5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "inline_invlist.c"
84 #include "unicode_constants.h"
87 /* At least one required character in the target string is expressible only in
89 static const char* const non_utf8_target_but_utf8_required
90 = "Can't match, because target string needs to be in UTF-8\n";
93 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
94 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
98 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
101 #define STATIC static
104 /* Valid only for non-utf8 strings: avoids the reginclass
105 * call if there are no complications: i.e., if everything matchable is
106 * straight forward in the bitmap */
107 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \
108 : ANYOF_BITMAP_TEST(p,*(c)))
114 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
115 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
117 #define HOPc(pos,off) \
118 (char *)(reginfo->is_utf8_target \
119 ? reghop3((U8*)pos, off, \
120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
123 #define HOPBACKc(pos, off) \
124 (char*)(reginfo->is_utf8_target \
125 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
126 : (pos - off >= reginfo->strbeg) \
130 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
131 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
133 /* lim must be +ve. Returns NULL on overshoot */
134 #define HOPMAYBE3(pos,off,lim) \
135 (reginfo->is_utf8_target \
136 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
137 : ((U8*)pos + off <= lim) \
141 /* like HOP3, but limits the result to <= lim even for the non-utf8 case.
142 * off must be >=0; args should be vars rather than expressions */
143 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
144 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
145 : (U8*)((pos + off) > lim ? lim : (pos + off)))
147 #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
148 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
150 #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
152 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
153 #define NEXTCHR_IS_EOS (nextchr < 0)
155 #define SET_nextchr \
156 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
158 #define SET_locinput(p) \
163 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
165 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
166 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
167 1, 0, invlist, &flags); \
172 /* If in debug mode, we test that a known character properly matches */
174 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
177 utf8_char_in_property) \
178 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \
179 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
181 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
184 utf8_char_in_property) \
185 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
188 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
189 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
191 PL_XPosix_ptrs[_CC_WORDCHAR], \
192 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
194 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
196 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
197 "_X_regular_begin", \
199 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \
200 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
203 COMBINING_GRAVE_ACCENT_UTF8); \
206 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
207 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
209 /* for use after a quantifier and before an EXACT-like node -- japhy */
210 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
212 * NOTE that *nothing* that affects backtracking should be in here, specifically
213 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
214 * node that is in between two EXACT like nodes when ascertaining what the required
215 * "follow" character is. This should probably be moved to regex compile time
216 * although it may be done at run time beause of the REF possibility - more
217 * investigation required. -- demerphq
219 #define JUMPABLE(rn) ( \
221 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
223 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
224 OP(rn) == PLUS || OP(rn) == MINMOD || \
226 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
228 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
230 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
233 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
234 we don't need this definition. XXX These are now out-of-sync*/
235 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
236 #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
237 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
240 /* ... so we use this as its faster. */
241 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==EXACTL )
242 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFLU8 || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
243 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
244 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
249 Search for mandatory following text node; for lookahead, the text must
250 follow but for lookbehind (rn->flags != 0) we skip to the next step.
252 #define FIND_NEXT_IMPT(rn) STMT_START { \
253 while (JUMPABLE(rn)) { \
254 const OPCODE type = OP(rn); \
255 if (type == SUSPEND || PL_regkind[type] == CURLY) \
256 rn = NEXTOPER(NEXTOPER(rn)); \
257 else if (type == PLUS) \
259 else if (type == IFMATCH) \
260 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
261 else rn += NEXT_OFF(rn); \
265 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
266 * These are for the pre-composed Hangul syllables, which are all in a
267 * contiguous block and arranged there in such a way so as to facilitate
268 * alorithmic determination of their characteristics. As such, they don't need
269 * a swash, but can be determined by simple arithmetic. Almost all are
270 * GCB=LVT, but every 28th one is a GCB=LV */
271 #define SBASE 0xAC00 /* Start of block */
272 #define SCount 11172 /* Length of block */
275 #define SLAB_FIRST(s) (&(s)->states[0])
276 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
278 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
279 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
280 static regmatch_state * S_push_slab(pTHX);
282 #define REGCP_PAREN_ELEMS 3
283 #define REGCP_OTHER_ELEMS 3
284 #define REGCP_FRAME_ELEMS 1
285 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
286 * are needed for the regexp context stack bookkeeping. */
289 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
291 const int retval = PL_savestack_ix;
292 const int paren_elems_to_push =
293 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
294 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
295 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
297 GET_RE_DEBUG_FLAGS_DECL;
299 PERL_ARGS_ASSERT_REGCPPUSH;
301 if (paren_elems_to_push < 0)
302 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
303 (int)paren_elems_to_push, (int)maxopenparen,
304 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
306 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
307 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
308 " out of range (%lu-%ld)",
310 (unsigned long)maxopenparen,
313 SSGROW(total_elems + REGCP_FRAME_ELEMS);
316 if ((int)maxopenparen > (int)parenfloor)
317 PerlIO_printf(Perl_debug_log,
318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
323 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
324 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
325 SSPUSHIV(rex->offs[p].end);
326 SSPUSHIV(rex->offs[p].start);
327 SSPUSHINT(rex->offs[p].start_tmp);
328 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
329 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
331 (IV)rex->offs[p].start,
332 (IV)rex->offs[p].start_tmp,
336 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
337 SSPUSHINT(maxopenparen);
338 SSPUSHINT(rex->lastparen);
339 SSPUSHINT(rex->lastcloseparen);
340 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
345 /* These are needed since we do not localize EVAL nodes: */
346 #define REGCP_SET(cp) \
348 PerlIO_printf(Perl_debug_log, \
349 " Setting an EVAL scope, savestack=%"IVdf"\n", \
350 (IV)PL_savestack_ix)); \
353 #define REGCP_UNWIND(cp) \
355 if (cp != PL_savestack_ix) \
356 PerlIO_printf(Perl_debug_log, \
357 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
358 (IV)(cp), (IV)PL_savestack_ix)); \
361 #define UNWIND_PAREN(lp, lcp) \
362 for (n = rex->lastparen; n > lp; n--) \
363 rex->offs[n].end = -1; \
364 rex->lastparen = n; \
365 rex->lastcloseparen = lcp;
369 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
373 GET_RE_DEBUG_FLAGS_DECL;
375 PERL_ARGS_ASSERT_REGCPPOP;
377 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
379 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
380 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
381 rex->lastcloseparen = SSPOPINT;
382 rex->lastparen = SSPOPINT;
383 *maxopenparen_p = SSPOPINT;
385 i -= REGCP_OTHER_ELEMS;
386 /* Now restore the parentheses context. */
388 if (i || rex->lastparen + 1 <= rex->nparens)
389 PerlIO_printf(Perl_debug_log,
390 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
395 paren = *maxopenparen_p;
396 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
398 rex->offs[paren].start_tmp = SSPOPINT;
399 rex->offs[paren].start = SSPOPIV;
401 if (paren <= rex->lastparen)
402 rex->offs[paren].end = tmps;
403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
406 (IV)rex->offs[paren].start,
407 (IV)rex->offs[paren].start_tmp,
408 (IV)rex->offs[paren].end,
409 (paren > rex->lastparen ? "(skipped)" : ""));
414 /* It would seem that the similar code in regtry()
415 * already takes care of this, and in fact it is in
416 * a better location to since this code can #if 0-ed out
417 * but the code in regtry() is needed or otherwise tests
418 * requiring null fields (pat.t#187 and split.t#{13,14}
419 * (as of patchlevel 7877) will fail. Then again,
420 * this code seems to be necessary or otherwise
421 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
422 * --jhi updated by dapm */
423 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
424 if (i > *maxopenparen_p)
425 rex->offs[i].start = -1;
426 rex->offs[i].end = -1;
427 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
428 " \\%"UVuf": %s ..-1 undeffing\n",
430 (i > *maxopenparen_p) ? "-1" : " "
436 /* restore the parens and associated vars at savestack position ix,
437 * but without popping the stack */
440 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
442 I32 tmpix = PL_savestack_ix;
443 PL_savestack_ix = ix;
444 regcppop(rex, maxopenparen_p);
445 PL_savestack_ix = tmpix;
448 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
451 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
453 /* Returns a boolean as to whether or not 'character' is a member of the
454 * Posix character class given by 'classnum' that should be equivalent to a
455 * value in the typedef '_char_class_number'.
457 * Ideally this could be replaced by a just an array of function pointers
458 * to the C library functions that implement the macros this calls.
459 * However, to compile, the precise function signatures are required, and
460 * these may vary from platform to to platform. To avoid having to figure
461 * out what those all are on each platform, I (khw) am using this method,
462 * which adds an extra layer of function call overhead (unless the C
463 * optimizer strips it away). But we don't particularly care about
464 * performance with locales anyway. */
466 switch ((_char_class_number) classnum) {
467 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
468 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
469 case _CC_ENUM_ASCII: return isASCII_LC(character);
470 case _CC_ENUM_BLANK: return isBLANK_LC(character);
471 case _CC_ENUM_CASED: return isLOWER_LC(character)
472 || isUPPER_LC(character);
473 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
474 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
475 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
476 case _CC_ENUM_LOWER: return isLOWER_LC(character);
477 case _CC_ENUM_PRINT: return isPRINT_LC(character);
478 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
479 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
480 case _CC_ENUM_SPACE: return isSPACE_LC(character);
481 case _CC_ENUM_UPPER: return isUPPER_LC(character);
482 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
483 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
484 default: /* VERTSPACE should never occur in locales */
485 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
488 NOT_REACHED; /* NOTREACHED */
493 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
495 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
496 * 'character' is a member of the Posix character class given by 'classnum'
497 * that should be equivalent to a value in the typedef
498 * '_char_class_number'.
500 * This just calls isFOO_lc on the code point for the character if it is in
501 * the range 0-255. Outside that range, all characters use Unicode
502 * rules, ignoring any locale. So use the Unicode function if this class
503 * requires a swash, and use the Unicode macro otherwise. */
505 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
507 if (UTF8_IS_INVARIANT(*character)) {
508 return isFOO_lc(classnum, *character);
510 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
511 return isFOO_lc(classnum,
512 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
515 if (classnum < _FIRST_NON_SWASH_CC) {
517 /* Initialize the swash unless done already */
518 if (! PL_utf8_swash_ptrs[classnum]) {
519 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
520 PL_utf8_swash_ptrs[classnum] =
521 _core_swash_init("utf8",
524 PL_XPosix_ptrs[classnum], &flags);
527 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
529 TRUE /* is UTF */ ));
532 switch ((_char_class_number) classnum) {
534 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
536 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
537 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
538 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
542 return FALSE; /* Things like CNTRL are always below 256 */
546 * pregexec and friends
549 #ifndef PERL_IN_XSUB_RE
551 - pregexec - match a regexp against a string
554 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
555 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
556 /* stringarg: the point in the string at which to begin matching */
557 /* strend: pointer to null at end of string */
558 /* strbeg: real beginning of string */
559 /* minend: end of match must be >= minend bytes after stringarg. */
560 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
561 * itself is accessed via the pointers above */
562 /* nosave: For optimizations. */
564 PERL_ARGS_ASSERT_PREGEXEC;
567 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
568 nosave ? 0 : REXEC_COPY_STR);
574 /* re_intuit_start():
576 * Based on some optimiser hints, try to find the earliest position in the
577 * string where the regex could match.
