in
477 * a better location to since this code can #if 0-ed out
478 * but the code in regtry() is needed or otherwise tests
479 * requiring null fields (pat.t#187 and split.t#{13,14}
480 * (as of patchlevel 7877) will fail. Then again,
481 * this code seems to be necessary or otherwise
482 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
483 * --jhi updated by dapm */
484 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
485 if (i > *maxopenparen_p)
486 rex->offs[i].start = -1;
487 rex->offs[i].end = -1;
488 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
489 " \\%"UVuf": %s ..-1 undeffing\n",
491 (i > *maxopenparen_p) ? "-1" : " "
497 /* restore the parens and associated vars at savestack position ix,
498 * but without popping the stack */
501 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
503 I32 tmpix = PL_savestack_ix;
504 PL_savestack_ix = ix;
505 regcppop(rex, maxopenparen_p);
506 PL_savestack_ix = tmpix;
509 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
512 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
514 /* Returns a boolean as to whether or not 'character' is a member of the
515 * Posix character class given by 'classnum' that should be equivalent to a
516 * value in the typedef '_char_class_number'.
518 * Ideally this could be replaced by a just an array of function pointers
519 * to the C library functions that implement the macros this calls.
520 * However, to compile, the precise function signatures are required, and
521 * these may vary from platform to to platform. To avoid having to figure
522 * out what those all are on each platform, I (khw) am using this method,
523 * which adds an extra layer of function call overhead. But we don't
524 * particularly care about performance with locales anyway. */
526 switch ((_char_class_number) classnum) {
527 case _CC_ENUM_ALNUMC: return isALNUMC_LC(character);
528 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
529 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
530 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
531 case _CC_ENUM_LOWER: return isLOWER_LC(character);
532 case _CC_ENUM_PRINT: return isPRINT_LC(character);
533 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
534 case _CC_ENUM_UPPER: return isUPPER_LC(character);
535 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
536 case _CC_ENUM_SPACE: return isSPACE_LC(character);
537 case _CC_ENUM_BLANK: return isBLANK_LC(character);
538 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
539 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
540 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
541 case _CC_ENUM_ASCII: return isASCII_LC(character);
542 default: /* VERTSPACE should never occur in locales */
543 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
546 assert(0); /* NOTREACHED */
551 * pregexec and friends
554 #ifndef PERL_IN_XSUB_RE
556 - pregexec - match a regexp against a string
559 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
560 char *strbeg, I32 minend, SV *screamer, U32 nosave)
561 /* stringarg: the point in the string at which to begin matching */
562 /* strend: pointer to null at end of string */
563 /* strbeg: real beginning of string */
564 /* minend: end of match must be >= minend bytes after stringarg. */
565 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
566 * itself is accessed via the pointers above */
567 /* nosave: For optimizations. */
569 PERL_ARGS_ASSERT_PREGEXEC;
572 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
573 nosave ? 0 : REXEC_COPY_STR);
578 * Need to implement the following flags for reg_anch:
580 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
582 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
583 * INTUIT_AUTORITATIVE_ML
584 * INTUIT_ONCE_NOML - Intuit can match in one location only.
587 * Another flag for this function: SECOND_TIME (so that float substrs
588 * with giant delta may be not rechecked).
591 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
593 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
594 Otherwise, only SvCUR(sv) is used to get strbeg. */
596 /* XXXX We assume that strpos is strbeg unless sv. */
598 /* XXXX Some places assume that there is a fixed substring.
599 An update may be needed if optimizer marks as "INTUITable"
600 RExen without fixed substrings. Similarly, it is assumed that
601 lengths of all the strings are no more than minlen, thus they
602 cannot come from lookahead.
603 (Or minlen should take into account lookahead.)
604 NOTE: Some of this comment is not correct. minlen does now take account
605 of lookahead/behind. Further research is required. -- demerphq
609 /* A failure to find a constant substring means that there is no need to make
610 an expensive call to REx engine, thus we celebrate a failure. Similarly,
611 finding a substring too deep into the string means that less calls to
612 regtry() should be needed.
614 REx compiler's optimizer found 4 possible hints:
615 a) Anchored substring;
617 c) Whether we are anchored (beginning-of-line or \G);
618 d) First node (of those at offset 0) which may distinguish positions;
619 We use a)b)d) and multiline-part of c), and try to find a position in the
620 string which does not contradict any of them.
623 /* Most of decisions we do here should have been done at compile time.
624 The nodes of the REx which we used for the search should have been
625 deleted from the finite automaton. */
628 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
629 char *strend, const U32 flags, re_scream_pos_data *data)
632 struct regexp *const prog = ReANY(rx);
634 /* Should be nonnegative! */
640 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
642 char *other_last = NULL; /* other substr checked before this */
643 char *check_at = NULL; /* check substr found at this pos */
644 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
645 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
646 RXi_GET_DECL(prog,progi);
648 const char * const i_strpos = strpos;
650 GET_RE_DEBUG_FLAGS_DECL;
652 PERL_ARGS_ASSERT_RE_INTUIT_START;
653 PERL_UNUSED_ARG(flags);
654 PERL_UNUSED_ARG(data);
656 RX_MATCH_UTF8_set(rx,utf8_target);
659 PL_reg_flags |= RF_utf8;
662 debug_start_match(rx, utf8_target, strpos, strend,
663 sv ? "Guessing start of match in sv for"
664 : "Guessing start of match in string for");
667 /* CHR_DIST() would be more correct here but it makes things slow. */
668 if (prog->minlen > strend - strpos) {
669 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
670 "String too short... [re_intuit_start]\n"));
674 /* XXX we need to pass strbeg as a separate arg: the following is
675 * guesswork and can be wrong... */
676 if (sv && SvPOK(sv)) {
677 char * p = SvPVX(sv);
678 STRLEN cur = SvCUR(sv);
679 if (p <= strpos && strpos < p + cur) {
681 assert(p <= strend && strend <= p + cur);
684 strbeg = strend - cur;
691 if (!prog->check_utf8 && prog->check_substr)
692 to_utf8_substr(prog);
693 check = prog->check_utf8;
695 if (!prog->check_substr && prog->check_utf8) {
696 if (! to_byte_substr(prog)) {
697 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
700 check = prog->check_substr;
702 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
703 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
704 || ( (prog->extflags & RXf_ANCH_BOL)
705 && !multiline ) ); /* Check after \n? */
708 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
709 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
710 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
712 && (strpos != strbeg)) {
713 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
716 if (prog->check_offset_min == prog->check_offset_max
717 && !(prog->extflags & RXf_CANY_SEEN)
718 && ! multiline) /* /m can cause \n's to match that aren't
719 accounted for in the string max length.
720 See [perl #115242] */
722 /* Substring at constant offset from beg-of-str... */
725 s = HOP3c(strpos, prog->check_offset_min, strend);
728 slen = SvCUR(check); /* >= 1 */
730 if ( strend - s > slen || strend - s < slen - 1
731 || (strend - s == slen && strend[-1] != '\n')) {
732 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
735 /* Now should match s[0..slen-2] */
737 if (slen && (*SvPVX_const(check) != *s
739 && memNE(SvPVX_const(check), s, slen)))) {
741 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
745 else if (*SvPVX_const(check) != *s
746 || ((slen = SvCUR(check)) > 1
747 && memNE(SvPVX_const(check), s, slen)))
750 goto success_at_start;
753 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
755 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
756 end_shift = prog->check_end_shift;
759 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
760 - (SvTAIL(check) != 0);
761 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
763 if (end_shift < eshift)
767 else { /* Can match at random position */
770 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
771 end_shift = prog->check_end_shift;
773 /* end shift should be non negative here */
776 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
778 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
779 (IV)end_shift, RX_PRECOMP(prog));
783 /* Find a possible match in the region s..strend by looking for
784 the "check" substring in the region corrected by start/end_shift. */
787 I32 srch_start_shift = start_shift;
788 I32 srch_end_shift = end_shift;
791 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
792 srch_end_shift -= ((strbeg - s) - srch_start_shift);
793 srch_start_shift = strbeg - s;
795 DEBUG_OPTIMISE_MORE_r({
796 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
797 (IV)prog->check_offset_min,
798 (IV)srch_start_shift,
800 (IV)prog->check_end_shift);
803 if (prog->extflags & RXf_CANY_SEEN) {
804 start_point= (U8*)(s + srch_start_shift);
805 end_point= (U8*)(strend - srch_end_shift);
807 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
808 end_point= HOP3(strend, -srch_end_shift, strbeg);
810 DEBUG_OPTIMISE_MORE_r({
811 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
812 (int)(end_point - start_point),
813 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
817 s = fbm_instr( start_point, end_point,
818 check, multiline ? FBMrf_MULTILINE : 0);
820 /* Update the count-of-usability, remove useless subpatterns,
824 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
825 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
826 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
827 (s ? "Found" : "Did not find"),
828 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
829 ? "anchored" : "floating"),
832 (s ? " at offset " : "...\n") );
837 /* Finish the diagnostic message */
838 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
840 /* XXX dmq: first branch is for positive lookbehind...
841 Our check string is offset from the beginning of the pattern.
842 So we need to do any stclass tests offset forward from that
851 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
852 Start with the other substr.
853 XXXX no SCREAM optimization yet - and a very coarse implementation
854 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
855 *always* match. Probably should be marked during compile...
856 Probably it is right to do no SCREAM here...
859 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
860 : (prog->float_substr && prog->anchored_substr))
862 /* Take into account the "other" substring. */
863 /* XXXX May be hopelessly wrong for UTF... */
866 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
869 char * const last = HOP3c(s, -start_shift, strbeg);
871 char * const saved_s = s;
874 t = s - prog->check_offset_max;
875 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
877 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
882 t = HOP3c(t, prog->anchored_offset, strend);
883 if (t < other_last) /* These positions already checked */
885 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
888 /* XXXX It is not documented what units *_offsets are in.
