5 * "One Ring to rule them all, One Ring to find them..."
8 /* This file contains functions for executing a regular expression. See
9 * also regcomp.c which funnily enough, contains functions for compiling
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_exec.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
19 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
20 * confused with the original package (see point 3 below). Thanks, Henry!
23 /* Additional note: this code is very heavily munged from Henry's version
24 * in places. In some spots I've traded clarity for efficiency, so don't
25 * blame Henry for some of the lack of readability.
28 /* The names of the functions have been changed from regcomp and
29 * regexec to pregcomp and pregexec in order to avoid conflicts
30 * with the POSIX routines of the same names.
33 #ifdef PERL_EXT_RE_BUILD
34 /* need to replace pregcomp et al, so enable that */
35 # ifndef PERL_IN_XSUB_RE
36 # define PERL_IN_XSUB_RE
38 /* need access to debugger hooks */
39 # if defined(PERL_EXT_RE_DEBUG) && !defined(DEBUGGING)
44 #ifdef PERL_IN_XSUB_RE
45 /* We *really* need to overwrite these symbols: */
46 # define Perl_regexec_flags my_regexec
47 # define Perl_regdump my_regdump
48 # define Perl_regprop my_regprop
49 # define Perl_re_intuit_start my_re_intuit_start
50 /* *These* symbols are masked to allow static link. */
51 # define Perl_pregexec my_pregexec
52 # define Perl_reginitcolors my_reginitcolors
53 # define Perl_regclass_swash my_regclass_swash
55 # define PERL_NO_GET_CONTEXT
59 * pregcomp and pregexec -- regsub and regerror are not used in perl
61 * Copyright (c) 1986 by University of Toronto.
62 * Written by Henry Spencer. Not derived from licensed software.
64 * Permission is granted to anyone to use this software for any
65 * purpose on any computer system, and to redistribute it freely,
66 * subject to the following restrictions:
68 * 1. The author is not responsible for the consequences of use of
69 * this software, no matter how awful, even if they arise
72 * 2. The origin of this software must not be misrepresented, either
73 * by explicit claim or by omission.
75 * 3. Altered versions must be plainly marked as such, and must not
76 * be misrepresented as being the original software.
78 **** Alterations to Henry's code are...
80 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
81 **** 2000, 2001, 2002, 2003, 2004, 2005, by Larry Wall and others
83 **** You may distribute under the terms of either the GNU General Public
84 **** License or the Artistic License, as specified in the README file.
86 * Beware that some of this code is subtly aware of the way operator
87 * precedence is structured in regular expressions. Serious changes in
88 * regular-expression syntax might require a total rethink.
91 #define PERL_IN_REGEXEC_C
96 #define RF_tainted 1 /* tainted information used? */
97 #define RF_warned 2 /* warned about big count? */
98 #define RF_evaled 4 /* Did an EVAL with setting? */
99 #define RF_utf8 8 /* String contains multibyte chars? */
101 #define UTF ((PL_reg_flags & RF_utf8) != 0)
103 #define RS_init 1 /* eval environment created */
104 #define RS_set 2 /* replsv value is set */
107 #define STATIC static
110 #define REGINCLASS(p,c) (ANYOF_FLAGS(p) ? reginclass(p,c,0,0) : ANYOF_BITMAP_TEST(p,*(c)))
116 #define CHR_SVLEN(sv) (do_utf8 ? sv_len_utf8(sv) : SvCUR(sv))
117 #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
119 #define reghop_c(pos,off) ((char*)reghop((U8*)pos, off))
120 #define reghopmaybe_c(pos,off) ((char*)reghopmaybe((U8*)pos, off))
121 #define HOP(pos,off) (PL_reg_match_utf8 ? reghop((U8*)pos, off) : (U8*)(pos + off))
122 #define HOPMAYBE(pos,off) (PL_reg_match_utf8 ? reghopmaybe((U8*)pos, off) : (U8*)(pos + off))
123 #define HOPc(pos,off) ((char*)HOP(pos,off))
124 #define HOPMAYBEc(pos,off) ((char*)HOPMAYBE(pos,off))
126 #define HOPBACK(pos, off) ( \
127 (PL_reg_match_utf8) \
128 ? reghopmaybe((U8*)pos, -off) \
129 : (pos - off >= PL_bostr) \
133 #define HOPBACKc(pos, off) (char*)HOPBACK(pos, off)
135 #define reghop3_c(pos,off,lim) ((char*)reghop3((U8*)pos, off, (U8*)lim))
136 #define reghopmaybe3_c(pos,off,lim) ((char*)reghopmaybe3((U8*)pos, off, (U8*)lim))
137 #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off))
138 #define HOPMAYBE3(pos,off,lim) (PL_reg_match_utf8 ? reghopmaybe3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off))
139 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
140 #define HOPMAYBE3c(pos,off,lim) ((char*)HOPMAYBE3(pos,off,lim))
142 #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
143 if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END
144 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
145 #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
146 #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
147 #define LOAD_UTF8_CHARCLASS_MARK() LOAD_UTF8_CHARCLASS(mark, "\xcd\x86")
149 /* for use after a quantifier and before an EXACT-like node -- japhy */
150 #define JUMPABLE(rn) ( \
151 OP(rn) == OPEN || OP(rn) == CLOSE || OP(rn) == EVAL || \
152 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
153 OP(rn) == PLUS || OP(rn) == MINMOD || \
154 (PL_regkind[(U8)OP(rn)] == CURLY && ARG1(rn) > 0) \
157 #define HAS_TEXT(rn) ( \
158 PL_regkind[(U8)OP(rn)] == EXACT || PL_regkind[(U8)OP(rn)] == REF \
162 Search for mandatory following text node; for lookahead, the text must
163 follow but for lookbehind (rn->flags != 0) we skip to the next step.
165 #define FIND_NEXT_IMPT(rn) STMT_START { \
166 while (JUMPABLE(rn)) \
167 if (OP(rn) == SUSPEND || PL_regkind[(U8)OP(rn)] == CURLY) \
168 rn = NEXTOPER(NEXTOPER(rn)); \
169 else if (OP(rn) == PLUS) \
171 else if (OP(rn) == IFMATCH) \
172 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
173 else rn += NEXT_OFF(rn); \
176 static void restore_pos(pTHX_ void *arg);
179 S_regcppush(pTHX_ I32 parenfloor)
181 const int retval = PL_savestack_ix;
182 #define REGCP_PAREN_ELEMS 4
183 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
186 if (paren_elems_to_push < 0)
187 Perl_croak(aTHX_ "panic: paren_elems_to_push < 0");
189 #define REGCP_OTHER_ELEMS 6
190 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS);
191 for (p = PL_regsize; p > parenfloor; p--) {
192 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
193 SSPUSHINT(PL_regendp[p]);
194 SSPUSHINT(PL_regstartp[p]);
195 SSPUSHPTR(PL_reg_start_tmp[p]);
198 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
199 SSPUSHINT(PL_regsize);
200 SSPUSHINT(*PL_reglastparen);
201 SSPUSHINT(*PL_reglastcloseparen);
202 SSPUSHPTR(PL_reginput);
203 #define REGCP_FRAME_ELEMS 2
204 /* REGCP_FRAME_ELEMS are part of the REGCP_OTHER_ELEMS and
205 * are needed for the regexp context stack bookkeeping. */
206 SSPUSHINT(paren_elems_to_push + REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
207 SSPUSHINT(SAVEt_REGCONTEXT); /* Magic cookie. */
212 /* These are needed since we do not localize EVAL nodes: */
213 # define REGCP_SET(cp) DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, \
214 " Setting an EVAL scope, savestack=%"IVdf"\n", \
215 (IV)PL_savestack_ix)); cp = PL_savestack_ix
217 # define REGCP_UNWIND(cp) DEBUG_EXECUTE_r(cp != PL_savestack_ix ? \
218 PerlIO_printf(Perl_debug_log, \
219 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
220 (IV)(cp), (IV)PL_savestack_ix) : 0); regcpblow(cp)
229 GET_RE_DEBUG_FLAGS_DECL;
231 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
233 assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
234 i = SSPOPINT; /* Parentheses elements to pop. */
235 input = (char *) SSPOPPTR;
236 *PL_reglastcloseparen = SSPOPINT;
237 *PL_reglastparen = SSPOPINT;
238 PL_regsize = SSPOPINT;
240 /* Now restore the parentheses context. */
241 for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS);
242 i > 0; i -= REGCP_PAREN_ELEMS) {
244 paren = (U32)SSPOPINT;
245 PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
246 PL_regstartp[paren] = SSPOPINT;
248 if (paren <= *PL_reglastparen)
249 PL_regendp[paren] = tmps;
251 PerlIO_printf(Perl_debug_log,
252 " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
253 (UV)paren, (IV)PL_regstartp[paren],
254 (IV)(PL_reg_start_tmp[paren] - PL_bostr),
255 (IV)PL_regendp[paren],
256 (paren > *PL_reglastparen ? "(no)" : ""));
260 if ((I32)(*PL_reglastparen + 1) <= PL_regnpar) {
261 PerlIO_printf(Perl_debug_log,
262 " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
263 (IV)(*PL_reglastparen + 1), (IV)PL_regnpar);
267 /* It would seem that the similar code in regtry()
268 * already takes care of this, and in fact it is in
269 * a better location to since this code can #if 0-ed out
270 * but the code in regtry() is needed or otherwise tests
271 * requiring null fields (pat.t#187 and split.t#{13,14}
272 * (as of patchlevel 7877) will fail. Then again,
273 * this code seems to be necessary or otherwise
274 * building DynaLoader will fail:
275 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
277 for (paren = *PL_reglastparen + 1; (I32)paren <= PL_regnpar; paren++) {
278 if ((I32)paren > PL_regsize)
279 PL_regstartp[paren] = -1;
280 PL_regendp[paren] = -1;
287 S_regcp_set_to(pTHX_ I32 ss)
289 const I32 tmp = PL_savestack_ix;
291 PL_savestack_ix = ss;
293 PL_savestack_ix = tmp;
297 typedef struct re_cc_state
301 struct re_cc_state *prev;
306 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
308 #define TRYPAREN(paren, n, input) { \
311 PL_regstartp[paren] = HOPc(input, -1) - PL_bostr; \
312 PL_regendp[paren] = input - PL_bostr; \
315 PL_regendp[paren] = -1; \
317 if (regmatch(next)) \
320 PL_regendp[paren] = -1; \
325 * pregexec and friends
329 - pregexec - match a regexp against a string
332 Perl_pregexec(pTHX_ register regexp *prog, char *stringarg, register char *strend,
333 char *strbeg, I32 minend, SV *screamer, U32 nosave)
334 /* strend: pointer to null at end of string */
335 /* strbeg: real beginning of string */
336 /* minend: end of match must be >=minend after stringarg. */
337 /* nosave: For optimizations. */
340 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
341 nosave ? 0 : REXEC_COPY_STR);
345 S_cache_re(pTHX_ regexp *prog)
347 PL_regprecomp = prog->precomp; /* Needed for FAIL. */
349 PL_regprogram = prog->program;
351 PL_regnpar = prog->nparens;
352 PL_regdata = prog->data;
357 * Need to implement the following flags for reg_anch:
359 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
361 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
362 * INTUIT_AUTORITATIVE_ML
363 * INTUIT_ONCE_NOML - Intuit can match in one location only.
366 * Another flag for this function: SECOND_TIME (so that float substrs
367 * with giant delta may be not rechecked).
370 /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
372 /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
373 Otherwise, only SvCUR(sv) is used to get strbeg. */
375 /* XXXX We assume that strpos is strbeg unless sv. */
377 /* XXXX Some places assume that there is a fixed substring.
378 An update may be needed if optimizer marks as "INTUITable"
379 RExen without fixed substrings. Similarly, it is assumed that
380 lengths of all the strings are no more than minlen, thus they
381 cannot come from lookahead.
382 (Or minlen should take into account lookahead.) */
384 /* A failure to find a constant substring means that there is no need to make
385 an expensive call to REx engine, thus we celebrate a failure. Similarly,
386 finding a substring too deep into the string means that less calls to
387 regtry() should be needed.
389 REx compiler's optimizer found 4 possible hints:
390 a) Anchored substring;
392 c) Whether we are anchored (beginning-of-line or \G);
393 d) First node (of those at offset 0) which may distingush positions;
394 We use a)b)d) and multiline-part of c), and try to find a position in the
395 string which does not contradict any of them.
398 /* Most of decisions we do here should have been done at compile time.
399 The nodes of the REx which we used for the search should have been
400 deleted from the finite automaton. */
403 Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
404 char *strend, U32 flags, re_scream_pos_data *data)
406 register I32 start_shift = 0;
407 /* Should be nonnegative! */
408 register I32 end_shift = 0;
413 const int do_utf8 = sv ? SvUTF8(sv) : 0; /* if no sv we have to assume bytes */
415 register char *other_last = Nullch; /* other substr checked before this */
416 char *check_at = Nullch; /* check substr found at this pos */
417 const I32 multiline = prog->reganch & PMf_MULTILINE;
419 const char * const i_strpos = strpos;
420 SV * const dsv = PERL_DEBUG_PAD_ZERO(0);
423 GET_RE_DEBUG_FLAGS_DECL;
425 RX_MATCH_UTF8_set(prog,do_utf8);
427 if (prog->reganch & ROPT_UTF8) {
428 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
429 "UTF-8 regex...\n"));
430 PL_reg_flags |= RF_utf8;
434 const char *s = PL_reg_match_utf8 ?
435 sv_uni_display(dsv, sv, 60, UNI_DISPLAY_REGEX) :
437 const int len = PL_reg_match_utf8 ?
