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)
352 const int retval = PL_savestack_ix;
353 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
354 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
355 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
357 GET_RE_DEBUG_FLAGS_DECL;
359 PERL_ARGS_ASSERT_REGCPPUSH;
361 if (paren_elems_to_push < 0)
362 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
363 paren_elems_to_push);
365 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
366 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
367 " out of range (%lu-%ld)",
368 total_elems, (unsigned long)PL_regsize, (long)parenfloor);
370 SSGROW(total_elems + REGCP_FRAME_ELEMS);
373 if ((int)PL_regsize > (int)parenfloor)
374 PerlIO_printf(Perl_debug_log,
375 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
380 for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
381 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
382 SSPUSHINT(rex->offs[p].end);
383 SSPUSHINT(rex->offs[p].start);
384 SSPUSHINT(rex->offs[p].start_tmp);
385 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
386 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
388 (IV)rex->offs[p].start,
389 (IV)rex->offs[p].start_tmp,
393 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
394 SSPUSHINT(PL_regsize);
395 SSPUSHINT(rex->lastparen);
396 SSPUSHINT(rex->lastcloseparen);
397 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
402 /* These are needed since we do not localize EVAL nodes: */
403 #define REGCP_SET(cp) \
405 PerlIO_printf(Perl_debug_log, \
406 " Setting an EVAL scope, savestack=%"IVdf"\n", \
407 (IV)PL_savestack_ix)); \
410 #define REGCP_UNWIND(cp) \
412 if (cp != PL_savestack_ix) \
413 PerlIO_printf(Perl_debug_log, \
414 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
415 (IV)(cp), (IV)PL_savestack_ix)); \
418 #define UNWIND_PAREN(lp, lcp) \
419 for (n = rex->lastparen; n > lp; n--) \
420 rex->offs[n].end = -1; \
421 rex->lastparen = n; \
422 rex->lastcloseparen = lcp;
426 S_regcppop(pTHX_ regexp *rex)
431 GET_RE_DEBUG_FLAGS_DECL;
433 PERL_ARGS_ASSERT_REGCPPOP;
435 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
437 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
438 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
439 rex->lastcloseparen = SSPOPINT;
440 rex->lastparen = SSPOPINT;
441 PL_regsize = SSPOPINT;
443 i -= REGCP_OTHER_ELEMS;
444 /* Now restore the parentheses context. */
446 if (i || rex->lastparen + 1 <= rex->nparens)
447 PerlIO_printf(Perl_debug_log,
448 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
454 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
456 rex->offs[paren].start_tmp = SSPOPINT;
457 rex->offs[paren].start = SSPOPINT;
459 if (paren <= rex->lastparen)
460 rex->offs[paren].end = tmps;
461 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
462 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
464 (IV)rex->offs[paren].start,
465 (IV)rex->offs[paren].start_tmp,
466 (IV)rex->offs[paren].end,
467 (paren > rex->lastparen ? "(skipped)" : ""));
472 /* It would seem that the similar code in regtry()
473 * already takes care of this, and in fact it is in
474 * a better location to since this code can #if 0-ed out
475 * but the code in regtry() is needed or otherwise tests
476 * requiring null fields (pat.t#187 and split.t#{13,14}
477 * (as of patchlevel 7877) will fail. Then again,
478 * this code seems to be necessary or otherwise
479 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
480 * --jhi updated by dapm */
481 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
483 rex->offs[i].start = -1;
484 rex->offs[i].end = -1;
485 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
486 " \\%"UVuf": %s ..-1 undeffing\n",
488 (i > PL_regsize) ? "-1" : " "
494 /* restore the parens and associated vars at savestack position ix,
495 * but without popping the stack */
498 S_regcp_restore(pTHX_ regexp *rex, I32 ix)
500 I32 tmpix = PL_savestack_ix;
501 PL_savestack_ix = ix;
503 PL_savestack_ix = tmpix;
506 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
509 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
511 /* Returns a boolean as to whether or not 'character' is a member of the
512 * Posix character class given by 'classnum' that should be equivalent to a
513 * value in the typedef '_char_class_number'.
515 * Ideally this could be replaced by a just an array of function pointers
516 * to the C library functions that implement the macros this calls.
517 * However, to compile, the precise function signatures are required, and
518 * these may vary from platform to to platform. To avoid having to figure
519 * out what those all are on each platform, I (khw) am using this method,
520 * which adds an extra layer of function call overhead. But we don't
521 * particularly care about performance with locales anyway. */
523 switch ((_char_class_number) classnum) {
524 case _CC_ENUM_ALNUMC: return isALNUMC_LC(character);
525 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
526 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
527 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
528 case _CC_ENUM_LOWER: return isLOWER_LC(character);
529 case _CC_ENUM_PRINT: return isPRINT_LC(character);
530 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
531 case _CC_ENUM_UPPER: return isUPPER_LC(character);
532 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
533 case _CC_ENUM_SPACE: return isSPACE_LC(character);
534 case _CC_ENUM_BLANK: return isBLANK_LC(character);
535 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
536 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
537 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
538 case _CC_ENUM_ASCII: return isASCII_LC(character);
539 default: /* VERTSPACE should never occur in locales */
540 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
543 assert(0); /* NOTREACHED */
548 * pregexec and friends
551 #ifndef PERL_IN_XSUB_RE
553 - pregexec - match a regexp against a string
556 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
557 char *strbeg, I32 minend, SV *screamer, U32 nosave)
558 /* stringarg: the point in the string at which to begin matching */
559 /* strend: pointer to null at end of string */
560 /* strbeg: real beginning of string */
561 /* minend: end of match must be >= minend bytes after stringarg. */
562 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
563 * itself is accessed via the pointers above */
564 /* nosave: For optimizations. */
566 PERL_ARGS_ASSERT_PREGEXEC;
569 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
570 nosave ? 0 : REXEC_COPY_STR);
575 * Need to implement the following flags for reg_anch:
577 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
579 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
580 * INTUIT_AUTORITATIVE_ML
581 * INTUIT_ONCE_NOML - Intuit can match in one location only.
584 * Another flag for this function: SECOND_TIME (so that float substrs
585 * with giant delta may be not rechecked).
588 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
590 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
591 Otherwise, only SvCUR(sv) is used to get strbeg. */
593 /* XXXX We assume that strpos is strbeg unless sv. */
595 /* XXXX Some places assume that there is a fixed substring.
596 An update may be needed if optimizer marks as "INTUITable"
597 RExen without fixed substrings. Similarly, it is assumed that
598 lengths of all the strings are no more than minlen, thus they
599 cannot come from lookahead.
600 (Or minlen should take into account lookahead.)
601 NOTE: Some of this comment is not correct. minlen does now take account
602 of lookahead/behind. Further research is required. -- demerphq
606 /* A failure to find a constant substring means that there is no need to make
607 an expensive call to REx engine, thus we celebrate a failure. Similarly,
608 finding a substring too deep into the string means that less calls to
609 regtry() should be needed.
611 REx compiler's optimizer found 4 possible hints:
612 a) Anchored substring;
614 c) Whether we are anchored (beginning-of-line or \G);
615 d) First node (of those at offset 0) which may distinguish positions;
616 We use a)b)d) and multiline-part of c), and try to find a position in the
617 string which does not contradict any of them.
620 /* Most of decisions we do here should have been done at compile time.
621 The nodes of the REx which we used for the search should have been
622 deleted from the finite automaton. */
625 Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
626 char *strend, const U32 flags, re_scream_pos_data *data)
629 struct regexp *const prog = ReANY(rx);
631 /* Should be nonnegative! */
637 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
639 char *other_last = NULL; /* other substr checked before this */
640 char *check_at = NULL; /* check substr found at this pos */
641 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
642 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
643 RXi_GET_DECL(prog,progi);
645 const char * const i_strpos = strpos;
647 GET_RE_DEBUG_FLAGS_DECL;
649 PERL_ARGS_ASSERT_RE_INTUIT_START;
650 PERL_UNUSED_ARG(flags);
651 PERL_UNUSED_ARG(data);
653 RX_MATCH_UTF8_set(rx,utf8_target);
656 PL_reg_flags |= RF_utf8;
659 debug_start_match(rx, utf8_target, strpos, strend,
660 sv ? "Guessing start of match in sv for"
661 : "Guessing start of match in string for");
664 /* CHR_DIST() would be more correct here but it makes things slow. */
665 if (prog->minlen > strend - strpos) {
666 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
667 "String too short... [re_intuit_start]\n"));
671 /* XXX we need to pass strbeg as a separate arg: the following is
672 * guesswork and can be wrong... */
673 if (sv && SvPOK(sv)) {
674 char * p = SvPVX(sv);
675 STRLEN cur = SvCUR(sv);
676 if (p <= strpos && strpos < p + cur) {
678 assert(p <= strend && strend <= p + cur);
681 strbeg = strend - cur;
688 if (!prog->check_utf8 && prog->check_substr)
689 to_utf8_substr(prog);
690 check = prog->check_utf8;
692 if (!prog->check_substr && prog->check_utf8) {
693 if (! to_byte_substr(prog)) {
694 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
697 check = prog->check_substr;
699 if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
700 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
701 || ( (prog->extflags & RXf_ANCH_BOL)
702 && !multiline ) ); /* Check after \n? */
705 if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
706 && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
707 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
709 && (strpos != strbeg)) {
710 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
713 if (prog->check_offset_min == prog->check_offset_max
714 && !(prog->extflags & RXf_CANY_SEEN)
715 && ! multiline) /* /m can cause \n's to match that aren't
716 accounted for in the string max length.