579 * rx: the regex to match against
580 * sv: the SV being matched: only used for utf8 flag; the string
581 * itself is accessed via the pointers below. Note that on
582 * something like an overloaded SV, SvPOK(sv) may be false
583 * and the string pointers may point to something unrelated to
585 * strbeg: real beginning of string
586 * strpos: the point in the string at which to begin matching
587 * strend: pointer to the byte following the last char of the string
588 * flags currently unused; set to 0
589 * data: currently unused; set to NULL
591 * The basic idea of re_intuit_start() is to use some known information
592 * about the pattern, namely:
594 * a) the longest known anchored substring (i.e. one that's at a
595 * constant offset from the beginning of the pattern; but not
596 * necessarily at a fixed offset from the beginning of the
598 * b) the longest floating substring (i.e. one that's not at a constant
599 * offset from the beginning of the pattern);
600 * c) Whether the pattern is anchored to the string; either
601 * an absolute anchor: /^../, or anchored to \n: /^.../m,
602 * or anchored to pos(): /\G/;
603 * d) A start class: a real or synthetic character class which
604 * represents which characters are legal at the start of the pattern;
606 * to either quickly reject the match, or to find the earliest position
607 * within the string at which the pattern might match, thus avoiding
608 * running the full NFA engine at those earlier locations, only to
609 * eventually fail and retry further along.
611 * Returns NULL if the pattern can't match, or returns the address within
612 * the string which is the earliest place the match could occur.
614 * The longest of the anchored and floating substrings is called 'check'
615 * and is checked first. The other is called 'other' and is checked
616 * second. The 'other' substring may not be present. For example,
618 * /(abc|xyz)ABC\d{0,3}DEFG/
622 * check substr (float) = "DEFG", offset 6..9 chars
623 * other substr (anchored) = "ABC", offset 3..3 chars
626 * Be aware that during the course of this function, sometimes 'anchored'
627 * refers to a substring being anchored relative to the start of the
628 * pattern, and sometimes to the pattern itself being anchored relative to
629 * the string. For example:
631 * /\dabc/: "abc" is anchored to the pattern;
632 * /^\dabc/: "abc" is anchored to the pattern and the string;
633 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
634 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
635 * but the pattern is anchored to the string.
639 Perl_re_intuit_start(pTHX_
642 const char * const strbeg,
646 re_scream_pos_data *data)
648 struct regexp *const prog = ReANY(rx);
649 SSize_t start_shift = prog->check_offset_min;
650 /* Should be nonnegative! */
651 SSize_t end_shift = 0;
652 /* current lowest pos in string where the regex can start matching */
653 char *rx_origin = strpos;
655 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
656 U8 other_ix = 1 - prog->substrs->check_ix;
658 char *other_last = strpos;/* latest pos 'other' substr already checked to */
659 char *check_at = NULL; /* check substr found at this pos */
660 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
661 RXi_GET_DECL(prog,progi);
662 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
663 regmatch_info *const reginfo = ®info_buf;
664 GET_RE_DEBUG_FLAGS_DECL;
666 PERL_ARGS_ASSERT_RE_INTUIT_START;
667 PERL_UNUSED_ARG(flags);
668 PERL_UNUSED_ARG(data);
670 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
671 "Intuit: trying to determine minimum start position...\n"));
673 /* for now, assume that all substr offsets are positive. If at some point
674 * in the future someone wants to do clever things with look-behind and
675 * -ve offsets, they'll need to fix up any code in this function
676 * which uses these offsets. See the thread beginning
677 * <20140113145929.GF27210@iabyn.com>
679 assert(prog->substrs->data[0].min_offset >= 0);
680 assert(prog->substrs->data[0].max_offset >= 0);
681 assert(prog->substrs->data[1].min_offset >= 0);
682 assert(prog->substrs->data[1].max_offset >= 0);
683 assert(prog->substrs->data[2].min_offset >= 0);
684 assert(prog->substrs->data[2].max_offset >= 0);
686 /* for now, assume that if both present, that the floating substring
687 * doesn't start before the anchored substring.
688 * If you break this assumption (e.g. doing better optimisations
689 * with lookahead/behind), then you'll need to audit the code in this
690 * function carefully first
693 ! ( (prog->anchored_utf8 || prog->anchored_substr)
694 && (prog->float_utf8 || prog->float_substr))
695 || (prog->float_min_offset >= prog->anchored_offset));
697 /* byte rather than char calculation for efficiency. It fails
698 * to quickly reject some cases that can't match, but will reject
699 * them later after doing full char arithmetic */
700 if (prog->minlen > strend - strpos) {
701 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
702 " String too short...\n"));
706 RX_MATCH_UTF8_set(rx,utf8_target);
707 reginfo->is_utf8_target = cBOOL(utf8_target);
708 reginfo->info_aux = NULL;
709 reginfo->strbeg = strbeg;
710 reginfo->strend = strend;
711 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
713 /* not actually used within intuit, but zero for safety anyway */
714 reginfo->poscache_maxiter = 0;
717 if (!prog->check_utf8 && prog->check_substr)
718 to_utf8_substr(prog);
719 check = prog->check_utf8;
721 if (!prog->check_substr && prog->check_utf8) {
722 if (! to_byte_substr(prog)) {
723 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
726 check = prog->check_substr;
729 /* dump the various substring data */
730 DEBUG_OPTIMISE_MORE_r({
732 for (i=0; i<=2; i++) {
733 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
734 : prog->substrs->data[i].substr);
738 PerlIO_printf(Perl_debug_log,
739 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
740 " useful=%"IVdf" utf8=%d [%s]\n",
742 (IV)prog->substrs->data[i].min_offset,
743 (IV)prog->substrs->data[i].max_offset,
744 (IV)prog->substrs->data[i].end_shift,
751 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
753 /* ml_anch: check after \n?
755 * A note about IMPLICIT: on an un-anchored pattern beginning
756 * with /.*.../, these flags will have been added by the
758 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
759 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
761 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
762 && !(prog->intflags & PREGf_IMPLICIT);
764 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
765 /* we are only allowed to match at BOS or \G */
767 /* trivially reject if there's a BOS anchor and we're not at BOS.
769 * Note that we don't try to do a similar quick reject for
770 * \G, since generally the caller will have calculated strpos
771 * based on pos() and gofs, so the string is already correctly
772 * anchored by definition; and handling the exceptions would
773 * be too fiddly (e.g. REXEC_IGNOREPOS).
775 if ( strpos != strbeg
776 && (prog->intflags & PREGf_ANCH_SBOL))
778 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
779 " Not at start...\n"));
783 /* in the presence of an anchor, the anchored (relative to the
784 * start of the regex) substr must also be anchored relative
785 * to strpos. So quickly reject if substr isn't found there.
786 * This works for \G too, because the caller will already have
787 * subtracted gofs from pos, and gofs is the offset from the
788 * \G to the start of the regex. For example, in /.abc\Gdef/,
789 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
790 * caller will have set strpos=pos()-4; we look for the substr
791 * at position pos()-4+1, which lines up with the "a" */
793 if (prog->check_offset_min == prog->check_offset_max
794 && !(prog->intflags & PREGf_CANY_SEEN))
796 /* Substring at constant offset from beg-of-str... */
797 SSize_t slen = SvCUR(check);
798 char *s = HOP3c(strpos, prog->check_offset_min, strend);
800 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
801 " Looking for check substr at fixed offset %"IVdf"...\n",
802 (IV)prog->check_offset_min));
805 /* In this case, the regex is anchored at the end too.
806 * Unless it's a multiline match, the lengths must match
807 * exactly, give or take a \n. NB: slen >= 1 since
808 * the last char of check is \n */
810 && ( strend - s > slen
811 || strend - s < slen - 1
812 || (strend - s == slen && strend[-1] != '\n')))
814 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
815 " String too long...\n"));
818 /* Now should match s[0..slen-2] */
821 if (slen && (*SvPVX_const(check) != *s
822 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
824 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
825 " String not equal...\n"));
830 goto success_at_start;
835 end_shift = prog->check_end_shift;
837 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
839 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
840 (IV)end_shift, RX_PRECOMP(prog));
845 /* This is the (re)entry point of the main loop in this function.
846 * The goal of this loop is to:
847 * 1) find the "check" substring in the region rx_origin..strend
848 * (adjusted by start_shift / end_shift). If not found, reject
850 * 2) If it exists, look for the "other" substr too if defined; for
851 * example, if the check substr maps to the anchored substr, then
852 * check the floating substr, and vice-versa. If not found, go
853 * back to (1) with rx_origin suitably incremented.
854 * 3) If we find an rx_origin position that doesn't contradict
855 * either of the substrings, then check the possible additional
856 * constraints on rx_origin of /^.../m or a known start class.
857 * If these fail, then depending on which constraints fail, jump
858 * back to here, or to various other re-entry points further along
859 * that skip some of the first steps.
860 * 4) If we pass all those tests, update the BmUSEFUL() count on the
861 * substring. If the start position was determined to be at the
862 * beginning of the string - so, not rejected, but not optimised,
863 * since we have to run regmatch from position 0 - decrement the
864 * BmUSEFUL() count. Otherwise increment it.
868 /* first, look for the 'check' substring */
874 DEBUG_OPTIMISE_MORE_r({
875 PerlIO_printf(Perl_debug_log,
876 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
877 " Start shift: %"IVdf" End shift %"IVdf
878 " Real end Shift: %"IVdf"\n",
879 (IV)(rx_origin - strpos),
880 (IV)prog->check_offset_min,
883 (IV)prog->check_end_shift);
886 if (prog->intflags & PREGf_CANY_SEEN) {
887 start_point= (U8*)(rx_origin + start_shift);
888 end_point= (U8*)(strend - end_shift);
889 if (start_point > end_point)
892 end_point = HOP3(strend, -end_shift, strbeg);
893 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
899 /* If the regex is absolutely anchored to either the start of the
900 * string (SBOL) or to pos() (ANCH_GPOS), then
901 * check_offset_max represents an upper bound on the string where
902 * the substr could start. For the ANCH_GPOS case, we assume that
903 * the caller of intuit will have already set strpos to
904 * pos()-gofs, so in this case strpos + offset_max will still be
905 * an upper bound on the substr.
908 && prog->intflags & PREGf_ANCH
909 && prog->check_offset_max != SSize_t_MAX)
911 SSize_t len = SvCUR(check) - !!SvTAIL(check);
912 const char * const anchor =
913 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
915 /* do a bytes rather than chars comparison. It's conservative;
916 * so it skips doing the HOP if the result can't possibly end
917 * up earlier than the old value of end_point.
919 if ((char*)end_point - anchor > prog->check_offset_max) {
920 end_point = HOP3lim((U8*)anchor,
921 prog->check_offset_max,
927 DEBUG_OPTIMISE_MORE_r({
928 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
929 (int)(end_point - start_point),
930 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
934 check_at = fbm_instr( start_point, end_point,
935 check, multiline ? FBMrf_MULTILINE : 0);
937 /* Update the count-of-usability, remove useless subpatterns,
941 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
942 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
943 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
944 (check_at ? "Found" : "Did not find"),
945 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
946 ? "anchored" : "floating"),
949 (check_at ? " at offset " : "...\n") );
954 /* Finish the diagnostic message */
955 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
957 /* set rx_origin to the minimum position where the regex could start
958 * matching, given the constraint of the just-matched check substring.
959 * But don't set it lower than previously.
962 if (check_at - rx_origin > prog->check_offset_max)
963 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
967 /* now look for the 'other' substring if defined */
969 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
970 : prog->substrs->data[other_ix].substr)
972 /* Take into account the "other" substring. */
976 struct reg_substr_datum *other;
979 other = &prog->substrs->data[other_ix];
981 /* if "other" is anchored:
982 * we've previously found a floating substr starting at check_at.