438 strlen(s) : strend - strpos;
441 if (PL_reg_match_utf8)
442 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
443 "UTF-8 target...\n"));
444 PerlIO_printf(Perl_debug_log,
445 "%sGuessing start of match, REx%s \"%s%.60s%s%s\" against \"%s%.*s%s%s\"...\n",
446 PL_colors[4], PL_colors[5], PL_colors[0],
449 (strlen(prog->precomp) > 60 ? "..." : ""),
451 (int)(len > 60 ? 60 : len),
453 (len > 60 ? "..." : "")
457 /* CHR_DIST() would be more correct here but it makes things slow. */
458 if (prog->minlen > strend - strpos) {
459 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
460 "String too short... [re_intuit_start]\n"));
463 strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
466 if (!prog->check_utf8 && prog->check_substr)
467 to_utf8_substr(prog);
468 check = prog->check_utf8;
470 if (!prog->check_substr && prog->check_utf8)
471 to_byte_substr(prog);
472 check = prog->check_substr;
474 if (check == &PL_sv_undef) {
475 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
476 "Non-utf string cannot match utf check string\n"));
479 if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
480 ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
481 || ( (prog->reganch & ROPT_ANCH_BOL)
482 && !multiline ) ); /* Check after \n? */
485 if ( !(prog->reganch & (ROPT_ANCH_GPOS /* Checked by the caller */
486 | ROPT_IMPLICIT)) /* not a real BOL */
487 /* SvCUR is not set on references: SvRV and SvPVX_const overlap */
489 && (strpos != strbeg)) {
490 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
493 if (prog->check_offset_min == prog->check_offset_max &&
494 !(prog->reganch & ROPT_CANY_SEEN)) {
495 /* Substring at constant offset from beg-of-str... */
498 s = HOP3c(strpos, prog->check_offset_min, strend);
500 slen = SvCUR(check); /* >= 1 */
502 if ( strend - s > slen || strend - s < slen - 1
503 || (strend - s == slen && strend[-1] != '\n')) {
504 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
507 /* Now should match s[0..slen-2] */
509 if (slen && (*SvPVX_const(check) != *s
511 && memNE(SvPVX_const(check), s, slen)))) {
513 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
517 else if (*SvPVX_const(check) != *s
518 || ((slen = SvCUR(check)) > 1
519 && memNE(SvPVX_const(check), s, slen)))
522 goto success_at_start;
525 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
527 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
528 end_shift = prog->minlen - start_shift -
529 CHR_SVLEN(check) + (SvTAIL(check) != 0);
531 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
532 - (SvTAIL(check) != 0);
533 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
535 if (end_shift < eshift)
539 else { /* Can match at random position */
542 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
543 /* Should be nonnegative! */
544 end_shift = prog->minlen - start_shift -
545 CHR_SVLEN(check) + (SvTAIL(check) != 0);
548 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
550 Perl_croak(aTHX_ "panic: end_shift");
554 /* Find a possible match in the region s..strend by looking for
555 the "check" substring in the region corrected by start/end_shift. */
556 if (flags & REXEC_SCREAM) {
557 I32 p = -1; /* Internal iterator of scream. */
558 I32 * const pp = data ? data->scream_pos : &p;
560 if (PL_screamfirst[BmRARE(check)] >= 0
561 || ( BmRARE(check) == '\n'
562 && (BmPREVIOUS(check) == SvCUR(check) - 1)
564 s = screaminstr(sv, check,
565 start_shift + (s - strbeg), end_shift, pp, 0);
568 /* we may be pointing at the wrong string */
569 if (s && RX_MATCH_COPIED(prog))
570 s = strbeg + (s - SvPVX_const(sv));
572 *data->scream_olds = s;
574 else if (prog->reganch & ROPT_CANY_SEEN)
575 s = fbm_instr((U8*)(s + start_shift),
576 (U8*)(strend - end_shift),
577 check, multiline ? FBMrf_MULTILINE : 0);
579 s = fbm_instr(HOP3(s, start_shift, strend),
580 HOP3(strend, -end_shift, strbeg),
581 check, multiline ? FBMrf_MULTILINE : 0);
583 /* Update the count-of-usability, remove useless subpatterns,
586 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s %s substr \"%s%.*s%s\"%s%s",
587 (s ? "Found" : "Did not find"),
588 (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) ? "anchored" : "floating"),
590 (int)(SvCUR(check) - (SvTAIL(check)!=0)),
592 PL_colors[1], (SvTAIL(check) ? "$" : ""),
593 (s ? " at offset " : "...\n") ) );
600 /* Finish the diagnostic message */
601 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
603 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
604 Start with the other substr.
605 XXXX no SCREAM optimization yet - and a very coarse implementation
606 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
607 *always* match. Probably should be marked during compile...
608 Probably it is right to do no SCREAM here...
611 if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) : (prog->float_substr && prog->anchored_substr)) {
612 /* Take into account the "other" substring. */
613 /* XXXX May be hopelessly wrong for UTF... */
616 if (check == (do_utf8 ? prog->float_utf8 : prog->float_substr)) {
619 char * const last = HOP3c(s, -start_shift, strbeg);
624 t = s - prog->check_offset_max;
625 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
627 || ((t = reghopmaybe3_c(s, -(prog->check_offset_max), strpos))
632 t = HOP3c(t, prog->anchored_offset, strend);
633 if (t < other_last) /* These positions already checked */
635 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
638 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
639 /* On end-of-str: see comment below. */
640 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
641 if (must == &PL_sv_undef) {
643 DEBUG_EXECUTE_r(must = prog->anchored_utf8); /* for debug */
648 HOP3(HOP3(last1, prog->anchored_offset, strend)
649 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
651 multiline ? FBMrf_MULTILINE : 0
653 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
654 "%s anchored substr \"%s%.*s%s\"%s",
655 (s ? "Found" : "Contradicts"),
658 - (SvTAIL(must)!=0)),
660 PL_colors[1], (SvTAIL(must) ? "$" : "")));
662 if (last1 >= last2) {
663 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
664 ", giving up...\n"));
667 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
668 ", trying floating at offset %ld...\n",
669 (long)(HOP3c(s1, 1, strend) - i_strpos)));
670 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
671 s = HOP3c(last, 1, strend);
675 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
676 (long)(s - i_strpos)));
677 t = HOP3c(s, -prog->anchored_offset, strbeg);
678 other_last = HOP3c(s, 1, strend);
686 else { /* Take into account the floating substring. */
691 t = HOP3c(s, -start_shift, strbeg);
693 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
694 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
695 last = HOP3c(t, prog->float_max_offset, strend);
696 s = HOP3c(t, prog->float_min_offset, strend);
699 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
700 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
701 /* fbm_instr() takes into account exact value of end-of-str
702 if the check is SvTAIL(ed). Since false positives are OK,
703 and end-of-str is not later than strend we are OK. */
704 if (must == &PL_sv_undef) {
706 DEBUG_EXECUTE_r(must = prog->float_utf8); /* for debug message */
709 s = fbm_instr((unsigned char*)s,
710 (unsigned char*)last + SvCUR(must)
712 must, multiline ? FBMrf_MULTILINE : 0);
713 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s floating substr \"%s%.*s%s\"%s",
714 (s ? "Found" : "Contradicts"),
716 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
718 PL_colors[1], (SvTAIL(must) ? "$" : "")));
721 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
722 ", giving up...\n"));
725 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
726 ", trying anchored starting at offset %ld...\n",
727 (long)(s1 + 1 - i_strpos)));
729 s = HOP3c(t, 1, strend);
733 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
734 (long)(s - i_strpos)));
735 other_last = s; /* Fix this later. --Hugo */
744 t = s - prog->check_offset_max;
745 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
747 || ((t = reghopmaybe3_c(s, -prog->check_offset_max, strpos))
749 /* Fixed substring is found far enough so that the match
750 cannot start at strpos. */
752 if (ml_anch && t[-1] != '\n') {
753 /* Eventually fbm_*() should handle this, but often
754 anchored_offset is not 0, so this check will not be wasted. */
755 /* XXXX In the code below we prefer to look for "^" even in
756 presence of anchored substrings. And we search even
757 beyond the found float position. These pessimizations
758 are historical artefacts only. */
760 while (t < strend - prog->minlen) {
762 if (t < check_at - prog->check_offset_min) {
763 if (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) {
764 /* Since we moved from the found position,
765 we definitely contradict the found anchored
766 substr. Due to the above check we do not
767 contradict "check" substr.
768 Thus we can arrive here only if check substr
769 is float. Redo checking for "other"=="fixed".
772 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
773 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
774 goto do_other_anchored;
776 /* We don't contradict the found floating substring. */
777 /* XXXX Why not check for STCLASS? */
779 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
780 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
783 /* Position contradicts check-string */
784 /* XXXX probably better to look for check-string
785 than for "\n", so one should lower the limit for t? */
786 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
787 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
788 other_last = strpos = s = t + 1;
793 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
794 PL_colors[0], PL_colors[1]));
798 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
799 PL_colors[0], PL_colors[1]));
803 ++BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
806 /* The found string does not prohibit matching at strpos,
807 - no optimization of calling REx engine can be performed,
808 unless it was an MBOL and we are not after MBOL,
809 or a future STCLASS check will fail this. */
811 /* Even in this situation we may use MBOL flag if strpos is offset
812 wrt the start of the string. */
813 if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
814 && (strpos != strbeg) && strpos[-1] != '\n'
815 /* May be due to an implicit anchor of m{.*foo} */
816 && !(prog->reganch & ROPT_IMPLICIT))
821 DEBUG_EXECUTE_r( if (ml_anch)
822 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
823 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
826 if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
828 prog->check_utf8 /* Could be deleted already */
829 && --BmUSEFUL(prog->check_utf8) < 0
830 && (prog->check_utf8 == prog->float_utf8)
832 prog->check_substr /* Could be deleted already */
833 && --BmUSEFUL(prog->check_substr) < 0
834 && (prog->check_substr == prog->float_substr)
837 /* If flags & SOMETHING - do not do it many times on the same match */
838 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
839 SvREFCNT_dec(do_utf8 ? prog->check_utf8 : prog->check_substr);
840 if (do_utf8 ? prog->check_substr : prog->check_utf8)
841 SvREFCNT_dec(do_utf8 ? prog->check_substr : prog->check_utf8);
842 prog->check_substr = prog->check_utf8 = Nullsv; /* disable */
843 prog->float_substr = prog->float_utf8 = Nullsv; /* clear */
844 check = Nullsv; /* abort */
846 /* XXXX This is a remnant of the old implementation. It
847 looks wasteful, since now INTUIT can use many
849 prog->reganch &= ~RE_USE_INTUIT;
856 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
857 if (prog->regstclass) {
858 /* minlen == 0 is possible if regstclass is \b or \B,
859 and the fixed substr is ''$.
860 Since minlen is already taken into account, s+1 is before strend;
861 accidentally, minlen >= 1 guaranties no false positives at s + 1
862 even for \b or \B. But (minlen? 1 : 0) below assumes that
863 regstclass does not come from lookahead... */
864 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
865 This leaves EXACTF only, which is dealt with in find_byclass(). */
866 const U8* const str = (U8*)STRING(prog->regstclass);
867 const int cl_l = (PL_regkind[(U8)OP(prog->regstclass)] == EXACT
868 ? CHR_DIST(str+STR_LEN(prog->regstclass), str)
870 const char * const endpos = (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
871 ? HOP3c(s, (prog->minlen ? cl_l : 0), strend)
872 : (prog->float_substr || prog->float_utf8
873 ? HOP3c(HOP3c(check_at, -start_shift, strbeg),
879 s = find_byclass(prog, prog->regstclass, s, endpos, 1);
882 const char *what = 0;
884 if (endpos == strend) {
885 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
886 "Could not match STCLASS...\n") );
889 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
890 "This position contradicts STCLASS...\n") );
891 if ((prog->reganch & ROPT_ANCH) && !ml_anch)
893 /* Contradict one of substrings */
894 if (prog->anchored_substr || prog->anchored_utf8) {
895 if ((do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) == check) {
896 DEBUG_EXECUTE_r( what = "anchored" );
898 s = HOP3c(t, 1, strend);
899 if (s + start_shift + end_shift > strend) {
900 /* XXXX Should be taken into account earlier? */
901 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
902 "Could not match STCLASS...\n") );
907 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
908 "Looking for %s substr starting at offset %ld...\n",
909 what, (long)(s + start_shift - i_strpos)) );
912 /* Have both, check_string is floating */
913 if (t + start_shift >= check_at) /* Contradicts floating=check */
914 goto retry_floating_check;
915 /* Recheck anchored substring, but not floating... */
919 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
920 "Looking for anchored substr starting at offset %ld...\n",
921 (long)(other_last - i_strpos)) );
922 goto do_other_anchored;
924 /* Another way we could have checked stclass at the
925 current position only: */
930 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
931 "Looking for /%s^%s/m starting at offset %ld...\n",
932 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
935 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
937 /* Check is floating subtring. */
938 retry_floating_check:
939 t = check_at - start_shift;
940 DEBUG_EXECUTE_r( what = "floating" );
941 goto hop_and_restart;
944 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
945 "By STCLASS: moving %ld --> %ld\n",
946 (long)(t - i_strpos), (long)(s - i_strpos))
950 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
951 "Does not contradict STCLASS...\n");
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
957 PL_colors[4], (check ? "Guessed" : "Giving up"),
958 PL_colors[5], (long)(s - i_strpos)) );
961 fail_finish: /* Substring not found */
962 if (prog->check_substr || prog->check_utf8) /* could be removed already */
963 BmUSEFUL(do_utf8 ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
965 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
966 PL_colors[4], PL_colors[5]));
970 /* We know what class REx starts with. Try to find this position... */
972 S_find_byclass(pTHX_ regexp * prog, regnode *c, char *s, const char *strend, I32 norun)
975 const I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
979 register STRLEN uskip;
983 register I32 tmp = 1; /* Scratch variable? */
984 register const bool do_utf8 = PL_reg_match_utf8;
986 /* We know what class it must start with. */
990 while (s + (uskip = UTF8SKIP(s)) <= strend) {
991 if ((ANYOF_FLAGS(c) & ANYOF_UNICODE) ||
992 !UTF8_IS_INVARIANT((U8)s[0]) ?
993 reginclass(c, (U8*)s, 0, do_utf8) :
994 REGINCLASS(c, (U8*)s)) {
995 if (tmp && (norun || regtry(prog, s)))
1006 while (s < strend) {
1009 if (REGINCLASS(c, (U8*)s) ||
1010 (ANYOF_FOLD_SHARP_S(c, s, strend) &&
1011 /* The assignment of 2 is intentional:
1012 * for the folded sharp s, the skip is 2. */
1013 (skip = SHARP_S_SKIP))) {
1014 if (tmp && (norun || regtry(prog, s)))
1026 while (s < strend) {
1027 if (tmp && (norun || regtry(prog, s)))
1036 ln = STR_LEN(c); /* length to match in octets/bytes */
1037 lnc = (I32) ln; /* length to match in characters */
1039 STRLEN ulen1, ulen2;
1041 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
1042 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
1043 const U32 uniflags = ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY;
1045 to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
1046 to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
1048 c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
1050 c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
1053 while (sm < ((U8 *) m + ln)) {
1068 c2 = PL_fold_locale[c1];
1070 e = HOP3c(strend, -((I32)lnc), s);
1073 e = s; /* Due to minlen logic of intuit() */
1075 /* The idea in the EXACTF* cases is to first find the
1076 * first character of the EXACTF* node and then, if
1077 * necessary, case-insensitively compare the full
1078 * text of the node. The c1 and c2 are the first
1079 * characters (though in Unicode it gets a bit
1080 * more complicated because there are more cases
1081 * than just upper and lower: one needs to use
1082 * the so-called folding case for case-insensitive
1083 * matching (called "loose matching" in Unicode).