889 We assume bytes, but this is clearly wrong.
890 Meaning this code needs to be carefully reviewed for errors.
894 /* On end-of-str: see comment below. */
895 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
896 if (must == &PL_sv_undef) {
898 DEBUG_r(must = prog->anchored_utf8); /* for debug */
903 HOP3(HOP3(last1, prog->anchored_offset, strend)
904 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
906 multiline ? FBMrf_MULTILINE : 0
909 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
910 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
911 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
912 (s ? "Found" : "Contradicts"),
913 quoted, RE_SV_TAIL(must));
918 if (last1 >= last2) {
919 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
920 ", giving up...\n"));
923 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
924 ", trying floating at offset %ld...\n",
925 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
926 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
927 s = HOP3c(last, 1, strend);
931 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
932 (long)(s - i_strpos)));
933 t = HOP3c(s, -prog->anchored_offset, strbeg);
934 other_last = HOP3c(s, 1, strend);
942 else { /* Take into account the floating substring. */
944 char * const saved_s = s;
947 t = HOP3c(s, -start_shift, strbeg);
949 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
950 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
951 last = HOP3c(t, prog->float_max_offset, strend);
952 s = HOP3c(t, prog->float_min_offset, strend);
955 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
956 must = utf8_target ? prog->float_utf8 : prog->float_substr;
957 /* fbm_instr() takes into account exact value of end-of-str
958 if the check is SvTAIL(ed). Since false positives are OK,
959 and end-of-str is not later than strend we are OK. */
960 if (must == &PL_sv_undef) {
962 DEBUG_r(must = prog->float_utf8); /* for debug message */
965 s = fbm_instr((unsigned char*)s,
966 (unsigned char*)last + SvCUR(must)
968 must, multiline ? FBMrf_MULTILINE : 0);
970 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
971 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
972 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
973 (s ? "Found" : "Contradicts"),
974 quoted, RE_SV_TAIL(must));
978 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
979 ", giving up...\n"));
982 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
983 ", trying anchored starting at offset %ld...\n",
984 (long)(saved_s + 1 - i_strpos)));
986 s = HOP3c(t, 1, strend);
990 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
991 (long)(s - i_strpos)));
992 other_last = s; /* Fix this later. --Hugo */
1002 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
1004 DEBUG_OPTIMISE_MORE_r(
1005 PerlIO_printf(Perl_debug_log,
1006 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
1007 (IV)prog->check_offset_min,
1008 (IV)prog->check_offset_max,
1016 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
1018 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
1021 /* Fixed substring is found far enough so that the match
1022 cannot start at strpos. */
1024 if (ml_anch && t[-1] != '\n') {
1025 /* Eventually fbm_*() should handle this, but often
1026 anchored_offset is not 0, so this check will not be wasted. */
1027 /* XXXX In the code below we prefer to look for "^" even in
1028 presence of anchored substrings. And we search even
1029 beyond the found float position. These pessimizations
1030 are historical artefacts only. */
1032 while (t < strend - prog->minlen) {
1034 if (t < check_at - prog->check_offset_min) {
1035 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1036 /* Since we moved from the found position,
1037 we definitely contradict the found anchored
1038 substr. Due to the above check we do not
1039 contradict "check" substr.
1040 Thus we can arrive here only if check substr
1041 is float. Redo checking for "other"=="fixed".
1044 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1045 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
1046 goto do_other_anchored;
1048 /* We don't contradict the found floating substring. */
1049 /* XXXX Why not check for STCLASS? */
1051 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1052 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1055 /* Position contradicts check-string */
1056 /* XXXX probably better to look for check-string
1057 than for "\n", so one should lower the limit for t? */
1058 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1059 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1060 other_last = strpos = s = t + 1;
1065 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1066 PL_colors[0], PL_colors[1]));
1070 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1071 PL_colors[0], PL_colors[1]));
1075 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1078 /* The found string does not prohibit matching at strpos,
1079 - no optimization of calling REx engine can be performed,
1080 unless it was an MBOL and we are not after MBOL,
1081 or a future STCLASS check will fail this. */
1083 /* Even in this situation we may use MBOL flag if strpos is offset
1084 wrt the start of the string. */
1085 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1086 && (strpos != strbeg) && strpos[-1] != '\n'
1087 /* May be due to an implicit anchor of m{.*foo} */
1088 && !(prog->intflags & PREGf_IMPLICIT))
1093 DEBUG_EXECUTE_r( if (ml_anch)
1094 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1095 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1098 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1100 prog->check_utf8 /* Could be deleted already */
1101 && --BmUSEFUL(prog->check_utf8) < 0
1102 && (prog->check_utf8 == prog->float_utf8)
1104 prog->check_substr /* Could be deleted already */
1105 && --BmUSEFUL(prog->check_substr) < 0
1106 && (prog->check_substr == prog->float_substr)
1109 /* If flags & SOMETHING - do not do it many times on the same match */
1110 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1111 /* XXX Does the destruction order has to change with utf8_target? */
1112 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1113 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1114 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1115 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1116 check = NULL; /* abort */
1118 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1119 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1120 if (prog->intflags & PREGf_IMPLICIT)
1121 prog->extflags &= ~RXf_ANCH_MBOL;
1122 /* XXXX This is a remnant of the old implementation. It
1123 looks wasteful, since now INTUIT can use many
1124 other heuristics. */
1125 prog->extflags &= ~RXf_USE_INTUIT;
1126 /* XXXX What other flags might need to be cleared in this branch? */
1132 /* Last resort... */
1133 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1134 /* trie stclasses are too expensive to use here, we are better off to
1135 leave it to regmatch itself */
1136 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1137 /* minlen == 0 is possible if regstclass is \b or \B,
1138 and the fixed substr is ''$.
1139 Since minlen is already taken into account, s+1 is before strend;
1140 accidentally, minlen >= 1 guaranties no false positives at s + 1
1141 even for \b or \B. But (minlen? 1 : 0) below assumes that
1142 regstclass does not come from lookahead... */
1143 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1144 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1145 const U8* const str = (U8*)STRING(progi->regstclass);
1146 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1147 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1150 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1151 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1152 else if (prog->float_substr || prog->float_utf8)
1153 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1157 if (checked_upto < s)
1159 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1160 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1163 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1168 const char *what = NULL;
1170 if (endpos == strend) {
1171 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1172 "Could not match STCLASS...\n") );
1175 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1176 "This position contradicts STCLASS...\n") );
1177 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1179 checked_upto = HOPBACKc(endpos, start_shift);
1180 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1181 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1182 /* Contradict one of substrings */
1183 if (prog->anchored_substr || prog->anchored_utf8) {
1184 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1185 DEBUG_EXECUTE_r( what = "anchored" );
1187 s = HOP3c(t, 1, strend);
1188 if (s + start_shift + end_shift > strend) {
1189 /* XXXX Should be taken into account earlier? */
1190 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1191 "Could not match STCLASS...\n") );
1196 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1197 "Looking for %s substr starting at offset %ld...\n",
1198 what, (long)(s + start_shift - i_strpos)) );
1201 /* Have both, check_string is floating */
1202 if (t + start_shift >= check_at) /* Contradicts floating=check */
1203 goto retry_floating_check;
1204 /* Recheck anchored substring, but not floating... */
1208 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1209 "Looking for anchored substr starting at offset %ld...\n",
1210 (long)(other_last - i_strpos)) );
1211 goto do_other_anchored;
1213 /* Another way we could have checked stclass at the
1214 current position only: */
1219 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1220 "Looking for /%s^%s/m starting at offset %ld...\n",
1221 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1224 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1226 /* Check is floating substring. */
1227 retry_floating_check:
1228 t = check_at - start_shift;
1229 DEBUG_EXECUTE_r( what = "floating" );
1230 goto hop_and_restart;
1233 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1234 "By STCLASS: moving %ld --> %ld\n",
1235 (long)(t - i_strpos), (long)(s - i_strpos))
1239 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1240 "Does not contradict STCLASS...\n");
1245 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1246 PL_colors[4], (check ? "Guessed" : "Giving up"),
1247 PL_colors[5], (long)(s - i_strpos)) );
1250 fail_finish: /* Substring not found */
1251 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1252 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1254 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1255 PL_colors[4], PL_colors[5]));
1259 #define DECL_TRIE_TYPE(scan) \
1260 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1261 trie_type = ((scan->flags == EXACT) \
1262 ? (utf8_target ? trie_utf8 : trie_plain) \
1263 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1265 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1266 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1268 switch (trie_type) { \
1269 case trie_utf8_fold: \
1270 if ( foldlen>0 ) { \
1271 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1276 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1277 len = UTF8SKIP(uc); \
1278 skiplen = UNISKIP( uvc ); \
1279 foldlen -= skiplen; \
1280 uscan = foldbuf + skiplen; \
1283 case trie_latin_utf8_fold: \
1284 if ( foldlen>0 ) { \
1285 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1291 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1292 skiplen = UNISKIP( uvc ); \
1293 foldlen -= skiplen; \
1294 uscan = foldbuf + skiplen; \
1298 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1305 charid = trie->charmap[ uvc ]; \
1309 if (widecharmap) { \
1310 SV** const svpp = hv_fetch(widecharmap, \
1311 (char*)&uvc, sizeof(UV), 0); \
1313 charid = (U16)SvIV(*svpp); \
1318 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1322 && (ln == 1 || folder(s, pat_string, ln)) \
1323 && (!reginfo || regtry(reginfo, &s)) ) \
1329 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1331 while (s < strend && s + (uskip = UTF8SKIP(s)) <= strend) { \
1337 #define REXEC_FBC_SCAN(CoDe) \
1339 while (s < strend) { \
1345 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1346 REXEC_FBC_UTF8_SCAN( \
1348 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1357 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1360 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1369 #define REXEC_FBC_TRYIT \
1370 if ((!reginfo || regtry(reginfo, &s))) \
1373 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1374 if (utf8_target) { \
1375 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1378 REXEC_FBC_CLASS_SCAN(CoNd); \
1381 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1382 if (utf8_target) { \