983 * This means that the regex origin must lie somewhere
984 * between min (rx_origin): HOP3(check_at, -check_offset_max)
985 * and max: HOP3(check_at, -check_offset_min)
986 * (except that min will be >= strpos)
987 * So the fixed substr must lie somewhere between
988 * HOP3(min, anchored_offset)
989 * HOP3(max, anchored_offset) + SvCUR(substr)
992 /* if "other" is floating
993 * Calculate last1, the absolute latest point where the
994 * floating substr could start in the string, ignoring any
995 * constraints from the earlier fixed match. It is calculated
998 * strend - prog->minlen (in chars) is the absolute latest
999 * position within the string where the origin of the regex
1000 * could appear. The latest start point for the floating
1001 * substr is float_min_offset(*) on from the start of the
1002 * regex. last1 simply combines thee two offsets.
1004 * (*) You might think the latest start point should be
1005 * float_max_offset from the regex origin, and technically
1006 * you'd be correct. However, consider
1008 * Here, float min, max are 3,5 and minlen is 7.
1009 * This can match either
1013 * In the first case, the regex matches minlen chars; in the
1014 * second, minlen+1, in the third, minlen+2.
1015 * In the first case, the floating offset is 3 (which equals
1016 * float_min), in the second, 4, and in the third, 5 (which
1017 * equals float_max). In all cases, the floating string bcd
1018 * can never start more than 4 chars from the end of the
1019 * string, which equals minlen - float_min. As the substring
1020 * starts to match more than float_min from the start of the
1021 * regex, it makes the regex match more than minlen chars,
1022 * and the two cancel each other out. So we can always use
1023 * float_min - minlen, rather than float_max - minlen for the
1024 * latest position in the string.
1026 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1027 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1030 assert(prog->minlen >= other->min_offset);
1031 last1 = HOP3c(strend,
1032 other->min_offset - prog->minlen, strbeg);
1034 if (other_ix) {/* i.e. if (other-is-float) */
1035 /* last is the latest point where the floating substr could
1036 * start, *given* any constraints from the earlier fixed
1037 * match. This constraint is that the floating string starts
1038 * <= float_max_offset chars from the regex origin (rx_origin).
1039 * If this value is less than last1, use it instead.
1041 assert(rx_origin <= last1);
1043 /* this condition handles the offset==infinity case, and
1044 * is a short-cut otherwise. Although it's comparing a
1045 * byte offset to a char length, it does so in a safe way,
1046 * since 1 char always occupies 1 or more bytes,
1047 * so if a string range is (last1 - rx_origin) bytes,
1048 * it will be less than or equal to (last1 - rx_origin)
1049 * chars; meaning it errs towards doing the accurate HOP3
1050 * rather than just using last1 as a short-cut */
1051 (last1 - rx_origin) < other->max_offset
1053 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1056 assert(strpos + start_shift <= check_at);
1057 last = HOP4c(check_at, other->min_offset - start_shift,
1061 s = HOP3c(rx_origin, other->min_offset, strend);
1062 if (s < other_last) /* These positions already checked */
1065 must = utf8_target ? other->utf8_substr : other->substr;
1066 assert(SvPOK(must));
1069 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1071 multiline ? FBMrf_MULTILINE : 0
1074 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1075 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1076 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1077 s ? "Found" : "Contradicts",
1078 other_ix ? "floating" : "anchored",
1079 quoted, RE_SV_TAIL(must));
1084 /* last1 is latest possible substr location. If we didn't
1085 * find it before there, we never will */
1086 if (last >= last1) {
1087 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1088 ", giving up...\n"));
1092 /* try to find the check substr again at a later
1093 * position. Maybe next time we'll find the "other" substr
1095 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1096 ", trying %s at offset %ld...\n",
1097 (other_ix ? "floating" : "anchored"),
1098 (long)(HOP3c(check_at, 1, strend) - strpos)));
1100 other_last = HOP3c(last, 1, strend) /* highest failure */;
1102 other_ix /* i.e. if other-is-float */
1103 ? HOP3c(rx_origin, 1, strend)
1104 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1108 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
1109 (long)(s - strpos)));
1111 if (other_ix) { /* if (other-is-float) */
1112 /* other_last is set to s, not s+1, since its possible for
1113 * a floating substr to fail first time, then succeed
1114 * second time at the same floating position; e.g.:
1115 * "-AB--AABZ" =~ /\wAB\d*Z/
1116 * The first time round, anchored and float match at
1117 * "-(AB)--AAB(Z)" then fail on the initial \w character
1118 * class. Second time round, they match at "-AB--A(AB)(Z)".
1123 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1124 other_last = HOP3c(s, 1, strend);
1129 DEBUG_OPTIMISE_MORE_r(
1130 PerlIO_printf(Perl_debug_log,
1131 " Check-only match: offset min:%"IVdf" max:%"IVdf
1132 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
1133 " strend-strpos:%"IVdf"\n",
1134 (IV)prog->check_offset_min,
1135 (IV)prog->check_offset_max,
1136 (IV)(check_at-strpos),
1137 (IV)(rx_origin-strpos),
1138 (IV)(rx_origin-check_at),
1144 postprocess_substr_matches:
1146 /* handle the extra constraint of /^.../m if present */
1148 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1151 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1152 " looking for /^/m anchor"));
1154 /* we have failed the constraint of a \n before rx_origin.
1155 * Find the next \n, if any, even if it's beyond the current
1156 * anchored and/or floating substrings. Whether we should be
1157 * scanning ahead for the next \n or the next substr is debatable.
1158 * On the one hand you'd expect rare substrings to appear less
1159 * often than \n's. On the other hand, searching for \n means
1160 * we're effectively flipping been check_substr and "\n" on each
1161 * iteration as the current "rarest" string candidate, which
1162 * means for example that we'll quickly reject the whole string if
1163 * hasn't got a \n, rather than trying every substr position
1167 s = HOP3c(strend, - prog->minlen, strpos);
1168 if (s <= rx_origin ||
1169 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1171 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1172 " Did not find /%s^%s/m...\n",
1173 PL_colors[0], PL_colors[1]));
1177 /* earliest possible origin is 1 char after the \n.
1178 * (since *rx_origin == '\n', it's safe to ++ here rather than
1179 * HOP(rx_origin, 1)) */
1182 if (prog->substrs->check_ix == 0 /* check is anchored */
1183 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1185 /* Position contradicts check-string; either because
1186 * check was anchored (and thus has no wiggle room),
1187 * or check was float and rx_origin is above the float range */
1188 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1189 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1190 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1194 /* if we get here, the check substr must have been float,
1195 * is in range, and we may or may not have had an anchored
1196 * "other" substr which still contradicts */
1197 assert(prog->substrs->check_ix); /* check is float */
1199 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1200 /* whoops, the anchored "other" substr exists, so we still
1201 * contradict. On the other hand, the float "check" substr
1202 * didn't contradict, so just retry the anchored "other"
1204 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1205 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1206 PL_colors[0], PL_colors[1],
1207 (long)(rx_origin - strpos),
1208 (long)(rx_origin - strpos + prog->anchored_offset)));
1209 goto do_other_substr;
1212 /* success: we don't contradict the found floating substring
1213 * (and there's no anchored substr). */
1214 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1215 " Found /%s^%s/m at offset %ld...\n",
1216 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1219 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1220 " (multiline anchor test skipped)\n"));
1226 /* if we have a starting character class, then test that extra constraint.
1227 * (trie stclasses are too expensive to use here, we are better off to
1228 * leave it to regmatch itself) */
1230 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1231 const U8* const str = (U8*)STRING(progi->regstclass);
1233 /* XXX this value could be pre-computed */
1234 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1235 ? (reginfo->is_utf8_pat
1236 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1237 : STR_LEN(progi->regstclass))
1241 /* latest pos that a matching float substr constrains rx start to */
1242 char *rx_max_float = NULL;
1244 /* if the current rx_origin is anchored, either by satisfying an
1245 * anchored substring constraint, or a /^.../m constraint, then we
1246 * can reject the current origin if the start class isn't found
1247 * at the current position. If we have a float-only match, then
1248 * rx_origin is constrained to a range; so look for the start class
1249 * in that range. if neither, then look for the start class in the
1250 * whole rest of the string */
1252 /* XXX DAPM it's not clear what the minlen test is for, and why
1253 * it's not used in the floating case. Nothing in the test suite
1254 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1255 * Here are some old comments, which may or may not be correct:
1257 * minlen == 0 is possible if regstclass is \b or \B,
1258 * and the fixed substr is ''$.
1259 * Since minlen is already taken into account, rx_origin+1 is
1260 * before strend; accidentally, minlen >= 1 guaranties no false
1261 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1262 * 0) below assumes that regstclass does not come from lookahead...
1263 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1264 * This leaves EXACTF-ish only, which are dealt with in
1268 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1269 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1270 else if (prog->float_substr || prog->float_utf8) {
1271 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1272 endpos= HOP3c(rx_max_float, cl_l, strend);
1277 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1278 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
1279 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1280 (IV)start_shift, (IV)(check_at - strbeg),
1281 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1283 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1286 if (endpos == strend) {
1287 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1288 " Could not match STCLASS...\n") );
1291 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1292 " This position contradicts STCLASS...\n") );
1293 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1294 && !(prog->intflags & PREGf_IMPLICIT))
1297 /* Contradict one of substrings */
1298 if (prog->anchored_substr || prog->anchored_utf8) {
1299 if (prog->substrs->check_ix == 1) { /* check is float */
1300 /* Have both, check_string is floating */
1301 assert(rx_origin + start_shift <= check_at);
1302 if (rx_origin + start_shift != check_at) {
1303 /* not at latest position float substr could match:
1304 * Recheck anchored substring, but not floating.
1305 * The condition above is in bytes rather than
1306 * chars for efficiency. It's conservative, in
1307 * that it errs on the side of doing 'goto
1308 * do_other_substr', where a more accurate
1309 * char-based calculation will be done */
1310 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1311 " Looking for anchored substr starting at offset %ld...\n",
1312 (long)(other_last - strpos)) );
1313 goto do_other_substr;
1321 /* In the presence of ml_anch, we might be able to
1322 * find another \n without breaking the current float
1325 /* strictly speaking this should be HOP3c(..., 1, ...),
1326 * but since we goto a block of code that's going to
1327 * search for the next \n if any, its safe here */
1329 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1330 " Looking for /%s^%s/m starting at offset %ld...\n",
1331 PL_colors[0], PL_colors[1],
1332 (long)(rx_origin - strpos)) );
1333 goto postprocess_substr_matches;
1336 /* strictly speaking this can never be true; but might
1337 * be if we ever allow intuit without substrings */
1338 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1341 rx_origin = rx_max_float;
1344 /* at this point, any matching substrings have been
1345 * contradicted. Start again... */
1347 rx_origin = HOP3c(rx_origin, 1, strend);
1349 /* uses bytes rather than char calculations for efficiency.