1084 * ibcmp_utf8() will do just that. */
1088 U8 tmpbuf [UTF8_MAXBYTES+1];
1089 STRLEN len, foldlen;
1090 const U32 uniflags = ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY;
1092 /* Upper and lower of 1st char are equal -
1093 * probably not a "letter". */
1095 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1099 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1100 m, (char **)0, ln, (bool)UTF))
1101 && (norun || regtry(prog, s)) )
1104 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1105 uvchr_to_utf8(tmpbuf, c);
1106 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1108 && (f == c1 || f == c2)
1109 && (ln == foldlen ||
1110 !ibcmp_utf8((char *) foldbuf,
1111 (char **)0, foldlen, do_utf8,
1113 (char **)0, ln, (bool)UTF))
1114 && (norun || regtry(prog, s)) )
1122 c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
1125 /* Handle some of the three Greek sigmas cases.
1126 * Note that not all the possible combinations
1127 * are handled here: some of them are handled
1128 * by the standard folding rules, and some of
1129 * them (the character class or ANYOF cases)
1130 * are handled during compiletime in
1131 * regexec.c:S_regclass(). */
1132 if (c == (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA ||
1133 c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA)
1134 c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA;
1136 if ( (c == c1 || c == c2)
1138 ibcmp_utf8(s, (char **)0, 0, do_utf8,
1139 m, (char **)0, ln, (bool)UTF))
1140 && (norun || regtry(prog, s)) )
1143 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
1144 uvchr_to_utf8(tmpbuf, c);
1145 f = to_utf8_fold(tmpbuf, foldbuf, &foldlen);
1147 && (f == c1 || f == c2)
1148 && (ln == foldlen ||
1149 !ibcmp_utf8((char *) foldbuf,
1150 (char **)0, foldlen, do_utf8,
1152 (char **)0, ln, (bool)UTF))
1153 && (norun || regtry(prog, s)) )
1164 && (ln == 1 || !(OP(c) == EXACTF
1166 : ibcmp_locale(s, m, ln)))
1167 && (norun || regtry(prog, s)) )
1173 if ( (*(U8*)s == c1 || *(U8*)s == c2)
1174 && (ln == 1 || !(OP(c) == EXACTF
1176 : ibcmp_locale(s, m, ln)))
1177 && (norun || regtry(prog, s)) )
1184 PL_reg_flags |= RF_tainted;
1191 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1193 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1195 tmp = ((OP(c) == BOUND ?
1196 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1197 LOAD_UTF8_CHARCLASS_ALNUM();
1198 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1199 if (tmp == !(OP(c) == BOUND ?
1200 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1201 isALNUM_LC_utf8((U8*)s)))
1204 if ((norun || regtry(prog, s)))
1211 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1212 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1213 while (s < strend) {
1215 !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
1217 if ((norun || regtry(prog, s)))
1223 if ((!prog->minlen && tmp) && (norun || regtry(prog, s)))
1227 PL_reg_flags |= RF_tainted;
1234 U8 *r = reghop3((U8*)s, -1, (U8*)PL_bostr);
1236 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
1238 tmp = ((OP(c) == NBOUND ?
1239 isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0);
1240 LOAD_UTF8_CHARCLASS_ALNUM();
1241 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1242 if (tmp == !(OP(c) == NBOUND ?
1243 swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) :
1244 isALNUM_LC_utf8((U8*)s)))
1246 else if ((norun || regtry(prog, s)))
1252 tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n';
1253 tmp = ((OP(c) == NBOUND ?
1254 isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
1255 while (s < strend) {
1257 !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
1259 else if ((norun || regtry(prog, s)))
1264 if ((!prog->minlen && !tmp) && (norun || regtry(prog, s)))
1269 LOAD_UTF8_CHARCLASS_ALNUM();
1270 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1271 if (swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1272 if (tmp && (norun || regtry(prog, s)))
1283 while (s < strend) {
1285 if (tmp && (norun || regtry(prog, s)))
1297 PL_reg_flags |= RF_tainted;
1299 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1300 if (isALNUM_LC_utf8((U8*)s)) {
1301 if (tmp && (norun || regtry(prog, s)))
1312 while (s < strend) {
1313 if (isALNUM_LC(*s)) {
1314 if (tmp && (norun || regtry(prog, s)))
1327 LOAD_UTF8_CHARCLASS_ALNUM();
1328 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1329 if (!swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) {
1330 if (tmp && (norun || regtry(prog, s)))
1341 while (s < strend) {
1343 if (tmp && (norun || regtry(prog, s)))
1355 PL_reg_flags |= RF_tainted;
1357 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1358 if (!isALNUM_LC_utf8((U8*)s)) {
1359 if (tmp && (norun || regtry(prog, s)))
1370 while (s < strend) {
1371 if (!isALNUM_LC(*s)) {
1372 if (tmp && (norun || regtry(prog, s)))
1385 LOAD_UTF8_CHARCLASS_SPACE();
1386 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1387 if (*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)) {
1388 if (tmp && (norun || regtry(prog, s)))
1399 while (s < strend) {
1401 if (tmp && (norun || regtry(prog, s)))
1413 PL_reg_flags |= RF_tainted;
1415 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1416 if (*s == ' ' || isSPACE_LC_utf8((U8*)s)) {
1417 if (tmp && (norun || regtry(prog, s)))
1428 while (s < strend) {
1429 if (isSPACE_LC(*s)) {
1430 if (tmp && (norun || regtry(prog, s)))
1443 LOAD_UTF8_CHARCLASS_SPACE();
1444 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1445 if (!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8))) {
1446 if (tmp && (norun || regtry(prog, s)))
1457 while (s < strend) {
1459 if (tmp && (norun || regtry(prog, s)))
1471 PL_reg_flags |= RF_tainted;
1473 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1474 if (!(*s == ' ' || isSPACE_LC_utf8((U8*)s))) {
1475 if (tmp && (norun || regtry(prog, s)))
1486 while (s < strend) {
1487 if (!isSPACE_LC(*s)) {
1488 if (tmp && (norun || regtry(prog, s)))
1501 LOAD_UTF8_CHARCLASS_DIGIT();
1502 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1503 if (swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1504 if (tmp && (norun || regtry(prog, s)))
1515 while (s < strend) {
1517 if (tmp && (norun || regtry(prog, s)))
1529 PL_reg_flags |= RF_tainted;
1531 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1532 if (isDIGIT_LC_utf8((U8*)s)) {
1533 if (tmp && (norun || regtry(prog, s)))
1544 while (s < strend) {
1545 if (isDIGIT_LC(*s)) {
1546 if (tmp && (norun || regtry(prog, s)))
1559 LOAD_UTF8_CHARCLASS_DIGIT();
1560 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1561 if (!swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) {
1562 if (tmp && (norun || regtry(prog, s)))
1573 while (s < strend) {
1575 if (tmp && (norun || regtry(prog, s)))
1587 PL_reg_flags |= RF_tainted;
1589 while (s + (uskip = UTF8SKIP(s)) <= strend) {
1590 if (!isDIGIT_LC_utf8((U8*)s)) {
1591 if (tmp && (norun || regtry(prog, s)))
1602 while (s < strend) {
1603 if (!isDIGIT_LC(*s)) {
1604 if (tmp && (norun || regtry(prog, s)))
1616 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
1625 - regexec_flags - match a regexp against a string
1628 Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char *strend,
1629 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
1630 /* strend: pointer to null at end of string */
1631 /* strbeg: real beginning of string */
1632 /* minend: end of match must be >=minend after stringarg. */
1633 /* data: May be used for some additional optimizations. */
1634 /* nosave: For optimizations. */
1637 register regnode *c;
1638 register char *startpos = stringarg;
1639 I32 minlen; /* must match at least this many chars */
1640 I32 dontbother = 0; /* how many characters not to try at end */
1641 I32 end_shift = 0; /* Same for the end. */ /* CC */
1642 I32 scream_pos = -1; /* Internal iterator of scream. */
1644 SV* oreplsv = GvSV(PL_replgv);
1645 const bool do_utf8 = DO_UTF8(sv);
1646 const I32 multiline = prog->reganch & PMf_MULTILINE;
1648 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
1649 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
1652 GET_RE_DEBUG_FLAGS_DECL;
1654 PERL_UNUSED_ARG(data);
1655 RX_MATCH_UTF8_set(prog,do_utf8);
1661 PL_regnarrate = DEBUG_r_TEST;
1664 /* Be paranoid... */
1665 if (prog == NULL || startpos == NULL) {
1666 Perl_croak(aTHX_ "NULL regexp parameter");
1670 minlen = prog->minlen;
1671 if (strend - startpos < minlen) {
1672 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1673 "String too short [regexec_flags]...\n"));
1677 /* Check validity of program. */
1678 if (UCHARAT(prog->program) != REG_MAGIC) {
1679 Perl_croak(aTHX_ "corrupted regexp program");
1683 PL_reg_eval_set = 0;
1686 if (prog->reganch & ROPT_UTF8)
1687 PL_reg_flags |= RF_utf8;
1689 /* Mark beginning of line for ^ and lookbehind. */
1690 PL_regbol = startpos;
1694 /* Mark end of line for $ (and such) */
1697 /* see how far we have to get to not match where we matched before */
1698 PL_regtill = startpos+minend;
1700 /* We start without call_cc context. */
1703 /* If there is a "must appear" string, look for it. */
1706 if (prog->reganch & ROPT_GPOS_SEEN) { /* Need to have PL_reg_ganch */
1709 if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
1710 PL_reg_ganch = startpos;
1711 else if (sv && SvTYPE(sv) >= SVt_PVMG
1713 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
1714 && mg->mg_len >= 0) {
1715 PL_reg_ganch = strbeg + mg->mg_len; /* Defined pos() */
1716 if (prog->reganch & ROPT_ANCH_GPOS) {
1717 if (s > PL_reg_ganch)
1722 else /* pos() not defined */
1723 PL_reg_ganch = strbeg;
1726 if (!(flags & REXEC_CHECKED) && (prog->check_substr != Nullsv || prog->check_utf8 != Nullsv)) {
1727 re_scream_pos_data d;
1729 d.scream_olds = &scream_olds;
1730 d.scream_pos = &scream_pos;
1731 s = re_intuit_start(prog, sv, s, strend, flags, &d);
1733 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
1734 goto phooey; /* not present */
1739 const char * const s0 = UTF
1740 ? pv_uni_display(dsv0, (U8*)prog->precomp, prog->prelen, 60,
1743 const int len0 = UTF ? SvCUR(dsv0) : prog->prelen;
1744 const char * const s1 = do_utf8 ? sv_uni_display(dsv1, sv, 60,
1745 UNI_DISPLAY_REGEX) : startpos;
1746 const int len1 = do_utf8 ? SvCUR(dsv1) : strend - startpos;
1749 PerlIO_printf(Perl_debug_log,
1750 "%sMatching REx%s \"%s%*.*s%s%s\" against \"%s%.*s%s%s\"\n",
1751 PL_colors[4], PL_colors[5], PL_colors[0],
1754 len0 > 60 ? "..." : "",
1756 (int)(len1 > 60 ? 60 : len1),
1758 (len1 > 60 ? "..." : "")
1762 /* Simplest case: anchored match need be tried only once. */
1763 /* [unless only anchor is BOL and multiline is set] */
1764 if (prog->reganch & (ROPT_ANCH & ~ROPT_ANCH_GPOS)) {
1765 if (s == startpos && regtry(prog, startpos))
1767 else if (multiline || (prog->reganch & ROPT_IMPLICIT)
1768 || (prog->reganch & ROPT_ANCH_MBOL)) /* XXXX SBOL? */
1773 dontbother = minlen - 1;
1774 end = HOP3c(strend, -dontbother, strbeg) - 1;
1775 /* for multiline we only have to try after newlines */
1776 if (prog->check_substr || prog->check_utf8) {
1780 if (regtry(prog, s))
1785 if (prog->reganch & RE_USE_INTUIT) {
1786 s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
1797 if (*s++ == '\n') { /* don't need PL_utf8skip here */
1798 if (regtry(prog, s))
1805 } else if (prog->reganch & ROPT_ANCH_GPOS) {
1806 if (regtry(prog, PL_reg_ganch))
1811 /* Messy cases: unanchored match. */
1812 if ((prog->anchored_substr || prog->anchored_utf8) && prog->reganch & ROPT_SKIP) {
1813 /* we have /x+whatever/ */
1814 /* it must be a one character string (XXXX Except UTF?) */
1819 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1820 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1821 ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0];
1824 while (s < strend) {
1826 DEBUG_EXECUTE_r( did_match = 1 );
1827 if (regtry(prog, s)) goto got_it;
1829 while (s < strend && *s == ch)
1836 while (s < strend) {
1838 DEBUG_EXECUTE_r( did_match = 1 );
1839 if (regtry(prog, s)) goto got_it;
1841 while (s < strend && *s == ch)
1847 DEBUG_EXECUTE_r(if (!did_match)
1848 PerlIO_printf(Perl_debug_log,
1849 "Did not find anchored character...\n")
1852 else if (prog->anchored_substr != Nullsv
1853 || prog->anchored_utf8 != Nullsv
1854 || ((prog->float_substr != Nullsv || prog->float_utf8 != Nullsv)
1855 && prog->float_max_offset < strend - s)) {
1860 char *last1; /* Last position checked before */
1864 if (prog->anchored_substr || prog->anchored_utf8) {
1865 if (!(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr))
1866 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1867 must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr;
1868 back_max = back_min = prog->anchored_offset;
1870 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1871 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1872 must = do_utf8 ? prog->float_utf8 : prog->float_substr;
1873 back_max = prog->float_max_offset;