717 See [perl #115242] */
719 /* Substring at constant offset from beg-of-str... */
722 s = HOP3c(strpos, prog->check_offset_min, strend);
725 slen = SvCUR(check); /* >= 1 */
727 if ( strend - s > slen || strend - s < slen - 1
728 || (strend - s == slen && strend[-1] != '\n')) {
729 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
732 /* Now should match s[0..slen-2] */
734 if (slen && (*SvPVX_const(check) != *s
736 && memNE(SvPVX_const(check), s, slen)))) {
738 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
742 else if (*SvPVX_const(check) != *s
743 || ((slen = SvCUR(check)) > 1
744 && memNE(SvPVX_const(check), s, slen)))
747 goto success_at_start;
750 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
752 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
753 end_shift = prog->check_end_shift;
756 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
757 - (SvTAIL(check) != 0);
758 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
760 if (end_shift < eshift)
764 else { /* Can match at random position */
767 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
768 end_shift = prog->check_end_shift;
770 /* end shift should be non negative here */
773 #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
775 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
776 (IV)end_shift, RX_PRECOMP(prog));
780 /* Find a possible match in the region s..strend by looking for
781 the "check" substring in the region corrected by start/end_shift. */
784 I32 srch_start_shift = start_shift;
785 I32 srch_end_shift = end_shift;
788 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
789 srch_end_shift -= ((strbeg - s) - srch_start_shift);
790 srch_start_shift = strbeg - s;
792 DEBUG_OPTIMISE_MORE_r({
793 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
794 (IV)prog->check_offset_min,
795 (IV)srch_start_shift,
797 (IV)prog->check_end_shift);
800 if (prog->extflags & RXf_CANY_SEEN) {
801 start_point= (U8*)(s + srch_start_shift);
802 end_point= (U8*)(strend - srch_end_shift);
804 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
805 end_point= HOP3(strend, -srch_end_shift, strbeg);
807 DEBUG_OPTIMISE_MORE_r({
808 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
809 (int)(end_point - start_point),
810 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
814 s = fbm_instr( start_point, end_point,
815 check, multiline ? FBMrf_MULTILINE : 0);
817 /* Update the count-of-usability, remove useless subpatterns,
821 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
822 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
823 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
824 (s ? "Found" : "Did not find"),
825 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
826 ? "anchored" : "floating"),
829 (s ? " at offset " : "...\n") );
834 /* Finish the diagnostic message */
835 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
837 /* XXX dmq: first branch is for positive lookbehind...
838 Our check string is offset from the beginning of the pattern.
839 So we need to do any stclass tests offset forward from that
848 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
849 Start with the other substr.
850 XXXX no SCREAM optimization yet - and a very coarse implementation
851 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
852 *always* match. Probably should be marked during compile...
853 Probably it is right to do no SCREAM here...
856 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
857 : (prog->float_substr && prog->anchored_substr))
859 /* Take into account the "other" substring. */
860 /* XXXX May be hopelessly wrong for UTF... */
863 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
866 char * const last = HOP3c(s, -start_shift, strbeg);
868 char * const saved_s = s;
871 t = s - prog->check_offset_max;
872 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
874 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
879 t = HOP3c(t, prog->anchored_offset, strend);
880 if (t < other_last) /* These positions already checked */
882 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
885 /* XXXX It is not documented what units *_offsets are in.
886 We assume bytes, but this is clearly wrong.
887 Meaning this code needs to be carefully reviewed for errors.
891 /* On end-of-str: see comment below. */
892 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
893 if (must == &PL_sv_undef) {
895 DEBUG_r(must = prog->anchored_utf8); /* for debug */
900 HOP3(HOP3(last1, prog->anchored_offset, strend)
901 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
903 multiline ? FBMrf_MULTILINE : 0
906 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
907 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
908 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
909 (s ? "Found" : "Contradicts"),
910 quoted, RE_SV_TAIL(must));
915 if (last1 >= last2) {
916 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
917 ", giving up...\n"));
920 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
921 ", trying floating at offset %ld...\n",
922 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
923 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
924 s = HOP3c(last, 1, strend);
928 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
929 (long)(s - i_strpos)));
930 t = HOP3c(s, -prog->anchored_offset, strbeg);
931 other_last = HOP3c(s, 1, strend);
939 else { /* Take into account the floating substring. */
941 char * const saved_s = s;
944 t = HOP3c(s, -start_shift, strbeg);
946 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
947 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
948 last = HOP3c(t, prog->float_max_offset, strend);
949 s = HOP3c(t, prog->float_min_offset, strend);
952 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
953 must = utf8_target ? prog->float_utf8 : prog->float_substr;
954 /* fbm_instr() takes into account exact value of end-of-str
955 if the check is SvTAIL(ed). Since false positives are OK,
956 and end-of-str is not later than strend we are OK. */
957 if (must == &PL_sv_undef) {
959 DEBUG_r(must = prog->float_utf8); /* for debug message */
962 s = fbm_instr((unsigned char*)s,
963 (unsigned char*)last + SvCUR(must)
965 must, multiline ? FBMrf_MULTILINE : 0);
967 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
968 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
969 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
970 (s ? "Found" : "Contradicts"),
971 quoted, RE_SV_TAIL(must));
975 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
976 ", giving up...\n"));
979 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
980 ", trying anchored starting at offset %ld...\n",
981 (long)(saved_s + 1 - i_strpos)));
983 s = HOP3c(t, 1, strend);
987 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
988 (long)(s - i_strpos)));
989 other_last = s; /* Fix this later. --Hugo */
999 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
1001 DEBUG_OPTIMISE_MORE_r(
1002 PerlIO_printf(Perl_debug_log,
1003 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
1004 (IV)prog->check_offset_min,
1005 (IV)prog->check_offset_max,
1013 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
1015 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
1018 /* Fixed substring is found far enough so that the match
1019 cannot start at strpos. */
1021 if (ml_anch && t[-1] != '\n') {
1022 /* Eventually fbm_*() should handle this, but often
1023 anchored_offset is not 0, so this check will not be wasted. */
1024 /* XXXX In the code below we prefer to look for "^" even in
1025 presence of anchored substrings. And we search even
1026 beyond the found float position. These pessimizations
1027 are historical artefacts only. */
1029 while (t < strend - prog->minlen) {
1031 if (t < check_at - prog->check_offset_min) {
1032 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1033 /* Since we moved from the found position,
1034 we definitely contradict the found anchored
1035 substr. Due to the above check we do not
1036 contradict "check" substr.
1037 Thus we can arrive here only if check substr
1038 is float. Redo checking for "other"=="fixed".
1041 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1042 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
1043 goto do_other_anchored;
1045 /* We don't contradict the found floating substring. */
1046 /* XXXX Why not check for STCLASS? */
1048 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
1049 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
1052 /* Position contradicts check-string */
1053 /* XXXX probably better to look for check-string
1054 than for "\n", so one should lower the limit for t? */
1055 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1056 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
1057 other_last = strpos = s = t + 1;
1062 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
1063 PL_colors[0], PL_colors[1]));
1067 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
1068 PL_colors[0], PL_colors[1]));
1072 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1075 /* The found string does not prohibit matching at strpos,
1076 - no optimization of calling REx engine can be performed,
1077 unless it was an MBOL and we are not after MBOL,
1078 or a future STCLASS check will fail this. */
1080 /* Even in this situation we may use MBOL flag if strpos is offset
1081 wrt the start of the string. */
1082 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
1083 && (strpos != strbeg) && strpos[-1] != '\n'
1084 /* May be due to an implicit anchor of m{.*foo} */
1085 && !(prog->intflags & PREGf_IMPLICIT))
1090 DEBUG_EXECUTE_r( if (ml_anch)
1091 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
1092 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
1095 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
1097 prog->check_utf8 /* Could be deleted already */
1098 && --BmUSEFUL(prog->check_utf8) < 0
1099 && (prog->check_utf8 == prog->float_utf8)
1101 prog->check_substr /* Could be deleted already */
1102 && --BmUSEFUL(prog->check_substr) < 0
1103 && (prog->check_substr == prog->float_substr)
1106 /* If flags & SOMETHING - do not do it many times on the same match */
1107 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
1108 /* XXX Does the destruction order has to change with utf8_target? */
1109 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1110 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1111 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1112 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1113 check = NULL; /* abort */
1115 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
1116 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1117 if (prog->intflags & PREGf_IMPLICIT)
1118 prog->extflags &= ~RXf_ANCH_MBOL;
1119 /* XXXX This is a remnant of the old implementation. It
1120 looks wasteful, since now INTUIT can use many
1121 other heuristics. */
1122 prog->extflags &= ~RXf_USE_INTUIT;
1123 /* XXXX What other flags might need to be cleared in this branch? */
1129 /* Last resort... */
1130 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1131 /* trie stclasses are too expensive to use here, we are better off to
1132 leave it to regmatch itself */
1133 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1134 /* minlen == 0 is possible if regstclass is \b or \B,
1135 and the fixed substr is ''$.