1384 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1387 REXEC_FBC_CLASS_SCAN(CoNd); \
1390 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1391 PL_reg_flags |= RF_tainted; \
1392 if (utf8_target) { \
1393 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1396 REXEC_FBC_CLASS_SCAN(CoNd); \
1399 #define DUMP_EXEC_POS(li,s,doutf8) \
1400 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1403 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1404 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1405 tmp = TEST_NON_UTF8(tmp); \
1406 REXEC_FBC_UTF8_SCAN( \
1407 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1416 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1417 if (s == PL_bostr) { \
1421 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1422 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1425 LOAD_UTF8_CHARCLASS_ALNUM(); \
1426 REXEC_FBC_UTF8_SCAN( \
1427 if (tmp == ! (TeSt2_UtF8)) { \
1436 /* The only difference between the BOUND and NBOUND cases is that
1437 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1438 * NBOUND. This is accomplished by passing it in either the if or else clause,
1439 * with the other one being empty */
1440 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1441 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1443 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1444 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1446 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1447 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1449 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1450 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1453 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1454 * be passed in completely with the variable name being tested, which isn't
1455 * such a clean interface, but this is easier to read than it was before. We
1456 * are looking for the boundary (or non-boundary between a word and non-word
1457 * character. The utf8 and non-utf8 cases have the same logic, but the details
1458 * must be different. Find the "wordness" of the character just prior to this
1459 * one, and compare it with the wordness of this one. If they differ, we have
1460 * a boundary. At the beginning of the string, pretend that the previous
1461 * character was a new-line */
1462 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1463 if (utf8_target) { \
1466 else { /* Not utf8 */ \
1467 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1468 tmp = TEST_NON_UTF8(tmp); \
1470 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1479 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1482 /* We know what class REx starts with. Try to find this position... */
1483 /* if reginfo is NULL, its a dryrun */
1484 /* annoyingly all the vars in this routine have different names from their counterparts
1485 in regmatch. /grrr */
1488 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1489 const char *strend, regmatch_info *reginfo)
1492 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1493 char *pat_string; /* The pattern's exactish string */
1494 char *pat_end; /* ptr to end char of pat_string */
1495 re_fold_t folder; /* Function for computing non-utf8 folds */
1496 const U8 *fold_array; /* array for folding ords < 256 */
1503 I32 tmp = 1; /* Scratch variable? */
1504 const bool utf8_target = PL_reg_match_utf8;
1505 UV utf8_fold_flags = 0;
1506 RXi_GET_DECL(prog,progi);
1508 PERL_ARGS_ASSERT_FIND_BYCLASS;
1510 /* We know what class it must start with. */
1514 REXEC_FBC_UTF8_CLASS_SCAN(
1515 reginclass(prog, c, (U8*)s, utf8_target));
1518 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1523 if (tmp && (!reginfo || regtry(reginfo, &s)))
1531 if (UTF_PATTERN || utf8_target) {
1532 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1533 goto do_exactf_utf8;
1535 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1536 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1537 goto do_exactf_non_utf8; /* isn't dealt with by these */
1542 /* regcomp.c already folded this if pattern is in UTF-8 */
1543 utf8_fold_flags = 0;
1544 goto do_exactf_utf8;
1546 fold_array = PL_fold;
1548 goto do_exactf_non_utf8;
1551 if (UTF_PATTERN || utf8_target) {
1552 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1553 goto do_exactf_utf8;
1555 fold_array = PL_fold_locale;
1556 folder = foldEQ_locale;
1557 goto do_exactf_non_utf8;
1561 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1563 goto do_exactf_utf8;
1565 case EXACTFU_TRICKYFOLD:
1567 if (UTF_PATTERN || utf8_target) {
1568 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1569 goto do_exactf_utf8;
1572 /* Any 'ss' in the pattern should have been replaced by regcomp,
1573 * so we don't have to worry here about this single special case
1574 * in the Latin1 range */
1575 fold_array = PL_fold_latin1;
1576 folder = foldEQ_latin1;
1580 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1581 are no glitches with fold-length differences
1582 between the target string and pattern */
1584 /* The idea in the non-utf8 EXACTF* cases is to first find the
1585 * first character of the EXACTF* node and then, if necessary,
1586 * case-insensitively compare the full text of the node. c1 is the
1587 * first character. c2 is its fold. This logic will not work for
1588 * Unicode semantics and the german sharp ss, which hence should
1589 * not be compiled into a node that gets here. */
1590 pat_string = STRING(c);
1591 ln = STR_LEN(c); /* length to match in octets/bytes */
1593 /* We know that we have to match at least 'ln' bytes (which is the
1594 * same as characters, since not utf8). If we have to match 3
1595 * characters, and there are only 2 availabe, we know without
1596 * trying that it will fail; so don't start a match past the
1597 * required minimum number from the far end */
1598 e = HOP3c(strend, -((I32)ln), s);
1600 if (!reginfo && e < s) {
1601 e = s; /* Due to minlen logic of intuit() */
1605 c2 = fold_array[c1];
1606 if (c1 == c2) { /* If char and fold are the same */
1607 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1610 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1618 /* If one of the operands is in utf8, we can't use the simpler folding
1619 * above, due to the fact that many different characters can have the
1620 * same fold, or portion of a fold, or different- length fold */
1621 pat_string = STRING(c);
1622 ln = STR_LEN(c); /* length to match in octets/bytes */
1623 pat_end = pat_string + ln;
1624 lnc = (UTF_PATTERN) /* length to match in characters */
1625 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1628 /* We have 'lnc' characters to match in the pattern, but because of
1629 * multi-character folding, each character in the target can match
1630 * up to 3 characters (Unicode guarantees it will never exceed
1631 * this) if it is utf8-encoded; and up to 2 if not (based on the
1632 * fact that the Latin 1 folds are already determined, and the
1633 * only multi-char fold in that range is the sharp-s folding to
1634 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1635 * string character. Adjust lnc accordingly, rounding up, so that
1636 * if we need to match at least 4+1/3 chars, that really is 5. */
1637 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1638 lnc = (lnc + expansion - 1) / expansion;
1640 /* As in the non-UTF8 case, if we have to match 3 characters, and
1641 * only 2 are left, it's guaranteed to fail, so don't start a
1642 * match that would require us to go beyond the end of the string
1644 e = HOP3c(strend, -((I32)lnc), s);
1646 if (!reginfo && e < s) {
1647 e = s; /* Due to minlen logic of intuit() */
1650 /* XXX Note that we could recalculate e to stop the loop earlier,
1651 * as the worst case expansion above will rarely be met, and as we
1652 * go along we would usually find that e moves further to the left.
1653 * This would happen only after we reached the point in the loop
1654 * where if there were no expansion we should fail. Unclear if
1655 * worth the expense */
1658 char *my_strend= (char *)strend;
1659 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1660 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1661 && (!reginfo || regtry(reginfo, &s)) )
1665 s += (utf8_target) ? UTF8SKIP(s) : 1;
1670 PL_reg_flags |= RF_tainted;
1671 FBC_BOUND(isALNUM_LC,
1672 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1673 isALNUM_LC_utf8((U8*)s));
1676 PL_reg_flags |= RF_tainted;
1677 FBC_NBOUND(isALNUM_LC,
1678 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1679 isALNUM_LC_utf8((U8*)s));
1682 FBC_BOUND(isWORDCHAR,
1684 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1687 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1689 isWORDCHAR_A((U8*)s));
1692 FBC_NBOUND(isWORDCHAR,
1694 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1697 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1699 isWORDCHAR_A((U8*)s));
1702 FBC_BOUND(isWORDCHAR_L1,
1704 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1707 FBC_NBOUND(isWORDCHAR_L1,
1709 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1712 REXEC_FBC_CSCAN_TAINT(
1713 isALNUM_LC_utf8((U8*)s),
1718 REXEC_FBC_CSCAN_PRELOAD(
1719 LOAD_UTF8_CHARCLASS_ALNUM(),
1720 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1721 isWORDCHAR_L1((U8) *s)
1725 REXEC_FBC_CSCAN_PRELOAD(
1726 LOAD_UTF8_CHARCLASS_ALNUM(),
1727 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1732 /* Don't need to worry about utf8, as it can match only a single
1733 * byte invariant character */
1734 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1737 REXEC_FBC_CSCAN_PRELOAD(
1738 LOAD_UTF8_CHARCLASS_ALNUM(),
1739 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1740 ! isWORDCHAR_L1((U8) *s)
1744 REXEC_FBC_CSCAN_PRELOAD(
1745 LOAD_UTF8_CHARCLASS_ALNUM(),
1746 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1757 REXEC_FBC_CSCAN_TAINT(
1758 !isALNUM_LC_utf8((U8*)s),
1764 is_XPERLSPACE_utf8(s),
1770 is_XPERLSPACE_utf8(s),
1775 /* Don't need to worry about utf8, as it can match only a single
1776 * byte invariant character */
1777 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1780 REXEC_FBC_CSCAN_TAINT(
1781 isSPACE_LC_utf8((U8*)s),
1787 ! is_XPERLSPACE_utf8(s),
1788 ! isSPACE_L1((U8) *s)
1793 ! is_XPERLSPACE_utf8(s),
1804 REXEC_FBC_CSCAN_TAINT(
1805 !isSPACE_LC_utf8((U8*)s),
1810 REXEC_FBC_CSCAN_PRELOAD(
1811 LOAD_UTF8_CHARCLASS_DIGIT(),
1812 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1817 /* Don't need to worry about utf8, as it can match only a single
1818 * byte invariant character */
1819 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1822 REXEC_FBC_CSCAN_TAINT(
1823 isDIGIT_LC_utf8((U8*)s),
1828 REXEC_FBC_CSCAN_PRELOAD(
1829 LOAD_UTF8_CHARCLASS_DIGIT(),
1830 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1841 REXEC_FBC_CSCAN_TAINT(
1842 !isDIGIT_LC_utf8((U8*)s),
1847 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1848 is_LNBREAK_latin1_safe(s, strend)
1853 is_VERTWS_utf8_safe(s, strend),
1854 is_VERTWS_latin1_safe(s, strend)
1859 !is_VERTWS_utf8_safe(s, strend),
1860 !is_VERTWS_latin1_safe(s, strend)
1865 is_HORIZWS_utf8_safe(s, strend),
1866 is_HORIZWS_latin1_safe(s, strend)
1871 !is_HORIZWS_utf8_safe(s, strend),
1872 !is_HORIZWS_latin1_safe(s, strend)
1876 /* Don't need to worry about utf8, as it can match only a single
1877 * byte invariant character. The flag in this node type is the
1878 * class number to pass to _generic_isCC() to build a mask for
1879 * searching in PL_charclass[] */
1880 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1884 !_generic_isCC_A(*s, FLAGS(c)),
1885 !_generic_isCC_A(*s, FLAGS(c))
1893 /* what trie are we using right now */
1894 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1895 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
1896 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1898 const char *last_start = strend - trie->minlen;
1900 const char *real_start = s;
1902 STRLEN maxlen = trie->maxlen;
1904 U8 **points; /* map of where we were in the input string
1905 when reading a given char. For ASCII this
1906 is unnecessary overhead as the relationship
1907 is always 1:1, but for Unicode, especially
1908 case folded Unicode this is not true. */
1909 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1913 GET_RE_DEBUG_FLAGS_DECL;
1915 /* We can't just allocate points here. We need to wrap it in
1916 * an SV so it gets freed properly if there is a croak while
1917 * running the match */
1920 sv_points=newSV(maxlen * sizeof(U8 *));
1921 SvCUR_set(sv_points,
1922 maxlen * sizeof(U8 *));
1923 SvPOK_on(sv_points);
1924 sv_2mortal(sv_points);
1925 points=(U8**)SvPV_nolen(sv_points );
1926 if ( trie_type != trie_utf8_fold
1927 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1930 bitmap=(U8*)trie->bitmap;
1932 bitmap=(U8*)ANYOF_BITMAP(c);
1934 /* this is the Aho-Corasick algorithm modified a touch
1935 to include special handling for long "unknown char" sequences.