1874 back_min = prog->float_min_offset;
1876 if (must == &PL_sv_undef)
1877 /* could not downgrade utf8 check substring, so must fail */
1880 last = HOP3c(strend, /* Cannot start after this */
1881 -(I32)(CHR_SVLEN(must)
1882 - (SvTAIL(must) != 0) + back_min), strbeg);
1885 last1 = HOPc(s, -1);
1887 last1 = s - 1; /* bogus */
1889 /* XXXX check_substr already used to find "s", can optimize if
1890 check_substr==must. */
1892 dontbother = end_shift;
1893 strend = HOPc(strend, -dontbother);
1894 while ( (s <= last) &&
1895 ((flags & REXEC_SCREAM)
1896 ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg,
1897 end_shift, &scream_pos, 0))
1898 : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend),
1899 (unsigned char*)strend, must,
1900 multiline ? FBMrf_MULTILINE : 0))) ) {
1901 /* we may be pointing at the wrong string */
1902 if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
1903 s = strbeg + (s - SvPVX_const(sv));
1904 DEBUG_EXECUTE_r( did_match = 1 );
1905 if (HOPc(s, -back_max) > last1) {
1906 last1 = HOPc(s, -back_min);
1907 s = HOPc(s, -back_max);
1910 char *t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
1912 last1 = HOPc(s, -back_min);
1916 while (s <= last1) {
1917 if (regtry(prog, s))
1923 while (s <= last1) {
1924 if (regtry(prog, s))
1930 DEBUG_EXECUTE_r(if (!did_match)
1931 PerlIO_printf(Perl_debug_log,
1932 "Did not find %s substr \"%s%.*s%s\"%s...\n",
1933 ((must == prog->anchored_substr || must == prog->anchored_utf8)
1934 ? "anchored" : "floating"),
1936 (int)(SvCUR(must) - (SvTAIL(must)!=0)),
1938 PL_colors[1], (SvTAIL(must) ? "$" : ""))
1942 else if ((c = prog->regstclass)) {
1944 I32 op = (U8)OP(prog->regstclass);
1945 /* don't bother with what can't match */
1946 if (PL_regkind[op] != EXACT && op != CANY)
1947 strend = HOPc(strend, -(minlen - 1));
1950 SV *prop = sv_newmortal();
1958 pv_uni_display(dsv0, (U8*)SvPVX_const(prop), SvCUR(prop), 60,
1959 UNI_DISPLAY_REGEX) :
1961 len0 = UTF ? SvCUR(dsv0) : SvCUR(prop);
1963 sv_uni_display(dsv1, sv, 60, UNI_DISPLAY_REGEX) : s;
1964 len1 = UTF ? SvCUR(dsv1) : strend - s;
1965 PerlIO_printf(Perl_debug_log,
1966 "Matching stclass \"%*.*s\" against \"%*.*s\"\n",
1970 if (find_byclass(prog, c, s, strend, 0))
1972 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass...\n"));
1976 if (prog->float_substr != Nullsv || prog->float_utf8 != Nullsv) {
1981 if (!(do_utf8 ? prog->float_utf8 : prog->float_substr))
1982 do_utf8 ? to_utf8_substr(prog) : to_byte_substr(prog);
1983 float_real = do_utf8 ? prog->float_utf8 : prog->float_substr;
1985 if (flags & REXEC_SCREAM) {
1986 last = screaminstr(sv, float_real, s - strbeg,
1987 end_shift, &scream_pos, 1); /* last one */
1989 last = scream_olds; /* Only one occurrence. */
1990 /* we may be pointing at the wrong string */
1991 else if (RX_MATCH_COPIED(prog))
1992 s = strbeg + (s - SvPVX_const(sv));
1996 const char * const little = SvPV_const(float_real, len);
1998 if (SvTAIL(float_real)) {
1999 if (memEQ(strend - len + 1, little, len - 1))
2000 last = strend - len + 1;
2001 else if (!multiline)
2002 last = memEQ(strend - len, little, len)
2003 ? strend - len : Nullch;
2009 last = rninstr(s, strend, little, little + len);
2011 last = strend; /* matching "$" */
2015 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2016 "%sCan't trim the tail, match fails (should not happen)%s\n",
2017 PL_colors[4], PL_colors[5]));
2018 goto phooey; /* Should not happen! */
2020 dontbother = strend - last + prog->float_min_offset;
2022 if (minlen && (dontbother < minlen))
2023 dontbother = minlen - 1;
2024 strend -= dontbother; /* this one's always in bytes! */
2025 /* We don't know much -- general case. */
2028 if (regtry(prog, s))
2037 if (regtry(prog, s))
2039 } while (s++ < strend);
2047 RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
2049 if (PL_reg_eval_set) {
2050 /* Preserve the current value of $^R */
2051 if (oreplsv != GvSV(PL_replgv))
2052 sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
2053 restored, the value remains
2055 restore_pos(aTHX_ 0);
2058 /* make sure $`, $&, $', and $digit will work later */
2059 if ( !(flags & REXEC_NOT_FIRST) ) {
2060 RX_MATCH_COPY_FREE(prog);
2061 if (flags & REXEC_COPY_STR) {
2062 I32 i = PL_regeol - startpos + (stringarg - strbeg);
2063 #ifdef PERL_OLD_COPY_ON_WRITE
2065 || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
2067 PerlIO_printf(Perl_debug_log,
2068 "Copy on write: regexp capture, type %d\n",
2071 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2072 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2073 assert (SvPOKp(prog->saved_copy));
2077 RX_MATCH_COPIED_on(prog);
2078 s = savepvn(strbeg, i);
2084 prog->subbeg = strbeg;
2085 prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
2092 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
2093 PL_colors[4], PL_colors[5]));
2094 if (PL_reg_eval_set)
2095 restore_pos(aTHX_ 0);
2100 - regtry - try match at specific point
2102 STATIC I32 /* 0 failure, 1 success */
2103 S_regtry(pTHX_ regexp *prog, char *startpos)
2109 GET_RE_DEBUG_FLAGS_DECL;
2112 PL_regindent = 0; /* XXXX Not good when matches are reenterable... */
2114 if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) {
2117 PL_reg_eval_set = RS_init;
2118 DEBUG_EXECUTE_r(DEBUG_s(
2119 PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n",
2120 (IV)(PL_stack_sp - PL_stack_base));
2122 SAVEI32(cxstack[cxstack_ix].blk_oldsp);
2123 cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base;
2124 /* Otherwise OP_NEXTSTATE will free whatever on stack now. */
2126 /* Apparently this is not needed, judging by wantarray. */
2127 /* SAVEI8(cxstack[cxstack_ix].blk_gimme);
2128 cxstack[cxstack_ix].blk_gimme = G_SCALAR; */
2131 /* Make $_ available to executed code. */
2132 if (PL_reg_sv != DEFSV) {
2137 if (!(SvTYPE(PL_reg_sv) >= SVt_PVMG && SvMAGIC(PL_reg_sv)
2138 && (mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global)))) {
2139 /* prepare for quick setting of pos */
2140 sv_magic(PL_reg_sv, (SV*)0,
2141 PERL_MAGIC_regex_global, Nullch, 0);
2142 mg = mg_find(PL_reg_sv, PERL_MAGIC_regex_global);
2146 PL_reg_oldpos = mg->mg_len;
2147 SAVEDESTRUCTOR_X(restore_pos, 0);
2149 if (!PL_reg_curpm) {
2150 Newxz(PL_reg_curpm, 1, PMOP);
2153 SV* repointer = newSViv(0);
2154 /* so we know which PL_regex_padav element is PL_reg_curpm */
2155 SvFLAGS(repointer) |= SVf_BREAK;
2156 av_push(PL_regex_padav,repointer);
2157 PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
2158 PL_regex_pad = AvARRAY(PL_regex_padav);
2162 PM_SETRE(PL_reg_curpm, prog);
2163 PL_reg_oldcurpm = PL_curpm;
2164 PL_curpm = PL_reg_curpm;
2165 if (RX_MATCH_COPIED(prog)) {
2166 /* Here is a serious problem: we cannot rewrite subbeg,
2167 since it may be needed if this match fails. Thus
2168 $` inside (?{}) could fail... */
2169 PL_reg_oldsaved = prog->subbeg;
2170 PL_reg_oldsavedlen = prog->sublen;
2171 #ifdef PERL_OLD_COPY_ON_WRITE
2172 PL_nrs = prog->saved_copy;
2174 RX_MATCH_COPIED_off(prog);
2177 PL_reg_oldsaved = Nullch;
2178 prog->subbeg = PL_bostr;
2179 prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
2181 prog->startp[0] = startpos - PL_bostr;
2182 PL_reginput = startpos;
2183 PL_regstartp = prog->startp;
2184 PL_regendp = prog->endp;
2185 PL_reglastparen = &prog->lastparen;
2186 PL_reglastcloseparen = &prog->lastcloseparen;
2187 prog->lastparen = 0;
2188 prog->lastcloseparen = 0;
2190 DEBUG_EXECUTE_r(PL_reg_starttry = startpos);
2191 if (PL_reg_start_tmpl <= prog->nparens) {
2192 PL_reg_start_tmpl = prog->nparens*3/2 + 3;
2193 if(PL_reg_start_tmp)
2194 Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2196 Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
2199 /* XXXX What this code is doing here?!!! There should be no need
2200 to do this again and again, PL_reglastparen should take care of
2203 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2204 * Actually, the code in regcppop() (which Ilya may be meaning by
2205 * PL_reglastparen), is not needed at all by the test suite
2206 * (op/regexp, op/pat, op/split), but that code is needed, oddly
2207 * enough, for building DynaLoader, or otherwise this
2208 * "Error: '*' not in typemap in DynaLoader.xs, line 164"
2209 * will happen. Meanwhile, this code *is* needed for the
2210 * above-mentioned test suite tests to succeed. The common theme
2211 * on those tests seems to be returning null fields from matches.
2216 if (prog->nparens) {
2217 for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) {
2224 if (regmatch(prog->program + 1)) {
2225 prog->endp[0] = PL_reginput - PL_bostr;
2228 REGCP_UNWIND(lastcp);
2232 #define RE_UNWIND_BRANCH 1
2233 #define RE_UNWIND_BRANCHJ 2
2237 typedef struct { /* XX: makes sense to enlarge it... */
2241 } re_unwind_generic_t;
2254 } re_unwind_branch_t;
2256 typedef union re_unwind_t {
2258 re_unwind_generic_t generic;
2259 re_unwind_branch_t branch;
2262 #define sayYES goto yes
2263 #define sayNO goto no
2264 #define sayNO_ANYOF goto no_anyof
2265 #define sayYES_FINAL goto yes_final
2266 #define sayYES_LOUD goto yes_loud
2267 #define sayNO_FINAL goto no_final
2268 #define sayNO_SILENT goto do_no
2269 #define saySAME(x) if (x) goto yes; else goto no
2271 #define POSCACHE_SUCCESS 0 /* caching success rather than failure */
2272 #define POSCACHE_SEEN 1 /* we know what we're caching */
2273 #define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */
2274 #define CACHEsayYES STMT_START { \
2275 if (cache_offset | cache_bit) { \
2276 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2277 PL_reg_poscache[0] |= (1<<POSCACHE_SUCCESS) || (1<<POSCACHE_SEEN); \
2278 else if (!(PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2279 /* cache records failure, but this is success */ \
2281 PerlIO_printf(Perl_debug_log, \
2282 "%*s (remove success from failure cache)\n", \
2283 REPORT_CODE_OFF+PL_regindent*2, "") \
2285 PL_reg_poscache[cache_offset] &= ~(1<<cache_bit); \
2290 #define CACHEsayNO STMT_START { \
2291 if (cache_offset | cache_bit) { \
2292 if (!(PL_reg_poscache[0] & (1<<POSCACHE_SEEN))) \
2293 PL_reg_poscache[0] |= (1<<POSCACHE_SEEN); \
2294 else if ((PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))) { \
2295 /* cache records success, but this is failure */ \
2297 PerlIO_printf(Perl_debug_log, \
2298 "%*s (remove failure from success cache)\n", \
2299 REPORT_CODE_OFF+PL_regindent*2, "") \
2301 PL_reg_poscache[cache_offset] &= ~(1<<cache_bit); \
2307 /* this is used to determine how far from the left messages like
2308 'failed...' are printed. Currently 29 makes these messages line
2309 up with the opcode they refer to. Earlier perls used 25 which
2310 left these messages outdented making reviewing a debug output
2313 #define REPORT_CODE_OFF 29
2316 /* Make sure there is a test for this +1 options in re_tests */
2317 #define TRIE_INITAL_ACCEPT_BUFFLEN 4;
2319 #define TRIE_CHECK_STATE_IS_ACCEPTING STMT_START { \
2320 if ( trie->states[ state ].wordnum ) { \
2321 if ( !accepted ) { \
2324 bufflen = TRIE_INITAL_ACCEPT_BUFFLEN ; \
2325 sv_accept_buff=NEWSV( 1234, \
2326 bufflen * sizeof(reg_trie_accepted) - 1 ); \
2327 SvCUR_set( sv_accept_buff, sizeof(reg_trie_accepted) ); \
2328 SvPOK_on( sv_accept_buff ); \
2329 sv_2mortal( sv_accept_buff ); \
2330 accept_buff = (reg_trie_accepted*)SvPV_nolen( sv_accept_buff );\
2332 if ( accepted >= bufflen ) { \
2334 accept_buff =(reg_trie_accepted*)SvGROW( sv_accept_buff, \
2335 bufflen * sizeof(reg_trie_accepted) ); \
2337 SvCUR_set( sv_accept_buff,SvCUR( sv_accept_buff ) \
2338 + sizeof( reg_trie_accepted ) ); \
2340 accept_buff[ accepted ].wordnum = trie->states[ state ].wordnum; \
2341 accept_buff[ accepted ].endpos = uc; \
2345 #define TRIE_HANDLE_CHAR STMT_START { \
2346 if ( uvc < 256 ) { \
2347 charid = trie->charmap[ uvc ]; \
2350 if( trie->widecharmap ) { \
2351 SV** svpp = (SV**)NULL; \
2352 svpp = hv_fetch( trie->widecharmap, (char*)&uvc, \
2353 sizeof( UV ), 0 ); \
2355 charid = (U16)SvIV( *svpp ); \
2360 ( base + charid > trie->uniquecharcount ) && \
2361 ( base + charid - 1 - trie->uniquecharcount < trie->lasttrans) && \
2362 trie->trans[ base + charid - 1 - trie->uniquecharcount ].check == state ) \
2364 state = trie->trans[ base + charid - 1 - trie->uniquecharcount ].next; \
2372 - regmatch - main matching routine
2374 * Conceptually the strategy is simple: check to see whether the current
2375 * node matches, call self recursively to see whether the rest matches,
2376 * and then act accordingly. In practice we make some effort to avoid
2377 * recursion, in particular by going through "ordinary" nodes (that don't
2378 * need to know whether the rest of the match failed) by a loop instead of
2381 /* [lwall] I've hoisted the register declarations to the outer block in order to
2382 * maybe save a little bit of pushing and popping on the stack. It also takes
2383 * advantage of machines that use a register save mask on subroutine entry.