1350 * It's conservative: it errs on the side of doing 'goto restart',
1351 * where there is code that does a proper char-based test */
1352 if (rx_origin + start_shift + end_shift > strend) {
1353 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1354 " Could not match STCLASS...\n") );
1357 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1358 " Looking for %s substr starting at offset %ld...\n",
1359 (prog->substrs->check_ix ? "floating" : "anchored"),
1360 (long)(rx_origin + start_shift - strpos)) );
1366 if (rx_origin != s) {
1367 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1368 " By STCLASS: moving %ld --> %ld\n",
1369 (long)(rx_origin - strpos), (long)(s - strpos))
1373 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1374 " Does not contradict STCLASS...\n");
1379 /* Decide whether using the substrings helped */
1381 if (rx_origin != strpos) {
1382 /* Fixed substring is found far enough so that the match
1383 cannot start at strpos. */
1385 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1386 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1389 /* The found rx_origin position does not prohibit matching at
1390 * strpos, so calling intuit didn't gain us anything. Decrement
1391 * the BmUSEFUL() count on the check substring, and if we reach
1393 if (!(prog->intflags & PREGf_NAUGHTY)
1395 prog->check_utf8 /* Could be deleted already */
1396 && --BmUSEFUL(prog->check_utf8) < 0
1397 && (prog->check_utf8 == prog->float_utf8)
1399 prog->check_substr /* Could be deleted already */
1400 && --BmUSEFUL(prog->check_substr) < 0
1401 && (prog->check_substr == prog->float_substr)
1404 /* If flags & SOMETHING - do not do it many times on the same match */
1405 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1406 /* XXX Does the destruction order has to change with utf8_target? */
1407 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1408 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1409 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1410 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1411 check = NULL; /* abort */
1412 /* XXXX This is a remnant of the old implementation. It
1413 looks wasteful, since now INTUIT can use many
1414 other heuristics. */
1415 prog->extflags &= ~RXf_USE_INTUIT;
1419 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1420 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1421 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1425 fail_finish: /* Substring not found */
1426 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1427 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1429 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1430 PL_colors[4], PL_colors[5]));
1435 #define DECL_TRIE_TYPE(scan) \
1436 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1437 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1438 trie_utf8l, trie_flu8 } \
1439 trie_type = ((scan->flags == EXACT) \
1440 ? (utf8_target ? trie_utf8 : trie_plain) \
1441 : (scan->flags == EXACTL) \
1442 ? (utf8_target ? trie_utf8l : trie_plain) \
1443 : (scan->flags == EXACTFA) \
1445 ? trie_utf8_exactfa_fold \
1446 : trie_latin_utf8_exactfa_fold) \
1447 : (scan->flags == EXACTFLU8 \
1451 : trie_latin_utf8_fold)))
1453 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1456 U8 flags = FOLD_FLAGS_FULL; \
1457 switch (trie_type) { \
1459 goto do_trie_utf8_fold; \
1460 case trie_utf8_exactfa_fold: \
1461 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1463 case trie_utf8_fold: \
1464 do_trie_utf8_fold: \
1465 if ( foldlen>0 ) { \
1466 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1471 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1472 len = UTF8SKIP(uc); \
1473 skiplen = UNISKIP( uvc ); \
1474 foldlen -= skiplen; \
1475 uscan = foldbuf + skiplen; \
1478 case trie_latin_utf8_exactfa_fold: \
1479 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1481 case trie_latin_utf8_fold: \
1482 if ( foldlen>0 ) { \
1483 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1489 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1490 skiplen = UNISKIP( uvc ); \
1491 foldlen -= skiplen; \
1492 uscan = foldbuf + skiplen; \
1497 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1504 charid = trie->charmap[ uvc ]; \
1508 if (widecharmap) { \
1509 SV** const svpp = hv_fetch(widecharmap, \
1510 (char*)&uvc, sizeof(UV), 0); \
1512 charid = (U16)SvIV(*svpp); \
1517 #define DUMP_EXEC_POS(li,s,doutf8) \
1518 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1521 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1525 && (ln == 1 || folder(s, pat_string, ln)) \
1526 && (reginfo->intuit || regtry(reginfo, &s)) )\
1532 #define REXEC_FBC_UTF8_SCAN(CODE) \
1534 while (s < strend) { \
1540 #define REXEC_FBC_SCAN(CODE) \
1542 while (s < strend) { \
1548 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1549 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1551 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1560 #define REXEC_FBC_CLASS_SCAN(COND) \
1561 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1563 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1572 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1573 if (utf8_target) { \
1574 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1577 REXEC_FBC_CLASS_SCAN(COND); \
1580 /* The three macros below are slightly different versions of the same logic.
1582 * The first is for /a and /aa when the target string is UTF-8. This can only
1583 * match ascii, but it must advance based on UTF-8. The other two handle the
1584 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1585 * for the boundary (or non-boundary) between a word and non-word character.
1586 * The utf8 and non-utf8 cases have the same logic, but the details must be
1587 * different. Find the "wordness" of the character just prior to this one, and
1588 * compare it with the wordness of this one. If they differ, we have a
1589 * boundary. At the beginning of the string, pretend that the previous
1590 * character was a new-line.
1592 * All these macros uncleanly have side-effects with each other and outside
1593 * variables. So far it's been too much trouble to clean-up
1595 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1596 * a word character or not.
1597 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1599 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1601 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1602 * are looking for a boundary or for a non-boundary. If we are looking for a
1603 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1604 * see if this tentative match actually works, and if so, to quit the loop
1605 * here. And vice-versa if we are looking for a non-boundary.
1607 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1608 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1609 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1610 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1611 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1612 * complement. But in that branch we complement tmp, meaning that at the
1613 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1614 * which means at the top of the loop in the next iteration, it is
1615 * TEST_NON_UTF8(s-1) */
1616 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1617 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1618 tmp = TEST_NON_UTF8(tmp); \
1619 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1620 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1622 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1629 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1630 * TEST_UTF8 is a macro that for the same input code points returns identically
1631 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1632 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1633 if (s == reginfo->strbeg) { \
1636 else { /* Back-up to the start of the previous character */ \
1637 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1638 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1639 0, UTF8_ALLOW_DEFAULT); \
1641 tmp = TEST_UV(tmp); \
1642 LOAD_UTF8_CHARCLASS_ALNUM(); \
1643 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1644 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1653 /* Like the above two macros. UTF8_CODE is the complete code for handling
1654 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1656 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1657 if (utf8_target) { \
1660 else { /* Not utf8 */ \
1661 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1662 tmp = TEST_NON_UTF8(tmp); \
1663 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1664 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1673 /* Here, things have been set up by the previous code so that tmp is the \
1674 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1675 * utf8ness of the target). We also have to check if this matches against \
1676 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1677 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1679 if (tmp == ! TEST_NON_UTF8('\n')) { \
1686 /* This is the macro to use when we want to see if something that looks like it
1687 * could match, actually does, and if so exits the loop */
1688 #define REXEC_FBC_TRYIT \
1689 if ((reginfo->intuit || regtry(reginfo, &s))) \
1692 /* The only difference between the BOUND and NBOUND cases is that
1693 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1694 * NBOUND. This is accomplished by passing it as either the if or else clause,
1695 * with the other one being empty (PLACEHOLDER is defined as empty).
1697 * The TEST_FOO parameters are for operating on different forms of input, but
1698 * all should be ones that return identically for the same underlying code
1700 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1702 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1703 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1705 #define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1707 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1708 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1710 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1712 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1713 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1715 #define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1717 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1718 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1721 /* We know what class REx starts with. Try to find this position... */
1722 /* if reginfo->intuit, its a dryrun */
1723 /* annoyingly all the vars in this routine have different names from their counterparts
1724 in regmatch. /grrr */
1726 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1727 const char *strend, regmatch_info *reginfo)
1730 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1731 char *pat_string; /* The pattern's exactish string */
1732 char *pat_end; /* ptr to end char of pat_string */
1733 re_fold_t folder; /* Function for computing non-utf8 folds */
1734 const U8 *fold_array; /* array for folding ords < 256 */
1740 I32 tmp = 1; /* Scratch variable? */
1741 const bool utf8_target = reginfo->is_utf8_target;
1742 UV utf8_fold_flags = 0;
1743 const bool is_utf8_pat = reginfo->is_utf8_pat;
1744 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1745 with a result inverts that result, as 0^1 =
1747 _char_class_number classnum;
1749 RXi_GET_DECL(prog,progi);
1751 PERL_ARGS_ASSERT_FIND_BYCLASS;
1753 /* We know what class it must start with. */
1758 REXEC_FBC_UTF8_CLASS_SCAN(
1759 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1762 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1767 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1774 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1775 assert(! is_utf8_pat);
1778 if (is_utf8_pat || utf8_target) {
1779 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1780 goto do_exactf_utf8;
1782 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1783 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1784 goto do_exactf_non_utf8; /* isn't dealt with by these */
1786 case EXACTF: /* This node only generated for non-utf8 patterns */
1787 assert(! is_utf8_pat);
1789 utf8_fold_flags = 0;
1790 goto do_exactf_utf8;
1792 fold_array = PL_fold;
1794 goto do_exactf_non_utf8;
1797 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1798 utf8_fold_flags = FOLDEQ_LOCALE;
1799 goto do_exactf_utf8;
1801 fold_array = PL_fold_locale;
1802 folder = foldEQ_locale;
1803 goto do_exactf_non_utf8;
1807 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1809 goto do_exactf_utf8;
1812 if (! utf8_target) { /* All code points in this node require
1813 UTF-8 to express. */
1816 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1817 goto do_exactf_utf8;
1820 if (is_utf8_pat || utf8_target) {
1821 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1822 goto do_exactf_utf8;
1825 /* Any 'ss' in the pattern should have been replaced by regcomp,
1826 * so we don't have to worry here about this single special case
1827 * in the Latin1 range */
1828 fold_array = PL_fold_latin1;
1829 folder = foldEQ_latin1;
1833 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1834 are no glitches with fold-length differences
1835 between the target string and pattern */
1837 /* The idea in the non-utf8 EXACTF* cases is to first find the
1838 * first character of the EXACTF* node and then, if necessary,
1839 * case-insensitively compare the full text of the node. c1 is the
1840 * first character. c2 is its fold. This logic will not work for
1841 * Unicode semantics and the german sharp ss, which hence should
1842 * not be compiled into a node that gets here. */
1843 pat_string = STRING(c);
1844 ln = STR_LEN(c); /* length to match in octets/bytes */
1846 /* We know that we have to match at least 'ln' bytes (which is the
1847 * same as characters, since not utf8). If we have to match 3
1848 * characters, and there are only 2 availabe, we know without
1849 * trying that it will fail; so don't start a match past the
1850 * required minimum number from the far end */
1851 e = HOP3c(strend, -((SSize_t)ln), s);
1853 if (reginfo->intuit && e < s) {
1854 e = s; /* Due to minlen logic of intuit() */
1858 c2 = fold_array[c1];
1859 if (c1 == c2) { /* If char and fold are the same */
1860 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1863 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1871 /* If one of the operands is in utf8, we can't use the simpler folding
1872 * above, due to the fact that many different characters can have the
1873 * same fold, or portion of a fold, or different- length fold */
1874 pat_string = STRING(c);
1875 ln = STR_LEN(c); /* length to match in octets/bytes */
1876 pat_end = pat_string + ln;
1877 lnc = is_utf8_pat /* length to match in characters */
1878 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1881 /* We have 'lnc' characters to match in the pattern, but because of
1882 * multi-character folding, each character in the target can match
1883 * up to 3 characters (Unicode guarantees it will never exceed
1884 * this) if it is utf8-encoded; and up to 2 if not (based on the
1885 * fact that the Latin 1 folds are already determined, and the
1886 * only multi-char fold in that range is the sharp-s folding to
1887 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1888 * string character. Adjust lnc accordingly, rounding up, so that
1889 * if we need to match at least 4+1/3 chars, that really is 5. */
1890 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1891 lnc = (lnc + expansion - 1) / expansion;
1893 /* As in the non-UTF8 case, if we have to match 3 characters, and
1894 * only 2 are left, it's guaranteed to fail, so don't start a
1895 * match that would require us to go beyond the end of the string
1897 e = HOP3c(strend, -((SSize_t)lnc), s);
1899 if (reginfo->intuit && e < s) {
1900 e = s; /* Due to minlen logic of intuit() */
1903 /* XXX Note that we could recalculate e to stop the loop earlier,
1904 * as the worst case expansion above will rarely be met, and as we
1905 * go along we would usually find that e moves further to the left.