1936 The basic idea being that we use AC as long as we are dealing
1937 with a possible matching char, when we encounter an unknown char
1938 (and we have not encountered an accepting state) we scan forward
1939 until we find a legal starting char.
1940 AC matching is basically that of trie matching, except that when
1941 we encounter a failing transition, we fall back to the current
1942 states "fail state", and try the current char again, a process
1943 we repeat until we reach the root state, state 1, or a legal
1944 transition. If we fail on the root state then we can either
1945 terminate if we have reached an accepting state previously, or
1946 restart the entire process from the beginning if we have not.
1949 while (s <= last_start) {
1950 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1958 U8 *uscan = (U8*)NULL;
1959 U8 *leftmost = NULL;
1961 U32 accepted_word= 0;
1965 while ( state && uc <= (U8*)strend ) {
1967 U32 word = aho->states[ state ].wordnum;
1971 DEBUG_TRIE_EXECUTE_r(
1972 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1973 dump_exec_pos( (char *)uc, c, strend, real_start,
1974 (char *)uc, utf8_target );
1975 PerlIO_printf( Perl_debug_log,
1976 " Scanning for legal start char...\n");
1980 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1984 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1990 if (uc >(U8*)last_start) break;
1994 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1995 if (!leftmost || lpos < leftmost) {
1996 DEBUG_r(accepted_word=word);
2002 points[pointpos++ % maxlen]= uc;
2003 if (foldlen || uc < (U8*)strend) {
2004 REXEC_TRIE_READ_CHAR(trie_type, trie,
2006 uscan, len, uvc, charid, foldlen,
2008 DEBUG_TRIE_EXECUTE_r({
2009 dump_exec_pos( (char *)uc, c, strend,
2010 real_start, s, utf8_target);
2011 PerlIO_printf(Perl_debug_log,
2012 " Charid:%3u CP:%4"UVxf" ",
2024 word = aho->states[ state ].wordnum;
2026 base = aho->states[ state ].trans.base;
2028 DEBUG_TRIE_EXECUTE_r({
2030 dump_exec_pos( (char *)uc, c, strend, real_start,
2032 PerlIO_printf( Perl_debug_log,
2033 "%sState: %4"UVxf", word=%"UVxf,
2034 failed ? " Fail transition to " : "",
2035 (UV)state, (UV)word);
2041 ( ((offset = base + charid
2042 - 1 - trie->uniquecharcount)) >= 0)
2043 && ((U32)offset < trie->lasttrans)
2044 && trie->trans[offset].check == state
2045 && (tmp=trie->trans[offset].next))
2047 DEBUG_TRIE_EXECUTE_r(
2048 PerlIO_printf( Perl_debug_log," - legal\n"));
2053 DEBUG_TRIE_EXECUTE_r(
2054 PerlIO_printf( Perl_debug_log," - fail\n"));
2056 state = aho->fail[state];
2060 /* we must be accepting here */
2061 DEBUG_TRIE_EXECUTE_r(
2062 PerlIO_printf( Perl_debug_log," - accepting\n"));
2071 if (!state) state = 1;
2074 if ( aho->states[ state ].wordnum ) {
2075 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2076 if (!leftmost || lpos < leftmost) {
2077 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2082 s = (char*)leftmost;
2083 DEBUG_TRIE_EXECUTE_r({
2085 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2086 (UV)accepted_word, (IV)(s - real_start)
2089 if (!reginfo || regtry(reginfo, &s)) {
2095 DEBUG_TRIE_EXECUTE_r({
2096 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2099 DEBUG_TRIE_EXECUTE_r(
2100 PerlIO_printf( Perl_debug_log,"No match.\n"));
2109 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2119 - regexec_flags - match a regexp against a string
2122 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2123 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2124 /* stringarg: the point in the string at which to begin matching */
2125 /* strend: pointer to null at end of string */
2126 /* strbeg: real beginning of string */
2127 /* minend: end of match must be >= minend bytes after stringarg. */
2128 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2129 * itself is accessed via the pointers above */
2130 /* data: May be used for some additional optimizations.
2131 Currently its only used, with a U32 cast, for transmitting
2132 the ganch offset when doing a /g match. This will change */
2133 /* nosave: For optimizations. */
2137 struct regexp *const prog = ReANY(rx);
2140 char *startpos = stringarg;
2141 I32 minlen; /* must match at least this many chars */
2142 I32 dontbother = 0; /* how many characters not to try at end */
2143 I32 end_shift = 0; /* Same for the end. */ /* CC */
2144 I32 scream_pos = -1; /* Internal iterator of scream. */
2145 char *scream_olds = NULL;
2146 const bool utf8_target = cBOOL(DO_UTF8(sv));
2148 RXi_GET_DECL(prog,progi);
2149 regmatch_info reginfo; /* create some info to pass to regtry etc */
2150 regexp_paren_pair *swap = NULL;
2151 GET_RE_DEBUG_FLAGS_DECL;
2153 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2154 PERL_UNUSED_ARG(data);
2156 /* Be paranoid... */
2157 if (prog == NULL || startpos == NULL) {
2158 Perl_croak(aTHX_ "NULL regexp parameter");
2162 multiline = prog->extflags & RXf_PMf_MULTILINE;
2163 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2165 RX_MATCH_UTF8_set(rx, utf8_target);
2167 debug_start_match(rx, utf8_target, startpos, strend,
2171 minlen = prog->minlen;
2173 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2174 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2175 "String too short [regexec_flags]...\n"));
2180 /* Check validity of program. */
2181 if (UCHARAT(progi->program) != REG_MAGIC) {
2182 Perl_croak(aTHX_ "corrupted regexp program");
2186 PL_reg_state.re_state_eval_setup_done = FALSE;
2190 PL_reg_flags |= RF_utf8;
2192 /* Mark beginning of line for ^ and lookbehind. */
2193 reginfo.bol = startpos; /* XXX not used ??? */
2197 /* Mark end of line for $ (and such) */
2200 /* see how far we have to get to not match where we matched before */
2201 reginfo.till = startpos+minend;
2203 /* If there is a "must appear" string, look for it. */
2206 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2208 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2209 reginfo.ganch = startpos + prog->gofs;
2210 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2211 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2212 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2214 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2215 && mg->mg_len >= 0) {
2216 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2217 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2218 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2220 if (prog->extflags & RXf_ANCH_GPOS) {
2221 if (s > reginfo.ganch)
2223 s = reginfo.ganch - prog->gofs;
2224 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2225 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2231 reginfo.ganch = strbeg + PTR2UV(data);
2232 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2233 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2235 } else { /* pos() not defined */
2236 reginfo.ganch = strbeg;
2237 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2238 "GPOS: reginfo.ganch = strbeg\n"));
2241 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2242 /* We have to be careful. If the previous successful match
2243 was from this regex we don't want a subsequent partially
2244 successful match to clobber the old results.
2245 So when we detect this possibility we add a swap buffer
2246 to the re, and switch the buffer each match. If we fail
2247 we switch it back, otherwise we leave it swapped.