2385 STATIC I32 /* 0 failure, 1 success */
2386 S_regmatch(pTHX_ regnode *prog)
2389 register regnode *scan; /* Current node. */
2390 regnode *next; /* Next node. */
2391 regnode *inner; /* Next node in internal branch. */
2392 register I32 nextchr; /* renamed nextchr - nextchar colides with
2393 function of same name */
2394 register I32 n; /* no or next */
2395 register I32 ln = 0; /* len or last */
2396 register char *s = Nullch; /* operand or save */
2397 register char *locinput = PL_reginput;
2398 register I32 c1 = 0, c2 = 0, paren; /* case fold search, parenth */
2399 int minmod = 0, sw = 0, logical = 0;
2402 /* used by the trie code */
2403 SV *sv_accept_buff = 0; /* accepting states we have traversed */
2404 reg_trie_accepted *accept_buff = 0; /* "" */
2405 reg_trie_data *trie; /* what trie are we using right now */
2406 U32 accepted = 0; /* how many accepting states we have seen*/
2409 I32 firstcp = PL_savestack_ix;
2411 register const bool do_utf8 = PL_reg_match_utf8;
2413 SV *dsv0 = PERL_DEBUG_PAD_ZERO(0);
2414 SV *dsv1 = PERL_DEBUG_PAD_ZERO(1);
2415 SV *dsv2 = PERL_DEBUG_PAD_ZERO(2);
2417 SV *re_debug_flags = NULL;
2419 U32 uniflags = ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY;
2428 /* Note that nextchr is a byte even in UTF */
2429 nextchr = UCHARAT(locinput);
2431 while (scan != NULL) {
2434 SV *prop = sv_newmortal();
2435 const int docolor = *PL_colors[0];
2436 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
2437 int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput);
2438 /* The part of the string before starttry has one color
2439 (pref0_len chars), between starttry and current
2440 position another one (pref_len - pref0_len chars),
2441 after the current position the third one.
2442 We assume that pref0_len <= pref_len, otherwise we
2443 decrease pref0_len. */
2444 int pref_len = (locinput - PL_bostr) > (5 + taill) - l
2445 ? (5 + taill) - l : locinput - PL_bostr;
2448 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
2450 pref0_len = pref_len - (locinput - PL_reg_starttry);
2451 if (l + pref_len < (5 + taill) && l < PL_regeol - locinput)
2452 l = ( PL_regeol - locinput > (5 + taill) - pref_len
2453 ? (5 + taill) - pref_len : PL_regeol - locinput);
2454 while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
2458 if (pref0_len > pref_len)
2459 pref0_len = pref_len;
2460 regprop(prop, scan);
2462 const char * const s0 =
2463 do_utf8 && OP(scan) != CANY ?
2464 pv_uni_display(dsv0, (U8*)(locinput - pref_len),
2465 pref0_len, 60, UNI_DISPLAY_REGEX) :
2466 locinput - pref_len;
2467 const int len0 = do_utf8 ? strlen(s0) : pref0_len;
2468 const char * const s1 = do_utf8 && OP(scan) != CANY ?
2469 pv_uni_display(dsv1, (U8*)(locinput - pref_len + pref0_len),
2470 pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) :
2471 locinput - pref_len + pref0_len;
2472 const int len1 = do_utf8 ? strlen(s1) : pref_len - pref0_len;
2473 const char * const s2 = do_utf8 && OP(scan) != CANY ?
2474 pv_uni_display(dsv2, (U8*)locinput,
2475 PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) :
2477 const int len2 = do_utf8 ? strlen(s2) : l;
2478 PerlIO_printf(Perl_debug_log,
2479 "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|%3"IVdf":%*s%s\n",
2480 (IV)(locinput - PL_bostr),
2487 (docolor ? "" : "> <"),
2491 15 - l - pref_len + 1,
2493 (IV)(scan - PL_regprogram), PL_regindent*2, "",
2498 next = scan + NEXT_OFF(scan);
2504 if (locinput == PL_bostr)
2506 /* regtill = regbol; */
2511 if (locinput == PL_bostr ||
2512 ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
2518 if (locinput == PL_bostr)
2522 if (locinput == PL_reg_ganch)
2528 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2533 if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
2535 if (PL_regeol - locinput > 1)
2539 if (PL_regeol != locinput)
2543 if (!nextchr && locinput >= PL_regeol)
2546 locinput += PL_utf8skip[nextchr];
2547 if (locinput > PL_regeol)
2549 nextchr = UCHARAT(locinput);
2552 nextchr = UCHARAT(++locinput);
2555 if (!nextchr && locinput >= PL_regeol)
2557 nextchr = UCHARAT(++locinput);
2560 if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
2563 locinput += PL_utf8skip[nextchr];
2564 if (locinput > PL_regeol)
2566 nextchr = UCHARAT(locinput);
2569 nextchr = UCHARAT(++locinput);
2575 traverse the TRIE keeping track of all accepting states
2576 we transition through until we get to a failing node.
2578 we use two slightly different pieces of code to handle
2579 the traversal depending on whether its case sensitive or
2580 not. we reuse the accept code however. (this should probably
2581 be turned into a macro.)
2587 U8 *uc = ( U8* )locinput;
2594 U8 *uscan = (U8*)NULL;
2598 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2600 while ( state && uc <= (U8*)PL_regeol ) {
2602 TRIE_CHECK_STATE_IS_ACCEPTING;
2604 base = trie->states[ state ].trans.base;
2606 DEBUG_TRIE_EXECUTE_r(
2607 PerlIO_printf( Perl_debug_log,
2608 "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ",
2609 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2610 (UV)state, (UV)base, (UV)accepted );
2617 uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags );
2622 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2623 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2624 uvc = to_uni_fold( uvc, foldbuf, &foldlen );
2625 foldlen -= UNISKIP( uvc );
2626 uscan = foldbuf + UNISKIP( uvc );
2638 DEBUG_TRIE_EXECUTE_r(
2639 PerlIO_printf( Perl_debug_log,
2640 "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n",
2641 charid, uvc, (UV)state, PL_colors[5] );
2650 /* unreached codepoint: we jump into the middle of the next case
2651 from previous if blocks */
2654 U8 *uc = (U8*)locinput;
2663 trie = (reg_trie_data*)PL_regdata->data[ ARG( scan ) ];
2665 while ( state && uc <= (U8*)PL_regeol ) {
2667 TRIE_CHECK_STATE_IS_ACCEPTING;
2669 base = trie->states[ state ].trans.base;
2671 DEBUG_TRIE_EXECUTE_r(
2672 PerlIO_printf( Perl_debug_log,
2673 "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ",
2674 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2675 (UV)state, (UV)base, (UV)accepted );
2681 uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags );
2692 DEBUG_TRIE_EXECUTE_r(
2693 PerlIO_printf( Perl_debug_log,
2694 "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n",
2695 charid, uvc, (UV)state, PL_colors[5] );
2705 There was at least one accepting state that we
2706 transitioned through. Presumably the number of accepting
2707 states is going to be low, typically one or two. So we
2708 simply scan through to find the one with lowest wordnum.
2709 Once we find it, we swap the last state into its place
2710 and decrement the size. We then try to match the rest of
2711 the pattern at the point where the word ends, if we
2712 succeed then we end the loop, otherwise the loop
2713 eventually terminates once all of the accepting states
2720 if ( accepted == 1 ) {
2722 SV **tmp = av_fetch( trie->words, accept_buff[ 0 ].wordnum-1, 0 );
2723 PerlIO_printf( Perl_debug_log,
2724 "%*s %sonly one match : #%d <%s>%s\n",
2725 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2726 accept_buff[ 0 ].wordnum,
2727 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",
2730 PL_reginput = (char *)accept_buff[ 0 ].endpos;
2731 /* in this case we free tmps/leave before we call regmatch
2732 as we wont be using accept_buff again. */
2735 gotit = regmatch( scan + NEXT_OFF( scan ) );
2738 PerlIO_printf( Perl_debug_log,"%*s %sgot %"IVdf" possible matches%s\n",
2739 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], (IV)accepted,
2742 while ( !gotit && accepted-- ) {
2745 for( cur = 1 ; cur <= accepted ; cur++ ) {
2746 DEBUG_TRIE_EXECUTE_r(
2747 PerlIO_printf( Perl_debug_log,
2748 "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n",
2749 REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4],
2750 (IV)best, accept_buff[ best ].wordnum, (IV)cur,
2751 accept_buff[ cur ].wordnum, PL_colors[5] );
2754 if ( accept_buff[ cur ].wordnum < accept_buff[ best ].wordnum )
2758 SV **tmp = av_fetch( trie->words, accept_buff[ best ].wordnum - 1, 0 );
2759 PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at 0x%p%s\n",
2760 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4],
2761 accept_buff[best].wordnum,
2762 tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr",scan,
2765 if ( best<accepted ) {
2766 reg_trie_accepted tmp = accept_buff[ best ];
2767 accept_buff[ best ] = accept_buff[ accepted ];
2768 accept_buff[ accepted ] = tmp;
2771 PL_reginput = (char *)accept_buff[ best ].endpos;
2774 as far as I can tell we only need the SAVETMPS/FREETMPS
2775 for re's with EVAL in them but I'm leaving them in for
2776 all until I can be sure.
2779 gotit = regmatch( scan + NEXT_OFF( scan ) ) ;
2792 /* unreached codepoint */
2796 if (do_utf8 != UTF) {
2797 /* The target and the pattern have differing utf8ness. */
2799 const char *e = s + ln;
2802 /* The target is utf8, the pattern is not utf8. */
2807 if (NATIVE_TO_UNI(*(U8*)s) !=
2808 utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
2816 /* The target is not utf8, the pattern is utf8. */
2821 if (NATIVE_TO_UNI(*((U8*)l)) !=
2822 utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
2830 nextchr = UCHARAT(locinput);
2833 /* The target and the pattern have the same utf8ness. */
2834 /* Inline the first character, for speed. */
2835 if (UCHARAT(s) != nextchr)
2837 if (PL_regeol - locinput < ln)
2839 if (ln > 1 && memNE(s, locinput, ln))
2842 nextchr = UCHARAT(locinput);
2845 PL_reg_flags |= RF_tainted;
2851 if (do_utf8 || UTF) {
2852 /* Either target or the pattern are utf8. */
2854 char *e = PL_regeol;
2856 if (ibcmp_utf8(s, 0, ln, (bool)UTF,
2857 l, &e, 0, do_utf8)) {
2858 /* One more case for the sharp s:
2859 * pack("U0U*", 0xDF) =~ /ss/i,
2860 * the 0xC3 0x9F are the UTF-8
2861 * byte sequence for the U+00DF. */
2863 toLOWER(s[0]) == 's' &&
2865 toLOWER(s[1]) == 's' &&
2872 nextchr = UCHARAT(locinput);
2876 /* Neither the target and the pattern are utf8. */
2878 /* Inline the first character, for speed. */
2879 if (UCHARAT(s) != nextchr &&
2880 UCHARAT(s) != ((OP(scan) == EXACTF)
2881 ? PL_fold : PL_fold_locale)[nextchr])
2883 if (PL_regeol - locinput < ln)
2885 if (ln > 1 && (OP(scan) == EXACTF
2886 ? ibcmp(s, locinput, ln)
2887 : ibcmp_locale(s, locinput, ln)))
2890 nextchr = UCHARAT(locinput);
2894 STRLEN inclasslen = PL_regeol - locinput;
2896 if (!reginclass(scan, (U8*)locinput, &inclasslen, do_utf8))
2898 if (locinput >= PL_regeol)
2900 locinput += inclasslen ? inclasslen : UTF8SKIP(locinput);
2901 nextchr = UCHARAT(locinput);
2906 nextchr = UCHARAT(locinput);
2907 if (!REGINCLASS(scan, (U8*)locinput))
2909 if (!nextchr && locinput >= PL_regeol)
2911 nextchr = UCHARAT(++locinput);
2915 /* If we might have the case of the German sharp s
2916 * in a casefolding Unicode character class. */
2918 if (ANYOF_FOLD_SHARP_S(scan, locinput, PL_regeol)) {
2919 locinput += SHARP_S_SKIP;
2920 nextchr = UCHARAT(locinput);
2926 PL_reg_flags |= RF_tainted;
2932 LOAD_UTF8_CHARCLASS_ALNUM();
2933 if (!(OP(scan) == ALNUM
2934 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2935 : isALNUM_LC_utf8((U8*)locinput)))
2939 locinput += PL_utf8skip[nextchr];
2940 nextchr = UCHARAT(locinput);
2943 if (!(OP(scan) == ALNUM
2944 ? isALNUM(nextchr) : isALNUM_LC(nextchr)))
2946 nextchr = UCHARAT(++locinput);
2949 PL_reg_flags |= RF_tainted;
2952 if (!nextchr && locinput >= PL_regeol)
2955 LOAD_UTF8_CHARCLASS_ALNUM();
2956 if (OP(scan) == NALNUM
2957 ? swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8)
2958 : isALNUM_LC_utf8((U8*)locinput))
2962 locinput += PL_utf8skip[nextchr];
2963 nextchr = UCHARAT(locinput);
2966 if (OP(scan) == NALNUM
2967 ? isALNUM(nextchr) : isALNUM_LC(nextchr))
2969 nextchr = UCHARAT(++locinput);
2973 PL_reg_flags |= RF_tainted;
2977 /* was last char in word? */
2979 if (locinput == PL_bostr)
2982 const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr);
2984 ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, 0);
2986 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
2987 ln = isALNUM_uni(ln);
2988 LOAD_UTF8_CHARCLASS_ALNUM();
2989 n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8);
2992 ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln));
2993 n = isALNUM_LC_utf8((U8*)locinput);
2997 ln = (locinput != PL_bostr) ?