1136 Since minlen is already taken into account, s+1 is before strend;
1137 accidentally, minlen >= 1 guaranties no false positives at s + 1
1138 even for \b or \B. But (minlen? 1 : 0) below assumes that
1139 regstclass does not come from lookahead... */
1140 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
1141 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
1142 const U8* const str = (U8*)STRING(progi->regstclass);
1143 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1144 ? CHR_DIST(str+STR_LEN(progi->regstclass), str)
1147 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1148 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1149 else if (prog->float_substr || prog->float_utf8)
1150 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1154 if (checked_upto < s)
1156 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1157 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1160 s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
1165 const char *what = NULL;
1167 if (endpos == strend) {
1168 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1169 "Could not match STCLASS...\n") );
1172 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1173 "This position contradicts STCLASS...\n") );
1174 if ((prog->extflags & RXf_ANCH) && !ml_anch)
1176 checked_upto = HOPBACKc(endpos, start_shift);
1177 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1178 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1179 /* Contradict one of substrings */
1180 if (prog->anchored_substr || prog->anchored_utf8) {
1181 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
1182 DEBUG_EXECUTE_r( what = "anchored" );
1184 s = HOP3c(t, 1, strend);
1185 if (s + start_shift + end_shift > strend) {
1186 /* XXXX Should be taken into account earlier? */
1187 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1188 "Could not match STCLASS...\n") );
1193 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1194 "Looking for %s substr starting at offset %ld...\n",
1195 what, (long)(s + start_shift - i_strpos)) );
1198 /* Have both, check_string is floating */
1199 if (t + start_shift >= check_at) /* Contradicts floating=check */
1200 goto retry_floating_check;
1201 /* Recheck anchored substring, but not floating... */
1205 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1206 "Looking for anchored substr starting at offset %ld...\n",
1207 (long)(other_last - i_strpos)) );
1208 goto do_other_anchored;
1210 /* Another way we could have checked stclass at the
1211 current position only: */
1216 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1217 "Looking for /%s^%s/m starting at offset %ld...\n",
1218 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
1221 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
1223 /* Check is floating substring. */
1224 retry_floating_check:
1225 t = check_at - start_shift;
1226 DEBUG_EXECUTE_r( what = "floating" );
1227 goto hop_and_restart;
1230 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1231 "By STCLASS: moving %ld --> %ld\n",
1232 (long)(t - i_strpos), (long)(s - i_strpos))
1236 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1237 "Does not contradict STCLASS...\n");
1242 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
1243 PL_colors[4], (check ? "Guessed" : "Giving up"),
1244 PL_colors[5], (long)(s - i_strpos)) );
1247 fail_finish: /* Substring not found */
1248 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1249 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1251 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1252 PL_colors[4], PL_colors[5]));
1256 #define DECL_TRIE_TYPE(scan) \
1257 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
1258 trie_type = ((scan->flags == EXACT) \
1259 ? (utf8_target ? trie_utf8 : trie_plain) \
1260 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1262 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
1263 uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
1265 switch (trie_type) { \
1266 case trie_utf8_fold: \
1267 if ( foldlen>0 ) { \
1268 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1273 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1274 len = UTF8SKIP(uc); \
1275 skiplen = UNISKIP( uvc ); \
1276 foldlen -= skiplen; \
1277 uscan = foldbuf + skiplen; \
1280 case trie_latin_utf8_fold: \
1281 if ( foldlen>0 ) { \
1282 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1288 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
1289 skiplen = UNISKIP( uvc ); \
1290 foldlen -= skiplen; \
1291 uscan = foldbuf + skiplen; \
1295 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1302 charid = trie->charmap[ uvc ]; \
1306 if (widecharmap) { \
1307 SV** const svpp = hv_fetch(widecharmap, \
1308 (char*)&uvc, sizeof(UV), 0); \
1310 charid = (U16)SvIV(*svpp); \
1315 #define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1319 && (ln == 1 || folder(s, pat_string, ln)) \
1320 && (!reginfo || regtry(reginfo, &s)) ) \
1326 #define REXEC_FBC_UTF8_SCAN(CoDe) \
1328 while (s < strend && s + (uskip = UTF8SKIP(s)) <= strend) { \
1334 #define REXEC_FBC_SCAN(CoDe) \
1336 while (s < strend) { \
1342 #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1343 REXEC_FBC_UTF8_SCAN( \
1345 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1354 #define REXEC_FBC_CLASS_SCAN(CoNd) \
1357 if (tmp && (!reginfo || regtry(reginfo, &s))) \
1366 #define REXEC_FBC_TRYIT \
1367 if ((!reginfo || regtry(reginfo, &s))) \
1370 #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
1371 if (utf8_target) { \
1372 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1375 REXEC_FBC_CLASS_SCAN(CoNd); \
1378 #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
1379 if (utf8_target) { \
1381 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1384 REXEC_FBC_CLASS_SCAN(CoNd); \
1387 #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
1388 PL_reg_flags |= RF_tainted; \
1389 if (utf8_target) { \
1390 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1393 REXEC_FBC_CLASS_SCAN(CoNd); \
1396 #define DUMP_EXEC_POS(li,s,doutf8) \
1397 dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
1400 #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1401 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1402 tmp = TEST_NON_UTF8(tmp); \
1403 REXEC_FBC_UTF8_SCAN( \
1404 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1413 #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
1414 if (s == PL_bostr) { \
1418 U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
1419 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1422 LOAD_UTF8_CHARCLASS_ALNUM(); \
1423 REXEC_FBC_UTF8_SCAN( \
1424 if (tmp == ! (TeSt2_UtF8)) { \
1433 /* The only difference between the BOUND and NBOUND cases is that
1434 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1435 * NBOUND. This is accomplished by passing it in either the if or else clause,
1436 * with the other one being empty */
1437 #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1438 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1440 #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1441 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1443 #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1444 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1446 #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1447 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1450 /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1451 * be passed in completely with the variable name being tested, which isn't
1452 * such a clean interface, but this is easier to read than it was before. We
1453 * are looking for the boundary (or non-boundary between a word and non-word
1454 * character. The utf8 and non-utf8 cases have the same logic, but the details
1455 * must be different. Find the "wordness" of the character just prior to this
1456 * one, and compare it with the wordness of this one. If they differ, we have
1457 * a boundary. At the beginning of the string, pretend that the previous
1458 * character was a new-line */
1459 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1460 if (utf8_target) { \
1463 else { /* Not utf8 */ \
1464 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
1465 tmp = TEST_NON_UTF8(tmp); \
1467 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1476 if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
1479 /* We know what class REx starts with. Try to find this position... */
1480 /* if reginfo is NULL, its a dryrun */
1481 /* annoyingly all the vars in this routine have different names from their counterparts
1482 in regmatch. /grrr */
1485 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1486 const char *strend, regmatch_info *reginfo)
1489 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1490 char *pat_string; /* The pattern's exactish string */
1491 char *pat_end; /* ptr to end char of pat_string */
1492 re_fold_t folder; /* Function for computing non-utf8 folds */
1493 const U8 *fold_array; /* array for folding ords < 256 */
1500 I32 tmp = 1; /* Scratch variable? */
1501 const bool utf8_target = PL_reg_match_utf8;
1502 UV utf8_fold_flags = 0;
1503 RXi_GET_DECL(prog,progi);
1505 PERL_ARGS_ASSERT_FIND_BYCLASS;
1507 /* We know what class it must start with. */
1511 REXEC_FBC_UTF8_CLASS_SCAN(
1512 reginclass(prog, c, (U8*)s, utf8_target));
1515 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1520 if (tmp && (!reginfo || regtry(reginfo, &s)))
1528 if (UTF_PATTERN || utf8_target) {
1529 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1530 goto do_exactf_utf8;
1532 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1533 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1534 goto do_exactf_non_utf8; /* isn't dealt with by these */
1539 /* regcomp.c already folded this if pattern is in UTF-8 */
1540 utf8_fold_flags = 0;
1541 goto do_exactf_utf8;
1543 fold_array = PL_fold;
1545 goto do_exactf_non_utf8;
1548 if (UTF_PATTERN || utf8_target) {
1549 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1550 goto do_exactf_utf8;
1552 fold_array = PL_fold_locale;
1553 folder = foldEQ_locale;
1554 goto do_exactf_non_utf8;
1558 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1560 goto do_exactf_utf8;
1562 case EXACTFU_TRICKYFOLD:
1564 if (UTF_PATTERN || utf8_target) {
1565 utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1566 goto do_exactf_utf8;
1569 /* Any 'ss' in the pattern should have been replaced by regcomp,
1570 * so we don't have to worry here about this single special case
1571 * in the Latin1 range */
1572 fold_array = PL_fold_latin1;
1573 folder = foldEQ_latin1;
1577 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1578 are no glitches with fold-length differences
1579 between the target string and pattern */
1581 /* The idea in the non-utf8 EXACTF* cases is to first find the
1582 * first character of the EXACTF* node and then, if necessary,
1583 * case-insensitively compare the full text of the node. c1 is the
1584 * first character. c2 is its fold. This logic will not work for
1585 * Unicode semantics and the german sharp ss, which hence should
1586 * not be compiled into a node that gets here. */
1587 pat_string = STRING(c);
1588 ln = STR_LEN(c); /* length to match in octets/bytes */
1590 /* We know that we have to match at least 'ln' bytes (which is the
1591 * same as characters, since not utf8). If we have to match 3
1592 * characters, and there are only 2 availabe, we know without
1593 * trying that it will fail; so don't start a match past the
1594 * required minimum number from the far end */
1595 e = HOP3c(strend, -((I32)ln), s);
1597 if (!reginfo && e < s) {
1598 e = s; /* Due to minlen logic of intuit() */
1602 c2 = fold_array[c1];
1603 if (c1 == c2) { /* If char and fold are the same */
1604 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1607 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1615 /* If one of the operands is in utf8, we can't use the simpler folding
1616 * above, due to the fact that many different characters can have the
1617 * same fold, or portion of a fold, or different- length fold */
1618 pat_string = STRING(c);
1619 ln = STR_LEN(c); /* length to match in octets/bytes */
1620 pat_end = pat_string + ln;
1621 lnc = (UTF_PATTERN) /* length to match in characters */
1622 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1625 /* We have 'lnc' characters to match in the pattern, but because of
1626 * multi-character folding, each character in the target can match
1627 * up to 3 characters (Unicode guarantees it will never exceed
1628 * this) if it is utf8-encoded; and up to 2 if not (based on the
1629 * fact that the Latin 1 folds are already determined, and the
1630 * only multi-char fold in that range is the sharp-s folding to
1631 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1632 * string character. Adjust lnc accordingly, rounding up, so that
1633 * if we need to match at least 4+1/3 chars, that really is 5. */
1634 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1635 lnc = (lnc + expansion - 1) / expansion;
1637 /* As in the non-UTF8 case, if we have to match 3 characters, and
1638 * only 2 are left, it's guaranteed to fail, so don't start a
1639 * match that would require us to go beyond the end of the string
1641 e = HOP3c(strend, -((I32)lnc), s);
1643 if (!reginfo && e < s) {
1644 e = s; /* Due to minlen logic of intuit() */
1647 /* XXX Note that we could recalculate e to stop the loop earlier,
1648 * as the worst case expansion above will rarely be met, and as we
1649 * go along we would usually find that e moves further to the left.
1650 * This would happen only after we reached the point in the loop
1651 * where if there were no expansion we should fail. Unclear if
1652 * worth the expense */
1655 char *my_strend= (char *)strend;
1656 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1657 pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
1658 && (!reginfo || regtry(reginfo, &s)) )
1662 s += (utf8_target) ? UTF8SKIP(s) : 1;
1667 PL_reg_flags |= RF_tainted;
1668 FBC_BOUND(isALNUM_LC,
1669 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1670 isALNUM_LC_utf8((U8*)s));
1673 PL_reg_flags |= RF_tainted;
1674 FBC_NBOUND(isALNUM_LC,
1675 isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
1676 isALNUM_LC_utf8((U8*)s));
1679 FBC_BOUND(isWORDCHAR,
1681 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1684 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1686 isWORDCHAR_A((U8*)s));
1689 FBC_NBOUND(isWORDCHAR,
1691 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1694 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1696 isWORDCHAR_A((U8*)s));
1699 FBC_BOUND(isWORDCHAR_L1,
1701 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1704 FBC_NBOUND(isWORDCHAR_L1,
1706 cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
1709 REXEC_FBC_CSCAN_TAINT(
1710 isALNUM_LC_utf8((U8*)s),
1715 REXEC_FBC_CSCAN_PRELOAD(
1716 LOAD_UTF8_CHARCLASS_ALNUM(),
1717 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1718 isWORDCHAR_L1((U8) *s)
1722 REXEC_FBC_CSCAN_PRELOAD(
1723 LOAD_UTF8_CHARCLASS_ALNUM(),
1724 swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1729 /* Don't need to worry about utf8, as it can match only a single
1730 * byte invariant character */
1731 REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
1734 REXEC_FBC_CSCAN_PRELOAD(
1735 LOAD_UTF8_CHARCLASS_ALNUM(),
1736 !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
1737 ! isWORDCHAR_L1((U8) *s)
1741 REXEC_FBC_CSCAN_PRELOAD(
1742 LOAD_UTF8_CHARCLASS_ALNUM(),
1743 !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
1754 REXEC_FBC_CSCAN_TAINT(
1755 !isALNUM_LC_utf8((U8*)s),
1761 is_XPERLSPACE_utf8(s),
1767 is_XPERLSPACE_utf8(s),
1772 /* Don't need to worry about utf8, as it can match only a single
1773 * byte invariant character */
1774 REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
1777 REXEC_FBC_CSCAN_TAINT(
1778 isSPACE_LC_utf8((U8*)s),
1784 ! is_XPERLSPACE_utf8(s),
1785 ! isSPACE_L1((U8) *s)
1790 ! is_XPERLSPACE_utf8(s),
1801 REXEC_FBC_CSCAN_TAINT(
1802 !isSPACE_LC_utf8((U8*)s),
1807 REXEC_FBC_CSCAN_PRELOAD(
1808 LOAD_UTF8_CHARCLASS_DIGIT(),
1809 swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1814 /* Don't need to worry about utf8, as it can match only a single
1815 * byte invariant character */
1816 REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
1819 REXEC_FBC_CSCAN_TAINT(
1820 isDIGIT_LC_utf8((U8*)s),
1825 REXEC_FBC_CSCAN_PRELOAD(
1826 LOAD_UTF8_CHARCLASS_DIGIT(),
1827 !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
1838 REXEC_FBC_CSCAN_TAINT(
1839 !isDIGIT_LC_utf8((U8*)s),
1844 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1845 is_LNBREAK_latin1_safe(s, strend)
1850 is_VERTWS_utf8_safe(s, strend),
1851 is_VERTWS_latin1_safe(s, strend)
1856 !is_VERTWS_utf8_safe(s, strend),
1857 !is_VERTWS_latin1_safe(s, strend)
1862 is_HORIZWS_utf8_safe(s, strend),
1863 is_HORIZWS_latin1_safe(s, strend)
1868 !is_HORIZWS_utf8_safe(s, strend),
1869 !is_HORIZWS_latin1_safe(s, strend)
1873 /* Don't need to worry about utf8, as it can match only a single
1874 * byte invariant character. The flag in this node type is the
1875 * class number to pass to _generic_isCC() to build a mask for
1876 * searching in PL_charclass[] */
1877 REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
1881 !_generic_isCC_A(*s, FLAGS(c)),
1882 !_generic_isCC_A(*s, FLAGS(c))
1890 /* what trie are we using right now */
1891 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1892 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
1893 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1895 const char *last_start = strend - trie->minlen;
1897 const char *real_start = s;
1899 STRLEN maxlen = trie->maxlen;
1901 U8 **points; /* map of where we were in the input string
1902 when reading a given char. For ASCII this
1903 is unnecessary overhead as the relationship
1904 is always 1:1, but for Unicode, especially
1905 case folded Unicode this is not true. */
1906 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1910 GET_RE_DEBUG_FLAGS_DECL;
1912 /* We can't just allocate points here. We need to wrap it in
1913 * an SV so it gets freed properly if there is a croak while
1914 * running the match */
1917 sv_points=newSV(maxlen * sizeof(U8 *));
1918 SvCUR_set(sv_points,
1919 maxlen * sizeof(U8 *));
1920 SvPOK_on(sv_points);
1921 sv_2mortal(sv_points);
1922 points=(U8**)SvPV_nolen(sv_points );
1923 if ( trie_type != trie_utf8_fold
1924 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1927 bitmap=(U8*)trie->bitmap;
1929 bitmap=(U8*)ANYOF_BITMAP(c);
1931 /* this is the Aho-Corasick algorithm modified a touch
1932 to include special handling for long "unknown char" sequences.
1933 The basic idea being that we use AC as long as we are dealing
1934 with a possible matching char, when we encounter an unknown char
1935 (and we have not encountered an accepting state) we scan forward
1936 until we find a legal starting char.
1937 AC matching is basically that of trie matching, except that when
1938 we encounter a failing transition, we fall back to the current
1939 states "fail state", and try the current char again, a process
1940 we repeat until we reach the root state, state 1, or a legal
1941 transition. If we fail on the root state then we can either
1942 terminate if we have reached an accepting state previously, or
1943 restart the entire process from the beginning if we have not.
1946 while (s <= last_start) {
1947 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1955 U8 *uscan = (U8*)NULL;
1956 U8 *leftmost = NULL;
1958 U32 accepted_word= 0;
1962 while ( state && uc <= (U8*)strend ) {
1964 U32 word = aho->states[ state ].wordnum;
1968 DEBUG_TRIE_EXECUTE_r(
1969 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1970 dump_exec_pos( (char *)uc, c, strend, real_start,
1971 (char *)uc, utf8_target );
1972 PerlIO_printf( Perl_debug_log,
1973 " Scanning for legal start char...\n");
1977 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1981 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1987 if (uc >(U8*)last_start) break;
1991 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1992 if (!leftmost || lpos < leftmost) {
1993 DEBUG_r(accepted_word=word);
1999 points[pointpos++ % maxlen]= uc;
2000 if (foldlen || uc < (U8*)strend) {
2001 REXEC_TRIE_READ_CHAR(trie_type, trie,
2003 uscan, len, uvc, charid, foldlen,
2005 DEBUG_TRIE_EXECUTE_r({
2006 dump_exec_pos( (char *)uc, c, strend,
2007 real_start, s, utf8_target);
2008 PerlIO_printf(Perl_debug_log,
2009 " Charid:%3u CP:%4"UVxf" ",
2021 word = aho->states[ state ].wordnum;
2023 base = aho->states[ state ].trans.base;
2025 DEBUG_TRIE_EXECUTE_r({
2027 dump_exec_pos( (char *)uc, c, strend, real_start,
2029 PerlIO_printf( Perl_debug_log,
2030 "%sState: %4"UVxf", word=%"UVxf,
2031 failed ? " Fail transition to " : "",
2032 (UV)state, (UV)word);
2038 ( ((offset = base + charid
2039 - 1 - trie->uniquecharcount)) >= 0)
2040 && ((U32)offset < trie->lasttrans)
2041 && trie->trans[offset].check == state
2042 && (tmp=trie->trans[offset].next))
2044 DEBUG_TRIE_EXECUTE_r(
2045 PerlIO_printf( Perl_debug_log," - legal\n"));
2050 DEBUG_TRIE_EXECUTE_r(
2051 PerlIO_printf( Perl_debug_log," - fail\n"));
2053 state = aho->fail[state];
2057 /* we must be accepting here */
2058 DEBUG_TRIE_EXECUTE_r(
2059 PerlIO_printf( Perl_debug_log," - accepting\n"));
2068 if (!state) state = 1;
2071 if ( aho->states[ state ].wordnum ) {
2072 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2073 if (!leftmost || lpos < leftmost) {
2074 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2079 s = (char*)leftmost;
2080 DEBUG_TRIE_EXECUTE_r({
2082 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2083 (UV)accepted_word, (IV)(s - real_start)
2086 if (!reginfo || regtry(reginfo, &s)) {
2092 DEBUG_TRIE_EXECUTE_r({
2093 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2096 DEBUG_TRIE_EXECUTE_r(
2097 PerlIO_printf( Perl_debug_log,"No match.\n"));
2106 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2116 - regexec_flags - match a regexp against a string
2119 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2120 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
2121 /* stringarg: the point in the string at which to begin matching */
2122 /* strend: pointer to null at end of string */
2123 /* strbeg: real beginning of string */
2124 /* minend: end of match must be >= minend bytes after stringarg. */
2125 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2126 * itself is accessed via the pointers above */
2127 /* data: May be used for some additional optimizations.
2128 Currently its only used, with a U32 cast, for transmitting
2129 the ganch offset when doing a /g match. This will change */
2130 /* nosave: For optimizations. */
2134 struct regexp *const prog = ReANY(rx);
2137 char *startpos = stringarg;
2138 I32 minlen; /* must match at least this many chars */
2139 I32 dontbother = 0; /* how many characters not to try at end */
2140 I32 end_shift = 0; /* Same for the end. */ /* CC */
2141 I32 scream_pos = -1; /* Internal iterator of scream. */
2142 char *scream_olds = NULL;
2143 const bool utf8_target = cBOOL(DO_UTF8(sv));
2145 RXi_GET_DECL(prog,progi);
2146 regmatch_info reginfo; /* create some info to pass to regtry etc */
2147 regexp_paren_pair *swap = NULL;
2148 GET_RE_DEBUG_FLAGS_DECL;
2150 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2151 PERL_UNUSED_ARG(data);
2153 /* Be paranoid... */
2154 if (prog == NULL || startpos == NULL) {
2155 Perl_croak(aTHX_ "NULL regexp parameter");
2159 multiline = prog->extflags & RXf_PMf_MULTILINE;
2160 reginfo.prog = rx; /* Yes, sorry that this is confusing. */
2162 RX_MATCH_UTF8_set(rx, utf8_target);
2164 debug_start_match(rx, utf8_target, startpos, strend,
2168 minlen = prog->minlen;
2170 if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2171 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2172 "String too short [regexec_flags]...\n"));
2177 /* Check validity of program. */
2178 if (UCHARAT(progi->program) != REG_MAGIC) {
2179 Perl_croak(aTHX_ "corrupted regexp program");
2183 PL_reg_state.re_state_eval_setup_done = FALSE;
2187 PL_reg_flags |= RF_utf8;
2189 /* Mark beginning of line for ^ and lookbehind. */
2190 reginfo.bol = startpos; /* XXX not used ??? */
2194 /* Mark end of line for $ (and such) */
2197 /* see how far we have to get to not match where we matched before */
2198 reginfo.till = startpos+minend;
2200 /* If there is a "must appear" string, look for it. */
2203 if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
2205 if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
2206 reginfo.ganch = startpos + prog->gofs;
2207 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2208 "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs));
2209 } else if (sv && SvTYPE(sv) >= SVt_PVMG
2211 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
2212 && mg->mg_len >= 0) {
2213 reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
2214 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2215 "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len));
2217 if (prog->extflags & RXf_ANCH_GPOS) {
2218 if (s > reginfo.ganch)
2220 s = reginfo.ganch - prog->gofs;
2221 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2222 "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs));
2228 reginfo.ganch = strbeg + PTR2UV(data);
2229 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2230 "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
2232 } else { /* pos() not defined */
2233 reginfo.ganch = strbeg;
2234 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2235 "GPOS: reginfo.ganch = strbeg\n"));
2238 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2239 /* We have to be careful. If the previous successful match
2240 was from this regex we don't want a subsequent partially
2241 successful match to clobber the old results.
2242 So when we detect this possibility we add a swap buffer
2243 to the re, and switch the buffer each match. If we fail
2244 we switch it back, otherwise we leave it swapped.
2247 /* do we need a save destructor here for eval dies? */
2248 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2249 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2250 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2256 if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
2257 re_scream_pos_data d;
2259 d.scream_olds = &scream_olds;
2260 d.scream_pos = &scream_pos;
2261 s = re_intuit_start(rx, sv, s, strend, flags, &d);
2263 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
2264 goto phooey; /* not present */
2270 /* Simplest case: anchored match need be tried only once. */
2271 /* [unless only anchor is BOL and multiline is set] */
2272 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
2273 if (s == startpos && regtry(®info, &startpos))
2275 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2276 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
2281 dontbother = minlen - 1;
2282 end = HOP3c(strend, -dontbother, strbeg) - 1;
2283 /* for multiline we only have to try after newlines */
2284 if (prog->check_substr || prog->check_utf8) {
2285 /* because of the goto we can not easily reuse the macros for bifurcating the
2286 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2289 goto after_try_utf8;
2291 if (regtry(®info, &s)) {
2298 if (prog->extflags & RXf_USE_INTUIT) {
2299 s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
2308 } /* end search for check string in unicode */
2310 if (s == startpos) {
2311 goto after_try_latin;
2314 if (regtry(®info, &s)) {
2321 if (prog->extflags & RXf_USE_INTUIT) {
2322 s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
2331 } /* end search for check string in latin*/
2332 } /* end search for check string */
2333 else { /* search for newline */
2335 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2338 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2339 while (s <= end) { /* note it could be possible to match at the end of the string */
2340 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2341 if (regtry(®info, &s))
2345 } /* end search for newline */
2346 } /* end anchored/multiline check string search */
2348 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
2350 /* the warning about reginfo.ganch being used without initialization
2351 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
2352 and we only enter this block when the same bit is set. */
2353 char *tmp_s = reginfo.ganch - prog->gofs;
2355 if (tmp_s >= strbeg && regtry(®info, &tmp_s))
2360 /* Messy cases: unanchored match. */
2361 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2362 /* we have /x+whatever/ */
2363 /* it must be a one character string (XXXX Except UTF_PATTERN?) */
2369 if (! prog->anchored_utf8) {
2370 to_utf8_substr(prog);
2372 ch = SvPVX_const(prog->anchored_utf8)[0];
2375 DEBUG_EXECUTE_r( did_match = 1 );
2376 if (regtry(®info, &s)) goto got_it;
2378 while (s < strend && *s == ch)
2385 if (! prog->anchored_substr) {
2386 if (! to_byte_substr(prog)) {
2387 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2390 ch = SvPVX_const(prog->anchored_substr)[0];
2393 DEBUG_EXECUTE_r( did_match = 1 );
2394 if (regtry(®info, &s)) goto got_it;
2396 while (s < strend && *s == ch)
2401 DEBUG_EXECUTE_r(if (!did_match)
2402 PerlIO_printf(Perl_debug_log,
2403 "Did not find anchored character...\n")
2406 else if (prog->anchored_substr != NULL
2407 || prog->anchored_utf8 != NULL
2408 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2409 && prog->float_max_offset < strend - s)) {
2414 char *last1; /* Last position checked before */
2418 if (prog->anchored_substr || prog->anchored_utf8) {
2420 if (! prog->anchored_utf8) {
2421 to_utf8_substr(prog);
2423 must = prog->anchored_utf8;
2426 if (! prog->anchored_substr) {
2427 if (! to_byte_substr(prog)) {
2428 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2431 must = prog->anchored_substr;
2433 back_max = back_min = prog->anchored_offset;
2436 if (! prog->float_utf8) {
2437 to_utf8_substr(prog);
2439 must = prog->float_utf8;
2442 if (! prog->float_substr) {
2443 if (! to_byte_substr(prog)) {
2444 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2447 must = prog->float_substr;
2449 back_max = prog->float_max_offset;
2450 back_min = prog->float_min_offset;
2456 last = HOP3c(strend, /* Cannot start after this */
2457 -(I32)(CHR_SVLEN(must)
2458 - (SvTAIL(must) != 0) + back_min), strbeg);
2461 last1 = HOPc(s, -1);
2463 last1 = s - 1; /* bogus */
2465 /* XXXX check_substr already used to find "s", can optimize if
2466 check_substr==must. */
2468 dontbother = end_shift;
2469 strend = HOPc(strend, -dontbother);
2470 while ( (s <= last) &&
2471 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
2472 (unsigned char*)strend, must,
2473 multiline ? FBMrf_MULTILINE : 0)) ) {
2474 DEBUG_EXECUTE_r( did_match = 1 );
2475 if (HOPc(s, -back_max) > last1) {
2476 last1 = HOPc(s, -back_min);
2477 s = HOPc(s, -back_max);
2480 char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
2482 last1 = HOPc(s, -back_min);
2486 while (s <= last1) {
2487 if (regtry(®info, &s))
2490 s++; /* to break out of outer loop */
2497 while (s <= last1) {
2498 if (regtry(®info, &s))
2504 DEBUG_EXECUTE_r(if (!did_match) {
2505 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2506 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2507 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2508 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2509 ? "anchored" : "floating"),
2510 quoted, RE_SV_TAIL(must));
2514 else if ( (c = progi->regstclass) ) {
2516 const OPCODE op = OP(progi->regstclass);
2517 /* don't bother with what can't match */
2518 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2519 strend = HOPc(strend, -(minlen - 1));
2522 SV * const prop = sv_newmortal();
2523 regprop(prog, prop, c);
2525 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2527 PerlIO_printf(Perl_debug_log,
2528 "Matching stclass %.*s against %s (%d bytes)\n",
2529 (int)SvCUR(prop), SvPVX_const(prop),
2530 quoted, (int)(strend - s));
2533 if (find_byclass(prog, c, s, strend, ®info))
2535 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2539 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2547 if (! prog->float_utf8) {
2548 to_utf8_substr(prog);
2550 float_real = prog->float_utf8;
2553 if (! prog->float_substr) {
2554 if (! to_byte_substr(prog)) {
2555 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2558 float_real = prog->float_substr;
2561 little = SvPV_const(float_real, len);
2562 if (SvTAIL(float_real)) {
2563 /* This means that float_real contains an artificial \n on
2564 * the end due to the presence of something like this:
2565 * /foo$/ where we can match both "foo" and "foo\n" at the
2566 * end of the string. So we have to compare the end of the
2567 * string first against the float_real without the \n and
2568 * then against the full float_real with the string. We
2569 * have to watch out for cases where the string might be
2570 * smaller than the float_real or the float_real without
2572 char *checkpos= strend - len;
2574 PerlIO_printf(Perl_debug_log,
2575 "%sChecking for float_real.%s\n",
2576 PL_colors[4], PL_colors[5]));
2577 if (checkpos + 1 < strbeg) {
2578 /* can't match, even if we remove the trailing \n
2579 * string is too short to match */
2581 PerlIO_printf(Perl_debug_log,
2582 "%sString shorter than required trailing substring, cannot match.%s\n",
2583 PL_colors[4], PL_colors[5]));
2585 } else if (memEQ(checkpos + 1, little, len - 1)) {
2586 /* can match, the end of the string matches without the
2588 last = checkpos + 1;
2589 } else if (checkpos < strbeg) {
2590 /* cant match, string is too short when the "\n" is
2593 PerlIO_printf(Perl_debug_log,
2594 "%sString does not contain required trailing substring, cannot match.%s\n",
2595 PL_colors[4], PL_colors[5]));
2597 } else if (!multiline) {
2598 /* non multiline match, so compare with the "\n" at the
2599 * end of the string */
2600 if (memEQ(checkpos, little, len)) {
2604 PerlIO_printf(Perl_debug_log,
2605 "%sString does not contain required trailing substring, cannot match.%s\n",
2606 PL_colors[4], PL_colors[5]));
2610 /* multiline match, so we have to search for a place
2611 * where the full string is located */
2617 last = rninstr(s, strend, little, little + len);
2619 last = strend; /* matching "$" */
2622 /* at one point this block contained a comment which was
2623 * probably incorrect, which said that this was a "should not
2624 * happen" case. Even if it was true when it was written I am
2625 * pretty sure it is not anymore, so I have removed the comment
2626 * and replaced it with this one. Yves */
2628 PerlIO_printf(Perl_debug_log,
2629 "String does not contain required substring, cannot match.\n"
2633 dontbother = strend - last + prog->float_min_offset;
2635 if (minlen && (dontbother < minlen))
2636 dontbother = minlen - 1;
2637 strend -= dontbother; /* this one's always in bytes! */
2638 /* We don't know much -- general case. */
2641 if (regtry(®info, &s))
2650 if (regtry(®info, &s))
2652 } while (s++ < strend);
2662 PerlIO_printf(Perl_debug_log,
2663 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2669 RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
2671 if (PL_reg_state.re_state_eval_setup_done)
2672 restore_pos(aTHX_ prog);
2673 if (RXp_PAREN_NAMES(prog))
2674 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
2676 /* make sure $`, $&, $', and $digit will work later */
2677 if ( !(flags & REXEC_NOT_FIRST) ) {
2678 if (flags & REXEC_COPY_STR) {
2682 PerlIO_printf(Perl_debug_log,
2683 "Copy on write: regexp capture, type %d\n",
2686 RX_MATCH_COPY_FREE(rx);
2687 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2688 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2689 assert (SvPOKp(prog->saved_copy));
2690 prog->sublen = PL_regeol - strbeg;
2691 prog->suboffset = 0;
2692 prog->subcoffset = 0;
2697 I32 max = PL_regeol - strbeg;
2700 if ( (flags & REXEC_COPY_SKIP_POST)
2701 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2702 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2703 ) { /* don't copy $' part of string */
2706 /* calculate the right-most part of the string covered
2707 * by a capture. Due to look-ahead, this may be to
2708 * the right of $&, so we have to scan all captures */
2709 while (n <= prog->lastparen) {
2710 if (prog->offs[n].end > max)
2711 max = prog->offs[n].end;
2715 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2716 ? prog->offs[0].start
2718 assert(max >= 0 && max <= PL_regeol - strbeg);
2721 if ( (flags & REXEC_COPY_SKIP_PRE)
2722 && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
2723 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2724 ) { /* don't copy $` part of string */
2727 /* calculate the left-most part of the string covered
2728 * by a capture. Due to look-behind, this may be to
2729 * the left of $&, so we have to scan all captures */
2730 while (min && n <= prog->lastparen) {
2731 if ( prog->offs[n].start != -1
2732 && prog->offs[n].start < min)
2734 min = prog->offs[n].start;
2738 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2739 && min > prog->offs[0].end
2741 min = prog->offs[0].end;
2745 assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
2748 if (RX_MATCH_COPIED(rx)) {
2749 if (sublen > prog->sublen)
2751 (char*)saferealloc(prog->subbeg, sublen+1);
2754 prog->subbeg = (char*)safemalloc(sublen+1);
2755 Copy(strbeg + min, prog->subbeg, sublen, char);
2756 prog->subbeg[sublen] = '\0';
2757 prog->suboffset = min;
2758 prog->sublen = sublen;
2759 RX_MATCH_COPIED_on(rx);
2761 prog->subcoffset = prog->suboffset;
2762 if (prog->suboffset && utf8_target) {
2763 /* Convert byte offset to chars.
2764 * XXX ideally should only compute this if @-/@+
2765 * has been seen, a la PL_sawampersand ??? */
2767 /* If there's a direct correspondence between the
2768 * string which we're matching and the original SV,
2769 * then we can use the utf8 len cache associated with
2770 * the SV. In particular, it means that under //g,
2771 * sv_pos_b2u() will use the previously cached
2772 * position to speed up working out the new length of
2773 * subcoffset, rather than counting from the start of
2774 * the string each time. This stops
2775 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2776 * from going quadratic */
2777 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2778 sv_pos_b2u(sv, &(prog->subcoffset));
2780 prog->subcoffset = utf8_length((U8*)strbeg,
2781 (U8*)(strbeg+prog->suboffset));
2785 RX_MATCH_COPY_FREE(rx);
2786 prog->subbeg = strbeg;
2787 prog->suboffset = 0;
2788 prog->subcoffset = 0;
2789 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2796 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2797 PL_colors[4], PL_colors[5]));
2798 if (PL_reg_state.re_state_eval_setup_done)
2799 restore_pos(aTHX_ prog);
2801 /* we failed :-( roll it back */
2802 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2803 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2808 Safefree(prog->offs);
2815 /* Set which rex is pointed to by PL_reg_state, handling ref counting.
2816 * Do inc before dec, in case old and new rex are the same */
2817 #define SET_reg_curpm(Re2) \
2818 if (PL_reg_state.re_state_eval_setup_done) { \
2819 (void)ReREFCNT_inc(Re2); \
2820 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2821 PM_SETRE((PL_reg_curpm), (Re2)); \
2826 - regtry - try match at specific point
2828 STATIC I32 /* 0 failure, 1 success */
2829 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
2833 REGEXP *const rx = reginfo->prog;
2834 regexp *const prog = ReANY(rx);
2836 RXi_GET_DECL(prog,progi);
2837 GET_RE_DEBUG_FLAGS_DECL;
2839 PERL_ARGS_ASSERT_REGTRY;
2841 reginfo->cutpoint=NULL;
2843 if ((prog->extflags & RXf_EVAL_SEEN)
2844 && !PL_reg_state.re_state_eval_setup_done)
2848 PL_reg_state.re_state_eval_setup_done = TRUE;
2850 /* Make $_ available to executed code. */
2851 if (reginfo->sv != DEFSV) {
2853 DEFSV_set(reginfo->sv);
2856 if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
2857 && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
2858 /* prepare for quick setting of pos */
2859 #ifdef PERL_OLD_COPY_ON_WRITE
2860 if (SvIsCOW(reginfo->sv))
2861 sv_force_normal_flags(reginfo->sv, 0);
2863 mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
2864 &PL_vtbl_mglob, NULL, 0);
2868 PL_reg_oldpos = mg->mg_len;
2869 SAVEDESTRUCTOR_X(restore_pos, prog);
2871 if (!PL_reg_curpm) {
2872 Newxz(PL_reg_curpm, 1, PMOP);
2875 SV* const repointer = &PL_sv_undef;
2876 /* this regexp is also owned by the new PL_reg_curpm, which
2877 will try to free it. */
2878 av_push(PL_regex_padav, repointer);
2879 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2880 PL_regex_pad = AvARRAY(PL_regex_padav);
2885 PL_reg_oldcurpm = PL_curpm;
2886 PL_curpm = PL_reg_curpm;
2887 if (RXp_MATCH_COPIED(prog)) {
2888 /* Here is a serious problem: we cannot rewrite subbeg,
2889 since it may be needed if this match fails. Thus
2890 $` inside (?{}) could fail... */
2891 PL_reg_oldsaved = prog->subbeg;
2892 PL_reg_oldsavedlen = prog->sublen;
2893 PL_reg_oldsavedoffset = prog->suboffset;
2894 PL_reg_oldsavedcoffset = prog->suboffset;
2896 PL_nrs = prog->saved_copy;
2898 RXp_MATCH_COPIED_off(prog);
2901 PL_reg_oldsaved = NULL;
2902 prog->subbeg = PL_bostr;
2903 prog->suboffset = 0;
2904 prog->subcoffset = 0;
2905 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2908 PL_reg_starttry = *startposp;
2910 prog->offs[0].start = *startposp - PL_bostr;
2911 prog->lastparen = 0;
2912 prog->lastcloseparen = 0;
2915 /* XXXX What this code is doing here?!!! There should be no need
2916 to do this again and again, prog->lastparen should take care of
2919 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2920 * Actually, the code in regcppop() (which Ilya may be meaning by
2921 * prog->lastparen), is not needed at all by the test suite
2922 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2923 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2924 * Meanwhile, this code *is* needed for the
2925 * above-mentioned test suite tests to succeed. The common theme
2926 * on those tests seems to be returning null fields from matches.
2927 * --jhi updated by dapm */
2929 if (prog->nparens) {
2930 regexp_paren_pair *pp = prog->offs;
2932 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
2940 result = regmatch(reginfo, *startposp, progi->program + 1);
2942 prog->offs[0].end = result;
2945 if (reginfo->cutpoint)
2946 *startposp= reginfo->cutpoint;
2947 REGCP_UNWIND(lastcp);
2952 #define sayYES goto yes
2953 #define sayNO goto no
2954 #define sayNO_SILENT goto no_silent
2956 /* we dont use STMT_START/END here because it leads to
2957 "unreachable code" warnings, which are bogus, but distracting. */
2958 #define CACHEsayNO \
2959 if (ST.cache_mask) \
2960 PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
2963 /* this is used to determine how far from the left messages like
2964 'failed...' are printed. It should be set such that messages
2965 are inline with the regop output that created them.
2967 #define REPORT_CODE_OFF 32
2970 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2971 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
2972 #define CHRTEST_NOT_A_CP_1 -999
2973 #define CHRTEST_NOT_A_CP_2 -998
2975 #define SLAB_FIRST(s) (&(s)->states[0])
2976 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
2978 /* grab a new slab and return the first slot in it */
2980 STATIC regmatch_state *
2983 #if PERL_VERSION < 9 && !defined(PERL_CORE)
2986 regmatch_slab *s = PL_regmatch_slab->next;
2988 Newx(s, 1, regmatch_slab);
2989 s->prev = PL_regmatch_slab;
2991 PL_regmatch_slab->next = s;
2993 PL_regmatch_slab = s;
2994 return SLAB_FIRST(s);
2998 /* push a new state then goto it */
3000 #define PUSH_STATE_GOTO(state, node, input) \
3001 pushinput = input; \
3003 st->resume_state = state; \
3006 /* push a new state with success backtracking, then goto it */
3008 #define PUSH_YES_STATE_GOTO(state, node, input) \
3009 pushinput = input; \
3011 st->resume_state = state; \
3012 goto push_yes_state;
3019 regmatch() - main matching routine
3021 This is basically one big switch statement in a loop. We execute an op,
3022 set 'next' to point the next op, and continue. If we come to a point which
3023 we may need to backtrack to on failure such as (A|B|C), we push a
3024 backtrack state onto the backtrack stack. On failure, we pop the top
3025 state, and re-enter the loop at the state indicated. If there are no more
3026 states to pop, we return failure.
3028 Sometimes we also need to backtrack on success; for example /A+/, where
3029 after successfully matching one A, we need to go back and try to
3030 match another one; similarly for lookahead assertions: if the assertion
3031 completes successfully, we backtrack to the state just before the assertion
3032 and then carry on. In these cases, the pushed state is marked as
3033 'backtrack on success too'. This marking is in fact done by a chain of
3034 pointers, each pointing to the previous 'yes' state. On success, we pop to
3035 the nearest yes state, discarding any intermediate failure-only states.
3036 Sometimes a yes state is pushed just to force some cleanup code to be
3037 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3038 it to free the inner regex.
3040 Note that failure backtracking rewinds the cursor position, while
3041 success backtracking leaves it alone.
3043 A pattern is complete when the END op is executed, while a subpattern
3044 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3045 ops trigger the "pop to last yes state if any, otherwise return true"
3048 A common convention in this function is to use A and B to refer to the two
3049 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3050 the subpattern to be matched possibly multiple times, while B is the entire
3051 rest of the pattern. Variable and state names reflect this convention.
3053 The states in the main switch are the union of ops and failure/success of
3054 substates associated with with that op. For example, IFMATCH is the op
3055 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3056 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3057 successfully matched A and IFMATCH_A_fail is a state saying that we have
3058 just failed to match A. Resume states always come in pairs. The backtrack
3059 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3060 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3061 on success or failure.
3063 The struct that holds a backtracking state is actually a big union, with
3064 one variant for each major type of op. The variable st points to the
3065 top-most backtrack struct. To make the code clearer, within each
3066 block of code we #define ST to alias the relevant union.
3068 Here's a concrete example of a (vastly oversimplified) IFMATCH
3074 #define ST st->u.ifmatch
3076 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3077 ST.foo = ...; // some state we wish to save
3079 // push a yes backtrack state with a resume value of
3080 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3082 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3085 case IFMATCH_A: // we have successfully executed A; now continue with B
3087 bar = ST.foo; // do something with the preserved value
3090 case IFMATCH_A_fail: // A failed, so the assertion failed
3091 ...; // do some housekeeping, then ...
3092 sayNO; // propagate the failure
3099 For any old-timers reading this who are familiar with the old recursive
3100 approach, the code above is equivalent to:
3102 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3111 ...; // do some housekeeping, then ...
3112 sayNO; // propagate the failure
3115 The topmost backtrack state, pointed to by st, is usually free. If you
3116 want to claim it, populate any ST.foo fields in it with values you wish to
3117 save, then do one of
3119 PUSH_STATE_GOTO(resume_state, node, newinput);
3120 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3122 which sets that backtrack state's resume value to 'resume_state', pushes a
3123 new free entry to the top of the backtrack stack, then goes to 'node'.
3124 On backtracking, the free slot is popped, and the saved state becomes the
3125 new free state. An ST.foo field in this new top state can be temporarily
3126 accessed to retrieve values, but once the main loop is re-entered, it
3127 becomes available for reuse.
3129 Note that the depth of the backtrack stack constantly increases during the
3130 left-to-right execution of the pattern, rather than going up and down with
3131 the pattern nesting. For example the stack is at its maximum at Z at the
3132 end of the pattern, rather than at X in the following:
3134 /(((X)+)+)+....(Y)+....Z/
3136 The only exceptions to this are lookahead/behind assertions and the cut,
3137 (?>A), which pop all the backtrack states associated with A before
3140 Backtrack state structs are allocated in slabs of about 4K in size.
3141 PL_regmatch_state and st always point to the currently active state,
3142 and PL_regmatch_slab points to the slab currently containing
3143 PL_regmatch_state. The first time regmatch() is called, the first slab is
3144 allocated, and is never freed until interpreter destruction. When the slab
3145 is full, a new one is allocated and chained to the end. At exit from
3146 regmatch(), slabs allocated since entry are freed.
3151 #define DEBUG_STATE_pp(pp) \
3153 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3154 PerlIO_printf(Perl_debug_log, \
3155 " %*s"pp" %s%s%s%s%s\n", \
3157 PL_reg_name[st->resume_state], \
3158 ((st==yes_state||st==mark_state) ? "[" : ""), \
3159 ((st==yes_state) ? "Y" : ""), \
3160 ((st==mark_state) ? "M" : ""), \
3161 ((st==yes_state||st==mark_state) ? "]" : "") \
3166 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3171 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3172 const char *start, const char *end, const char *blurb)
3174 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3176 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3181 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3182 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3184 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3185 start, end - start, 60);
3187 PerlIO_printf(Perl_debug_log,
3188 "%s%s REx%s %s against %s\n",
3189 PL_colors[4], blurb, PL_colors[5], s0, s1);
3191 if (utf8_target||utf8_pat)
3192 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3193 utf8_pat ? "pattern" : "",
3194 utf8_pat && utf8_target ? " and " : "",
3195 utf8_target ? "string" : ""
3201 S_dump_exec_pos(pTHX_ const char *locinput,
3202 const regnode *scan,
3203 const char *loc_regeol,
3204 const char *loc_bostr,
3205 const char *loc_reg_starttry,
3206 const bool utf8_target)
3208 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3209 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3210 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3211 /* The part of the string before starttry has one color
3212 (pref0_len chars), between starttry and current
3213 position another one (pref_len - pref0_len chars),
3214 after the current position the third one.
3215 We assume that pref0_len <= pref_len, otherwise we
3216 decrease pref0_len. */
3217 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3218 ? (5 + taill) - l : locinput - loc_bostr;
3221 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3223 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3225 pref0_len = pref_len - (locinput - loc_reg_starttry);
3226 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3227 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3228 ? (5 + taill) - pref_len : loc_regeol - locinput);
3229 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3233 if (pref0_len > pref_len)
3234 pref0_len = pref_len;
3236 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3238 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3239 (locinput - pref_len),pref0_len, 60, 4, 5);
3241 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3242 (locinput - pref_len + pref0_len),
3243 pref_len - pref0_len, 60, 2, 3);
3245 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3246 locinput, loc_regeol - locinput, 10, 0, 1);
3248 const STRLEN tlen=len0+len1+len2;
3249 PerlIO_printf(Perl_debug_log,
3250 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3251 (IV)(locinput - loc_bostr),
3254 (docolor ? "" : "> <"),
3256 (int)(tlen > 19 ? 0 : 19 - tlen),
3263 /* reg_check_named_buff_matched()
3264 * Checks to see if a named buffer has matched. The data array of
3265 * buffer numbers corresponding to the buffer is expected to reside
3266 * in the regexp->data->data array in the slot stored in the ARG() of
3267 * node involved. Note that this routine doesn't actually care about the
3268 * name, that information is not preserved from compilation to execution.
3269 * Returns the index of the leftmost defined buffer with the given name
3270 * or 0 if non of the buffers matched.
3273 S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3276 RXi_GET_DECL(rex,rexi);
3277 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3278 I32 *nums=(I32*)SvPVX(sv_dat);
3280 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3282 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3283 if ((I32)rex->lastparen >= nums[n] &&
3284 rex->offs[nums[n]].end != -1)
3293 /* free all slabs above current one - called during LEAVE_SCOPE */
3296 S_clear_backtrack_stack(pTHX_ void *p)
3298 regmatch_slab *s = PL_regmatch_slab->next;
3303 PL_regmatch_slab->next = NULL;
3305 regmatch_slab * const osl = s;
3311 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, U8* c1_utf8, int *c2p, U8* c2_utf8)
3313 /* This function determines if there are one or two characters that match
3314 * the first character of the passed-in EXACTish node <text_node>, and if
3315 * so, returns them in the passed-in pointers.
3317 * If it determines that no possible character in the target string can
3318 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3319 * the first character in <text_node> requires UTF-8 to represent, and the
3320 * target string isn't in UTF-8.)
3322 * If there are more than two characters that could match the beginning of
3323 * <text_node>, or if more context is required to determine a match or not,
3324 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3326 * The motiviation behind this function is to allow the caller to set up
3327 * tight loops for matching. If <text_node> is of type EXACT, there is
3328 * only one possible character that can match its first character, and so
3329 * the situation is quite simple. But things get much more complicated if
3330 * folding is involved. It may be that the first character of an EXACTFish
3331 * node doesn't participate in any possible fold, e.g., punctuation, so it
3332 * can be matched only by itself. The vast majority of characters that are
3333 * in folds match just two things, their lower and upper-case equivalents.
3334 * But not all are like that; some have multiple possible matches, or match
3335 * sequences of more than one character. This function sorts all that out.
3337 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3338 * loop of trying to match A*, we know we can't exit where the thing
3339 * following it isn't a B. And something can't be a B unless it is the
3340 * beginning of B. By putting a quick test for that beginning in a tight
3341 * loop, we can rule out things that can't possibly be B without having to
3342 * break out of the loop, thus avoiding work. Similarly, if A is a single
3343 * character, we can make a tight loop matching A*, using the outputs of
3346 * If the target string to match isn't in UTF-8, and there aren't
3347 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3348 * the one or two possible octets (which are characters in this situation)
3349 * that can match. In all cases, if there is only one character that can
3350 * match, *<c1p> and *<c2p> will be identical.
3352 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3353 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3354 * can match the beginning of <text_node>. They should be declared with at
3355 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3356 * undefined what these contain.) If one or both of the buffers are
3357 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3358 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3359 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3360 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3361 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3363 const bool utf8_target = PL_reg_match_utf8;
3365 UV c1 = CHRTEST_NOT_A_CP_1;
3366 UV c2 = CHRTEST_NOT_A_CP_2;
3367 bool use_chrtest_void = FALSE;
3369 /* Used when we have both utf8 input and utf8 output, to avoid converting
3370 * to/from code points */
3371 bool utf8_has_been_setup = FALSE;
3375 U8 *pat = (U8*)STRING(text_node);
3377 if (OP(text_node) == EXACT) {
3379 /* In an exact node, only one thing can be matched, that first
3380 * character. If both the pat and the target are UTF-8, we can just
3381 * copy the input to the output, avoiding finding the code point of
3383 if (! UTF_PATTERN) {
3386 else if (utf8_target) {
3387 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3388 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3389 utf8_has_been_setup = TRUE;
3392 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3395 else /* an EXACTFish node */
3397 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3398 pat + STR_LEN(text_node)))
3400 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3401 pat + STR_LEN(text_node))))
3403 /* Multi-character folds require more context to sort out. Also
3404 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3405 * handled outside this routine */
3406 use_chrtest_void = TRUE;
3408 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3409 c1 = (UTF_PATTERN) ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3411 /* Load the folds hash, if not already done */
3413 if (! PL_utf8_foldclosures) {
3414 if (! PL_utf8_tofold) {
3415 U8 dummy[UTF8_MAXBYTES+1];
3417 /* Force loading this by folding an above-Latin1 char */
3418 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3419 assert(PL_utf8_tofold); /* Verify that worked */
3421 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3424 /* The fold closures data structure is a hash with the keys being
3425 * the UTF-8 of every character that is folded to, like 'k', and
3426 * the values each an array of all code points that fold to its
3427 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3429 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3434 /* Not found in the hash, therefore there are no folds
3435 * containing it, so there is only a single character that
3439 else { /* Does participate in folds */
3440 AV* list = (AV*) *listp;
3441 if (av_len(list) != 1) {
3443 /* If there aren't exactly two folds to this, it is outside
3444 * the scope of this function */
3445 use_chrtest_void = TRUE;
3447 else { /* There are two. Get them */
3448 SV** c_p = av_fetch(list, 0, FALSE);
3450 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3454 c_p = av_fetch(list, 1, FALSE);
3456 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3460 /* Folds that cross the 255/256 boundary are forbidden if
3461 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3462 * pattern character is above 256, and its only other match
3463 * is below 256, the only legal match will be to itself.
3464 * We have thrown away the original, so have to compute
3465 * which is the one above 255 */
3466 if ((c1 < 256) != (c2 < 256)) {
3467 if (OP(text_node) == EXACTFL
3468 || (OP(text_node) == EXACTFA
3469 && (isASCII(c1) || isASCII(c2))))
3482 else /* Here, c1 is < 255 */
3484 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3485 && OP(text_node) != EXACTFL
3486 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
3488 /* Here, there could be something above Latin1 in the target which
3489 * folds to this character in the pattern. All such cases except
3490 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3491 * involved in their folds, so are outside the scope of this
3493 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3494 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3497 use_chrtest_void = TRUE;
3500 else { /* Here nothing above Latin1 can fold to the pattern character */
3501 switch (OP(text_node)) {
3503 case EXACTFL: /* /l rules */
3504 c2 = PL_fold_locale[c1];
3508 if (! utf8_target) { /* /d rules */
3513 /* /u rules for all these. This happens to work for
3514 * EXACTFA as nothing in Latin1 folds to ASCII */
3516 case EXACTFU_TRICKYFOLD:
3519 c2 = PL_fold_latin1[c1];
3523 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3524 assert(0); /* NOTREACHED */
3529 /* Here have figured things out. Set up the returns */
3530 if (use_chrtest_void) {
3531 *c2p = *c1p = CHRTEST_VOID;
3533 else if (utf8_target) {
3534 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3535 uvchr_to_utf8(c1_utf8, c1);
3536 uvchr_to_utf8(c2_utf8, c2);
3539 /* Invariants are stored in both the utf8 and byte outputs; Use
3540 * negative numbers otherwise for the byte ones. Make sure that the
3541 * byte ones are the same iff the utf8 ones are the same */
3542 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3543 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3546 ? CHRTEST_NOT_A_CP_1
3547 : CHRTEST_NOT_A_CP_2;
3549 else if (c1 > 255) {
3550 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3555 *c1p = *c2p = c2; /* c2 is the only representable value */
3557 else { /* c1 is representable; see about c2 */
3559 *c2p = (c2 < 256) ? c2 : c1;
3565 /* returns -1 on failure, $+[0] on success */
3567 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3569 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3573 const bool utf8_target = PL_reg_match_utf8;
3574 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3575 REGEXP *rex_sv = reginfo->prog;
3576 regexp *rex = ReANY(rex_sv);
3577 RXi_GET_DECL(rex,rexi);
3579 /* the current state. This is a cached copy of PL_regmatch_state */
3581 /* cache heavy used fields of st in registers */
3584 U32 n = 0; /* general value; init to avoid compiler warning */
3585 I32 ln = 0; /* len or last; init to avoid compiler warning */
3586 char *locinput = startpos;
3587 char *pushinput; /* where to continue after a PUSH */
3588 I32 nextchr; /* is always set to UCHARAT(locinput) */
3590 bool result = 0; /* return value of S_regmatch */
3591 int depth = 0; /* depth of backtrack stack */
3592 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3593 const U32 max_nochange_depth =
3594 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3595 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3596 regmatch_state *yes_state = NULL; /* state to pop to on success of
3598 /* mark_state piggy backs on the yes_state logic so that when we unwind
3599 the stack on success we can update the mark_state as we go */
3600 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3601 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3602 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3604 bool no_final = 0; /* prevent failure from backtracking? */
3605 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3606 char *startpoint = locinput;
3607 SV *popmark = NULL; /* are we looking for a mark? */
3608 SV *sv_commit = NULL; /* last mark name seen in failure */
3609 SV *sv_yes_mark = NULL; /* last mark name we have seen
3610 during a successful match */
3611 U32 lastopen = 0; /* last open we saw */
3612 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3613 SV* const oreplsv = GvSV(PL_replgv);
3614 /* these three flags are set by various ops to signal information to
3615 * the very next op. They have a useful lifetime of exactly one loop
3616 * iteration, and are not preserved or restored by state pushes/pops
3618 bool sw = 0; /* the condition value in (?(cond)a|b) */
3619 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3620 int logical = 0; /* the following EVAL is:
3624 or the following IFMATCH/UNLESSM is:
3625 false: plain (?=foo)
3626 true: used as a condition: (?(?=foo))
3628 PAD* last_pad = NULL;
3630 I32 gimme = G_SCALAR;
3631 CV *caller_cv = NULL; /* who called us */
3632 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3633 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3636 GET_RE_DEBUG_FLAGS_DECL;
3639 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3640 multicall_oldcatch = 0;
3641 multicall_cv = NULL;
3643 PERL_UNUSED_VAR(multicall_cop);
3644 PERL_UNUSED_VAR(newsp);
3647 PERL_ARGS_ASSERT_REGMATCH;
3649 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3650 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3652 /* on first ever call to regmatch, allocate first slab */
3653 if (!PL_regmatch_slab) {
3654 Newx(PL_regmatch_slab, 1, regmatch_slab);
3655 PL_regmatch_slab->prev = NULL;
3656 PL_regmatch_slab->next = NULL;
3657 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3660 oldsave = PL_savestack_ix;
3661 SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
3662 SAVEVPTR(PL_regmatch_slab);
3663 SAVEVPTR(PL_regmatch_state);
3665 /* grab next free state slot */
3666 st = ++PL_regmatch_state;
3667 if (st > SLAB_LAST(PL_regmatch_slab))
3668 st = PL_regmatch_state = S_push_slab(aTHX);
3670 /* Note that nextchr is a byte even in UTF */
3673 while (scan != NULL) {
3676 SV * const prop = sv_newmortal();
3677 regnode *rnext=regnext(scan);
3678 DUMP_EXEC_POS( locinput, scan, utf8_target );
3679 regprop(rex, prop, scan);
3681 PerlIO_printf(Perl_debug_log,
3682 "%3"IVdf":%*s%s(%"IVdf")\n",
3683 (IV)(scan - rexi->program), depth*2, "",
3685 (PL_regkind[OP(scan)] == END || !rnext) ?
3686 0 : (IV)(rnext - rexi->program));
3689 next = scan + NEXT_OFF(scan);
3692 state_num = OP(scan);
3697 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
3699 switch (state_num) {
3700 case BOL: /* /^../ */
3701 if (locinput == PL_bostr)
3703 /* reginfo->till = reginfo->bol; */
3708 case MBOL: /* /^../m */
3709 if (locinput == PL_bostr ||
3710 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
3716 case SBOL: /* /^../s */
3717 if (locinput == PL_bostr)
3722 if (locinput == reginfo->ganch)
3726 case KEEPS: /* \K */
3727 /* update the startpoint */
3728 st->u.keeper.val = rex->offs[0].start;
3729 rex->offs[0].start = locinput - PL_bostr;
3730 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
3731 assert(0); /*NOTREACHED*/
3732 case KEEPS_next_fail:
3733 /* rollback the start point change */
3734 rex->offs[0].start = st->u.keeper.val;
3736 assert(0); /*NOTREACHED*/
3738 case EOL: /* /..$/ */
3741 case MEOL: /* /..$/m */
3742 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3746 case SEOL: /* /..$/s */
3748 if (!NEXTCHR_IS_EOS && nextchr != '\n')
3750 if (PL_regeol - locinput > 1)
3755 if (!NEXTCHR_IS_EOS)
3759 case SANY: /* /./s */
3762 goto increment_locinput;
3770 case REG_ANY: /* /./ */
3771 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
3773 goto increment_locinput;
3777 #define ST st->u.trie
3778 case TRIEC: /* (ab|cd) with known charclass */
3779 /* In this case the charclass data is available inline so
3780 we can fail fast without a lot of extra overhead.
3782 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
3784 PerlIO_printf(Perl_debug_log,
3785 "%*s %sfailed to match trie start class...%s\n",
3786 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3789 assert(0); /* NOTREACHED */
3792 case TRIE: /* (ab|cd) */
3793 /* the basic plan of execution of the trie is:
3794 * At the beginning, run though all the states, and
3795 * find the longest-matching word. Also remember the position
3796 * of the shortest matching word. For example, this pattern:
3799 * when matched against the string "abcde", will generate
3800 * accept states for all words except 3, with the longest
3801 * matching word being 4, and the shortest being 2 (with
3802 * the position being after char 1 of the string).
3804 * Then for each matching word, in word order (i.e. 1,2,4,5),
3805 * we run the remainder of the pattern; on each try setting
3806 * the current position to the character following the word,
3807 * returning to try the next word on failure.
3809 * We avoid having to build a list of words at runtime by
3810 * using a compile-time structure, wordinfo[].prev, which
3811 * gives, for each word, the previous accepting word (if any).
3812 * In the case above it would contain the mappings 1->2, 2->0,
3813 * 3->0, 4->5, 5->1. We can use this table to generate, from
3814 * the longest word (4 above), a list of all words, by
3815 * following the list of prev pointers; this gives us the
3816 * unordered list 4,5,1,2. Then given the current word we have
3817 * just tried, we can go through the list and find the
3818 * next-biggest word to try (so if we just failed on word 2,
3819 * the next in the list is 4).
3821 * Since at runtime we don't record the matching position in
3822 * the string for each word, we have to work that out for
3823 * each word we're about to process. The wordinfo table holds
3824 * the character length of each word; given that we recorded
3825 * at the start: the position of the shortest word and its
3826 * length in chars, we just need to move the pointer the
3827 * difference between the two char lengths. Depending on
3828 * Unicode status and folding, that's cheap or expensive.
3830 * This algorithm is optimised for the case where are only a
3831 * small number of accept states, i.e. 0,1, or maybe 2.
3832 * With lots of accepts states, and having to try all of them,
3833 * it becomes quadratic on number of accept states to find all
3838 /* what type of TRIE am I? (utf8 makes this contextual) */
3839 DECL_TRIE_TYPE(scan);
3841 /* what trie are we using right now */
3842 reg_trie_data * const trie
3843 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
3844 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3845 U32 state = trie->startstate;
3848 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
3850 if (trie->states[ state ].wordnum) {
3852 PerlIO_printf(Perl_debug_log,
3853 "%*s %smatched empty string...%s\n",
3854 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3860 PerlIO_printf(Perl_debug_log,
3861 "%*s %sfailed to match trie start class...%s\n",
3862 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3869 U8 *uc = ( U8* )locinput;
3873 U8 *uscan = (U8*)NULL;
3874 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
3875 U32 charcount = 0; /* how many input chars we have matched */
3876 U32 accepted = 0; /* have we seen any accepting states? */
3878 ST.jump = trie->jump;
3881 ST.longfold = FALSE; /* char longer if folded => it's harder */
3884 /* fully traverse the TRIE; note the position of the
3885 shortest accept state and the wordnum of the longest
3888 while ( state && uc <= (U8*)PL_regeol ) {
3889 U32 base = trie->states[ state ].trans.base;
3893 wordnum = trie->states[ state ].wordnum;
3895 if (wordnum) { /* it's an accept state */
3898 /* record first match position */
3900 ST.firstpos = (U8*)locinput;
3905 ST.firstchars = charcount;
3908 if (!ST.nextword || wordnum < ST.nextword)
3909 ST.nextword = wordnum;
3910 ST.topword = wordnum;
3913 DEBUG_TRIE_EXECUTE_r({
3914 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
3915 PerlIO_printf( Perl_debug_log,
3916 "%*s %sState: %4"UVxf" Accepted: %c ",
3917 2+depth * 2, "", PL_colors[4],
3918 (UV)state, (accepted ? 'Y' : 'N'));
3921 /* read a char and goto next state */
3922 if ( base && (foldlen || uc < (U8*)PL_regeol)) {
3924 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
3925 uscan, len, uvc, charid, foldlen,