2250 /* do we need a save destructor here for eval dies? */
2251 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2252 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2253 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2259 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2260 re_scream_pos_data d;
2262 d.scream_olds = &scream_olds;
2263 d.scream_pos = &scream_pos;
2264 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2266 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2267 goto phooey; /* not present */
2273 /* Simplest case: anchored match need be tried only once. */
2274 /* [unless only anchor is BOL and multiline is set] */
2275 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2276 if (s == startpos && regtry(®info, &startpos))
2278 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2279 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2284 dontbother = minlen - 1;
2285 end = HOP3c(strend, -dontbother, strbeg) - 1;
2286 /* for multiline we only have to try after newlines */
2287 if (prog->check_substr || prog->check_utf8) {
2288 /* because of the goto we can not easily reuse the macros for bifurcating the
2289 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2292 goto after_try_utf8;
2294 if (regtry(®info, &s)) {
2301 if (prog->extflags & RXf_USE_INTUIT) {
2302 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2311 } /* end search for check string in unicode */
2313 if (s == startpos) {
2314 goto after_try_latin;
2317 if (regtry(®info, &s)) {
2324 if (prog->extflags & RXf_USE_INTUIT) {
2325 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2334 } /* end search for check string in latin*/
2335 } /* end search for check string */
2336 else { /* search for newline */
2338 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2341 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2342 while (s <= end) { /* note it could be possible to match at the end of the string */
2343 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2344 if (regtry(®info, &s))
2348 } /* end search for newline */
2349 } /* end anchored/multiline check string search */
2351 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2353 /* the warning about reginfo.ganch being used without initialization
2354 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2355 and we only enter this block when the same bit is set. */
2356 char *tmp_s = reginfo.ganch - prog->gofs;
2358 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2363 /* Messy cases: unanchored match. */
2364 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2365 /* we have /x+whatever/ */
2366 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2372 if (! prog->anchored_utf8) {
2373 to_utf8_substr(prog);
2375 ch = SvPVX_const(prog->anchored_utf8)[0];
2378 DEBUG_EXECUTE_r( did_match = 1 );
2379 if (regtry(®info, &s)) goto got_it;
2381 while (s < strend && *s == ch)
2388 if (! prog->anchored_substr) {
2389 if (! to_byte_substr(prog)) {
2390 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2393 ch = SvPVX_const(prog->anchored_substr)[0];
2396 DEBUG_EXECUTE_r( did_match = 1 );
2397 if (regtry(®info, &s)) goto got_it;
2399 while (s < strend && *s == ch)
2404 DEBUG_EXECUTE_r(if (!did_match)
2405 PerlIO_printf(Perl_debug_log,
2406 "Did not find anchored character...\n")
2409 else if (prog->anchored_substr != NULL
2410 || prog->anchored_utf8 != NULL
2411 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2412 && prog->float_max_offset < strend - s)) {
2417 char *last1; /* Last position checked before */
2421 if (prog->anchored_substr || prog->anchored_utf8) {
2423 if (! prog->anchored_utf8) {
2424 to_utf8_substr(prog);
2426 must = prog->anchored_utf8;
2429 if (! prog->anchored_substr) {
2430 if (! to_byte_substr(prog)) {
2431 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2434 must = prog->anchored_substr;
2436 back_max = back_min = prog->anchored_offset;
2439 if (! prog->float_utf8) {
2440 to_utf8_substr(prog);
2442 must = prog->float_utf8;
2445 if (! prog->float_substr) {
2446 if (! to_byte_substr(prog)) {
2447 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2450 must = prog->float_substr;
2452 back_max = prog->float_max_offset;
2453 back_min = prog->float_min_offset;
2459 last = HOP3c(strend, /* Cannot start after this */
2460 -(I32)(CHR_SVLEN(must)
2461 - (SvTAIL(must) != 0) + back_min), strbeg);
2464 last1 = HOPc(s, -1);
2466 last1 = s - 1; /* bogus */
2468 /* XXXX check_substr already used to find "s", can optimize if
2469 check_substr==must. */
2471 dontbother = end_shift;
2472 strend = HOPc(strend, -dontbother);
2473 while ( (s <= last) &&
2474 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2475 (unsigned char*)strend, must,
2476 multiline ? FBMrf_MULTILINE : 0)) ) {
2477 DEBUG_EXECUTE_r( did_match = 1 );
2478 if (HOPc(s, -back_max) > last1) {
2479 last1 = HOPc(s, -back_min);
2480 s = HOPc(s, -back_max);
2483 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2485 last1 = HOPc(s, -back_min);
2489 while (s <= last1) {
2490 if (regtry(®info, &s))
2493 s++; /* to break out of outer loop */
2500 while (s <= last1) {
2501 if (regtry(®info, &s))
2507 DEBUG_EXECUTE_r(if (!did_match) {
2508 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2509 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2510 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2511 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2512 ? "anchored" : "floating"),
2513 quoted, RE_SV_TAIL(must));
2517 else if ( (c = progi->regstclass) ) {
2519 const OPCODE op = OP(progi->regstclass);
2520 /* don't bother with what can't match */
2521 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2522 strend = HOPc(strend, -(minlen - 1));
2525 SV * const prop = sv_newmortal();
2526 regprop(prog, prop, c);
2528 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2530 PerlIO_printf(Perl_debug_log,
2531 "Matching stclass %.*s against %s (%d bytes)\n",
2532 (int)SvCUR(prop), SvPVX_const(prop),
2533 quoted, (int)(strend - s));
2536 if (find_byclass(prog, c, s, strend, ®info))
2538 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2542 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2550 if (! prog->float_utf8) {
2551 to_utf8_substr(prog);
2553 float_real = prog->float_utf8;
2556 if (! prog->float_substr) {
2557 if (! to_byte_substr(prog)) {
2558 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2561 float_real = prog->float_substr;
2564 little = SvPV_const(float_real, len);
2565 if (SvTAIL(float_real)) {
2566 /* This means that float_real contains an artificial \n on
2567 * the end due to the presence of something like this:
2568 * /foo$/ where we can match both "foo" and "foo\n" at the
2569 * end of the string. So we have to compare the end of the
2570 * string first against the float_real without the \n and
2571 * then against the full float_real with the string. We
2572 * have to watch out for cases where the string might be
2573 * smaller than the float_real or the float_real without
2575 char *checkpos= strend - len;
2577 PerlIO_printf(Perl_debug_log,
2578 "%sChecking for float_real.%s\n",
2579 PL_colors[4], PL_colors[5]));
2580 if (checkpos + 1 < strbeg) {
2581 /* can't match, even if we remove the trailing \n
2582 * string is too short to match */
2584 PerlIO_printf(Perl_debug_log,
2585 "%sString shorter than required trailing substring, cannot match.%s\n",
2586 PL_colors[4], PL_colors[5]));
2588 } else if (memEQ(checkpos + 1, little, len - 1)) {
2589 /* can match, the end of the string matches without the
2591 last = checkpos + 1;
2592 } else if (checkpos < strbeg) {
2593 /* cant match, string is too short when the "\n" is
2596 PerlIO_printf(Perl_debug_log,
2597 "%sString does not contain required trailing substring, cannot match.%s\n",
2598 PL_colors[4], PL_colors[5]));
2600 } else if (!multiline) {
2601 /* non multiline match, so compare with the "\n" at the
2602 * end of the string */
2603 if (memEQ(checkpos, little, len)) {
2607 PerlIO_printf(Perl_debug_log,
2608 "%sString does not contain required trailing substring, cannot match.%s\n",
2609 PL_colors[4], PL_colors[5]));
2613 /* multiline match, so we have to search for a place
2614 * where the full string is located */
2620 last = rninstr(s, strend, little, little + len);
2622 last = strend; /* matching "$" */
2625 /* at one point this block contained a comment which was
2626 * probably incorrect, which said that this was a "should not
2627 * happen" case. Even if it was true when it was written I am
2628 * pretty sure it is not anymore, so I have removed the comment
2629 * and replaced it with this one. Yves */
2631 PerlIO_printf(Perl_debug_log,
2632 "String does not contain required substring, cannot match.\n"
2636 dontbother = strend - last + prog->float_min_offset;
2638 if (minlen && (dontbother < minlen))
2639 dontbother = minlen - 1;
2640 strend -= dontbother; /* this one's always in bytes! */
2641 /* We don't know much -- general case. */
2644 if (regtry(®info, &s))
2653 if (regtry(®info, &s))
2655 } while (s++ < strend);
2665 PerlIO_printf(Perl_debug_log,
2666 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2672 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2674 if (PL_reg_state.re_state_eval_setup_done)
2675 restore_pos(aTHX_ prog);
2676 if (RXp_PAREN_NAMES(prog))
2677 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2679 /* make sure $`, $&, $', and $digit will work later */
2680 if ( !(flags & REXEC_NOT_FIRST) ) {
2681 if (flags & REXEC_COPY_STR) {
2685 PerlIO_printf(Perl_debug_log,
2686 "Copy on write: regexp capture, type %d\n",
2689 RX_MATCH_COPY_FREE(rx);
2690 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2691 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2692 assert (SvPOKp(prog->saved_copy));
2693 prog->sublen = PL_regeol - strbeg;
2694 prog->suboffset = 0;
2695 prog->subcoffset = 0;
2700 I32 max = PL_regeol - strbeg;
2703 if ( (flags & REXEC_COPY_SKIP_POST)
2704 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2705 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2706 ) { /* don't copy $' part of string */
2709 /* calculate the right-most part of the string covered
2710 * by a capture. Due to look-ahead, this may be to
2711 * the right of $&, so we have to scan all captures */
2712 while (n <= prog->lastparen) {
2713 if (prog->offs[n].end > max)
2714 max = prog->offs[n].end;
2718 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2719 ? prog->offs[0].start
2721 assert(max >= 0 && max <= PL_regeol - strbeg);
2724 if ( (flags & REXEC_COPY_SKIP_PRE)
2725 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2726 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2727 ) { /* don't copy $` part of string */
2730 /* calculate the left-most part of the string covered
2731 * by a capture. Due to look-behind, this may be to
2732 * the left of $&, so we have to scan all captures */
2733 while (min && n <= prog->lastparen) {
2734 if ( prog->offs[n].start != -1
2735 && prog->offs[n].start < min)
2737 min = prog->offs[n].start;
2741 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2742 && min > prog->offs[0].end
2744 min = prog->offs[0].end;
2748 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2751 if (RX_MATCH_COPIED(rx)) {
2752 if (sublen > prog->sublen)
2754 (char*)saferealloc(prog->subbeg, sublen+1);
2757 prog->subbeg = (char*)safemalloc(sublen+1);
2758 Copy(strbeg + min, prog->subbeg, sublen, char);
2759 prog->subbeg[sublen] = '\0';
2760 prog->suboffset = min;
2761 prog->sublen = sublen;
2762 RX_MATCH_COPIED_on(rx);
2764 prog->subcoffset = prog->suboffset;
2765 if (prog->suboffset && utf8_target) {
2766 /* Convert byte offset to chars.