2998 UCHARAT(locinput - 1) : '\n';
2999 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
3001 n = isALNUM(nextchr);
3004 ln = isALNUM_LC(ln);
3005 n = isALNUM_LC(nextchr);
3008 if (((!ln) == (!n)) == (OP(scan) == BOUND ||
3009 OP(scan) == BOUNDL))
3013 PL_reg_flags |= RF_tainted;
3019 if (UTF8_IS_CONTINUED(nextchr)) {
3020 LOAD_UTF8_CHARCLASS_SPACE();
3021 if (!(OP(scan) == SPACE
3022 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3023 : isSPACE_LC_utf8((U8*)locinput)))
3027 locinput += PL_utf8skip[nextchr];
3028 nextchr = UCHARAT(locinput);
3031 if (!(OP(scan) == SPACE
3032 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3034 nextchr = UCHARAT(++locinput);
3037 if (!(OP(scan) == SPACE
3038 ? isSPACE(nextchr) : isSPACE_LC(nextchr)))
3040 nextchr = UCHARAT(++locinput);
3044 PL_reg_flags |= RF_tainted;
3047 if (!nextchr && locinput >= PL_regeol)
3050 LOAD_UTF8_CHARCLASS_SPACE();
3051 if (OP(scan) == NSPACE
3052 ? swash_fetch(PL_utf8_space, (U8*)locinput, do_utf8)
3053 : isSPACE_LC_utf8((U8*)locinput))
3057 locinput += PL_utf8skip[nextchr];
3058 nextchr = UCHARAT(locinput);
3061 if (OP(scan) == NSPACE
3062 ? isSPACE(nextchr) : isSPACE_LC(nextchr))
3064 nextchr = UCHARAT(++locinput);
3067 PL_reg_flags |= RF_tainted;
3073 LOAD_UTF8_CHARCLASS_DIGIT();
3074 if (!(OP(scan) == DIGIT
3075 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3076 : isDIGIT_LC_utf8((U8*)locinput)))
3080 locinput += PL_utf8skip[nextchr];
3081 nextchr = UCHARAT(locinput);
3084 if (!(OP(scan) == DIGIT
3085 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
3087 nextchr = UCHARAT(++locinput);
3090 PL_reg_flags |= RF_tainted;
3093 if (!nextchr && locinput >= PL_regeol)
3096 LOAD_UTF8_CHARCLASS_DIGIT();
3097 if (OP(scan) == NDIGIT
3098 ? swash_fetch(PL_utf8_digit, (U8*)locinput, do_utf8)
3099 : isDIGIT_LC_utf8((U8*)locinput))
3103 locinput += PL_utf8skip[nextchr];
3104 nextchr = UCHARAT(locinput);
3107 if (OP(scan) == NDIGIT
3108 ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
3110 nextchr = UCHARAT(++locinput);
3113 if (locinput >= PL_regeol)
3116 LOAD_UTF8_CHARCLASS_MARK();
3117 if (swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3119 locinput += PL_utf8skip[nextchr];
3120 while (locinput < PL_regeol &&
3121 swash_fetch(PL_utf8_mark,(U8*)locinput, do_utf8))
3122 locinput += UTF8SKIP(locinput);
3123 if (locinput > PL_regeol)
3128 nextchr = UCHARAT(locinput);
3131 PL_reg_flags |= RF_tainted;
3135 n = ARG(scan); /* which paren pair */
3136 ln = PL_regstartp[n];
3137 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3138 if ((I32)*PL_reglastparen < n || ln == -1)
3139 sayNO; /* Do not match unless seen CLOSEn. */
3140 if (ln == PL_regendp[n])
3144 if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */
3146 const char *e = PL_bostr + PL_regendp[n];
3148 * Note that we can't do the "other character" lookup trick as
3149 * in the 8-bit case (no pun intended) because in Unicode we
3150 * have to map both upper and title case to lower case.
3152 if (OP(scan) == REFF) {
3154 STRLEN ulen1, ulen2;
3155 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3156 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3160 toLOWER_utf8((U8*)s, tmpbuf1, &ulen1);
3161 toLOWER_utf8((U8*)l, tmpbuf2, &ulen2);
3162 if (ulen1 != ulen2 || memNE((char *)tmpbuf1, (char *)tmpbuf2, ulen1))
3169 nextchr = UCHARAT(locinput);
3173 /* Inline the first character, for speed. */
3174 if (UCHARAT(s) != nextchr &&
3176 (UCHARAT(s) != ((OP(scan) == REFF
3177 ? PL_fold : PL_fold_locale)[nextchr]))))
3179 ln = PL_regendp[n] - ln;
3180 if (locinput + ln > PL_regeol)
3182 if (ln > 1 && (OP(scan) == REF
3183 ? memNE(s, locinput, ln)
3185 ? ibcmp(s, locinput, ln)
3186 : ibcmp_locale(s, locinput, ln))))
3189 nextchr = UCHARAT(locinput);
3200 OP_4tree *oop = PL_op;
3201 COP *ocurcop = PL_curcop;
3204 struct regexp *oreg = PL_reg_re;
3207 PL_op = (OP_4tree*)PL_regdata->data[n];
3208 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) );
3209 PAD_SAVE_LOCAL(old_comppad, (PAD*)PL_regdata->data[n + 2]);
3210 PL_regendp[0] = PL_reg_magic->mg_len = locinput - PL_bostr;
3214 CALLRUNOPS(aTHX); /* Scalar context. */
3217 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
3225 PAD_RESTORE_LOCAL(old_comppad);
3226 PL_curcop = ocurcop;
3228 if (logical == 2) { /* Postponed subexpression. */
3230 MAGIC *mg = Null(MAGIC*);
3232 CHECKPOINT cp, lastcp;
3236 if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
3237 mg = mg_find(sv, PERL_MAGIC_qr);
3238 else if (SvSMAGICAL(ret)) {
3239 if (SvGMAGICAL(ret))
3240 sv_unmagic(ret, PERL_MAGIC_qr);
3242 mg = mg_find(ret, PERL_MAGIC_qr);
3246 re = (regexp *)mg->mg_obj;
3247 (void)ReREFCNT_inc(re);
3251 const char *t = SvPV_const(ret, len);
3253 char * const oprecomp = PL_regprecomp;
3254 const I32 osize = PL_regsize;
3255 const I32 onpar = PL_regnpar;
3258 if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8;
3259 re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm);
3261 & (SVs_TEMP | SVs_PADTMP | SVf_READONLY
3263 sv_magic(ret,(SV*)ReREFCNT_inc(re),
3265 PL_regprecomp = oprecomp;
3270 PerlIO_printf(Perl_debug_log,
3271 "Entering embedded \"%s%.60s%s%s\"\n",
3275 (strlen(re->precomp) > 60 ? "..." : ""))
3278 state.prev = PL_reg_call_cc;
3279 state.cc = PL_regcc;
3280 state.re = PL_reg_re;
3284 cp = regcppush(0); /* Save *all* the positions. */
3287 state.ss = PL_savestack_ix;
3288 *PL_reglastparen = 0;
3289 *PL_reglastcloseparen = 0;
3290 PL_reg_call_cc = &state;
3291 PL_reginput = locinput;
3292 toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^
3293 ((re->reganch & ROPT_UTF8) != 0);
3294 if (toggleutf) PL_reg_flags ^= RF_utf8;
3296 /* XXXX This is too dramatic a measure... */
3299 if (regmatch(re->program + 1)) {
3300 /* Even though we succeeded, we need to restore
3301 global variables, since we may be wrapped inside
3302 SUSPEND, thus the match may be not finished yet. */
3304 /* XXXX Do this only if SUSPENDed? */
3305 PL_reg_call_cc = state.prev;
3306 PL_regcc = state.cc;
3307 PL_reg_re = state.re;
3308 cache_re(PL_reg_re);
3309 if (toggleutf) PL_reg_flags ^= RF_utf8;
3311 /* XXXX This is too dramatic a measure... */
3314 /* These are needed even if not SUSPEND. */
3320 REGCP_UNWIND(lastcp);
3322 PL_reg_call_cc = state.prev;
3323 PL_regcc = state.cc;
3324 PL_reg_re = state.re;
3325 cache_re(PL_reg_re);
3326 if (toggleutf) PL_reg_flags ^= RF_utf8;
3328 /* XXXX This is too dramatic a measure... */
3338 sv_setsv(save_scalar(PL_replgv), ret);
3344 n = ARG(scan); /* which paren pair */
3345 PL_reg_start_tmp[n] = locinput;
3350 n = ARG(scan); /* which paren pair */
3351 PL_regstartp[n] = PL_reg_start_tmp[n] - PL_bostr;
3352 PL_regendp[n] = locinput - PL_bostr;
3353 if (n > (I32)*PL_reglastparen)
3354 *PL_reglastparen = n;
3355 *PL_reglastcloseparen = n;
3358 n = ARG(scan); /* which paren pair */
3359 sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1);
3362 PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
3364 next = NEXTOPER(NEXTOPER(scan));
3366 next = scan + ARG(scan);
3367 if (OP(next) == IFTHEN) /* Fake one. */
3368 next = NEXTOPER(NEXTOPER(next));
3372 logical = scan->flags;
3374 /*******************************************************************
3375 PL_regcc contains infoblock about the innermost (...)* loop, and
3376 a pointer to the next outer infoblock.
3378 Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
3380 1) After matching X, regnode for CURLYX is processed;
3382 2) This regnode creates infoblock on the stack, and calls
3383 regmatch() recursively with the starting point at WHILEM node;
3385 3) Each hit of WHILEM node tries to match A and Z (in the order
3386 depending on the current iteration, min/max of {min,max} and
3387 greediness). The information about where are nodes for "A"
3388 and "Z" is read from the infoblock, as is info on how many times "A"
3389 was already matched, and greediness.
3391 4) After A matches, the same WHILEM node is hit again.
3393 5) Each time WHILEM is hit, PL_regcc is the infoblock created by CURLYX
3394 of the same pair. Thus when WHILEM tries to match Z, it temporarily
3395 resets PL_regcc, since this Y(A)*Z can be a part of some other loop:
3396 as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
3397 of the external loop.
3399 Currently present infoblocks form a tree with a stem formed by PL_curcc
3400 and whatever it mentions via ->next, and additional attached trees
3401 corresponding to temporarily unset infoblocks as in "5" above.
3403 In the following picture infoblocks for outer loop of
3404 (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
3405 is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
3406 infoblocks are drawn below the "reset" infoblock.
3408 In fact in the picture below we do not show failed matches for Z and T
3409 by WHILEM blocks. [We illustrate minimal matches, since for them it is
3410 more obvious *why* one needs to *temporary* unset infoblocks.]
3412 Matched REx position InfoBlocks Comment
3416 Y A)*?Z)*?T x <- O <- I
3417 YA )*?Z)*?T x <- O <- I
3418 YA A)*?Z)*?T x <- O <- I
3419 YAA )*?Z)*?T x <- O <- I
3420 YAA Z)*?T x <- O # Temporary unset I
3423 YAAZ Y(A)*?Z)*?T x <- O
3426 YAAZY (A)*?Z)*?T x <- O
3429 YAAZY A)*?Z)*?T x <- O <- I
3432 YAAZYA )*?Z)*?T x <- O <- I
3435 YAAZYA Z)*?T x <- O # Temporary unset I
3441 YAAZYAZ T x # Temporary unset O
3448 *******************************************************************/
3451 CHECKPOINT cp = PL_savestack_ix;
3452 /* No need to save/restore up to this paren */
3453 I32 parenfloor = scan->flags;
3455 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
3457 cc.oldcc = PL_regcc;
3459 /* XXXX Probably it is better to teach regpush to support
3460 parenfloor > PL_regsize... */
3461 if (parenfloor > (I32)*PL_reglastparen)
3462 parenfloor = *PL_reglastparen; /* Pessimization... */
3463 cc.parenfloor = parenfloor;
3465 cc.min = ARG1(scan);
3466 cc.max = ARG2(scan);
3467 cc.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3471 PL_reginput = locinput;
3472 n = regmatch(PREVOPER(next)); /* start on the WHILEM */
3474 PL_regcc = cc.oldcc;
3480 * This is really hard to understand, because after we match
3481 * what we're trying to match, we must make sure the rest of
3482 * the REx is going to match for sure, and to do that we have
3483 * to go back UP the parse tree by recursing ever deeper. And
3484 * if it fails, we have to reset our parent's current state
3485 * that we can try again after backing off.
3488 CHECKPOINT cp, lastcp;
3489 CURCUR* cc = PL_regcc;
3490 char *lastloc = cc->lastloc; /* Detection of 0-len. */
3491 I32 cache_offset = 0, cache_bit = 0;
3493 n = cc->cur + 1; /* how many we know we matched */
3494 PL_reginput = locinput;
3497 PerlIO_printf(Perl_debug_log,
3498 "%*s %ld out of %ld..%ld cc=%"UVxf"\n",
3499 REPORT_CODE_OFF+PL_regindent*2, "",
3500 (long)n, (long)cc->min,
3501 (long)cc->max, PTR2UV(cc))
3504 /* If degenerate scan matches "", assume scan done. */
3506 if (locinput == cc->lastloc && n >= cc->min) {
3507 PL_regcc = cc->oldcc;
3511 PerlIO_printf(Perl_debug_log,
3512 "%*s empty match detected, try continuation...\n",
3513 REPORT_CODE_OFF+PL_regindent*2, "")
3515 if (regmatch(cc->next))
3523 /* First just match a string of min scans. */
3527 cc->lastloc = locinput;
3528 if (regmatch(cc->scan))
3531 cc->lastloc = lastloc;
3536 /* Check whether we already were at this position.