1906 * This would happen only after we reached the point in the loop
1907 * where if there were no expansion we should fail. Unclear if
1908 * worth the expense */
1911 char *my_strend= (char *)strend;
1912 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1913 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1914 && (reginfo->intuit || regtry(reginfo, &s)) )
1918 s += (utf8_target) ? UTF8SKIP(s) : 1;
1924 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1927 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1930 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1933 FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1936 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1939 FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1942 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1945 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1948 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1949 is_LNBREAK_latin1_safe(s, strend)
1953 /* The argument to all the POSIX node types is the class number to pass to
1954 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1961 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1962 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1977 /* The complement of something that matches only ASCII matches all
1978 * non-ASCII, plus everything in ASCII that isn't in the class. */
1979 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
1980 || ! _generic_isCC_A(*s, FLAGS(c)));
1989 /* Don't need to worry about utf8, as it can match only a single
1990 * byte invariant character. */
1991 REXEC_FBC_CLASS_SCAN(
1992 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
2000 if (! utf8_target) {
2001 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
2007 classnum = (_char_class_number) FLAGS(c);
2008 if (classnum < _FIRST_NON_SWASH_CC) {
2009 while (s < strend) {
2011 /* We avoid loading in the swash as long as possible, but
2012 * should we have to, we jump to a separate loop. This
2013 * extra 'if' statement is what keeps this code from being
2014 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
2015 if (UTF8_IS_ABOVE_LATIN1(*s)) {
2016 goto found_above_latin1;
2018 if ((UTF8_IS_INVARIANT(*s)
2019 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2021 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2022 && to_complement ^ cBOOL(
2023 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2027 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2039 else switch (classnum) { /* These classes are implemented as
2041 case _CC_ENUM_SPACE: /* XXX would require separate code if we
2042 revert the change of \v matching this */
2045 case _CC_ENUM_PSXSPC:
2046 REXEC_FBC_UTF8_CLASS_SCAN(
2047 to_complement ^ cBOOL(isSPACE_utf8(s)));
2050 case _CC_ENUM_BLANK:
2051 REXEC_FBC_UTF8_CLASS_SCAN(
2052 to_complement ^ cBOOL(isBLANK_utf8(s)));
2055 case _CC_ENUM_XDIGIT:
2056 REXEC_FBC_UTF8_CLASS_SCAN(
2057 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2060 case _CC_ENUM_VERTSPACE:
2061 REXEC_FBC_UTF8_CLASS_SCAN(
2062 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2065 case _CC_ENUM_CNTRL:
2066 REXEC_FBC_UTF8_CLASS_SCAN(
2067 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2071 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2072 NOT_REACHED; /* NOTREACHED */
2077 found_above_latin1: /* Here we have to load a swash to get the result
2078 for the current code point */
2079 if (! PL_utf8_swash_ptrs[classnum]) {
2080 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2081 PL_utf8_swash_ptrs[classnum] =
2082 _core_swash_init("utf8",
2085 PL_XPosix_ptrs[classnum], &flags);
2088 /* This is a copy of the loop above for swash classes, though using the
2089 * FBC macro instead of being expanded out. Since we've loaded the
2090 * swash, we don't have to check for that each time through the loop */
2091 REXEC_FBC_UTF8_CLASS_SCAN(
2092 to_complement ^ cBOOL(_generic_utf8(
2095 swash_fetch(PL_utf8_swash_ptrs[classnum],
2103 /* what trie are we using right now */
2104 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2105 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2106 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2108 const char *last_start = strend - trie->minlen;
2110 const char *real_start = s;
2112 STRLEN maxlen = trie->maxlen;
2114 U8 **points; /* map of where we were in the input string
2115 when reading a given char. For ASCII this
2116 is unnecessary overhead as the relationship
2117 is always 1:1, but for Unicode, especially
2118 case folded Unicode this is not true. */
2119 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2123 GET_RE_DEBUG_FLAGS_DECL;
2125 /* We can't just allocate points here. We need to wrap it in
2126 * an SV so it gets freed properly if there is a croak while
2127 * running the match */
2130 sv_points=newSV(maxlen * sizeof(U8 *));
2131 SvCUR_set(sv_points,
2132 maxlen * sizeof(U8 *));
2133 SvPOK_on(sv_points);
2134 sv_2mortal(sv_points);
2135 points=(U8**)SvPV_nolen(sv_points );
2136 if ( trie_type != trie_utf8_fold
2137 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2140 bitmap=(U8*)trie->bitmap;
2142 bitmap=(U8*)ANYOF_BITMAP(c);
2144 /* this is the Aho-Corasick algorithm modified a touch
2145 to include special handling for long "unknown char" sequences.
2146 The basic idea being that we use AC as long as we are dealing
2147 with a possible matching char, when we encounter an unknown char
2148 (and we have not encountered an accepting state) we scan forward
2149 until we find a legal starting char.
2150 AC matching is basically that of trie matching, except that when
2151 we encounter a failing transition, we fall back to the current
2152 states "fail state", and try the current char again, a process
2153 we repeat until we reach the root state, state 1, or a legal
2154 transition. If we fail on the root state then we can either
2155 terminate if we have reached an accepting state previously, or
2156 restart the entire process from the beginning if we have not.
2159 while (s <= last_start) {
2160 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2168 U8 *uscan = (U8*)NULL;
2169 U8 *leftmost = NULL;
2171 U32 accepted_word= 0;
2175 while ( state && uc <= (U8*)strend ) {
2177 U32 word = aho->states[ state ].wordnum;
2181 DEBUG_TRIE_EXECUTE_r(
2182 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2183 dump_exec_pos( (char *)uc, c, strend, real_start,
2184 (char *)uc, utf8_target );
2185 PerlIO_printf( Perl_debug_log,
2186 " Scanning for legal start char...\n");
2190 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2194 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2200 if (uc >(U8*)last_start) break;
2204 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2205 if (!leftmost || lpos < leftmost) {
2206 DEBUG_r(accepted_word=word);
2212 points[pointpos++ % maxlen]= uc;
2213 if (foldlen || uc < (U8*)strend) {
2214 REXEC_TRIE_READ_CHAR(trie_type, trie,
2216 uscan, len, uvc, charid, foldlen,
2218 DEBUG_TRIE_EXECUTE_r({
2219 dump_exec_pos( (char *)uc, c, strend,
2220 real_start, s, utf8_target);
2221 PerlIO_printf(Perl_debug_log,
2222 " Charid:%3u CP:%4"UVxf" ",
2234 word = aho->states[ state ].wordnum;
2236 base = aho->states[ state ].trans.base;
2238 DEBUG_TRIE_EXECUTE_r({
2240 dump_exec_pos( (char *)uc, c, strend, real_start,
2242 PerlIO_printf( Perl_debug_log,
2243 "%sState: %4"UVxf", word=%"UVxf,
2244 failed ? " Fail transition to " : "",
2245 (UV)state, (UV)word);
2251 ( ((offset = base + charid
2252 - 1 - trie->uniquecharcount)) >= 0)
2253 && ((U32)offset < trie->lasttrans)
2254 && trie->trans[offset].check == state
2255 && (tmp=trie->trans[offset].next))
2257 DEBUG_TRIE_EXECUTE_r(
2258 PerlIO_printf( Perl_debug_log," - legal\n"));
2263 DEBUG_TRIE_EXECUTE_r(
2264 PerlIO_printf( Perl_debug_log," - fail\n"));
2266 state = aho->fail[state];
2270 /* we must be accepting here */
2271 DEBUG_TRIE_EXECUTE_r(
2272 PerlIO_printf( Perl_debug_log," - accepting\n"));
2281 if (!state) state = 1;
2284 if ( aho->states[ state ].wordnum ) {
2285 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2286 if (!leftmost || lpos < leftmost) {
2287 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2292 s = (char*)leftmost;
2293 DEBUG_TRIE_EXECUTE_r({
2295 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2296 (UV)accepted_word, (IV)(s - real_start)
2299 if (reginfo->intuit || regtry(reginfo, &s)) {
2305 DEBUG_TRIE_EXECUTE_r({
2306 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2309 DEBUG_TRIE_EXECUTE_r(
2310 PerlIO_printf( Perl_debug_log,"No match.\n"));
2319 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2326 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2327 * flags have same meanings as with regexec_flags() */
2330 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2337 struct regexp *const prog = ReANY(rx);
2339 if (flags & REXEC_COPY_STR) {
2343 PerlIO_printf(Perl_debug_log,
2344 "Copy on write: regexp capture, type %d\n",
2347 /* Create a new COW SV to share the match string and store
2348 * in saved_copy, unless the current COW SV in saved_copy
2349 * is valid and suitable for our purpose */
2350 if (( prog->saved_copy
2351 && SvIsCOW(prog->saved_copy)
2352 && SvPOKp(prog->saved_copy)
2355 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2357 /* just reuse saved_copy SV */
2358 if (RXp_MATCH_COPIED(prog)) {
2359 Safefree(prog->subbeg);
2360 RXp_MATCH_COPIED_off(prog);
2364 /* create new COW SV to share string */
2365 RX_MATCH_COPY_FREE(rx);
2366 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2368 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2369 assert (SvPOKp(prog->saved_copy));
2370 prog->sublen = strend - strbeg;
2371 prog->suboffset = 0;
2372 prog->subcoffset = 0;
2377 SSize_t max = strend - strbeg;
2380 if ( (flags & REXEC_COPY_SKIP_POST)
2381 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2382 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2383 ) { /* don't copy $' part of string */
2386 /* calculate the right-most part of the string covered
2387 * by a capture. Due to look-ahead, this may be to
2388 * the right of $&, so we have to scan all captures */
2389 while (n <= prog->lastparen) {
2390 if (prog->offs[n].end > max)
2391 max = prog->offs[n].end;
2395 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2396 ? prog->offs[0].start
2398 assert(max >= 0 && max <= strend - strbeg);
2401 if ( (flags & REXEC_COPY_SKIP_PRE)
2402 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2403 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2404 ) { /* don't copy $` part of string */
2407 /* calculate the left-most part of the string covered
2408 * by a capture. Due to look-behind, this may be to
2409 * the left of $&, so we have to scan all captures */
2410 while (min && n <= prog->lastparen) {
2411 if ( prog->offs[n].start != -1
2412 && prog->offs[n].start < min)
2414 min = prog->offs[n].start;
2418 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2419 && min > prog->offs[0].end
2421 min = prog->offs[0].end;
2425 assert(min >= 0 && min <= max && min <= strend - strbeg);
2428 if (RX_MATCH_COPIED(rx)) {
2429 if (sublen > prog->sublen)
2431 (char*)saferealloc(prog->subbeg, sublen+1);
2434 prog->subbeg = (char*)safemalloc(sublen+1);
2435 Copy(strbeg + min, prog->subbeg, sublen, char);
2436 prog->subbeg[sublen] = '\0';
2437 prog->suboffset = min;
2438 prog->sublen = sublen;
2439 RX_MATCH_COPIED_on(rx);
2441 prog->subcoffset = prog->suboffset;
2442 if (prog->suboffset && utf8_target) {
2443 /* Convert byte offset to chars.
2444 * XXX ideally should only compute this if @-/@+
2445 * has been seen, a la PL_sawampersand ??? */
2447 /* If there's a direct correspondence between the
2448 * string which we're matching and the original SV,
2449 * then we can use the utf8 len cache associated with
2450 * the SV. In particular, it means that under //g,
2451 * sv_pos_b2u() will use the previously cached
2452 * position to speed up working out the new length of
2453 * subcoffset, rather than counting from the start of
2454 * the string each time. This stops
2455 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2456 * from going quadratic */
2457 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2458 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2459 SV_GMAGIC|SV_CONST_RETURN);
2461 prog->subcoffset = utf8_length((U8*)strbeg,
2462 (U8*)(strbeg+prog->suboffset));
2466 RX_MATCH_COPY_FREE(rx);
2467 prog->subbeg = strbeg;
2468 prog->suboffset = 0;
2469 prog->subcoffset = 0;
2470 prog->sublen = strend - strbeg;
2478 - regexec_flags - match a regexp against a string
2481 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2482 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2483 /* stringarg: the point in the string at which to begin matching */
2484 /* strend: pointer to null at end of string */
2485 /* strbeg: real beginning of string */
2486 /* minend: end of match must be >= minend bytes after stringarg. */
2487 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2488 * itself is accessed via the pointers above */
2489 /* data: May be used for some additional optimizations.
2490 Currently unused. */
2491 /* flags: For optimizations. See REXEC_* in regexp.h */
2494 struct regexp *const prog = ReANY(rx);
2498 SSize_t minlen; /* must match at least this many chars */
2499 SSize_t dontbother = 0; /* how many characters not to try at end */
2500 const bool utf8_target = cBOOL(DO_UTF8(sv));
2502 RXi_GET_DECL(prog,progi);
2503 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2504 regmatch_info *const reginfo = ®info_buf;
2505 regexp_paren_pair *swap = NULL;
2507 GET_RE_DEBUG_FLAGS_DECL;
2509 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2510 PERL_UNUSED_ARG(data);
2512 /* Be paranoid... */
2513 if (prog == NULL || stringarg == NULL) {
2514 Perl_croak(aTHX_ "NULL regexp parameter");
2518 debug_start_match(rx, utf8_target, stringarg, strend,
2522 startpos = stringarg;
2524 if (prog->intflags & PREGf_GPOS_SEEN) {
2527 /* set reginfo->ganch, the position where \G can match */
2530 (flags & REXEC_IGNOREPOS)
2531 ? stringarg /* use start pos rather than pos() */
2532 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2533 /* Defined pos(): */
2534 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2535 : strbeg; /* pos() not defined; use start of string */
2537 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2538 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
2540 /* in the presence of \G, we may need to start looking earlier in
2541 * the string than the suggested start point of stringarg:
2542 * if prog->gofs is set, then that's a known, fixed minimum
2545 * /ab|c\G/: gofs = 1
2546 * or if the minimum offset isn't known, then we have to go back
2547 * to the start of the string, e.g. /w+\G/
2550 if (prog->intflags & PREGf_ANCH_GPOS) {
2551 startpos = reginfo->ganch - prog->gofs;
2553 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2555 DEBUG_r(PerlIO_printf(Perl_debug_log,
2556 "fail: ganch-gofs before earliest possible start\n"));
2560 else if (prog->gofs) {
2561 if (startpos - prog->gofs < strbeg)
2564 startpos -= prog->gofs;
2566 else if (prog->intflags & PREGf_GPOS_FLOAT)
2570 minlen = prog->minlen;
2571 if ((startpos + minlen) > strend || startpos < strbeg) {
2572 DEBUG_r(PerlIO_printf(Perl_debug_log,
2573 "Regex match can't succeed, so not even tried\n"));
2577 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2578 * which will call destuctors to reset PL_regmatch_state, free higher
2579 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2580 * regmatch_info_aux_eval */
2582 oldsave = PL_savestack_ix;
2586 if ((prog->extflags & RXf_USE_INTUIT)
2587 && !(flags & REXEC_CHECKED))
2589 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2594 if (prog->extflags & RXf_CHECK_ALL) {
2595 /* we can match based purely on the result of INTUIT.
2596 * Set up captures etc just for $& and $-[0]
2597 * (an intuit-only match wont have $1,$2,..) */
2598 assert(!prog->nparens);
2600 /* s/// doesn't like it if $& is earlier than where we asked it to
2601 * start searching (which can happen on something like /.\G/) */
2602 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2605 /* this should only be possible under \G */
2606 assert(prog->intflags & PREGf_GPOS_SEEN);
2607 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2608 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2612 /* match via INTUIT shouldn't have any captures.
2613 * Let @-, @+, $^N know */
2614 prog->lastparen = prog->lastcloseparen = 0;
2615 RX_MATCH_UTF8_set(rx, utf8_target);
2616 prog->offs[0].start = s - strbeg;
2617 prog->offs[0].end = utf8_target
2618 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2619 : s - strbeg + prog->minlenret;
2620 if ( !(flags & REXEC_NOT_FIRST) )
2621 S_reg_set_capture_string(aTHX_ rx,
2623 sv, flags, utf8_target);
2629 multiline = prog->extflags & RXf_PMf_MULTILINE;
2631 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2632 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2633 "String too short [regexec_flags]...\n"));
2637 /* Check validity of program. */
2638 if (UCHARAT(progi->program) != REG_MAGIC) {
2639 Perl_croak(aTHX_ "corrupted regexp program");
2642 RX_MATCH_TAINTED_off(rx);
2643 RX_MATCH_UTF8_set(rx, utf8_target);
2645 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2646 reginfo->intuit = 0;
2647 reginfo->is_utf8_target = cBOOL(utf8_target);
2648 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2649 reginfo->warned = FALSE;
2650 reginfo->strbeg = strbeg;
2652 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2653 reginfo->strend = strend;
2654 /* see how far we have to get to not match where we matched before */
2655 reginfo->till = stringarg + minend;
2657 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
2658 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2659 S_cleanup_regmatch_info_aux has executed (registered by
2660 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2661 magic belonging to this SV.
2662 Not newSVsv, either, as it does not COW.
2664 reginfo->sv = newSV(0);
2665 SvSetSV_nosteal(reginfo->sv, sv);
2666 SAVEFREESV(reginfo->sv);
2669 /* reserve next 2 or 3 slots in PL_regmatch_state:
2670 * slot N+0: may currently be in use: skip it
2671 * slot N+1: use for regmatch_info_aux struct
2672 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2673 * slot N+3: ready for use by regmatch()
2677 regmatch_state *old_regmatch_state;
2678 regmatch_slab *old_regmatch_slab;
2679 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2681 /* on first ever match, allocate first slab */
2682 if (!PL_regmatch_slab) {
2683 Newx(PL_regmatch_slab, 1, regmatch_slab);
2684 PL_regmatch_slab->prev = NULL;
2685 PL_regmatch_slab->next = NULL;
2686 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2689 old_regmatch_state = PL_regmatch_state;
2690 old_regmatch_slab = PL_regmatch_slab;
2692 for (i=0; i <= max; i++) {
2694 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2696 reginfo->info_aux_eval =
2697 reginfo->info_aux->info_aux_eval =
2698 &(PL_regmatch_state->u.info_aux_eval);
2700 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2701 PL_regmatch_state = S_push_slab(aTHX);
2704 /* note initial PL_regmatch_state position; at end of match we'll
2705 * pop back to there and free any higher slabs */
2707 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2708 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2709 reginfo->info_aux->poscache = NULL;
2711 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2713 if ((prog->extflags & RXf_EVAL_SEEN))
2714 S_setup_eval_state(aTHX_ reginfo);
2716 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2719 /* If there is a "must appear" string, look for it. */
2721 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2722 /* We have to be careful. If the previous successful match
2723 was from this regex we don't want a subsequent partially
2724 successful match to clobber the old results.
2725 So when we detect this possibility we add a swap buffer
2726 to the re, and switch the buffer each match. If we fail,
2727 we switch it back; otherwise we leave it swapped.
2730 /* do we need a save destructor here for eval dies? */
2731 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2732 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2733 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2740 /* Simplest case: anchored match need be tried only once. */
2741 /* [unless only anchor is MBOL - implying multiline is set] */
2742 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
2743 if (s == startpos && regtry(reginfo, &s))
2745 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
2750 dontbother = minlen - 1;
2751 end = HOP3c(strend, -dontbother, strbeg) - 1;
2752 /* for multiline we only have to try after newlines */
2753 if (prog->check_substr || prog->check_utf8) {
2754 /* because of the goto we can not easily reuse the macros for bifurcating the
2755 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2758 goto after_try_utf8;
2760 if (regtry(reginfo, &s)) {
2767 if (prog->extflags & RXf_USE_INTUIT) {
2768 s = re_intuit_start(rx, sv, strbeg,
2769 s + UTF8SKIP(s), strend, flags, NULL);
2778 } /* end search for check string in unicode */
2780 if (s == startpos) {
2781 goto after_try_latin;
2784 if (regtry(reginfo, &s)) {
2791 if (prog->extflags & RXf_USE_INTUIT) {
2792 s = re_intuit_start(rx, sv, strbeg,
2793 s + 1, strend, flags, NULL);
2802 } /* end search for check string in latin*/
2803 } /* end search for check string */
2804 else { /* search for newline */
2806 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2809 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2810 while (s <= end) { /* note it could be possible to match at the end of the string */
2811 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2812 if (regtry(reginfo, &s))
2816 } /* end search for newline */
2817 } /* end anchored/multiline check string search */
2819 } else if (prog->intflags & PREGf_ANCH_GPOS)
2821 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2822 assert(prog->intflags & PREGf_GPOS_SEEN);
2823 /* For anchored \G, the only position it can match from is
2824 * (ganch-gofs); we already set startpos to this above; if intuit
2825 * moved us on from there, we can't possibly succeed */
2826 assert(startpos == reginfo->ganch - prog->gofs);
2827 if (s == startpos && regtry(reginfo, &s))
2832 /* Messy cases: unanchored match. */
2833 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2834 /* we have /x+whatever/ */
2835 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2841 if (! prog->anchored_utf8) {
2842 to_utf8_substr(prog);
2844 ch = SvPVX_const(prog->anchored_utf8)[0];
2847 DEBUG_EXECUTE_r( did_match = 1 );
2848 if (regtry(reginfo, &s)) goto got_it;
2850 while (s < strend && *s == ch)
2857 if (! prog->anchored_substr) {
2858 if (! to_byte_substr(prog)) {
2859 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2862 ch = SvPVX_const(prog->anchored_substr)[0];
2865 DEBUG_EXECUTE_r( did_match = 1 );
2866 if (regtry(reginfo, &s)) goto got_it;
2868 while (s < strend && *s == ch)
2873 DEBUG_EXECUTE_r(if (!did_match)
2874 PerlIO_printf(Perl_debug_log,
2875 "Did not find anchored character...\n")
2878 else if (prog->anchored_substr != NULL
2879 || prog->anchored_utf8 != NULL
2880 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2881 && prog->float_max_offset < strend - s)) {
2886 char *last1; /* Last position checked before */
2890 if (prog->anchored_substr || prog->anchored_utf8) {
2892 if (! prog->anchored_utf8) {
2893 to_utf8_substr(prog);
2895 must = prog->anchored_utf8;
2898 if (! prog->anchored_substr) {
2899 if (! to_byte_substr(prog)) {
2900 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2903 must = prog->anchored_substr;
2905 back_max = back_min = prog->anchored_offset;
2908 if (! prog->float_utf8) {
2909 to_utf8_substr(prog);
2911 must = prog->float_utf8;
2914 if (! prog->float_substr) {
2915 if (! to_byte_substr(prog)) {
2916 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2919 must = prog->float_substr;
2921 back_max = prog->float_max_offset;
2922 back_min = prog->float_min_offset;
2928 last = HOP3c(strend, /* Cannot start after this */
2929 -(SSize_t)(CHR_SVLEN(must)
2930 - (SvTAIL(must) != 0) + back_min), strbeg);
2932 if (s > reginfo->strbeg)
2933 last1 = HOPc(s, -1);
2935 last1 = s - 1; /* bogus */
2937 /* XXXX check_substr already used to find "s", can optimize if
2938 check_substr==must. */
2940 strend = HOPc(strend, -dontbother);
2941 while ( (s <= last) &&
2942 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
2943 (unsigned char*)strend, must,
2944 multiline ? FBMrf_MULTILINE : 0)) ) {
2945 DEBUG_EXECUTE_r( did_match = 1 );
2946 if (HOPc(s, -back_max) > last1) {
2947 last1 = HOPc(s, -back_min);
2948 s = HOPc(s, -back_max);
2951 char * const t = (last1 >= reginfo->strbeg)
2952 ? HOPc(last1, 1) : last1 + 1;
2954 last1 = HOPc(s, -back_min);
2958 while (s <= last1) {
2959 if (regtry(reginfo, &s))
2962 s++; /* to break out of outer loop */
2969 while (s <= last1) {
2970 if (regtry(reginfo, &s))
2976 DEBUG_EXECUTE_r(if (!did_match) {
2977 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2978 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2979 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2980 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2981 ? "anchored" : "floating"),
2982 quoted, RE_SV_TAIL(must));
2986 else if ( (c = progi->regstclass) ) {
2988 const OPCODE op = OP(progi->regstclass);
2989 /* don't bother with what can't match */
2990 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2991 strend = HOPc(strend, -(minlen - 1));
2994 SV * const prop = sv_newmortal();
2995 regprop(prog, prop, c, reginfo, NULL);
2997 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2999 PerlIO_printf(Perl_debug_log,
3000 "Matching stclass %.*s against %s (%d bytes)\n",
3001 (int)SvCUR(prop), SvPVX_const(prop),
3002 quoted, (int)(strend - s));
3005 if (find_byclass(prog, c, s, strend, reginfo))
3007 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
3011 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3019 if (! prog->float_utf8) {
3020 to_utf8_substr(prog);
3022 float_real = prog->float_utf8;
3025 if (! prog->float_substr) {
3026 if (! to_byte_substr(prog)) {
3027 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3030 float_real = prog->float_substr;
3033 little = SvPV_const(float_real, len);
3034 if (SvTAIL(float_real)) {
3035 /* This means that float_real contains an artificial \n on
3036 * the end due to the presence of something like this:
3037 * /foo$/ where we can match both "foo" and "foo\n" at the
3038 * end of the string. So we have to compare the end of the
3039 * string first against the float_real without the \n and
3040 * then against the full float_real with the string. We
3041 * have to watch out for cases where the string might be
3042 * smaller than the float_real or the float_real without
3044 char *checkpos= strend - len;
3046 PerlIO_printf(Perl_debug_log,
3047 "%sChecking for float_real.%s\n",
3048 PL_colors[4], PL_colors[5]));
3049 if (checkpos + 1 < strbeg) {
3050 /* can't match, even if we remove the trailing \n
3051 * string is too short to match */
3053 PerlIO_printf(Perl_debug_log,
3054 "%sString shorter than required trailing substring, cannot match.%s\n",
3055 PL_colors[4], PL_colors[5]));
3057 } else if (memEQ(checkpos + 1, little, len - 1)) {
3058 /* can match, the end of the string matches without the
3060 last = checkpos + 1;
3061 } else if (checkpos < strbeg) {
3062 /* cant match, string is too short when the "\n" is
3065 PerlIO_printf(Perl_debug_log,
3066 "%sString does not contain required trailing substring, cannot match.%s\n",
3067 PL_colors[4], PL_colors[5]));
3069 } else if (!multiline) {
3070 /* non multiline match, so compare with the "\n" at the
3071 * end of the string */
3072 if (memEQ(checkpos, little, len)) {
3076 PerlIO_printf(Perl_debug_log,
3077 "%sString does not contain required trailing substring, cannot match.%s\n",
3078 PL_colors[4], PL_colors[5]));
3082 /* multiline match, so we have to search for a place
3083 * where the full string is located */
3089 last = rninstr(s, strend, little, little + len);
3091 last = strend; /* matching "$" */
3094 /* at one point this block contained a comment which was
3095 * probably incorrect, which said that this was a "should not
3096 * happen" case. Even if it was true when it was written I am
3097 * pretty sure it is not anymore, so I have removed the comment
3098 * and replaced it with this one. Yves */
3100 PerlIO_printf(Perl_debug_log,
3101 "%sString does not contain required substring, cannot match.%s\n",
3102 PL_colors[4], PL_colors[5]
3106 dontbother = strend - last + prog->float_min_offset;
3108 if (minlen && (dontbother < minlen))
3109 dontbother = minlen - 1;
3110 strend -= dontbother; /* this one's always in bytes! */
3111 /* We don't know much -- general case. */
3114 if (regtry(reginfo, &s))
3123 if (regtry(reginfo, &s))
3125 } while (s++ < strend);
3133 /* s/// doesn't like it if $& is earlier than where we asked it to
3134 * start searching (which can happen on something like /.\G/) */
3135 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3136 && (prog->offs[0].start < stringarg - strbeg))
3138 /* this should only be possible under \G */
3139 assert(prog->intflags & PREGf_GPOS_SEEN);
3140 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3141 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3147 PerlIO_printf(Perl_debug_log,
3148 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3155 /* clean up; this will trigger destructors that will free all slabs
3156 * above the current one, and cleanup the regmatch_info_aux
3157 * and regmatch_info_aux_eval sructs */
3159 LEAVE_SCOPE(oldsave);
3161 if (RXp_PAREN_NAMES(prog))
3162 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3164 /* make sure $`, $&, $', and $digit will work later */
3165 if ( !(flags & REXEC_NOT_FIRST) )
3166 S_reg_set_capture_string(aTHX_ rx,
3167 strbeg, reginfo->strend,
3168 sv, flags, utf8_target);
3173 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
3174 PL_colors[4], PL_colors[5]));
3176 /* clean up; this will trigger destructors that will free all slabs
3177 * above the current one, and cleanup the regmatch_info_aux
3178 * and regmatch_info_aux_eval sructs */
3180 LEAVE_SCOPE(oldsave);
3183 /* we failed :-( roll it back */
3184 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3185 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3190 Safefree(prog->offs);
3197 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3198 * Do inc before dec, in case old and new rex are the same */
3199 #define SET_reg_curpm(Re2) \
3200 if (reginfo->info_aux_eval) { \
3201 (void)ReREFCNT_inc(Re2); \
3202 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3203 PM_SETRE((PL_reg_curpm), (Re2)); \
3208 - regtry - try match at specific point
3210 STATIC I32 /* 0 failure, 1 success */
3211 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3214 REGEXP *const rx = reginfo->prog;
3215 regexp *const prog = ReANY(rx);
3217 RXi_GET_DECL(prog,progi);
3218 GET_RE_DEBUG_FLAGS_DECL;
3220 PERL_ARGS_ASSERT_REGTRY;
3222 reginfo->cutpoint=NULL;
3224 prog->offs[0].start = *startposp - reginfo->strbeg;
3225 prog->lastparen = 0;
3226 prog->lastcloseparen = 0;
3228 /* XXXX What this code is doing here?!!! There should be no need
3229 to do this again and again, prog->lastparen should take care of
3232 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3233 * Actually, the code in regcppop() (which Ilya may be meaning by
3234 * prog->lastparen), is not needed at all by the test suite
3235 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3236 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3237 * Meanwhile, this code *is* needed for the
3238 * above-mentioned test suite tests to succeed. The common theme
3239 * on those tests seems to be returning null fields from matches.
3240 * --jhi updated by dapm */
3242 if (prog->nparens) {
3243 regexp_paren_pair *pp = prog->offs;
3245 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3253 result = regmatch(reginfo, *startposp, progi->program + 1);
3255 prog->offs[0].end = result;
3258 if (reginfo->cutpoint)
3259 *startposp= reginfo->cutpoint;
3260 REGCP_UNWIND(lastcp);
3265 #define sayYES goto yes
3266 #define sayNO goto no
3267 #define sayNO_SILENT goto no_silent
3269 /* we dont use STMT_START/END here because it leads to
3270 "unreachable code" warnings, which are bogus, but distracting. */
3271 #define CACHEsayNO \
3272 if (ST.cache_mask) \
3273 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3276 /* this is used to determine how far from the left messages like
3277 'failed...' are printed. It should be set such that messages
3278 are inline with the regop output that created them.
3280 #define REPORT_CODE_OFF 32
3283 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3284 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3285 #define CHRTEST_NOT_A_CP_1 -999
3286 #define CHRTEST_NOT_A_CP_2 -998
3288 /* grab a new slab and return the first slot in it */
3290 STATIC regmatch_state *
3293 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3296 regmatch_slab *s = PL_regmatch_slab->next;
3298 Newx(s, 1, regmatch_slab);
3299 s->prev = PL_regmatch_slab;
3301 PL_regmatch_slab->next = s;
3303 PL_regmatch_slab = s;
3304 return SLAB_FIRST(s);
3308 /* push a new state then goto it */
3310 #define PUSH_STATE_GOTO(state, node, input) \
3311 pushinput = input; \
3313 st->resume_state = state; \
3316 /* push a new state with success backtracking, then goto it */
3318 #define PUSH_YES_STATE_GOTO(state, node, input) \
3319 pushinput = input; \
3321 st->resume_state = state; \
3322 goto push_yes_state;
3329 regmatch() - main matching routine
3331 This is basically one big switch statement in a loop. We execute an op,
3332 set 'next' to point the next op, and continue. If we come to a point which
3333 we may need to backtrack to on failure such as (A|B|C), we push a
3334 backtrack state onto the backtrack stack. On failure, we pop the top
3335 state, and re-enter the loop at the state indicated. If there are no more
3336 states to pop, we return failure.
3338 Sometimes we also need to backtrack on success; for example /A+/, where
3339 after successfully matching one A, we need to go back and try to
3340 match another one; similarly for lookahead assertions: if the assertion
3341 completes successfully, we backtrack to the state just before the assertion
3342 and then carry on. In these cases, the pushed state is marked as
3343 'backtrack on success too'. This marking is in fact done by a chain of
3344 pointers, each pointing to the previous 'yes' state. On success, we pop to
3345 the nearest yes state, discarding any intermediate failure-only states.
3346 Sometimes a yes state is pushed just to force some cleanup code to be
3347 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3348 it to free the inner regex.
3350 Note that failure backtracking rewinds the cursor position, while
3351 success backtracking leaves it alone.
3353 A pattern is complete when the END op is executed, while a subpattern
3354 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3355 ops trigger the "pop to last yes state if any, otherwise return true"
3358 A common convention in this function is to use A and B to refer to the two
3359 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3360 the subpattern to be matched possibly multiple times, while B is the entire
3361 rest of the pattern. Variable and state names reflect this convention.
3363 The states in the main switch are the union of ops and failure/success of
3364 substates associated with with that op. For example, IFMATCH is the op
3365 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3366 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3367 successfully matched A and IFMATCH_A_fail is a state saying that we have
3368 just failed to match A. Resume states always come in pairs. The backtrack
3369 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3370 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3371 on success or failure.
3373 The struct that holds a backtracking state is actually a big union, with
3374 one variant for each major type of op. The variable st points to the
3375 top-most backtrack struct. To make the code clearer, within each
3376 block of code we #define ST to alias the relevant union.
3378 Here's a concrete example of a (vastly oversimplified) IFMATCH
3384 #define ST st->u.ifmatch
3386 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3387 ST.foo = ...; // some state we wish to save
3389 // push a yes backtrack state with a resume value of
3390 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3392 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3395 case IFMATCH_A: // we have successfully executed A; now continue with B
3397 bar = ST.foo; // do something with the preserved value
3400 case IFMATCH_A_fail: // A failed, so the assertion failed
3401 ...; // do some housekeeping, then ...
3402 sayNO; // propagate the failure
3409 For any old-timers reading this who are familiar with the old recursive
3410 approach, the code above is equivalent to:
3412 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3421 ...; // do some housekeeping, then ...
3422 sayNO; // propagate the failure
3425 The topmost backtrack state, pointed to by st, is usually free. If you
3426 want to claim it, populate any ST.foo fields in it with values you wish to
3427 save, then do one of
3429 PUSH_STATE_GOTO(resume_state, node, newinput);
3430 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3432 which sets that backtrack state's resume value to 'resume_state', pushes a
3433 new free entry to the top of the backtrack stack, then goes to 'node'.
3434 On backtracking, the free slot is popped, and the saved state becomes the
3435 new free state. An ST.foo field in this new top state can be temporarily
3436 accessed to retrieve values, but once the main loop is re-entered, it
3437 becomes available for reuse.
3439 Note that the depth of the backtrack stack constantly increases during the
3440 left-to-right execution of the pattern, rather than going up and down with
3441 the pattern nesting. For example the stack is at its maximum at Z at the
3442 end of the pattern, rather than at X in the following:
3444 /(((X)+)+)+....(Y)+....Z/
3446 The only exceptions to this are lookahead/behind assertions and the cut,
3447 (?>A), which pop all the backtrack states associated with A before
3450 Backtrack state structs are allocated in slabs of about 4K in size.
3451 PL_regmatch_state and st always point to the currently active state,
3452 and PL_regmatch_slab points to the slab currently containing
3453 PL_regmatch_state. The first time regmatch() is called, the first slab is
3454 allocated, and is never freed until interpreter destruction. When the slab
3455 is full, a new one is allocated and chained to the end. At exit from
3456 regmatch(), slabs allocated since entry are freed.
3461 #define DEBUG_STATE_pp(pp) \
3463 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3464 PerlIO_printf(Perl_debug_log, \
3465 " %*s"pp" %s%s%s%s%s\n", \
3467 PL_reg_name[st->resume_state], \
3468 ((st==yes_state||st==mark_state) ? "[" : ""), \
3469 ((st==yes_state) ? "Y" : ""), \
3470 ((st==mark_state) ? "M" : ""), \
3471 ((st==yes_state||st==mark_state) ? "]" : "") \
3476 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3481 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3482 const char *start, const char *end, const char *blurb)
3484 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3486 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3491 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3492 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3494 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3495 start, end - start, 60);
3497 PerlIO_printf(Perl_debug_log,
3498 "%s%s REx%s %s against %s\n",
3499 PL_colors[4], blurb, PL_colors[5], s0, s1);
3501 if (utf8_target||utf8_pat)
3502 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3503 utf8_pat ? "pattern" : "",
3504 utf8_pat && utf8_target ? " and " : "",
3505 utf8_target ? "string" : ""
3511 S_dump_exec_pos(pTHX_ const char *locinput,
3512 const regnode *scan,
3513 const char *loc_regeol,
3514 const char *loc_bostr,
3515 const char *loc_reg_starttry,
3516 const bool utf8_target)
3518 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3519 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3520 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3521 /* The part of the string before starttry has one color
3522 (pref0_len chars), between starttry and current
3523 position another one (pref_len - pref0_len chars),
3524 after the current position the third one.
3525 We assume that pref0_len <= pref_len, otherwise we
3526 decrease pref0_len. */
3527 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3528 ? (5 + taill) - l : locinput - loc_bostr;
3531 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3533 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3535 pref0_len = pref_len - (locinput - loc_reg_starttry);
3536 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3537 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3538 ? (5 + taill) - pref_len : loc_regeol - locinput);
3539 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3543 if (pref0_len > pref_len)
3544 pref0_len = pref_len;
3546 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3548 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3549 (locinput - pref_len),pref0_len, 60, 4, 5);
3551 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3552 (locinput - pref_len + pref0_len),
3553 pref_len - pref0_len, 60, 2, 3);
3555 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3556 locinput, loc_regeol - locinput, 10, 0, 1);
3558 const STRLEN tlen=len0+len1+len2;
3559 PerlIO_printf(Perl_debug_log,
3560 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3561 (IV)(locinput - loc_bostr),
3564 (docolor ? "" : "> <"),
3566 (int)(tlen > 19 ? 0 : 19 - tlen),
3573 /* reg_check_named_buff_matched()
3574 * Checks to see if a named buffer has matched. The data array of
3575 * buffer numbers corresponding to the buffer is expected to reside
3576 * in the regexp->data->data array in the slot stored in the ARG() of
3577 * node involved. Note that this routine doesn't actually care about the
3578 * name, that information is not preserved from compilation to execution.
3579 * Returns the index of the leftmost defined buffer with the given name
3580 * or 0 if non of the buffers matched.
3583 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3586 RXi_GET_DECL(rex,rexi);
3587 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3588 I32 *nums=(I32*)SvPVX(sv_dat);
3590 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3592 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3593 if ((I32)rex->lastparen >= nums[n] &&
3594 rex->offs[nums[n]].end != -1)
3604 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3605 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3607 /* This function determines if there are one or two characters that match
3608 * the first character of the passed-in EXACTish node <text_node>, and if
3609 * so, returns them in the passed-in pointers.
3611 * If it determines that no possible character in the target string can
3612 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3613 * the first character in <text_node> requires UTF-8 to represent, and the
3614 * target string isn't in UTF-8.)
3616 * If there are more than two characters that could match the beginning of
3617 * <text_node>, or if more context is required to determine a match or not,
3618 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3620 * The motiviation behind this function is to allow the caller to set up
3621 * tight loops for matching. If <text_node> is of type EXACT, there is
3622 * only one possible character that can match its first character, and so
3623 * the situation is quite simple. But things get much more complicated if
3624 * folding is involved. It may be that the first character of an EXACTFish
3625 * node doesn't participate in any possible fold, e.g., punctuation, so it
3626 * can be matched only by itself. The vast majority of characters that are
3627 * in folds match just two things, their lower and upper-case equivalents.
3628 * But not all are like that; some have multiple possible matches, or match
3629 * sequences of more than one character. This function sorts all that out.
3631 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3632 * loop of trying to match A*, we know we can't exit where the thing
3633 * following it isn't a B. And something can't be a B unless it is the
3634 * beginning of B. By putting a quick test for that beginning in a tight
3635 * loop, we can rule out things that can't possibly be B without having to
3636 * break out of the loop, thus avoiding work. Similarly, if A is a single
3637 * character, we can make a tight loop matching A*, using the outputs of
3640 * If the target string to match isn't in UTF-8, and there aren't
3641 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3642 * the one or two possible octets (which are characters in this situation)
3643 * that can match. In all cases, if there is only one character that can
3644 * match, *<c1p> and *<c2p> will be identical.
3646 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3647 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3648 * can match the beginning of <text_node>. They should be declared with at
3649 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3650 * undefined what these contain.) If one or both of the buffers are
3651 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3652 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3653 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3654 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3655 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3657 const bool utf8_target = reginfo->is_utf8_target;
3659 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
3660 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
3661 bool use_chrtest_void = FALSE;
3662 const bool is_utf8_pat = reginfo->is_utf8_pat;
3664 /* Used when we have both utf8 input and utf8 output, to avoid converting
3665 * to/from code points */
3666 bool utf8_has_been_setup = FALSE;
3670 U8 *pat = (U8*)STRING(text_node);
3671 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
3673 if (OP(text_node) == EXACT || OP(text_node) == EXACTL) {
3675 /* In an exact node, only one thing can be matched, that first
3676 * character. If both the pat and the target are UTF-8, we can just
3677 * copy the input to the output, avoiding finding the code point of
3682 else if (utf8_target) {
3683 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3684 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3685 utf8_has_been_setup = TRUE;
3688 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3691 else { /* an EXACTFish node */
3692 U8 *pat_end = pat + STR_LEN(text_node);
3694 /* An EXACTFL node has at least some characters unfolded, because what
3695 * they match is not known until now. So, now is the time to fold
3696 * the first few of them, as many as are needed to determine 'c1' and
3697 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
3698 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
3699 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
3700 * need to fold as many characters as a single character can fold to,
3701 * so that later we can check if the first ones are such a multi-char
3702 * fold. But, in such a pattern only locale-problematic characters
3703 * aren't folded, so we can skip this completely if the first character
3704 * in the node isn't one of the tricky ones */
3705 if (OP(text_node) == EXACTFL) {
3707 if (! is_utf8_pat) {
3708 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
3710 folded[0] = folded[1] = 's';
3712 pat_end = folded + 2;
3715 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
3720 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
3722 *(d++) = (U8) toFOLD_LC(*s);
3727 _to_utf8_fold_flags(s,
3730 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
3741 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
3742 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
3744 /* Multi-character folds require more context to sort out. Also
3745 * PL_utf8_foldclosures used below doesn't handle them, so have to
3746 * be handled outside this routine */
3747 use_chrtest_void = TRUE;
3749 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3750 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3752 /* Load the folds hash, if not already done */
3754 if (! PL_utf8_foldclosures) {
3755 _load_PL_utf8_foldclosures();
3758 /* The fold closures data structure is a hash with the keys
3759 * being the UTF-8 of every character that is folded to, like
3760 * 'k', and the values each an array of all code points that
3761 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
3762 * Multi-character folds are not included */
3763 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3768 /* Not found in the hash, therefore there are no folds
3769 * containing it, so there is only a single character that
3773 else { /* Does participate in folds */
3774 AV* list = (AV*) *listp;
3775 if (av_tindex(list) != 1) {
3777 /* If there aren't exactly two folds to this, it is
3778 * outside the scope of this function */
3779 use_chrtest_void = TRUE;
3781 else { /* There are two. Get them */
3782 SV** c_p = av_fetch(list, 0, FALSE);
3784 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3788 c_p = av_fetch(list, 1, FALSE);
3790 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3794 /* Folds that cross the 255/256 boundary are forbidden
3795 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
3796 * one is ASCIII. Since the pattern character is above
3797 * 255, and its only other match is below 256, the only
3798 * legal match will be to itself. We have thrown away
3799 * the original, so have to compute which is the one
3801 if ((c1 < 256) != (c2 < 256)) {
3802 if ((OP(text_node) == EXACTFL
3803 && ! IN_UTF8_CTYPE_LOCALE)
3804 || ((OP(text_node) == EXACTFA
3805 || OP(text_node) == EXACTFA_NO_TRIE)
3806 && (isASCII(c1) || isASCII(c2))))
3819 else /* Here, c1 is <= 255 */
3821 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3822 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
3823 && ((OP(text_node) != EXACTFA
3824 && OP(text_node) != EXACTFA_NO_TRIE)
3827 /* Here, there could be something above Latin1 in the target
3828 * which folds to this character in the pattern. All such
3829 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
3830 * than two characters involved in their folds, so are outside
3831 * the scope of this function */
3832 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3833 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3836 use_chrtest_void = TRUE;
3839 else { /* Here nothing above Latin1 can fold to the pattern
3841 switch (OP(text_node)) {
3843 case EXACTFL: /* /l rules */
3844 c2 = PL_fold_locale[c1];
3847 case EXACTF: /* This node only generated for non-utf8
3849 assert(! is_utf8_pat);
3850 if (! utf8_target) { /* /d rules */
3855 /* /u rules for all these. This happens to work for
3856 * EXACTFA as nothing in Latin1 folds to ASCII */
3857 case EXACTFA_NO_TRIE: /* This node only generated for
3858 non-utf8 patterns */
3859 assert(! is_utf8_pat);
3864 c2 = PL_fold_latin1[c1];
3868 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3869 NOT_REACHED; /* NOTREACHED */
3875 /* Here have figured things out. Set up the returns */
3876 if (use_chrtest_void) {
3877 *c2p = *c1p = CHRTEST_VOID;
3879 else if (utf8_target) {
3880 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3881 uvchr_to_utf8(c1_utf8, c1);
3882 uvchr_to_utf8(c2_utf8, c2);
3885 /* Invariants are stored in both the utf8 and byte outputs; Use
3886 * negative numbers otherwise for the byte ones. Make sure that the
https://perl5.git.perl.org / perl5.git / blob