2767 * XXX ideally should only compute this if @-/@+
2768 * has been seen, a la PL_sawampersand ??? */
2770 /* If there's a direct correspondence between the
2771 * string which we're matching and the original SV,
2772 * then we can use the utf8 len cache associated with
2773 * the SV. In particular, it means that under //g,
2774 * sv_pos_b2u() will use the previously cached
2775 * position to speed up working out the new length of
2776 * subcoffset, rather than counting from the start of
2777 * the string each time. This stops
2778 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2779 * from going quadratic */
2780 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2781 sv_pos_b2u(sv, &(prog->subcoffset));
2783 prog->subcoffset = utf8_length((U8*)strbeg,
2784 (U8*)(strbeg+prog->suboffset));
2788 RX_MATCH_COPY_FREE(rx);
2789 prog->subbeg = strbeg;
2790 prog->suboffset = 0;
2791 prog->subcoffset = 0;
2792 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2799 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2800 PL_colors[4], PL_colors[5]));
2801 if (PL_reg_state.re_state_eval_setup_done)
2802 restore_pos(aTHX_ prog);
2804 /* we failed :-( roll it back */
2805 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2806 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2811 Safefree(prog->offs);
2818 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2819 * Do inc before dec, in case old and new rex are the same */
2820 #define SET_reg_curpm(Re2) \
2821 if (PL_reg_state.re_state_eval_setup_done) { \
2822 (void)ReREFCNT_inc(Re2); \
2823 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2824 PM_SETRE((PL_reg_curpm), (Re2)); \
2829 - regtry - try match at specific point
2831 STATIC I32 /* 0 failure, 1 success */
2832 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2836 REGEXP *const rx = reginfo->prog;
2837 regexp *const prog = ReANY(rx);
2839 RXi_GET_DECL(prog,progi);
2840 GET_RE_DEBUG_FLAGS_DECL;
2842 PERL_ARGS_ASSERT_REGTRY;
2844 reginfo->cutpoint=NULL;
2846 if ((prog->extflags & RXf_EVAL_SEEN)
2847 && !PL_reg_state.re_state_eval_setup_done)
2851 PL_reg_state.re_state_eval_setup_done = TRUE;
2853 /* Make $_ available to executed code. */
2854 if (reginfo->sv != DEFSV) {
2856 DEFSV_set(reginfo->sv);
2859 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2860 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2861 /* prepare for quick setting of pos */
2862 #ifdef PERL_OLD_COPY_ON_WRITE
2863 if (SvIsCOW(reginfo->sv))
2864 sv_force_normal_flags(reginfo->sv, 0);
2866 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2867 &PL_vtbl_mglob, NULL, 0);
2871 PL_reg_oldpos = mg->mg_len;
2872 SAVEDESTRUCTOR_X(restore_pos, prog);
2874 if (!PL_reg_curpm) {
2875 Newxz(PL_reg_curpm, 1, PMOP);
2878 SV* const repointer = &PL_sv_undef;
2879 /* this regexp is also owned by the new PL_reg_curpm, which
2880 will try to free it. */
2881 av_push(PL_regex_padav, repointer);
2882 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2883 PL_regex_pad = AvARRAY(PL_regex_padav);
2888 PL_reg_oldcurpm = PL_curpm;
2889 PL_curpm = PL_reg_curpm;
2890 if (RXp_MATCH_COPIED(prog)) {
2891 /* Here is a serious problem: we cannot rewrite subbeg,
2892 since it may be needed if this match fails. Thus
2893 $` inside (?{}) could fail... */
2894 PL_reg_oldsaved = prog->subbeg;
2895 PL_reg_oldsavedlen = prog->sublen;
2896 PL_reg_oldsavedoffset = prog->suboffset;
2897 PL_reg_oldsavedcoffset = prog->suboffset;
2899 PL_nrs = prog->saved_copy;
2901 RXp_MATCH_COPIED_off(prog);
2904 PL_reg_oldsaved = NULL;
2905 prog->subbeg = PL_bostr;
2906 prog->suboffset = 0;
2907 prog->subcoffset = 0;
2908 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2911 PL_reg_starttry = *startposp;
2913 prog->offs[0].start = *startposp - PL_bostr;
2914 prog->lastparen = 0;
2915 prog->lastcloseparen = 0;
2917 /* XXXX What this code is doing here?!!! There should be no need
2918 to do this again and again, prog->lastparen should take care of
2921 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2922 * Actually, the code in regcppop() (which Ilya may be meaning by
2923 * prog->lastparen), is not needed at all by the test suite
2924 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2925 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2926 * Meanwhile, this code *is* needed for the
2927 * above-mentioned test suite tests to succeed. The common theme
2928 * on those tests seems to be returning null fields from matches.
2929 * --jhi updated by dapm */
2931 if (prog->nparens) {
2932 regexp_paren_pair *pp = prog->offs;
2934 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2942 result = regmatch(reginfo, *startposp, progi->program + 1);
2944 prog->offs[0].end = result;
2947 if (reginfo->cutpoint)
2948 *startposp= reginfo->cutpoint;
2949 REGCP_UNWIND(lastcp);
2954 #define sayYES goto yes
2955 #define sayNO goto no
2956 #define sayNO_SILENT goto no_silent
2958 /* we dont use STMT_START/END here because it leads to
2959 "unreachable code" warnings, which are bogus, but distracting. */
2960 #define CACHEsayNO \
2961 if (ST.cache_mask) \
2962 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2965 /* this is used to determine how far from the left messages like
2966 'failed...' are printed. It should be set such that messages
2967 are inline with the regop output that created them.
2969 #define REPORT_CODE_OFF 32
2972 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2973 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2974 #define CHRTEST_NOT_A_CP_1 -999
2975 #define CHRTEST_NOT_A_CP_2 -998
2977 #define SLAB_FIRST(s) (&(s)->states[0])
2978 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2980 /* grab a new slab and return the first slot in it */
2982 STATIC regmatch_state *
2985 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2988 regmatch_slab *s = PL_regmatch_slab->next;
2990 Newx(s, 1, regmatch_slab);
2991 s->prev = PL_regmatch_slab;
2993 PL_regmatch_slab->next = s;
2995 PL_regmatch_slab = s;
2996 return SLAB_FIRST(s);
3000 /* push a new state then goto it */
3002 #define PUSH_STATE_GOTO(state, node, input) \
3003 pushinput = input; \
3005 st->resume_state = state; \
3008 /* push a new state with success backtracking, then goto it */
3010 #define PUSH_YES_STATE_GOTO(state, node, input) \
3011 pushinput = input; \
3013 st->resume_state = state; \
3014 goto push_yes_state;
3021 regmatch() - main matching routine
3023 This is basically one big switch statement in a loop. We execute an op,
3024 set 'next' to point the next op, and continue. If we come to a point which
3025 we may need to backtrack to on failure such as (A|B|C), we push a
3026 backtrack state onto the backtrack stack. On failure, we pop the top
3027 state, and re-enter the loop at the state indicated. If there are no more
3028 states to pop, we return failure.
3030 Sometimes we also need to backtrack on success; for example /A+/, where
3031 after successfully matching one A, we need to go back and try to
3032 match another one; similarly for lookahead assertions: if the assertion
3033 completes successfully, we backtrack to the state just before the assertion
3034 and then carry on. In these cases, the pushed state is marked as
3035 'backtrack on success too'. This marking is in fact done by a chain of
3036 pointers, each pointing to the previous 'yes' state. On success, we pop to
3037 the nearest yes state, discarding any intermediate failure-only states.
3038 Sometimes a yes state is pushed just to force some cleanup code to be
3039 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3040 it to free the inner regex.
3042 Note that failure backtracking rewinds the cursor position, while
3043 success backtracking leaves it alone.
3045 A pattern is complete when the END op is executed, while a subpattern
3046 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3047 ops trigger the "pop to last yes state if any, otherwise return true"
3050 A common convention in this function is to use A and B to refer to the two
3051 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3052 the subpattern to be matched possibly multiple times, while B is the entire
3053 rest of the pattern. Variable and state names reflect this convention.
3055 The states in the main switch are the union of ops and failure/success of
3056 substates associated with with that op. For example, IFMATCH is the op
3057 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3058 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3059 successfully matched A and IFMATCH_A_fail is a state saying that we have
3060 just failed to match A. Resume states always come in pairs. The backtrack
3061 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3062 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3063 on success or failure.
3065 The struct that holds a backtracking state is actually a big union, with
3066 one variant for each major type of op. The variable st points to the
3067 top-most backtrack struct. To make the code clearer, within each
3068 block of code we #define ST to alias the relevant union.
3070 Here's a concrete example of a (vastly oversimplified) IFMATCH
3076 #define ST st->u.ifmatch
3078 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3079 ST.foo = ...; // some state we wish to save
3081 // push a yes backtrack state with a resume value of
3082 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3084 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3087 case IFMATCH_A: // we have successfully executed A; now continue with B
3089 bar = ST.foo; // do something with the preserved value
3092 case IFMATCH_A_fail: // A failed, so the assertion failed
3093 ...; // do some housekeeping, then ...
3094 sayNO; // propagate the failure
3101 For any old-timers reading this who are familiar with the old recursive
3102 approach, the code above is equivalent to:
3104 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3113 ...; // do some housekeeping, then ...
3114 sayNO; // propagate the failure
3117 The topmost backtrack state, pointed to by st, is usually free. If you
3118 want to claim it, populate any ST.foo fields in it with values you wish to
3119 save, then do one of
3121 PUSH_STATE_GOTO(resume_state, node, newinput);
3122 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3124 which sets that backtrack state's resume value to 'resume_state', pushes a
3125 new free entry to the top of the backtrack stack, then goes to 'node'.
3126 On backtracking, the free slot is popped, and the saved state becomes the
3127 new free state. An ST.foo field in this new top state can be temporarily
3128 accessed to retrieve values, but once the main loop is re-entered, it
3129 becomes available for reuse.
3131 Note that the depth of the backtrack stack constantly increases during the
3132 left-to-right execution of the pattern, rather than going up and down with
3133 the pattern nesting. For example the stack is at its maximum at Z at the
3134 end of the pattern, rather than at X in the following:
3136 /(((X)+)+)+....(Y)+....Z/
3138 The only exceptions to this are lookahead/behind assertions and the cut,
3139 (?>A), which pop all the backtrack states associated with A before
3142 Backtrack state structs are allocated in slabs of about 4K in size.
3143 PL_regmatch_state and st always point to the currently active state,
3144 and PL_regmatch_slab points to the slab currently containing
3145 PL_regmatch_state. The first time regmatch() is called, the first slab is
3146 allocated, and is never freed until interpreter destruction. When the slab
3147 is full, a new one is allocated and chained to the end. At exit from
3148 regmatch(), slabs allocated since entry are freed.
3153 #define DEBUG_STATE_pp(pp) \
3155 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3156 PerlIO_printf(Perl_debug_log, \
3157 " %*s"pp" %s%s%s%s%s\n", \
3159 PL_reg_name[st->resume_state], \
3160 ((st==yes_state||st==mark_state) ? "[" : ""), \
3161 ((st==yes_state) ? "Y" : ""), \
3162 ((st==mark_state) ? "M" : ""), \
3163 ((st==yes_state||st==mark_state) ? "]" : "") \
3168 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3173 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3174 const char *start, const char *end, const char *blurb)
3176 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3178 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3183 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3184 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3186 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3187 start, end - start, 60);
3189 PerlIO_printf(Perl_debug_log,
3190 "%s%s REx%s %s against %s\n",
3191 PL_colors[4], blurb, PL_colors[5], s0, s1);
3193 if (utf8_target||utf8_pat)
3194 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3195 utf8_pat ? "pattern" : "",
3196 utf8_pat && utf8_target ? " and " : "",
3197 utf8_target ? "string" : ""
3203 S_dump_exec_pos(pTHX_ const char *locinput,
3204 const regnode *scan,
3205 const char *loc_regeol,
3206 const char *loc_bostr,
3207 const char *loc_reg_starttry,
3208 const bool utf8_target)
3210 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3211 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3212 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3213 /* The part of the string before starttry has one color
3214 (pref0_len chars), between starttry and current
3215 position another one (pref_len - pref0_len chars),
3216 after the current position the third one.
3217 We assume that pref0_len <= pref_len, otherwise we
3218 decrease pref0_len. */
3219 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3220 ? (5 + taill) - l : locinput - loc_bostr;
3223 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3225 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3227 pref0_len = pref_len - (locinput - loc_reg_starttry);
3228 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3229 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3230 ? (5 + taill) - pref_len : loc_regeol - locinput);
3231 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3235 if (pref0_len > pref_len)
3236 pref0_len = pref_len;
3238 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3240 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3241 (locinput - pref_len),pref0_len, 60, 4, 5);
3243 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3244 (locinput - pref_len + pref0_len),
3245 pref_len - pref0_len, 60, 2, 3);
3247 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3248 locinput, loc_regeol - locinput, 10, 0, 1);
3250 const STRLEN tlen=len0+len1+len2;
3251 PerlIO_printf(Perl_debug_log,
3252 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3253 (IV)(locinput - loc_bostr),
3256 (docolor ? "" : "> <"),
3258 (int)(tlen > 19 ? 0 : 19 - tlen),
3265 /* reg_check_named_buff_matched()
3266 * Checks to see if a named buffer has matched. The data array of
3267 * buffer numbers corresponding to the buffer is expected to reside
3268 * in the regexp->data->data array in the slot stored in the ARG() of
3269 * node involved. Note that this routine doesn't actually care about the
3270 * name, that information is not preserved from compilation to execution.
3271 * Returns the index of the leftmost defined buffer with the given name
3272 * or 0 if non of the buffers matched.
3275 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3278 RXi_GET_DECL(rex,rexi);
3279 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3280 I32 *nums=(I32*)SvPVX(sv_dat);
3282 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3284 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3285 if ((I32)rex->lastparen >= nums[n] &&
3286 rex->offs[nums[n]].end != -1)
3295 /* free all slabs above current one - called during LEAVE_SCOPE */
3298 S_clear_backtrack_stack(pTHX_ void *p)
3300 regmatch_slab *s = PL_regmatch_slab->next;
3305 PL_regmatch_slab->next = NULL;
3307 regmatch_slab * const osl = s;
3313 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, U8* c1_utf8, int *c2p, U8* c2_utf8)
3315 /* This function determines if there are one or two characters that match
3316 * the first character of the passed-in EXACTish node <text_node>, and if
3317 * so, returns them in the passed-in pointers.
3319 * If it determines that no possible character in the target string can
3320 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3321 * the first character in <text_node> requires UTF-8 to represent, and the
3322 * target string isn't in UTF-8.)
3324 * If there are more than two characters that could match the beginning of
3325 * <text_node>, or if more context is required to determine a match or not,
3326 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3328 * The motiviation behind this function is to allow the caller to set up
3329 * tight loops for matching. If <text_node> is of type EXACT, there is
3330 * only one possible character that can match its first character, and so
3331 * the situation is quite simple. But things get much more complicated if
3332 * folding is involved. It may be that the first character of an EXACTFish
3333 * node doesn't participate in any possible fold, e.g., punctuation, so it
3334 * can be matched only by itself. The vast majority of characters that are
3335 * in folds match just two things, their lower and upper-case equivalents.
3336 * But not all are like that; some have multiple possible matches, or match
3337 * sequences of more than one character. This function sorts all that out.
3339 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3340 * loop of trying to match A*, we know we can't exit where the thing
3341 * following it isn't a B. And something can't be a B unless it is the
3342 * beginning of B. By putting a quick test for that beginning in a tight
3343 * loop, we can rule out things that can't possibly be B without having to
3344 * break out of the loop, thus avoiding work. Similarly, if A is a single
3345 * character, we can make a tight loop matching A*, using the outputs of
3348 * If the target string to match isn't in UTF-8, and there aren't
3349 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3350 * the one or two possible octets (which are characters in this situation)
3351 * that can match. In all cases, if there is only one character that can
3352 * match, *<c1p> and *<c2p> will be identical.
3354 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3355 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3356 * can match the beginning of <text_node>. They should be declared with at
3357 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3358 * undefined what these contain.) If one or both of the buffers are
3359 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3360 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3361 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3362 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3363 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3365 const bool utf8_target = PL_reg_match_utf8;
3367 UV c1 = CHRTEST_NOT_A_CP_1;
3368 UV c2 = CHRTEST_NOT_A_CP_2;
3369 bool use_chrtest_void = FALSE;
3371 /* Used when we have both utf8 input and utf8 output, to avoid converting
3372 * to/from code points */
3373 bool utf8_has_been_setup = FALSE;
3377 U8 *pat = (U8*)STRING(text_node);
3379 if (OP(text_node) == EXACT) {
3381 /* In an exact node, only one thing can be matched, that first
3382 * character. If both the pat and the target are UTF-8, we can just
3383 * copy the input to the output, avoiding finding the code point of
3385 if (! UTF_PATTERN) {
3388 else if (utf8_target) {
3389 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3390 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3391 utf8_has_been_setup = TRUE;
3394 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3397 else /* an EXACTFish node */
3399 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3400 pat + STR_LEN(text_node)))
3402 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3403 pat + STR_LEN(text_node))))
3405 /* Multi-character folds require more context to sort out. Also
3406 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3407 * handled outside this routine */
3408 use_chrtest_void = TRUE;
3410 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3411 c1 = (UTF_PATTERN) ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3413 /* Load the folds hash, if not already done */
3415 if (! PL_utf8_foldclosures) {
3416 if (! PL_utf8_tofold) {
3417 U8 dummy[UTF8_MAXBYTES+1];
3419 /* Force loading this by folding an above-Latin1 char */
3420 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3421 assert(PL_utf8_tofold); /* Verify that worked */
3423 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3426 /* The fold closures data structure is a hash with the keys being
3427 * the UTF-8 of every character that is folded to, like 'k', and
3428 * the values each an array of all code points that fold to its
3429 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3431 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3436 /* Not found in the hash, therefore there are no folds
3437 * containing it, so there is only a single character that
3441 else { /* Does participate in folds */
3442 AV* list = (AV*) *listp;
3443 if (av_len(list) != 1) {
3445 /* If there aren't exactly two folds to this, it is outside
3446 * the scope of this function */
3447 use_chrtest_void = TRUE;
3449 else { /* There are two. Get them */
3450 SV** c_p = av_fetch(list, 0, FALSE);
3452 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3456 c_p = av_fetch(list, 1, FALSE);
3458 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3462 /* Folds that cross the 255/256 boundary are forbidden if
3463 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3464 * pattern character is above 256, and its only other match
3465 * is below 256, the only legal match will be to itself.
3466 * We have thrown away the original, so have to compute
3467 * which is the one above 255 */
3468 if ((c1 < 256) != (c2 < 256)) {
3469 if (OP(text_node) == EXACTFL
3470 || (OP(text_node) == EXACTFA
3471 && (isASCII(c1) || isASCII(c2))))
3484 else /* Here, c1 is < 255 */
3486 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3487 && OP(text_node) != EXACTFL
3488 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3490 /* Here, there could be something above Latin1 in the target which
3491 * folds to this character in the pattern. All such cases except
3492 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3493 * involved in their folds, so are outside the scope of this
3495 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3496 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3499 use_chrtest_void = TRUE;
3502 else { /* Here nothing above Latin1 can fold to the pattern character */
3503 switch (OP(text_node)) {
3505 case EXACTFL: /* /l rules */
3506 c2 = PL_fold_locale[c1];
3510 if (! utf8_target) { /* /d rules */
3515 /* /u rules for all these. This happens to work for
3516 * EXACTFA as nothing in Latin1 folds to ASCII */
3518 case EXACTFU_TRICKYFOLD:
3521 c2 = PL_fold_latin1[c1];
3525 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3526 assert(0); /* NOTREACHED */
3531 /* Here have figured things out. Set up the returns */
3532 if (use_chrtest_void) {
3533 *c2p = *c1p = CHRTEST_VOID;
3535 else if (utf8_target) {
3536 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3537 uvchr_to_utf8(c1_utf8, c1);
3538 uvchr_to_utf8(c2_utf8, c2);
3541 /* Invariants are stored in both the utf8 and byte outputs; Use
3542 * negative numbers otherwise for the byte ones. Make sure that the
3543 * byte ones are the same iff the utf8 ones are the same */
3544 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3545 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3548 ? CHRTEST_NOT_A_CP_1
3549 : CHRTEST_NOT_A_CP_2;
3551 else if (c1 > 255) {
3552 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3557 *c1p = *c2p = c2; /* c2 is the only representable value */
3559 else { /* c1 is representable; see about c2 */
3561 *c2p = (c2 < 256) ? c2 : c1;
3567 /* returns -1 on failure, $+[0] on success */
3569 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3571 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3575 const bool utf8_target = PL_reg_match_utf8;
3576 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3577 REGEXP *rex_sv = reginfo->prog;
3578 regexp *rex = ReANY(rex_sv);
3579 RXi_GET_DECL(rex,rexi);
3581 /* the current state. This is a cached copy of PL_regmatch_state */
3583 /* cache heavy used fields of st in registers */
3586 U32 n = 0; /* general value; init to avoid compiler warning */
3587 I32 ln = 0; /* len or last; init to avoid compiler warning */
3588 char *locinput = startpos;
3589 char *pushinput; /* where to continue after a PUSH */
3590 I32 nextchr; /* is always set to UCHARAT(locinput) */
3592 bool result = 0; /* return value of S_regmatch */
3593 int depth = 0; /* depth of backtrack stack */
3594 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3595 const U32 max_nochange_depth =
3596 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3597 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3598 regmatch_state *yes_state = NULL; /* state to pop to on success of
3600 /* mark_state piggy backs on the yes_state logic so that when we unwind
3601 the stack on success we can update the mark_state as we go */
3602 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3603 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3604 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3606 bool no_final = 0; /* prevent failure from backtracking? */
3607 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3608 char *startpoint = locinput;
3609 SV *popmark = NULL; /* are we looking for a mark? */
3610 SV *sv_commit = NULL; /* last mark name seen in failure */
3611 SV *sv_yes_mark = NULL; /* last mark name we have seen
3612 during a successful match */
3613 U32 lastopen = 0; /* last open we saw */
3614 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3615 SV* const oreplsv = GvSV(PL_replgv);
3616 /* these three flags are set by various ops to signal information to
3617 * the very next op. They have a useful lifetime of exactly one loop
3618 * iteration, and are not preserved or restored by state pushes/pops
3620 bool sw = 0; /* the condition value in (?(cond)a|b) */
3621 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3622 int logical = 0; /* the following EVAL is:
3626 or the following IFMATCH/UNLESSM is:
3627 false: plain (?=foo)
3628 true: used as a condition: (?(?=foo))
3630 PAD* last_pad = NULL;
3632 I32 gimme = G_SCALAR;
3633 CV *caller_cv = NULL; /* who called us */
3634 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3635 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3636 U32 maxopenparen = 0; /* max '(' index seen so far */
3639 GET_RE_DEBUG_FLAGS_DECL;
3642 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3643 multicall_oldcatch = 0;
3644 multicall_cv = NULL;
3646 PERL_UNUSED_VAR(multicall_cop);
3647 PERL_UNUSED_VAR(newsp);
3650 PERL_ARGS_ASSERT_REGMATCH;
3652 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3653 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3655 /* on first ever call to regmatch, allocate first slab */
3656 if (!PL_regmatch_slab) {
3657 Newx(PL_regmatch_slab, 1, regmatch_slab);
3658 PL_regmatch_slab->prev = NULL;
3659 PL_regmatch_slab->next = NULL;
3660 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3663 oldsave = PL_savestack_ix;
3664 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3665 SAVEVPTR(PL_regmatch_slab);
3666 SAVEVPTR(PL_regmatch_state);
3668 /* grab next free state slot */
3669 st = ++PL_regmatch_state;
3670 if (st > SLAB_LAST(PL_regmatch_slab))
3671 st = PL_regmatch_state = S_push_slab(aTHX);
3673 /* Note that nextchr is a byte even in UTF */
3676 while (scan != NULL) {
3679 SV * const prop = sv_newmortal();
3680 regnode *rnext=regnext(scan);
3681 DUMP_EXEC_POS( locinput, scan, utf8_target );
3682 regprop(rex, prop, scan);
3684 PerlIO_printf(Perl_debug_log,
3685 "%3"IVdf":%*s%s(%"IVdf")\n",
3686 (IV)(scan - rexi->program), depth*2, "",
3688 (PL_regkind[OP(scan)] == END || !rnext) ?
3689 0 : (IV)(rnext - rexi->program));
3692 next = scan + NEXT_OFF(scan);
3695 state_num = OP(scan);
3700 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3702 switch (state_num) {
3703 case BOL: /* /^../ */
3704 if (locinput == PL_bostr)
3706 /* reginfo->till = reginfo->bol; */
3711 case MBOL: /* /^../m */
3712 if (locinput == PL_bostr ||
3713 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3719 case SBOL: /* /^../s */
3720 if (locinput == PL_bostr)
3725 if (locinput == reginfo->ganch)
3729 case KEEPS: /* \K */
3730 /* update the startpoint */
3731 st->u.keeper.val = rex->offs[0].start;
3732 rex->offs[0].start = locinput - PL_bostr;
3733 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3734 assert(0); /*NOTREACHED*/
3735 case KEEPS_next_fail:
3736 /* rollback the start point change */
3737 rex->offs[0].start = st->u.keeper.val;
3739 assert(0); /*NOTREACHED*/
3741 case EOL: /* /..$/ */
3744 case MEOL: /* /..$/m */
3745 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3749 case SEOL: /* /..$/s */
3751 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3753 if (PL_regeol - locinput > 1)
3758 if (!NEXTCHR_IS_EOS)
3762 case SANY: /* /./s */
3765 goto increment_locinput;
3773 case REG_ANY: /* /./ */
3774 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3776 goto increment_locinput;
3780 #define ST st->u.trie
3781 case TRIEC: /* (ab|cd) with known charclass */
3782 /* In this case the charclass data is available inline so
3783 we can fail fast without a lot of extra overhead.
3785 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3787 PerlIO_printf(Perl_debug_log,
3788 "%*s %sfailed to match trie start class...%s\n",
3789 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3792 assert(0); /* NOTREACHED */
3795 case TRIE: /* (ab|cd) */
3796 /* the basic plan of execution of the trie is:
3797 * At the beginning, run though all the states, and
3798 * find the longest-matching word. Also remember the position
3799 * of the shortest matching word. For example, this pattern:
3802 * when matched against the string "abcde", will generate
3803 * accept states for all words except 3, with the longest
3804 * matching word being 4, and the shortest being 2 (with
3805 * the position being after char 1 of the string).
3807 * Then for each matching word, in word order (i.e. 1,2,4,5),
3808 * we run the remainder of the pattern; on each try setting
3809 * the current position to the character following the word,
3810 * returning to try the next word on failure.
3812 * We avoid having to build a list of words at runtime by
3813 * using a compile-time structure, wordinfo[].prev, which
3814 * gives, for each word, the previous accepting word (if any).
3815 * In the case above it would contain the mappings 1->2, 2->0,
3816 * 3->0, 4->5, 5->1. We can use this table to generate, from
3817 * the longest word (4 above), a list of all words, by
3818 * following the list of prev pointers; this gives us the
3819 * unordered list 4,5,1,2. Then given the current word we have
3820 * just tried, we can go through the list and find the
3821 * next-biggest word to try (so if we just failed on word 2,
3822 * the next in the list is 4).
3824 * Since at runtime we don't record the matching position in
3825 * the string for each word, we have to work that out for
3826 * each word we're about to process. The wordinfo table holds
3827 * the character length of each word; given that we recorded
3828 * at the start: the position of the shortest word and its
3829 * length in chars, we just need to move the pointer the
3830 * difference between the two char lengths. Depending on
3831 * Unicode status and folding, that's cheap or expensive.
3833 * This algorithm is optimised for the case where are only a
3834 * small number of accept states, i.e. 0,1, or maybe 2.
3835 * With lots of accepts states, and having to try all of them,
3836 * it becomes quadratic on number of accept states to find all
3841 /* what type of TRIE am I? (utf8 makes this contextual) */
3842 DECL_TRIE_TYPE(scan);
3844 /* what trie are we using right now */
3845 reg_trie_data * const trie
3846 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3847 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3848 U32 state = trie->startstate;
3851 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3853 if (trie->states[ state ].wordnum) {
3855 PerlIO_printf(Perl_debug_log,
3856 "%*s %smatched empty string...%s\n",
3857 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3863 PerlIO_printf(Perl_debug_log,
3864 "%*s %sfailed to match trie start class...%s\n",
3865 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3872 U8 *uc = ( U8* )locinput;
3876 U8 *uscan = (U8*)NULL;
3877 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3878 U32 charcount = 0; /* how many input chars we have matched */
3879 U32 accepted = 0; /* have we seen any accepting states? */
3881 ST.jump = trie->jump;
3884 ST.longfold = FALSE; /* char longer if folded => it's harder */
3887 /* fully traverse the TRIE; note the position of the
3888 shortest accept state and the wordnum of the longest
3891 while ( state && uc <= (U8*)PL_regeol ) {
3892 U32 base = trie->states[ state ].trans.base;
3896 wordnum = trie->states[ state ].wordnum;
3898 if (wordnum) { /* it's an accept state */
3901 /* record first match position */
3903 ST.firstpos = (U8*)locinput;
3908 ST.firstchars = charcount;
3911 if (!ST.nextword || wordnum < ST.nextword)
3912 ST.nextword = wordnum;
3913 ST.topword = wordnum;
3916 DEBUG_TRIE_EXECUTE_r({
3917 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3918 PerlIO_printf( Perl_debug_log,
3919 "%*s %sState: %4"UVxf" Accepted: %c ",
3920 2+depth * 2, "", PL_colors[4],
3921 (UV)state, (accepted ? 'Y' : 'N'));
3924 /* read a char and goto next state */
3925 if ( base && (foldlen || uc < (U8*)PL_regeol)) {
3927 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,