3537 Postpone detection until we know the match is not
3538 *that* much linear. */
3539 if (!PL_reg_maxiter) {
3540 PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
3541 PL_reg_leftiter = PL_reg_maxiter;
3543 if (PL_reg_leftiter-- == 0) {
3544 const I32 size = (PL_reg_maxiter + 7 + POSCACHE_START)/8;
3545 if (PL_reg_poscache) {
3546 if ((I32)PL_reg_poscache_size < size) {
3547 Renew(PL_reg_poscache, size, char);
3548 PL_reg_poscache_size = size;
3550 Zero(PL_reg_poscache, size, char);
3553 PL_reg_poscache_size = size;
3554 Newxz(PL_reg_poscache, size, char);
3557 PerlIO_printf(Perl_debug_log,
3558 "%sDetected a super-linear match, switching on caching%s...\n",
3559 PL_colors[4], PL_colors[5])
3562 if (PL_reg_leftiter < 0) {
3563 cache_offset = locinput - PL_bostr;
3565 cache_offset = (scan->flags & 0xf) - 1 + POSCACHE_START
3566 + cache_offset * (scan->flags>>4);
3567 cache_bit = cache_offset % 8;
3569 if (PL_reg_poscache[cache_offset] & (1<<cache_bit)) {
3571 PerlIO_printf(Perl_debug_log,
3572 "%*s already tried at this position...\n",
3573 REPORT_CODE_OFF+PL_regindent*2, "")
3575 if (PL_reg_poscache[0] & (1<<POSCACHE_SUCCESS))
3576 /* cache records success */
3579 /* cache records failure */
3582 PL_reg_poscache[cache_offset] |= (1<<cache_bit);
3586 /* Prefer next over scan for minimal matching. */
3589 PL_regcc = cc->oldcc;
3592 cp = regcppush(cc->parenfloor);
3594 if (regmatch(cc->next)) {
3596 CACHEsayYES; /* All done. */
3598 REGCP_UNWIND(lastcp);
3604 if (n >= cc->max) { /* Maximum greed exceeded? */
3605 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3606 && !(PL_reg_flags & RF_warned)) {
3607 PL_reg_flags |= RF_warned;
3608 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3609 "Complex regular subexpression recursion",
3616 PerlIO_printf(Perl_debug_log,
3617 "%*s trying longer...\n",
3618 REPORT_CODE_OFF+PL_regindent*2, "")
3620 /* Try scanning more and see if it helps. */
3621 PL_reginput = locinput;
3623 cc->lastloc = locinput;
3624 cp = regcppush(cc->parenfloor);
3626 if (regmatch(cc->scan)) {
3630 REGCP_UNWIND(lastcp);
3633 cc->lastloc = lastloc;
3637 /* Prefer scan over next for maximal matching. */
3639 if (n < cc->max) { /* More greed allowed? */
3640 cp = regcppush(cc->parenfloor);
3642 cc->lastloc = locinput;
3644 if (regmatch(cc->scan)) {
3648 REGCP_UNWIND(lastcp);
3649 regcppop(); /* Restore some previous $<digit>s? */
3650 PL_reginput = locinput;
3652 PerlIO_printf(Perl_debug_log,
3653 "%*s failed, try continuation...\n",
3654 REPORT_CODE_OFF+PL_regindent*2, "")
3657 if (ckWARN(WARN_REGEXP) && n >= REG_INFTY
3658 && !(PL_reg_flags & RF_warned)) {
3659 PL_reg_flags |= RF_warned;
3660 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded",
3661 "Complex regular subexpression recursion",
3665 /* Failed deeper matches of scan, so see if this one works. */
3666 PL_regcc = cc->oldcc;
3669 if (regmatch(cc->next))
3675 cc->lastloc = lastloc;
3680 next = scan + ARG(scan);
3683 inner = NEXTOPER(NEXTOPER(scan));
3686 inner = NEXTOPER(scan);
3690 if (OP(next) != c1) /* No choice. */
3691 next = inner; /* Avoid recursion. */
3693 const I32 lastparen = *PL_reglastparen;
3695 re_unwind_branch_t *uw;
3697 /* Put unwinding data on stack */
3698 unwind1 = SSNEWt(1,re_unwind_branch_t);
3699 uw = SSPTRt(unwind1,re_unwind_branch_t);
3702 uw->type = ((c1 == BRANCH)
3704 : RE_UNWIND_BRANCHJ);
3705 uw->lastparen = lastparen;
3707 uw->locinput = locinput;
3708 uw->nextchr = nextchr;
3710 uw->regindent = ++PL_regindent;
3713 REGCP_SET(uw->lastcp);
3715 /* Now go into the first branch */
3728 /* We suppose that the next guy does not need
3729 backtracking: in particular, it is of constant non-zero length,
3730 and has no parenths to influence future backrefs. */
3731 ln = ARG1(scan); /* min to match */
3732 n = ARG2(scan); /* max to match */
3733 paren = scan->flags;
3735 if (paren > PL_regsize)
3737 if (paren > (I32)*PL_reglastparen)
3738 *PL_reglastparen = paren;
3740 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3742 scan += NEXT_OFF(scan); /* Skip former OPEN. */
3743 PL_reginput = locinput;
3746 if (ln && regrepeat_hard(scan, ln, &l) < ln)
3748 locinput = PL_reginput;
3749 if (HAS_TEXT(next) || JUMPABLE(next)) {
3750 regnode *text_node = next;
3752 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3754 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3756 if (PL_regkind[(U8)OP(text_node)] == REF) {
3760 else { c1 = (U8)*STRING(text_node); }
3761 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3763 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3764 c2 = PL_fold_locale[c1];
3773 while (n >= ln || (n == REG_INFTY && ln > 0)) { /* ln overflow ? */
3774 /* If it could work, try it. */
3776 UCHARAT(PL_reginput) == c1 ||
3777 UCHARAT(PL_reginput) == c2)
3781 PL_regstartp[paren] =
3782 HOPc(PL_reginput, -l) - PL_bostr;
3783 PL_regendp[paren] = PL_reginput - PL_bostr;
3786 PL_regendp[paren] = -1;
3790 REGCP_UNWIND(lastcp);
3792 /* Couldn't or didn't -- move forward. */
3793 PL_reginput = locinput;
3794 if (regrepeat_hard(scan, 1, &l)) {
3796 locinput = PL_reginput;
3803 n = regrepeat_hard(scan, n, &l);
3804 locinput = PL_reginput;
3806 PerlIO_printf(Perl_debug_log,
3807 "%*s matched %"IVdf" times, len=%"IVdf"...\n",
3808 (int)(REPORT_CODE_OFF+PL_regindent*2), "",
3812 if (HAS_TEXT(next) || JUMPABLE(next)) {
3813 regnode *text_node = next;
3815 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3817 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3819 if (PL_regkind[(U8)OP(text_node)] == REF) {
3823 else { c1 = (U8)*STRING(text_node); }
3825 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3827 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3828 c2 = PL_fold_locale[c1];
3839 /* If it could work, try it. */
3841 UCHARAT(PL_reginput) == c1 ||
3842 UCHARAT(PL_reginput) == c2)
3845 PerlIO_printf(Perl_debug_log,
3846 "%*s trying tail with n=%"IVdf"...\n",
3847 (int)(REPORT_CODE_OFF+PL_regindent*2), "", (IV)n)
3851 PL_regstartp[paren] = HOPc(PL_reginput, -l) - PL_bostr;
3852 PL_regendp[paren] = PL_reginput - PL_bostr;
3855 PL_regendp[paren] = -1;
3859 REGCP_UNWIND(lastcp);
3861 /* Couldn't or didn't -- back up. */
3863 locinput = HOPc(locinput, -l);
3864 PL_reginput = locinput;
3871 paren = scan->flags; /* Which paren to set */
3872 if (paren > PL_regsize)
3874 if (paren > (I32)*PL_reglastparen)
3875 *PL_reglastparen = paren;
3876 ln = ARG1(scan); /* min to match */
3877 n = ARG2(scan); /* max to match */
3878 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
3882 ln = ARG1(scan); /* min to match */
3883 n = ARG2(scan); /* max to match */
3884 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
3889 scan = NEXTOPER(scan);
3895 scan = NEXTOPER(scan);
3899 * Lookahead to avoid useless match attempts
3900 * when we know what character comes next.
3904 * Used to only do .*x and .*?x, but now it allows
3905 * for )'s, ('s and (?{ ... })'s to be in the way
3906 * of the quantifier and the EXACT-like node. -- japhy
3909 if (HAS_TEXT(next) || JUMPABLE(next)) {
3911 regnode *text_node = next;
3913 if (! HAS_TEXT(text_node)) FIND_NEXT_IMPT(text_node);
3915 if (! HAS_TEXT(text_node)) c1 = c2 = -1000;
3917 if (PL_regkind[(U8)OP(text_node)] == REF) {
3919 goto assume_ok_easy;
3921 else { s = (U8*)STRING(text_node); }
3925 if (OP(text_node) == EXACTF || OP(text_node) == REFF)
3927 else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL)
3928 c2 = PL_fold_locale[c1];
3931 if (OP(text_node) == EXACTF || OP(text_node) == REFF) {
3932 STRLEN ulen1, ulen2;
3933 U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
3934 U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
3936 to_utf8_lower((U8*)s, tmpbuf1, &ulen1);
3937 to_utf8_upper((U8*)s, tmpbuf2, &ulen2);
3939 c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0,
3941 c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0,
3945 c2 = c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0,
3954 PL_reginput = locinput;
3958 if (ln && regrepeat(scan, ln) < ln)
3960 locinput = PL_reginput;
3963 char *e; /* Should not check after this */
3964 char *old = locinput;
3967 if (n == REG_INFTY) {
3970 while (UTF8_IS_CONTINUATION(*(U8*)e))
3976 m >0 && e + UTF8SKIP(e) <= PL_regeol; m--)
3980 e = locinput + n - ln;
3985 /* Find place 'next' could work */
3988 while (locinput <= e &&
3989 UCHARAT(locinput) != c1)
3992 while (locinput <= e
3993 && UCHARAT(locinput) != c1
3994 && UCHARAT(locinput) != c2)
3997 count = locinput - old;
4002 /* count initialised to
4003 * utf8_distance(old, locinput) */
4004 while (locinput <= e &&
4005 utf8n_to_uvchr((U8*)locinput,
4006 UTF8_MAXBYTES, &len,
4007 uniflags) != (UV)c1) {
4013 /* count initialised to
4014 * utf8_distance(old, locinput) */
4015 while (locinput <= e) {
4016 UV c = utf8n_to_uvchr((U8*)locinput,
4017 UTF8_MAXBYTES, &len,
4019 if (c == (UV)c1 || c == (UV)c2)
4028 /* PL_reginput == old now */
4029 if (locinput != old) {
4030 ln = 1; /* Did some */
4031 if (regrepeat(scan, count) < count)
4034 /* PL_reginput == locinput now */
4035 TRYPAREN(paren, ln, locinput);
4036 PL_reginput = locinput; /* Could be reset... */
4037 REGCP_UNWIND(lastcp);
4038 /* Couldn't or didn't -- move forward. */
4041 locinput += UTF8SKIP(locinput);
4048 while (n >= ln || (n == REG_INFTY && ln > 0)) { /* ln overflow ? */
4052 c = utf8n_to_uvchr((U8*)PL_reginput,
4056 c = UCHARAT(PL_reginput);
4057 /* If it could work, try it. */
4058 if (c == (UV)c1 || c == (UV)c2)
4060 TRYPAREN(paren, ln, PL_reginput);
4061 REGCP_UNWIND(lastcp);
4064 /* If it could work, try it. */
4065 else if (c1 == -1000)
4067 TRYPAREN(paren, ln, PL_reginput);
4068 REGCP_UNWIND(lastcp);
4070 /* Couldn't or didn't -- move forward. */
4071 PL_reginput = locinput;
4072 if (regrepeat(scan, 1)) {
4074 locinput = PL_reginput;
4082 n = regrepeat(scan, n);
4083 locinput = PL_reginput;
4084 if (ln < n && PL_regkind[(U8)OP(next)] == EOL &&
4085 (OP(next) != MEOL ||
4086 OP(next) == SEOL || OP(next) == EOS))
4088 ln = n; /* why back off? */
4089 /* ...because $ and \Z can match before *and* after
4090 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
4091 We should back off by one in this case. */
4092 if (UCHARAT(PL_reginput - 1) == '\n' && OP(next) != EOS)
4101 c = utf8n_to_uvchr((U8*)PL_reginput,
4105 c = UCHARAT(PL_reginput);
4107 /* If it could work, try it. */
4108 if (c1 == -1000 || c == (UV)c1 || c == (UV)c2)
4110 TRYPAREN(paren, n, PL_reginput);
4111 REGCP_UNWIND(lastcp);
4113 /* Couldn't or didn't -- back up. */
4115 PL_reginput = locinput = HOPc(locinput, -1);
4123 c = utf8n_to_uvchr((U8*)PL_reginput,
4127 c = UCHARAT(PL_reginput);
4129 /* If it could work, try it. */
4130 if (c1 == -1000 || c == (UV)c1 || c == (UV)c2)
4132 TRYPAREN(paren, n, PL_reginput);
4133 REGCP_UNWIND(lastcp);
4135 /* Couldn't or didn't -- back up. */
4137 PL_reginput = locinput = HOPc(locinput, -1);
4144 if (PL_reg_call_cc) {
4145 re_cc_state *cur_call_cc = PL_reg_call_cc;
4146 CURCUR *cctmp = PL_regcc;
4147 regexp *re = PL_reg_re;
4148 CHECKPOINT cp, lastcp;
4150 cp = regcppush(0); /* Save *all* the positions. */
4152 regcp_set_to(PL_reg_call_cc->ss); /* Restore parens of
4154 PL_reginput = locinput; /* Make position available to
4156 cache_re(PL_reg_call_cc->re);
4157 PL_regcc = PL_reg_call_cc->cc;
4158 PL_reg_call_cc = PL_reg_call_cc->prev;
4159 if (regmatch(cur_call_cc->node)) {
4160 PL_reg_call_cc = cur_call_cc;
4164 REGCP_UNWIND(lastcp);
4166 PL_reg_call_cc = cur_call_cc;
4172 PerlIO_printf(Perl_debug_log,
4173 "%*s continuation failed...\n",
4174 REPORT_CODE_OFF+PL_regindent*2, "")
4178 if (locinput < PL_regtill) {
4179 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
4180 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
4182 (long)(locinput - PL_reg_starttry),
4183 (long)(PL_regtill - PL_reg_starttry),
4185 sayNO_FINAL; /* Cannot match: too short. */
4187 PL_reginput = locinput; /* put where regtry can find it */
4188 sayYES_FINAL; /* Success! */
4190 PL_reginput = locinput; /* put where regtry can find it */
4191 sayYES_LOUD; /* Success! */
4194 PL_reginput = locinput;
4199 s = HOPBACKc(locinput, scan->flags);
4205 PL_reginput = locinput;
4210 s = HOPBACKc(locinput, scan->flags);
4216 PL_reginput = locinput;
4219 inner = NEXTOPER(NEXTOPER(scan));
4220 if (regmatch(inner) != n) {
4235 if (OP(scan) == SUSPEND) {
4236 locinput = PL_reginput;
4237 nextchr = UCHARAT(locinput);
4242 next = scan + ARG(scan);
4247 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
4248 PTR2UV(scan), OP(scan));
4249 Perl_croak(aTHX_ "regexp memory corruption");
4256 * We get here only if there's trouble -- normally "case END" is
4257 * the terminating point.
4259 Perl_croak(aTHX_ "corrupted regexp pointers");
4265 PerlIO_printf(Perl_debug_log,
4266 "%*s %scould match...%s\n",
4267 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])
4271 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
4272 PL_colors[4], PL_colors[5]));
4278 #if 0 /* Breaks $^R */
4286 PerlIO_printf(Perl_debug_log,
4287 "%*s %sfailed...%s\n",
4288 REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])
4294 re_unwind_t *uw = SSPTRt(unwind,re_unwind_t);
4297 case RE_UNWIND_BRANCH:
4298 case RE_UNWIND_BRANCHJ:
4300 re_unwind_branch_t *uwb = &(uw->branch);
4301 const I32 lastparen = uwb->lastparen;
4303 REGCP_UNWIND(uwb->lastcp);
4304 for (n = *PL_reglastparen; n > lastparen; n--)
4306 *PL_reglastparen = n;
4307 scan = next = uwb->next;
4309 OP(scan) != (uwb->type == RE_UNWIND_BRANCH
4310 ? BRANCH : BRANCHJ) ) { /* Failure */
4317 /* Have more choice yet. Reuse the same uwb. */
4318 if ((n = (uwb->type == RE_UNWIND_BRANCH
4319 ? NEXT_OFF(next) : ARG(next))))
4322 next = NULL; /* XXXX Needn't unwinding in this case... */
4324 next = NEXTOPER(scan);
4325 if (uwb->type == RE_UNWIND_BRANCHJ)
4326 next = NEXTOPER(next);
4327 locinput = uwb->locinput;
4328 nextchr = uwb->nextchr;
4330 PL_regindent = uwb->regindent;
4337 Perl_croak(aTHX_ "regexp unwind memory corruption");
4348 - regrepeat - repeatedly match something simple, report how many
4351 * [This routine now assumes that it will only match on things of length 1.
4352 * That was true before, but now we assume scan - reginput is the count,
4353 * rather than incrementing count on every character. [Er, except utf8.]]
4356 S_regrepeat(pTHX_ const regnode *p, I32 max)
4359 register char *scan;
4361 register char *loceol = PL_regeol;
4362 register I32 hardcount = 0;
4363 register bool do_utf8 = PL_reg_match_utf8;
4366 if (max == REG_INFTY)
4368 else if (max < loceol - scan)
4369 loceol = scan + max;
4374 while (scan < loceol && hardcount < max && *scan != '\n') {
4375 scan += UTF8SKIP(scan);
4379 while (scan < loceol && *scan != '\n')
4386 while (scan < loceol && hardcount < max) {
4387 scan += UTF8SKIP(scan);
4397 case EXACT: /* length of string is 1 */
4399 while (scan < loceol && UCHARAT(scan) == c)
4402 case EXACTF: /* length of string is 1 */
4404 while (scan < loceol &&
4405 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
4408 case EXACTFL: /* length of string is 1 */
4409 PL_reg_flags |= RF_tainted;
4411 while (scan < loceol &&
4412 (UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
4418 while (hardcount < max && scan < loceol &&
4419 reginclass(p, (U8*)scan, 0, do_utf8)) {
4420 scan += UTF8SKIP(scan);
4424 while (scan < loceol && REGINCLASS(p, (U8*)scan))
4431 LOAD_UTF8_CHARCLASS_ALNUM();
4432 while (hardcount < max && scan < loceol &&
4433 swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4434 scan += UTF8SKIP(scan);
4438 while (scan < loceol && isALNUM(*scan))
4443 PL_reg_flags |= RF_tainted;
4446 while (hardcount < max && scan < loceol &&
4447 isALNUM_LC_utf8((U8*)scan)) {
4448 scan += UTF8SKIP(scan);
4452 while (scan < loceol && isALNUM_LC(*scan))
4459 LOAD_UTF8_CHARCLASS_ALNUM();
4460 while (hardcount < max && scan < loceol &&
4461 !swash_fetch(PL_utf8_alnum, (U8*)scan, do_utf8)) {
4462 scan += UTF8SKIP(scan);
4466 while (scan < loceol && !isALNUM(*scan))
4471 PL_reg_flags |= RF_tainted;
4474 while (hardcount < max && scan < loceol &&
4475 !isALNUM_LC_utf8((U8*)scan)) {
4476 scan += UTF8SKIP(scan);
4480 while (scan < loceol && !isALNUM_LC(*scan))
4487 LOAD_UTF8_CHARCLASS_SPACE();
4488 while (hardcount < max && scan < loceol &&
4490 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4491 scan += UTF8SKIP(scan);
4495 while (scan < loceol && isSPACE(*scan))
4500 PL_reg_flags |= RF_tainted;
4503 while (hardcount < max && scan < loceol &&
4504 (*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4505 scan += UTF8SKIP(scan);
4509 while (scan < loceol && isSPACE_LC(*scan))
4516 LOAD_UTF8_CHARCLASS_SPACE();
4517 while (hardcount < max && scan < loceol &&
4519 swash_fetch(PL_utf8_space,(U8*)scan, do_utf8))) {
4520 scan += UTF8SKIP(scan);
4524 while (scan < loceol && !isSPACE(*scan))
4529 PL_reg_flags |= RF_tainted;
4532 while (hardcount < max && scan < loceol &&
4533 !(*scan == ' ' || isSPACE_LC_utf8((U8*)scan))) {
4534 scan += UTF8SKIP(scan);
4538 while (scan < loceol && !isSPACE_LC(*scan))
4545 LOAD_UTF8_CHARCLASS_DIGIT();
4546 while (hardcount < max && scan < loceol &&
4547 swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
4548 scan += UTF8SKIP(scan);
4552 while (scan < loceol && isDIGIT(*scan))
4559 LOAD_UTF8_CHARCLASS_DIGIT();
4560 while (hardcount < max && scan < loceol &&
4561 !swash_fetch(PL_utf8_digit, (U8*)scan, do_utf8)) {
4562 scan += UTF8SKIP(scan);
4566 while (scan < loceol && !isDIGIT(*scan))
4570 default: /* Called on something of 0 width. */
4571 break; /* So match right here or not at all. */
4577 c = scan - PL_reginput;
4581 SV *re_debug_flags = NULL;
4582 SV *prop = sv_newmortal();
4586 PerlIO_printf(Perl_debug_log,
4587 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
4588 REPORT_CODE_OFF+1, "", SvPVX_const(prop),(IV)c,(IV)max);
4596 - regrepeat_hard - repeatedly match something, report total lenth and length
4598 * The repeater is supposed to have constant non-zero length.
4602 S_regrepeat_hard(pTHX_ regnode *p, I32 max, I32 *lp)
4604 register char *scan = Nullch;
4605 register char *start;
4606 register char *loceol = PL_regeol;
4608 I32 count = 0, res = 1;
4613 start = PL_reginput;
4614 if (PL_reg_match_utf8) {
4615 while (PL_reginput < loceol && (scan = PL_reginput, res = regmatch(p))) {
4618 while (start < PL_reginput) {
4620 start += UTF8SKIP(start);
4631 while (PL_reginput < loceol && (scan = PL_reginput, res = regmatch(p))) {
4633 *lp = l = PL_reginput - start;
4634 if (max != REG_INFTY && l*max < loceol - scan)
4635 loceol = scan + l*max;
4648 - regclass_swash - prepare the utf8 swash
4652 Perl_regclass_swash(pTHX_ register const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
4658 if (PL_regdata && PL_regdata->count) {
4659 const U32 n = ARG(node);
4661 if (PL_regdata->what[n] == 's') {
4662 SV * const rv = (SV*)PL_regdata->data[n];
4663 AV * const av = (AV*)SvRV((SV*)rv);
4664 SV **const ary = AvARRAY(av);
4667 /* See the end of regcomp.c:S_regclass() for
4668 * documentation of these array elements. */
4671 a = SvROK(ary[1]) ? &ary[1] : 0;
4672 b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
4676 else if (si && doinit) {
4677 sw = swash_init("utf8", "", si, 1, 0);
4678 (void)av_store(av, 1, sw);
4694 - reginclass - determine if a character falls into a character class
4696 The n is the ANYOF regnode, the p is the target string, lenp
4697 is pointer to the maximum length of how far to go in the p
4698 (if the lenp is zero, UTF8SKIP(p) is used),
4699 do_utf8 tells whether the target string is in UTF-8.
4704 S_reginclass(pTHX_ register const regnode *n, register const U8* p, STRLEN* lenp, register bool do_utf8)
4707 const char flags = ANYOF_FLAGS(n);
4713 if (do_utf8 && !UTF8_IS_INVARIANT(c))
4714 c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
4715 ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
4717 plen = lenp ? *lenp : UNISKIP(NATIVE_TO_UNI(c));
4718 if (do_utf8 || (flags & ANYOF_UNICODE)) {
4721 if (do_utf8 && !ANYOF_RUNTIME(n)) {
4722 if (len != (STRLEN)-1 && c < 256 && ANYOF_BITMAP_TEST(n, c))
4725 if (!match && do_utf8 && (flags & ANYOF_UNICODE_ALL) && c >= 256)
4729 SV * const sw = regclass_swash(n, TRUE, 0, (SV**)&av);
4732 if (swash_fetch(sw, p, do_utf8))
4734 else if (flags & ANYOF_FOLD) {
4735 if (!match && lenp && av) {
4737 for (i = 0; i <= av_len(av); i++) {
4738 SV* const sv = *av_fetch(av, i, FALSE);
4740 const char * const s = SvPV_const(sv, len);
4742 if (len <= plen && memEQ(s, (char*)p, len)) {
4750 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4753 to_utf8_fold(p, tmpbuf, &tmplen);
4754 if (swash_fetch(sw, tmpbuf, do_utf8))
4760 if (match && lenp && *lenp == 0)
4761 *lenp = UNISKIP(NATIVE_TO_UNI(c));
4763 if (!match && c < 256) {
4764 if (ANYOF_BITMAP_TEST(n, c))
4766 else if (flags & ANYOF_FOLD) {
4769 if (flags & ANYOF_LOCALE) {
4770 PL_reg_flags |= RF_tainted;
4771 f = PL_fold_locale[c];
4775 if (f != c && ANYOF_BITMAP_TEST(n, f))
4779 if (!match && (flags & ANYOF_CLASS)) {
4780 PL_reg_flags |= RF_tainted;
4782 (ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) ||
4783 (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
4784 (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) ||
4785 (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
4786 (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
4787 (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
4788 (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
4789 (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
4790 (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) ||
4791 (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
4792 (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) ||
4793 (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) ||
4794 (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
4795 (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
4796 (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
4797 (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
4798 (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) ||
4799 (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
4800 (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) ||
4801 (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
4802 (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
4803 (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
4804 (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) ||
4805 (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
4806 (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) ||
4807 (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) ||
4808 (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) ||
4809 (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) ||
4810 (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) ||
4811 (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c))
4812 ) /* How's that for a conditional? */
4819 return (flags & ANYOF_INVERT) ? !match : match;
4823 S_reghop(pTHX_ U8 *s, I32 off)
4825 return S_reghop3(aTHX_ s, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr));
4829 S_reghop3(pTHX_ U8 *s, I32 off, U8* lim)
4832 while (off-- && s < lim) {
4833 /* XXX could check well-formedness here */
4841 if (UTF8_IS_CONTINUED(*s)) {
4842 while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s))
4845 /* XXX could check well-formedness here */
4853 S_reghopmaybe(pTHX_ U8 *s, I32 off)
4855 return S_reghopmaybe3(aTHX_ s, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr));
4859 S_reghopmaybe3(pTHX_ U8* s, I32 off, U8* lim)
4862 while (off-- && s < lim) {
4863 /* XXX could check well-formedness here */
4873 if (UTF8_IS_CONTINUED(*s)) {
4874 while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s))
4877 /* XXX could check well-formedness here */
4889 restore_pos(pTHX_ void *arg)
4891 PERL_UNUSED_ARG(arg);
4892 if (PL_reg_eval_set) {
4893 if (PL_reg_oldsaved) {
4894 PL_reg_re->subbeg = PL_reg_oldsaved;
4895 PL_reg_re->sublen = PL_reg_oldsavedlen;
4896 #ifdef PERL_OLD_COPY_ON_WRITE
4897 PL_reg_re->saved_copy = PL_nrs;
4899 RX_MATCH_COPIED_on(PL_reg_re);
4901 PL_reg_magic->mg_len = PL_reg_oldpos;
4902 PL_reg_eval_set = 0;
4903 PL_curpm = PL_reg_oldcurpm;
4908 S_to_utf8_substr(pTHX_ register regexp *prog)
4910 if (prog->float_substr && !prog->float_utf8) {
4912 prog->float_utf8 = sv = newSVsv(prog->float_substr);
4913 sv_utf8_upgrade(sv);
4914 if (SvTAIL(prog->float_substr))
4916 if (prog->float_substr == prog->check_substr)
4917 prog->check_utf8 = sv;
4919 if (prog->anchored_substr && !prog->anchored_utf8) {
4921 prog->anchored_utf8 = sv = newSVsv(prog->anchored_substr);
4922 sv_utf8_upgrade(sv);
4923 if (SvTAIL(prog->anchored_substr))
4925 if (prog->anchored_substr == prog->check_substr)
4926 prog->check_utf8 = sv;
4931 S_to_byte_substr(pTHX_ register regexp *prog)
4933 if (prog->float_utf8 && !prog->float_substr) {
4935 prog->float_substr = sv = newSVsv(prog->float_utf8);
4936 if (sv_utf8_downgrade(sv, TRUE)) {
4937 if (SvTAIL(prog->float_utf8))
4941 prog->float_substr = sv = &PL_sv_undef;
4943 if (prog->float_utf8 == prog->check_utf8)
4944 prog->check_substr = sv;
4946 if (prog->anchored_utf8 && !prog->anchored_substr) {
4948 prog->anchored_substr = sv = newSVsv(prog->anchored_utf8);
4949 if (sv_utf8_downgrade(sv, TRUE)) {
4950 if (SvTAIL(prog->anchored_utf8))
4954 prog->anchored_substr = sv = &PL_sv_undef;
4956 if (prog->anchored_utf8 == prog->check_utf8)
4957 prog->check_substr = sv;
4963 * c-indentation-style: bsd
4965 * indent-tabs-mode: t
4968 * ex: set ts=8 sts=4 sw=4 noet: