5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
8 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
9 * confused with the original package (see point 3 below). Thanks, Henry!
12 /* Additional note: this code is very heavily munged from Henry's version
13 * in places. In some spots I've traded clarity for efficiency, so don't
14 * blame Henry for some of the lack of readability.
17 /* The names of the functions have been changed from regcomp and
18 * regexec to pregcomp and pregexec in order to avoid conflicts
19 * with the POSIX routines of the same names.
22 #ifdef PERL_EXT_RE_BUILD
23 /* need to replace pregcomp et al, so enable that */
24 # ifndef PERL_IN_XSUB_RE
25 # define PERL_IN_XSUB_RE
27 /* need access to debugger hooks */
28 # if defined(PERL_EXT_RE_DEBUG) && !defined(DEBUGGING)
33 #ifdef PERL_IN_XSUB_RE
34 /* We *really* need to overwrite these symbols: */
35 # define Perl_pregcomp my_regcomp
36 # define Perl_regdump my_regdump
37 # define Perl_regprop my_regprop
38 # define Perl_pregfree my_regfree
39 # define Perl_re_intuit_string my_re_intuit_string
40 /* *These* symbols are masked to allow static link. */
41 # define Perl_regnext my_regnext
42 # define Perl_save_re_context my_save_re_context
43 # define Perl_reginitcolors my_reginitcolors
45 # define PERL_NO_GET_CONTEXT
50 * pregcomp and pregexec -- regsub and regerror are not used in perl
52 * Copyright (c) 1986 by University of Toronto.
53 * Written by Henry Spencer. Not derived from licensed software.
55 * Permission is granted to anyone to use this software for any
56 * purpose on any computer system, and to redistribute it freely,
57 * subject to the following restrictions:
59 * 1. The author is not responsible for the consequences of use of
60 * this software, no matter how awful, even if they arise
63 * 2. The origin of this software must not be misrepresented, either
64 * by explicit claim or by omission.
66 * 3. Altered versions must be plainly marked as such, and must not
67 * be misrepresented as being the original software.
70 **** Alterations to Henry's code are...
72 **** Copyright (c) 1991-2000, Larry Wall
74 **** You may distribute under the terms of either the GNU General Public
75 **** License or the Artistic License, as specified in the README file.
78 * Beware that some of this code is subtly aware of the way operator
79 * precedence is structured in regular expressions. Serious changes in
80 * regular-expression syntax might require a total rethink.
83 #define PERL_IN_REGCOMP_C
86 #ifdef PERL_IN_XSUB_RE
87 # if defined(PERL_CAPI) || defined(PERL_OBJECT)
102 # if defined(BUGGY_MSC6)
103 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
104 # pragma optimize("a",off)
105 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
106 # pragma optimize("w",on )
107 # endif /* BUGGY_MSC6 */
111 #define STATIC static
114 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
115 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
116 ((*s) == '{' && regcurly(s)))
118 #define PERL_META "^$.[()|?+*\\"
120 #define META "^$.[()|?+*\\"
124 #undef SPSTART /* dratted cpp namespace... */
127 * Flags to be passed up and down.
129 #define WORST 0 /* Worst case. */
130 #define HASWIDTH 0x1 /* Known to match non-null strings. */
131 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
132 #define SPSTART 0x4 /* Starts with * or +. */
133 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
135 /* Length of a variant. */
137 typedef struct scan_data_t {
143 I32 last_end; /* min value, <0 unless valid. */
146 SV **longest; /* Either &l_fixed, or &l_float. */
150 I32 offset_float_min;
151 I32 offset_float_max;
154 struct regnode_charclass_class *start_class;
158 * Forward declarations for pregcomp()'s friends.
161 static scan_data_t zero_scan_data = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
164 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
165 #define SF_BEFORE_SEOL 0x1
166 #define SF_BEFORE_MEOL 0x2
167 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
168 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
171 # define SF_FIX_SHIFT_EOL (0+2)
172 # define SF_FL_SHIFT_EOL (0+4)
174 # define SF_FIX_SHIFT_EOL (+2)
175 # define SF_FL_SHIFT_EOL (+4)
178 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
179 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
181 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
182 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
183 #define SF_IS_INF 0x40
184 #define SF_HAS_PAR 0x80
185 #define SF_IN_PAR 0x100
186 #define SF_HAS_EVAL 0x200
187 #define SCF_DO_SUBSTR 0x400
188 #define SCF_DO_STCLASS_AND 0x0800
189 #define SCF_DO_STCLASS_OR 0x1000
190 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
193 #define UTF (PL_reg_flags & RF_utf8)
194 #define LOC (PL_regflags & PMf_LOCALE)
195 #define FOLD (PL_regflags & PMf_FOLD)
197 #define OOB_CHAR8 1234
198 #define OOB_UTF8 123456
199 #define OOB_NAMEDCLASS -1
201 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
202 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
205 /* length of regex to show in messages that don't mark a position within */
206 #define RegexLengthToShowInErrorMessages 127
209 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
210 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
211 * op/pragma/warn/regcomp.
213 #define MARKER1 "HERE" /* marker as it appears in the description */
214 #define MARKER2 " << HERE " /* marker as it appears within the regex */
216 #define REPORT_LOCATION " before " MARKER1 " mark in regex m/%.*s" MARKER2 "%s/"
219 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
220 * arg. Show regex, up to a maximum length. If it's too long, chop and add
225 char *ellipses = ""; \
226 unsigned len = strlen(PL_regprecomp); \
229 SAVEDESTRUCTOR_X(clear_re,(void*)PL_regcomp_rx); \
231 if (len > RegexLengthToShowInErrorMessages) { \
232 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
233 len = RegexLengthToShowInErrorMessages - 10; \
236 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
237 msg, (int)len, PL_regprecomp, ellipses); \
241 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
242 * args. Show regex, up to a maximum length. If it's too long, chop and add
245 #define FAIL2(pat,msg) \
247 char *ellipses = ""; \
248 unsigned len = strlen(PL_regprecomp); \
251 SAVEDESTRUCTOR_X(clear_re,(void*)PL_regcomp_rx); \
253 if (len > RegexLengthToShowInErrorMessages) { \
254 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
255 len = RegexLengthToShowInErrorMessages - 10; \
258 S_re_croak2(aTHX_ pat, " in regex m/%.*s%s/", \
259 msg, (int)len, PL_regprecomp, ellipses); \
264 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
266 #define Simple_vFAIL(m) \
268 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-PL_regcomp_parse); \
270 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
271 m, (int)offset, PL_regprecomp, PL_regprecomp + offset); \
275 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
280 SAVEDESTRUCTOR_X(clear_re,(void*)PL_regcomp_rx); \
285 * Like Simple_vFAIL(), but accepts two arguments.
287 #define Simple_vFAIL2(m,a1) \
289 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-PL_regcomp_parse); \
291 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
292 (int)offset, PL_regprecomp, PL_regprecomp + offset); \
296 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
298 #define vFAIL2(m,a1) \
301 SAVEDESTRUCTOR_X(clear_re,(void*)PL_regcomp_rx); \
302 Simple_vFAIL2(m, a1); \
307 * Like Simple_vFAIL(), but accepts three arguments.
309 #define Simple_vFAIL3(m, a1, a2) \
311 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-PL_regcomp_parse); \
313 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
314 (int)offset, PL_regprecomp, PL_regprecomp + offset); \
318 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
320 #define vFAIL3(m,a1,a2) \
323 SAVEDESTRUCTOR_X(clear_re,(void*)PL_regcomp_rx); \
324 Simple_vFAIL3(m, a1, a2); \
328 * Like Simple_vFAIL(), but accepts four arguments.
330 #define Simple_vFAIL4(m, a1, a2, a3) \
332 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-PL_regcomp_parse); \
334 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3,\
335 (int)offset, PL_regprecomp, PL_regprecomp + offset); \
339 * Like Simple_vFAIL(), but accepts five arguments.
341 #define Simple_vFAIL5(m, a1, a2, a3, a4) \
343 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-PL_regcomp_parse); \
344 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, a4,\
345 (int)offset, PL_regprecomp, PL_regprecomp + offset); \
349 #define vWARN(loc,m) \
351 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-(loc)); \
352 Perl_warner(aTHX_ WARN_REGEXP, "%s" REPORT_LOCATION,\
353 m, (int)offset, PL_regprecomp, PL_regprecomp + offset); \
357 #define vWARN2(loc, m, a1) \
359 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-(loc)); \
360 Perl_warner(aTHX_ WARN_REGEXP, m REPORT_LOCATION,\
362 (int)offset, PL_regprecomp, PL_regprecomp + offset); \
365 #define vWARN3(loc, m, a1, a2) \
367 unsigned offset = strlen(PL_regprecomp) - (PL_regxend - (loc)); \
368 Perl_warner(aTHX_ WARN_REGEXP, m REPORT_LOCATION, \
370 (int)offset, PL_regprecomp, PL_regprecomp + offset); \
373 #define vWARN4(loc, m, a1, a2, a3) \
375 unsigned offset = strlen(PL_regprecomp)-(PL_regxend-(loc)); \
376 Perl_warner(aTHX_ WARN_REGEXP, m REPORT_LOCATION,\
378 (int)offset, PL_regprecomp, PL_regprecomp + offset); \
383 /* Allow for side effects in s */
384 #define REGC(c,s) STMT_START { if (!SIZE_ONLY) *(s) = (c); else (s);} STMT_END
386 static void clear_re(pTHXo_ void *r);
388 /* Mark that we cannot extend a found fixed substring at this point.
389 Updata the longest found anchored substring and the longest found
390 floating substrings if needed. */
393 S_scan_commit(pTHX_ scan_data_t *data)
396 STRLEN l = CHR_SVLEN(data->last_found);
397 STRLEN old_l = CHR_SVLEN(*data->longest);
399 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
400 sv_setsv(*data->longest, data->last_found);
401 if (*data->longest == data->longest_fixed) {
402 data->offset_fixed = l ? data->last_start_min : data->pos_min;
403 if (data->flags & SF_BEFORE_EOL)
405 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
407 data->flags &= ~SF_FIX_BEFORE_EOL;
410 data->offset_float_min = l ? data->last_start_min : data->pos_min;
411 data->offset_float_max = (l
412 ? data->last_start_max
413 : data->pos_min + data->pos_delta);
414 if (data->flags & SF_BEFORE_EOL)
416 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
418 data->flags &= ~SF_FL_BEFORE_EOL;
421 SvCUR_set(data->last_found, 0);
423 data->flags &= ~SF_BEFORE_EOL;
426 /* Can match anything (initialization) */
428 S_cl_anything(pTHX_ struct regnode_charclass_class *cl)
430 ANYOF_CLASS_ZERO(cl);
431 ANYOF_BITMAP_SETALL(cl);
432 cl->flags = ANYOF_EOS;
434 cl->flags |= ANYOF_LOCALE;
437 /* Can match anything (initialization) */
439 S_cl_is_anything(pTHX_ struct regnode_charclass_class *cl)
443 for (value = 0; value <= ANYOF_MAX; value += 2)
444 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
446 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
447 if (!ANYOF_BITMAP_BYTE(cl, value))
452 /* Can match anything (initialization) */
454 S_cl_init(pTHX_ struct regnode_charclass_class *cl)
456 Zero(cl, 1, struct regnode_charclass_class);
462 S_cl_init_zero(pTHX_ struct regnode_charclass_class *cl)
464 Zero(cl, 1, struct regnode_charclass_class);
468 cl->flags |= ANYOF_LOCALE;
471 /* 'And' a given class with another one. Can create false positives */
472 /* We assume that cl is not inverted */
474 S_cl_and(pTHX_ struct regnode_charclass_class *cl,
475 struct regnode_charclass_class *and_with)
477 if (!(and_with->flags & ANYOF_CLASS)
478 && !(cl->flags & ANYOF_CLASS)
479 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
480 && !(and_with->flags & ANYOF_FOLD)
481 && !(cl->flags & ANYOF_FOLD)) {
484 if (and_with->flags & ANYOF_INVERT)
485 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
486 cl->bitmap[i] &= ~and_with->bitmap[i];
488 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
489 cl->bitmap[i] &= and_with->bitmap[i];
490 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
491 if (!(and_with->flags & ANYOF_EOS))
492 cl->flags &= ~ANYOF_EOS;
495 /* 'OR' a given class with another one. Can create false positives */
496 /* We assume that cl is not inverted */
498 S_cl_or(pTHX_ struct regnode_charclass_class *cl, struct regnode_charclass_class *or_with)
500 if (or_with->flags & ANYOF_INVERT) {
502 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
503 * <= (B1 | !B2) | (CL1 | !CL2)
504 * which is wasteful if CL2 is small, but we ignore CL2:
505 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
506 * XXXX Can we handle case-fold? Unclear:
507 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
508 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
510 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
511 && !(or_with->flags & ANYOF_FOLD)
512 && !(cl->flags & ANYOF_FOLD) ) {
515 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
516 cl->bitmap[i] |= ~or_with->bitmap[i];
517 } /* XXXX: logic is complicated otherwise */
522 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
523 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
524 && (!(or_with->flags & ANYOF_FOLD)
525 || (cl->flags & ANYOF_FOLD)) ) {
528 /* OR char bitmap and class bitmap separately */
529 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
530 cl->bitmap[i] |= or_with->bitmap[i];
531 if (or_with->flags & ANYOF_CLASS) {
532 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
533 cl->classflags[i] |= or_with->classflags[i];
534 cl->flags |= ANYOF_CLASS;
537 else { /* XXXX: logic is complicated, leave it along for a moment. */
541 if (or_with->flags & ANYOF_EOS)
542 cl->flags |= ANYOF_EOS;
545 /* REx optimizer. Converts nodes into quickier variants "in place".
546 Finds fixed substrings. */
548 /* Stops at toplevel WHILEM as well as at `last'. At end *scanp is set
549 to the position after last scanned or to NULL. */
552 S_study_chunk(pTHX_ regnode **scanp, I32 *deltap, regnode *last, scan_data_t *data, U32 flags)
553 /* scanp: Start here (read-write). */
554 /* deltap: Write maxlen-minlen here. */
555 /* last: Stop before this one. */
558 I32 min = 0, pars = 0, code;
559 regnode *scan = *scanp, *next;
561 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
562 int is_inf_internal = 0; /* The studied chunk is infinite */
563 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
564 scan_data_t data_fake;
565 struct regnode_charclass_class and_with; /* Valid if flags & SCF_DO_STCLASS_OR */
567 while (scan && OP(scan) != END && scan < last) {
568 /* Peephole optimizer: */
570 if (PL_regkind[(U8)OP(scan)] == EXACT) {
571 /* Merge several consecutive EXACTish nodes into one. */
572 regnode *n = regnext(scan);
575 regnode *stop = scan;
578 next = scan + NODE_SZ_STR(scan);
579 /* Skip NOTHING, merge EXACT*. */
581 ( PL_regkind[(U8)OP(n)] == NOTHING ||
582 (stringok && (OP(n) == OP(scan))))
584 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
585 if (OP(n) == TAIL || n > next)
587 if (PL_regkind[(U8)OP(n)] == NOTHING) {
588 NEXT_OFF(scan) += NEXT_OFF(n);
589 next = n + NODE_STEP_REGNODE;
597 int oldl = STR_LEN(scan);
598 regnode *nnext = regnext(n);
600 if (oldl + STR_LEN(n) > U8_MAX)
602 NEXT_OFF(scan) += NEXT_OFF(n);
603 STR_LEN(scan) += STR_LEN(n);
604 next = n + NODE_SZ_STR(n);
605 /* Now we can overwrite *n : */
606 Move(STRING(n), STRING(scan) + oldl,
617 n = scan + NODE_SZ_STR(scan);
619 if (PL_regkind[(U8)OP(n)] != NOTHING || OP(n) == NOTHING) {
627 /* Follow the next-chain of the current node and optimize
628 away all the NOTHINGs from it. */
629 if (OP(scan) != CURLYX) {
630 int max = (reg_off_by_arg[OP(scan)]
632 /* I32 may be smaller than U16 on CRAYs! */
633 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
634 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
638 /* Skip NOTHING and LONGJMP. */
639 while ((n = regnext(n))
640 && ((PL_regkind[(U8)OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
641 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
644 if (reg_off_by_arg[OP(scan)])
647 NEXT_OFF(scan) = off;
649 /* The principal pseudo-switch. Cannot be a switch, since we
650 look into several different things. */
651 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
652 || OP(scan) == IFTHEN || OP(scan) == SUSPEND) {
653 next = regnext(scan);
656 if (OP(next) == code || code == IFTHEN || code == SUSPEND) {
657 I32 max1 = 0, min1 = I32_MAX, num = 0;
658 struct regnode_charclass_class accum;
660 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
661 scan_commit(data); /* Cannot merge strings after this. */
662 if (flags & SCF_DO_STCLASS)
663 cl_init_zero(&accum);
664 while (OP(scan) == code) {
665 I32 deltanext, minnext, f = 0;
666 struct regnode_charclass_class this_class;
671 data_fake.whilem_c = data->whilem_c;
672 next = regnext(scan);
673 scan = NEXTOPER(scan);
675 scan = NEXTOPER(scan);
676 if (flags & SCF_DO_STCLASS) {
677 cl_init(&this_class);
678 data_fake.start_class = &this_class;
679 f = SCF_DO_STCLASS_AND;
681 /* we suppose the run is continuous, last=next...*/
682 minnext = study_chunk(&scan, &deltanext, next,
686 if (max1 < minnext + deltanext)
687 max1 = minnext + deltanext;
688 if (deltanext == I32_MAX)
689 is_inf = is_inf_internal = 1;
691 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
693 if (data && (data_fake.flags & SF_HAS_EVAL))
694 data->flags |= SF_HAS_EVAL;
696 data->whilem_c = data_fake.whilem_c;
697 if (flags & SCF_DO_STCLASS)
698 cl_or(&accum, &this_class);
702 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
704 if (flags & SCF_DO_SUBSTR) {
705 data->pos_min += min1;
706 data->pos_delta += max1 - min1;
707 if (max1 != min1 || is_inf)
708 data->longest = &(data->longest_float);
711 delta += max1 - min1;
712 if (flags & SCF_DO_STCLASS_OR) {
713 cl_or(data->start_class, &accum);
715 cl_and(data->start_class, &and_with);
716 flags &= ~SCF_DO_STCLASS;
719 else if (flags & SCF_DO_STCLASS_AND) {
721 cl_and(data->start_class, &accum);
722 flags &= ~SCF_DO_STCLASS;
725 /* Switch to OR mode: cache the old value of
726 * data->start_class */
727 StructCopy(data->start_class, &and_with,
728 struct regnode_charclass_class);
729 flags &= ~SCF_DO_STCLASS_AND;
730 StructCopy(&accum, data->start_class,
731 struct regnode_charclass_class);
732 flags |= SCF_DO_STCLASS_OR;
733 data->start_class->flags |= ANYOF_EOS;
737 else if (code == BRANCHJ) /* single branch is optimized. */
738 scan = NEXTOPER(NEXTOPER(scan));
739 else /* single branch is optimized. */
740 scan = NEXTOPER(scan);
743 else if (OP(scan) == EXACT) {
744 I32 l = STR_LEN(scan);
746 unsigned char *s = (unsigned char *)STRING(scan);
747 unsigned char *e = s + l;
756 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
757 /* The code below prefers earlier match for fixed
758 offset, later match for variable offset. */
759 if (data->last_end == -1) { /* Update the start info. */
760 data->last_start_min = data->pos_min;
761 data->last_start_max = is_inf
762 ? I32_MAX : data->pos_min + data->pos_delta;
764 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
765 data->last_end = data->pos_min + l;
766 data->pos_min += l; /* As in the first entry. */
767 data->flags &= ~SF_BEFORE_EOL;
769 if (flags & SCF_DO_STCLASS_AND) {
770 /* Check whether it is compatible with what we know already! */
773 if (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
774 && !ANYOF_BITMAP_TEST(data->start_class, *STRING(scan))
775 && (!(data->start_class->flags & ANYOF_FOLD)
776 || !ANYOF_BITMAP_TEST(data->start_class,
777 PL_fold[*(U8*)STRING(scan)])))
779 ANYOF_CLASS_ZERO(data->start_class);
780 ANYOF_BITMAP_ZERO(data->start_class);
782 ANYOF_BITMAP_SET(data->start_class, *STRING(scan));
783 data->start_class->flags &= ~ANYOF_EOS;
785 else if (flags & SCF_DO_STCLASS_OR) {
786 /* false positive possible if the class is case-folded */
787 ANYOF_BITMAP_SET(data->start_class, *STRING(scan));
788 data->start_class->flags &= ~ANYOF_EOS;
789 cl_and(data->start_class, &and_with);
791 flags &= ~SCF_DO_STCLASS;
793 else if (PL_regkind[(U8)OP(scan)] == EXACT) { /* But OP != EXACT! */
794 I32 l = STR_LEN(scan);
796 /* Search for fixed substrings supports EXACT only. */
797 if (flags & SCF_DO_SUBSTR)
800 unsigned char *s = (unsigned char *)STRING(scan);
801 unsigned char *e = s + l;
810 if (data && (flags & SCF_DO_SUBSTR))
812 if (flags & SCF_DO_STCLASS_AND) {
813 /* Check whether it is compatible with what we know already! */
816 if (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
817 && !ANYOF_BITMAP_TEST(data->start_class, *STRING(scan))
818 && !ANYOF_BITMAP_TEST(data->start_class,
819 PL_fold[*(U8*)STRING(scan)]))
821 ANYOF_CLASS_ZERO(data->start_class);
822 ANYOF_BITMAP_ZERO(data->start_class);
824 ANYOF_BITMAP_SET(data->start_class, *STRING(scan));
825 data->start_class->flags &= ~ANYOF_EOS;
826 data->start_class->flags |= ANYOF_FOLD;
827 if (OP(scan) == EXACTFL)
828 data->start_class->flags |= ANYOF_LOCALE;
831 else if (flags & SCF_DO_STCLASS_OR) {
832 if (data->start_class->flags & ANYOF_FOLD) {
833 /* false positive possible if the class is case-folded.
834 Assume that the locale settings are the same... */
835 ANYOF_BITMAP_SET(data->start_class, *STRING(scan));
836 data->start_class->flags &= ~ANYOF_EOS;
838 cl_and(data->start_class, &and_with);
840 flags &= ~SCF_DO_STCLASS;
842 else if (strchr((char*)PL_varies,OP(scan))) {
843 I32 mincount, maxcount, minnext, deltanext, fl;
844 I32 f = flags, pos_before = 0;
845 regnode *oscan = scan;
846 struct regnode_charclass_class this_class;
847 struct regnode_charclass_class *oclass = NULL;
849 switch (PL_regkind[(U8)OP(scan)]) {
850 case WHILEM: /* End of (?:...)* . */
851 scan = NEXTOPER(scan);
854 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
855 next = NEXTOPER(scan);
856 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
858 maxcount = REG_INFTY;
859 next = regnext(scan);
860 scan = NEXTOPER(scan);
864 if (flags & SCF_DO_SUBSTR)
869 if (flags & SCF_DO_STCLASS) {
871 maxcount = REG_INFTY;
872 next = regnext(scan);
873 scan = NEXTOPER(scan);
876 is_inf = is_inf_internal = 1;
877 scan = regnext(scan);
878 if (flags & SCF_DO_SUBSTR) {
879 scan_commit(data); /* Cannot extend fixed substrings */
880 data->longest = &(data->longest_float);
882 goto optimize_curly_tail;
884 mincount = ARG1(scan);
885 maxcount = ARG2(scan);
886 next = regnext(scan);
887 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
889 if (flags & SCF_DO_SUBSTR) {
890 if (mincount == 0) scan_commit(data); /* Cannot extend fixed substrings */
891 pos_before = data->pos_min;
895 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
897 data->flags |= SF_IS_INF;
899 if (flags & SCF_DO_STCLASS) {
900 cl_init(&this_class);
901 oclass = data->start_class;
902 data->start_class = &this_class;
903 f |= SCF_DO_STCLASS_AND;
904 f &= ~SCF_DO_STCLASS_OR;
907 /* This will finish on WHILEM, setting scan, or on NULL: */
908 minnext = study_chunk(&scan, &deltanext, last, data,
910 ? (f & ~SCF_DO_SUBSTR) : f);
912 if (flags & SCF_DO_STCLASS)
913 data->start_class = oclass;
914 if (mincount == 0 || minnext == 0) {
915 if (flags & SCF_DO_STCLASS_OR) {
916 cl_or(data->start_class, &this_class);
918 else if (flags & SCF_DO_STCLASS_AND) {
919 /* Switch to OR mode: cache the old value of
920 * data->start_class */
921 StructCopy(data->start_class, &and_with,
922 struct regnode_charclass_class);
923 flags &= ~SCF_DO_STCLASS_AND;
924 StructCopy(&this_class, data->start_class,
925 struct regnode_charclass_class);
926 flags |= SCF_DO_STCLASS_OR;
927 data->start_class->flags |= ANYOF_EOS;
929 } else { /* Non-zero len */
930 if (flags & SCF_DO_STCLASS_OR) {
931 cl_or(data->start_class, &this_class);
932 cl_and(data->start_class, &and_with);
934 else if (flags & SCF_DO_STCLASS_AND)
935 cl_and(data->start_class, &this_class);
936 flags &= ~SCF_DO_STCLASS;
938 if (!scan) /* It was not CURLYX, but CURLY. */
940 if (ckWARN(WARN_REGEXP) && (minnext + deltanext == 0)
941 && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
942 && maxcount <= REG_INFTY/3) /* Complement check for big count */
944 vWARN(PL_regcomp_parse,
945 "Quantifier unexpected on zero-length expression");
948 min += minnext * mincount;
949 is_inf_internal |= ((maxcount == REG_INFTY
950 && (minnext + deltanext) > 0)
951 || deltanext == I32_MAX);
952 is_inf |= is_inf_internal;
953 delta += (minnext + deltanext) * maxcount - minnext * mincount;
955 /* Try powerful optimization CURLYX => CURLYN. */
956 if ( OP(oscan) == CURLYX && data
957 && data->flags & SF_IN_PAR
958 && !(data->flags & SF_HAS_EVAL)
959 && !deltanext && minnext == 1 ) {
960 /* Try to optimize to CURLYN. */
961 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
962 regnode *nxt1 = nxt, *nxt2;
966 if (!strchr((char*)PL_simple,OP(nxt))
967 && !(PL_regkind[(U8)OP(nxt)] == EXACT
968 && STR_LEN(nxt) == 1))
972 if (OP(nxt) != CLOSE)
974 /* Now we know that nxt2 is the only contents: */
975 oscan->flags = ARG(nxt);
977 OP(nxt1) = NOTHING; /* was OPEN. */
979 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
980 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
981 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
982 OP(nxt) = OPTIMIZED; /* was CLOSE. */
983 OP(nxt + 1) = OPTIMIZED; /* was count. */
984 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
989 /* Try optimization CURLYX => CURLYM. */
990 if ( OP(oscan) == CURLYX && data
991 && !(data->flags & SF_HAS_PAR)
992 && !(data->flags & SF_HAS_EVAL)
994 /* XXXX How to optimize if data == 0? */
995 /* Optimize to a simpler form. */
996 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
1000 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
1001 && (OP(nxt2) != WHILEM))
1003 OP(nxt2) = SUCCEED; /* Whas WHILEM */
1004 /* Need to optimize away parenths. */
1005 if (data->flags & SF_IN_PAR) {
1006 /* Set the parenth number. */
1007 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
1009 if (OP(nxt) != CLOSE)
1010 FAIL("Panic opt close");
1011 oscan->flags = ARG(nxt);
1012 OP(nxt1) = OPTIMIZED; /* was OPEN. */
1013 OP(nxt) = OPTIMIZED; /* was CLOSE. */
1015 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
1016 OP(nxt + 1) = OPTIMIZED; /* was count. */
1017 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
1018 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
1021 while ( nxt1 && (OP(nxt1) != WHILEM)) {
1022 regnode *nnxt = regnext(nxt1);
1025 if (reg_off_by_arg[OP(nxt1)])
1026 ARG_SET(nxt1, nxt2 - nxt1);
1027 else if (nxt2 - nxt1 < U16_MAX)
1028 NEXT_OFF(nxt1) = nxt2 - nxt1;
1030 OP(nxt) = NOTHING; /* Cannot beautify */
1035 /* Optimize again: */
1036 study_chunk(&nxt1, &deltanext, nxt, NULL, 0);
1041 else if (OP(oscan) == CURLYX && data && ++data->whilem_c < 16) {
1042 /* This stays as CURLYX, and can put the count/of pair. */
1043 /* Find WHILEM (as in regexec.c) */
1044 regnode *nxt = oscan + NEXT_OFF(oscan);
1046 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
1048 PREVOPER(nxt)->flags = data->whilem_c
1049 | (PL_reg_whilem_seen << 4); /* On WHILEM */
1051 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
1053 if (flags & SCF_DO_SUBSTR) {
1054 SV *last_str = Nullsv;
1055 int counted = mincount != 0;
1057 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
1058 I32 b = pos_before >= data->last_start_min
1059 ? pos_before : data->last_start_min;
1061 char *s = SvPV(data->last_found, l);
1062 I32 old = b - data->last_start_min;
1065 old = utf8_hop((U8*)s, old) - (U8*)s;
1068 /* Get the added string: */
1069 last_str = newSVpvn(s + old, l);
1070 if (deltanext == 0 && pos_before == b) {
1071 /* What was added is a constant string */
1073 SvGROW(last_str, (mincount * l) + 1);
1074 repeatcpy(SvPVX(last_str) + l,
1075 SvPVX(last_str), l, mincount - 1);
1076 SvCUR(last_str) *= mincount;
1077 /* Add additional parts. */
1078 SvCUR_set(data->last_found,
1079 SvCUR(data->last_found) - l);
1080 sv_catsv(data->last_found, last_str);
1081 data->last_end += l * (mincount - 1);
1084 /* start offset must point into the last copy */
1085 data->last_start_min += minnext * (mincount - 1);
1086 data->last_start_max += is_inf ? 0 : (maxcount - 1)
1087 * (minnext + data->pos_delta);
1090 /* It is counted once already... */
1091 data->pos_min += minnext * (mincount - counted);
1092 data->pos_delta += - counted * deltanext +
1093 (minnext + deltanext) * maxcount - minnext * mincount;
1094 if (mincount != maxcount) {
1095 /* Cannot extend fixed substrings found inside
1098 if (mincount && last_str) {
1099 sv_setsv(data->last_found, last_str);
1100 data->last_end = data->pos_min;
1101 data->last_start_min =
1102 data->pos_min - CHR_SVLEN(last_str);
1103 data->last_start_max = is_inf
1105 : data->pos_min + data->pos_delta
1106 - CHR_SVLEN(last_str);
1108 data->longest = &(data->longest_float);
1110 SvREFCNT_dec(last_str);
1112 if (data && (fl & SF_HAS_EVAL))
1113 data->flags |= SF_HAS_EVAL;
1114 optimize_curly_tail:
1115 if (OP(oscan) != CURLYX) {
1116 while (PL_regkind[(U8)OP(next = regnext(oscan))] == NOTHING
1118 NEXT_OFF(oscan) += NEXT_OFF(next);
1121 default: /* REF and CLUMP only? */
1122 if (flags & SCF_DO_SUBSTR) {
1123 scan_commit(data); /* Cannot expect anything... */
1124 data->longest = &(data->longest_float);
1126 is_inf = is_inf_internal = 1;
1127 if (flags & SCF_DO_STCLASS_OR)
1128 cl_anything(data->start_class);
1129 flags &= ~SCF_DO_STCLASS;
1133 else if (strchr((char*)PL_simple,OP(scan)) || PL_regkind[(U8)OP(scan)] == ANYUTF8) {
1136 if (flags & SCF_DO_SUBSTR) {
1141 if (flags & SCF_DO_STCLASS) {
1142 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
1144 /* Some of the logic below assumes that switching
1145 locale on will only add false positives. */
1146 switch (PL_regkind[(U8)OP(scan)]) {
1163 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
1164 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
1165 cl_anything(data->start_class);
1168 if (OP(scan) == SANY)
1170 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
1171 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
1172 || (data->start_class->flags & ANYOF_CLASS));
1173 cl_anything(data->start_class);
1175 if (flags & SCF_DO_STCLASS_AND || !value)
1176 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
1179 if (flags & SCF_DO_STCLASS_AND)
1180 cl_and(data->start_class,
1181 (struct regnode_charclass_class*)scan);
1183 cl_or(data->start_class,
1184 (struct regnode_charclass_class*)scan);
1187 if (flags & SCF_DO_STCLASS_AND) {
1188 if (!(data->start_class->flags & ANYOF_LOCALE)) {
1189 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
1190 for (value = 0; value < 256; value++)
1191 if (!isALNUM(value))
1192 ANYOF_BITMAP_CLEAR(data->start_class, value);
1196 if (data->start_class->flags & ANYOF_LOCALE)
1197 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
1199 for (value = 0; value < 256; value++)
1201 ANYOF_BITMAP_SET(data->start_class, value);
1206 if (flags & SCF_DO_STCLASS_AND) {
1207 if (data->start_class->flags & ANYOF_LOCALE)
1208 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
1211 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
1212 data->start_class->flags |= ANYOF_LOCALE;
1216 if (flags & SCF_DO_STCLASS_AND) {
1217 if (!(data->start_class->flags & ANYOF_LOCALE)) {
1218 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
1219 for (value = 0; value < 256; value++)
1221 ANYOF_BITMAP_CLEAR(data->start_class, value);
1225 if (data->start_class->flags & ANYOF_LOCALE)
1226 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
1228 for (value = 0; value < 256; value++)
1229 if (!isALNUM(value))
1230 ANYOF_BITMAP_SET(data->start_class, value);
1235 if (flags & SCF_DO_STCLASS_AND) {
1236 if (data->start_class->flags & ANYOF_LOCALE)
1237 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
1240 data->start_class->flags |= ANYOF_LOCALE;
1241 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
1245 if (flags & SCF_DO_STCLASS_AND) {
1246 if (!(data->start_class->flags & ANYOF_LOCALE)) {
1247 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
1248 for (value = 0; value < 256; value++)
1249 if (!isSPACE(value))
1250 ANYOF_BITMAP_CLEAR(data->start_class, value);
1254 if (data->start_class->flags & ANYOF_LOCALE)
1255 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
1257 for (value = 0; value < 256; value++)
1259 ANYOF_BITMAP_SET(data->start_class, value);
1264 if (flags & SCF_DO_STCLASS_AND) {
1265 if (data->start_class->flags & ANYOF_LOCALE)
1266 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
1269 data->start_class->flags |= ANYOF_LOCALE;
1270 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
1274 if (flags & SCF_DO_STCLASS_AND) {
1275 if (!(data->start_class->flags & ANYOF_LOCALE)) {
1276 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
1277 for (value = 0; value < 256; value++)
1279 ANYOF_BITMAP_CLEAR(data->start_class, value);
1283 if (data->start_class->flags & ANYOF_LOCALE)
1284 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
1286 for (value = 0; value < 256; value++)
1287 if (!isSPACE(value))
1288 ANYOF_BITMAP_SET(data->start_class, value);
1293 if (flags & SCF_DO_STCLASS_AND) {
1294 if (data->start_class->flags & ANYOF_LOCALE) {
1295 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
1296 for (value = 0; value < 256; value++)
1297 if (!isSPACE(value))
1298 ANYOF_BITMAP_CLEAR(data->start_class, value);
1302 data->start_class->flags |= ANYOF_LOCALE;
1303 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
1307 if (flags & SCF_DO_STCLASS_AND) {
1308 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
1309 for (value = 0; value < 256; value++)
1310 if (!isDIGIT(value))
1311 ANYOF_BITMAP_CLEAR(data->start_class, value);
1314 if (data->start_class->flags & ANYOF_LOCALE)
1315 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
1317 for (value = 0; value < 256; value++)
1319 ANYOF_BITMAP_SET(data->start_class, value);
1324 if (flags & SCF_DO_STCLASS_AND) {
1325 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
1326 for (value = 0; value < 256; value++)
1328 ANYOF_BITMAP_CLEAR(data->start_class, value);
1331 if (data->start_class->flags & ANYOF_LOCALE)
1332 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
1334 for (value = 0; value < 256; value++)
1335 if (!isDIGIT(value))
1336 ANYOF_BITMAP_SET(data->start_class, value);
1341 if (flags & SCF_DO_STCLASS_OR)
1342 cl_and(data->start_class, &and_with);
1343 flags &= ~SCF_DO_STCLASS;
1346 else if (PL_regkind[(U8)OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
1347 data->flags |= (OP(scan) == MEOL
1351 else if ( PL_regkind[(U8)OP(scan)] == BRANCHJ
1352 /* Lookbehind, or need to calculate parens/evals/stclass: */
1353 && (scan->flags || data || (flags & SCF_DO_STCLASS))
1354 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
1355 /* Lookahead/lookbehind */
1356 I32 deltanext, minnext;
1358 struct regnode_charclass_class intrnl;
1361 data_fake.flags = 0;
1363 data_fake.whilem_c = data->whilem_c;
1364 if ( flags & SCF_DO_STCLASS && !scan->flags
1365 && OP(scan) == IFMATCH ) { /* Lookahead */
1367 data_fake.start_class = &intrnl;
1368 f = SCF_DO_STCLASS_AND;
1370 next = regnext(scan);
1371 nscan = NEXTOPER(NEXTOPER(scan));
1372 minnext = study_chunk(&nscan, &deltanext, last, &data_fake, f);
1375 vFAIL("Variable length lookbehind not implemented");
1377 else if (minnext > U8_MAX) {
1378 vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
1380 scan->flags = minnext;
1382 if (data && data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
1384 if (data && (data_fake.flags & SF_HAS_EVAL))
1385 data->flags |= SF_HAS_EVAL;
1387 data->whilem_c = data_fake.whilem_c;
1389 int was = (data->start_class->flags & ANYOF_EOS);
1391 cl_and(data->start_class, &intrnl);
1393 data->start_class->flags |= ANYOF_EOS;
1396 else if (OP(scan) == OPEN) {
1399 else if (OP(scan) == CLOSE && ARG(scan) == is_par) {
1400 next = regnext(scan);
1402 if ( next && (OP(next) != WHILEM) && next < last)
1403 is_par = 0; /* Disable optimization */
1405 else if (OP(scan) == EVAL) {
1407 data->flags |= SF_HAS_EVAL;
1409 else if (OP(scan) == LOGICAL && scan->flags == 2) { /* Embedded follows */
1410 if (flags & SCF_DO_SUBSTR) {
1412 data->longest = &(data->longest_float);
1414 is_inf = is_inf_internal = 1;
1415 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
1416 cl_anything(data->start_class);
1417 flags &= ~SCF_DO_STCLASS;
1419 /* Else: zero-length, ignore. */
1420 scan = regnext(scan);
1425 *deltap = is_inf_internal ? I32_MAX : delta;
1426 if (flags & SCF_DO_SUBSTR && is_inf)
1427 data->pos_delta = I32_MAX - data->pos_min;
1428 if (is_par > U8_MAX)
1430 if (is_par && pars==1 && data) {
1431 data->flags |= SF_IN_PAR;
1432 data->flags &= ~SF_HAS_PAR;
1434 else if (pars && data) {
1435 data->flags |= SF_HAS_PAR;
1436 data->flags &= ~SF_IN_PAR;
1438 if (flags & SCF_DO_STCLASS_OR)
1439 cl_and(data->start_class, &and_with);
1444 S_add_data(pTHX_ I32 n, char *s)
1447 if (PL_regcomp_rx->data) {
1448 Renewc(PL_regcomp_rx->data,
1449 sizeof(*PL_regcomp_rx->data) + sizeof(void*) * (PL_regcomp_rx->data->count + n - 1),
1450 char, struct reg_data);
1451 Renew(PL_regcomp_rx->data->what, PL_regcomp_rx->data->count + n, U8);
1452 PL_regcomp_rx->data->count += n;
1455 Newc(1207, PL_regcomp_rx->data, sizeof(*PL_regcomp_rx->data) + sizeof(void*) * (n - 1),
1456 char, struct reg_data);
1457 New(1208, PL_regcomp_rx->data->what, n, U8);
1458 PL_regcomp_rx->data->count = n;
1460 Copy(s, PL_regcomp_rx->data->what + PL_regcomp_rx->data->count - n, n, U8);
1461 return PL_regcomp_rx->data->count - n;
1465 Perl_reginitcolors(pTHX)
1469 char *s = PerlEnv_getenv("PERL_RE_COLORS");
1472 PL_colors[0] = s = savepv(s);
1474 s = strchr(s, '\t');
1480 PL_colors[i] = s = "";
1484 PL_colors[i++] = "";
1491 - pregcomp - compile a regular expression into internal code
1493 * We can't allocate space until we know how big the compiled form will be,
1494 * but we can't compile it (and thus know how big it is) until we've got a
1495 * place to put the code. So we cheat: we compile it twice, once with code
1496 * generation turned off and size counting turned on, and once "for real".
1497 * This also means that we don't allocate space until we are sure that the
1498 * thing really will compile successfully, and we never have to move the
1499 * code and thus invalidate pointers into it. (Note that it has to be in
1500 * one piece because free() must be able to free it all.) [NB: not true in perl]
1502 * Beware that the optimization-preparation code in here knows about some
1503 * of the structure of the compiled regexp. [I'll say.]
1506 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
1519 FAIL("NULL regexp argument");
1521 if (pm->op_pmdynflags & PMdf_UTF8) {
1522 PL_reg_flags |= RF_utf8;
1527 PL_regprecomp = savepvn(exp, xend - exp);
1528 DEBUG_r(if (!PL_colorset) reginitcolors());
1529 DEBUG_r(PerlIO_printf(Perl_debug_log, "%sCompiling REx%s `%s%*s%s'\n",
1530 PL_colors[4],PL_colors[5],PL_colors[0],
1531 (int)(xend - exp), PL_regprecomp, PL_colors[1]));
1532 PL_regflags = pm->op_pmflags;
1536 PL_seen_zerolen = *exp == '^' ? -1 : 0;
1540 /* First pass: determine size, legality. */
1541 PL_regcomp_parse = exp;
1546 PL_regcode = &PL_regdummy;
1547 PL_reg_whilem_seen = 0;
1548 #if 0 /* REGC() is (currently) a NOP at the first pass.
1549 * Clever compilers notice this and complain. --jhi */
1550 REGC((U8)REG_MAGIC, (char*)PL_regcode);
1552 if (reg(0, &flags) == NULL) {
1553 Safefree(PL_regprecomp);
1554 PL_regprecomp = Nullch;
1557 DEBUG_r(PerlIO_printf(Perl_debug_log, "size %"IVdf" ", (IV)PL_regsize));
1559 /* Small enough for pointer-storage convention?
1560 If extralen==0, this means that we will not need long jumps. */
1561 if (PL_regsize >= 0x10000L && PL_extralen)
1562 PL_regsize += PL_extralen;
1565 if (PL_reg_whilem_seen > 15)
1566 PL_reg_whilem_seen = 15;
1568 /* Allocate space and initialize. */
1569 Newc(1001, r, sizeof(regexp) + (unsigned)PL_regsize * sizeof(regnode),
1572 FAIL("Regexp out of space");
1575 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
1576 Zero(r, sizeof(regexp) + (unsigned)PL_regsize * sizeof(regnode), char);
1579 r->prelen = xend - exp;
1580 r->precomp = PL_regprecomp;
1582 r->reganch = pm->op_pmflags & PMf_COMPILETIME;
1583 r->nparens = PL_regnpar - 1; /* set early to validate backrefs */
1585 r->substrs = 0; /* Useful during FAIL. */
1586 r->startp = 0; /* Useful during FAIL. */
1587 r->endp = 0; /* Useful during FAIL. */
1591 /* Second pass: emit code. */
1592 PL_regcomp_parse = exp;
1596 PL_regcode = r->program;
1597 /* Store the count of eval-groups for security checks: */
1598 PL_regcode->next_off = ((PL_seen_evals > U16_MAX) ? U16_MAX : PL_seen_evals);
1599 REGC((U8)REG_MAGIC, (char*) PL_regcode++);
1601 if (reg(0, &flags) == NULL)
1604 /* Dig out information for optimizations. */
1605 r->reganch = pm->op_pmflags & PMf_COMPILETIME; /* Again? */
1606 pm->op_pmflags = PL_regflags;
1608 r->reganch |= ROPT_UTF8;
1609 r->regstclass = NULL;
1610 if (PL_regnaughty >= 10) /* Probably an expensive pattern. */
1611 r->reganch |= ROPT_NAUGHTY;
1612 scan = r->program + 1; /* First BRANCH. */
1614 /* XXXX To minimize changes to RE engine we always allocate
1615 3-units-long substrs field. */
1616 Newz(1004, r->substrs, 1, struct reg_substr_data);
1618 StructCopy(&zero_scan_data, &data, scan_data_t);
1619 /* XXXX Should not we check for something else? Usually it is OPEN1... */
1620 if (OP(scan) != BRANCH) { /* Only one top-level choice. */
1622 STRLEN longest_float_length, longest_fixed_length;
1623 struct regnode_charclass_class ch_class;
1627 /* Skip introductions and multiplicators >= 1. */
1628 while ((OP(first) == OPEN && (sawopen = 1)) ||
1629 /* An OR of *one* alternative - should not happen now. */
1630 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
1631 (OP(first) == PLUS) ||
1632 (OP(first) == MINMOD) ||
1633 /* An {n,m} with n>0 */
1634 (PL_regkind[(U8)OP(first)] == CURLY && ARG1(first) > 0) ) {
1635 if (OP(first) == PLUS)
1638 first += regarglen[(U8)OP(first)];
1639 first = NEXTOPER(first);
1642 /* Starting-point info. */
1644 if (PL_regkind[(U8)OP(first)] == EXACT) {
1645 if (OP(first) == EXACT); /* Empty, get anchored substr later. */
1646 else if ((OP(first) == EXACTF || OP(first) == EXACTFL)
1648 r->regstclass = first;
1650 else if (strchr((char*)PL_simple,OP(first)))
1651 r->regstclass = first;
1652 else if (PL_regkind[(U8)OP(first)] == BOUND ||
1653 PL_regkind[(U8)OP(first)] == NBOUND)
1654 r->regstclass = first;
1655 else if (PL_regkind[(U8)OP(first)] == BOL) {
1656 r->reganch |= (OP(first) == MBOL
1658 : (OP(first) == SBOL
1661 first = NEXTOPER(first);
1664 else if (OP(first) == GPOS) {
1665 r->reganch |= ROPT_ANCH_GPOS;
1666 first = NEXTOPER(first);
1669 else if ((OP(first) == STAR &&
1670 PL_regkind[(U8)OP(NEXTOPER(first))] == REG_ANY) &&
1671 !(r->reganch & ROPT_ANCH) )
1673 /* turn .* into ^.* with an implied $*=1 */
1674 int type = OP(NEXTOPER(first));
1676 if (type == REG_ANY || type == ANYUTF8)
1677 type = ROPT_ANCH_MBOL;
1679 type = ROPT_ANCH_SBOL;
1681 r->reganch |= type | ROPT_IMPLICIT;
1682 first = NEXTOPER(first);
1685 if (sawplus && (!sawopen || !PL_regsawback)
1686 && !(PL_regseen & REG_SEEN_EVAL)) /* May examine pos and $& */
1687 /* x+ must match at the 1st pos of run of x's */
1688 r->reganch |= ROPT_SKIP;
1690 /* Scan is after the zeroth branch, first is atomic matcher. */
1691 DEBUG_r(PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
1692 (IV)(first - scan + 1)));
1694 * If there's something expensive in the r.e., find the
1695 * longest literal string that must appear and make it the
1696 * regmust. Resolve ties in favor of later strings, since
1697 * the regstart check works with the beginning of the r.e.
1698 * and avoiding duplication strengthens checking. Not a
1699 * strong reason, but sufficient in the absence of others.
1700 * [Now we resolve ties in favor of the earlier string if
1701 * it happens that c_offset_min has been invalidated, since the
1702 * earlier string may buy us something the later one won't.]
1706 data.longest_fixed = newSVpvn("",0);
1707 data.longest_float = newSVpvn("",0);
1708 data.last_found = newSVpvn("",0);
1709 data.longest = &(data.longest_fixed);
1711 if (!r->regstclass) {
1713 data.start_class = &ch_class;
1714 stclass_flag = SCF_DO_STCLASS_AND;
1715 } else /* XXXX Check for BOUND? */
1718 minlen = study_chunk(&first, &fake, scan + PL_regsize, /* Up to end */
1719 &data, SCF_DO_SUBSTR | stclass_flag);
1720 if ( PL_regnpar == 1 && data.longest == &(data.longest_fixed)
1721 && data.last_start_min == 0 && data.last_end > 0
1723 && (!(PL_regseen & REG_SEEN_GPOS) || (r->reganch & ROPT_ANCH_GPOS)))
1724 r->reganch |= ROPT_CHECK_ALL;
1726 SvREFCNT_dec(data.last_found);
1728 longest_float_length = CHR_SVLEN(data.longest_float);
1729 if (longest_float_length
1730 || (data.flags & SF_FL_BEFORE_EOL
1731 && (!(data.flags & SF_FL_BEFORE_MEOL)
1732 || (PL_regflags & PMf_MULTILINE)))) {
1735 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
1736 && data.offset_fixed == data.offset_float_min
1737 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
1738 goto remove_float; /* As in (a)+. */
1740 r->float_substr = data.longest_float;
1741 r->float_min_offset = data.offset_float_min;
1742 r->float_max_offset = data.offset_float_max;
1743 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
1744 && (!(data.flags & SF_FL_BEFORE_MEOL)
1745 || (PL_regflags & PMf_MULTILINE)));
1746 fbm_compile(r->float_substr, t ? FBMcf_TAIL : 0);
1750 r->float_substr = Nullsv;
1751 SvREFCNT_dec(data.longest_float);
1752 longest_float_length = 0;
1755 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
1756 if (longest_fixed_length
1757 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
1758 && (!(data.flags & SF_FIX_BEFORE_MEOL)
1759 || (PL_regflags & PMf_MULTILINE)))) {
1762 r->anchored_substr = data.longest_fixed;
1763 r->anchored_offset = data.offset_fixed;
1764 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
1765 && (!(data.flags & SF_FIX_BEFORE_MEOL)
1766 || (PL_regflags & PMf_MULTILINE)));
1767 fbm_compile(r->anchored_substr, t ? FBMcf_TAIL : 0);
1770 r->anchored_substr = Nullsv;
1771 SvREFCNT_dec(data.longest_fixed);
1772 longest_fixed_length = 0;
1775 && (OP(r->regstclass) == REG_ANY || OP(r->regstclass) == ANYUTF8
1776 || OP(r->regstclass) == SANYUTF8 || OP(r->regstclass) == SANY))
1777 r->regstclass = NULL;
1778 if ((!r->anchored_substr || r->anchored_offset) && stclass_flag
1779 && !(data.start_class->flags & ANYOF_EOS)
1780 && !cl_is_anything(data.start_class)) {
1782 I32 n = add_data(1, "f");
1784 New(1006, PL_regcomp_rx->data->data[n], 1,
1785 struct regnode_charclass_class);
1786 StructCopy(data.start_class,
1787 (struct regnode_charclass_class*)PL_regcomp_rx->data->data[n],
1788 struct regnode_charclass_class);
1789 r->regstclass = (regnode*)PL_regcomp_rx->data->data[n];
1790 r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
1791 DEBUG_r((sv = sv_newmortal(),
1792 regprop(sv, (regnode*)data.start_class),
1793 PerlIO_printf(Perl_debug_log, "synthetic stclass `%s'.\n",
1797 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
1798 if (longest_fixed_length > longest_float_length) {
1799 r->check_substr = r->anchored_substr;
1800 r->check_offset_min = r->check_offset_max = r->anchored_offset;
1801 if (r->reganch & ROPT_ANCH_SINGLE)
1802 r->reganch |= ROPT_NOSCAN;
1805 r->check_substr = r->float_substr;
1806 r->check_offset_min = data.offset_float_min;
1807 r->check_offset_max = data.offset_float_max;
1809 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
1810 This should be changed ASAP! */
1811 if (r->check_substr && !(r->reganch & ROPT_ANCH_GPOS)) {
1812 r->reganch |= RE_USE_INTUIT;
1813 if (SvTAIL(r->check_substr))
1814 r->reganch |= RE_INTUIT_TAIL;
1818 /* Several toplevels. Best we can is to set minlen. */
1820 struct regnode_charclass_class ch_class;
1822 DEBUG_r(PerlIO_printf(Perl_debug_log, "\n"));
1823 scan = r->program + 1;
1825 data.start_class = &ch_class;
1826 minlen = study_chunk(&scan, &fake, scan + PL_regsize, &data, SCF_DO_STCLASS_AND);
1827 r->check_substr = r->anchored_substr = r->float_substr = Nullsv;
1828 if (!(data.start_class->flags & ANYOF_EOS)
1829 && !cl_is_anything(data.start_class)) {
1831 I32 n = add_data(1, "f");
1833 New(1006, PL_regcomp_rx->data->data[n], 1,
1834 struct regnode_charclass_class);
1835 StructCopy(data.start_class,
1836 (struct regnode_charclass_class*)PL_regcomp_rx->data->data[n],
1837 struct regnode_charclass_class);
1838 r->regstclass = (regnode*)PL_regcomp_rx->data->data[n];
1839 r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
1840 DEBUG_r((sv = sv_newmortal(),
1841 regprop(sv, (regnode*)data.start_class),
1842 PerlIO_printf(Perl_debug_log, "synthetic stclass `%s'.\n",
1848 if (PL_regseen & REG_SEEN_GPOS)
1849 r->reganch |= ROPT_GPOS_SEEN;
1850 if (PL_regseen & REG_SEEN_LOOKBEHIND)
1851 r->reganch |= ROPT_LOOKBEHIND_SEEN;
1852 if (PL_regseen & REG_SEEN_EVAL)
1853 r->reganch |= ROPT_EVAL_SEEN;
1854 Newz(1002, r->startp, PL_regnpar, I32);
1855 Newz(1002, r->endp, PL_regnpar, I32);
1856 DEBUG_r(regdump(r));
1861 - reg - regular expression, i.e. main body or parenthesized thing
1863 * Caller must absorb opening parenthesis.
1865 * Combining parenthesis handling with the base level of regular expression
1866 * is a trifle forced, but the need to tie the tails of the branches to what
1867 * follows makes it hard to avoid.
1870 S_reg(pTHX_ I32 paren, I32 *flagp)
1871 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
1874 register regnode *ret; /* Will be the head of the group. */
1875 register regnode *br;
1876 register regnode *lastbr;
1877 register regnode *ender = 0;
1878 register I32 parno = 0;
1879 I32 flags, oregflags = PL_regflags, have_branch = 0, open = 0;
1880 char *oregcomp_parse = PL_regcomp_parse;
1883 *flagp = 0; /* Tentatively. */
1885 /* Make an OPEN node, if parenthesized. */
1887 if (*PL_regcomp_parse == '?') {
1888 U16 posflags = 0, negflags = 0;
1889 U16 *flagsp = &posflags;
1891 char *seqstart = PL_regcomp_parse;
1894 paren = *PL_regcomp_parse++;
1895 ret = NULL; /* For look-ahead/behind. */
1898 PL_regseen |= REG_SEEN_LOOKBEHIND;
1899 if (*PL_regcomp_parse == '!')
1901 if (*PL_regcomp_parse != '=' && *PL_regcomp_parse != '!')
1912 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
1915 while (*PL_regcomp_parse && *PL_regcomp_parse != ')')
1917 if (*PL_regcomp_parse != ')')
1918 FAIL("Sequence (?#... not terminated");
1924 vWARN(PL_regcomp_parse, "(?p{}) is deprecated - use (??{})");
1928 paren = *PL_regcomp_parse++;
1933 I32 count = 1, n = 0;
1935 char *s = PL_regcomp_parse;
1937 OP_4tree *sop, *rop;
1940 PL_regseen |= REG_SEEN_EVAL;
1941 while (count && (c = *PL_regcomp_parse)) {
1942 if (c == '\\' && PL_regcomp_parse[1])
1950 if (*PL_regcomp_parse != ')')
1952 PL_regcomp_parse = s;
1953 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
1958 if (PL_regcomp_parse - 1 - s)
1959 sv = newSVpvn(s, PL_regcomp_parse - 1 - s);
1961 sv = newSVpvn("", 0);
1964 Perl_save_re_context(aTHX);
1965 rop = sv_compile_2op(sv, &sop, "re", &av);
1968 n = add_data(3, "nop");
1969 PL_regcomp_rx->data->data[n] = (void*)rop;
1970 PL_regcomp_rx->data->data[n+1] = (void*)sop;
1971 PL_regcomp_rx->data->data[n+2] = (void*)av;
1974 else { /* First pass */
1975 if (PL_reginterp_cnt < ++PL_seen_evals
1976 && PL_curcop != &PL_compiling)
1977 /* No compiled RE interpolated, has runtime
1978 components ===> unsafe. */
1979 FAIL("Eval-group not allowed at runtime, use re 'eval'");
1981 FAIL("Eval-group in insecure regular expression");
1986 ret = reg_node(LOGICAL);
1989 regtail(ret, reganode(EVAL, n));
1992 return reganode(EVAL, n);
1996 if (PL_regcomp_parse[0] == '?') {
1997 if (PL_regcomp_parse[1] == '=' || PL_regcomp_parse[1] == '!'
1998 || PL_regcomp_parse[1] == '<'
1999 || PL_regcomp_parse[1] == '{') { /* Lookahead or eval. */
2002 ret = reg_node(LOGICAL);
2005 regtail(ret, reg(1, &flag));
2009 else if (PL_regcomp_parse[0] >= '1' && PL_regcomp_parse[0] <= '9' ) {
2010 parno = atoi(PL_regcomp_parse++);
2012 while (isDIGIT(*PL_regcomp_parse))
2014 ret = reganode(GROUPP, parno);
2015 if ((c = *nextchar()) != ')')
2016 vFAIL("Switch condition not recognized");
2018 regtail(ret, reganode(IFTHEN, 0));
2019 br = regbranch(&flags, 1);
2021 br = reganode(LONGJMP, 0);
2023 regtail(br, reganode(LONGJMP, 0));
2028 lastbr = reganode(IFTHEN, 0); /* Fake one for optimizer. */
2029 regbranch(&flags, 1);
2030 regtail(ret, lastbr);
2038 vFAIL("Switch (?(condition)... contains too many branches");
2039 ender = reg_node(TAIL);
2042 regtail(lastbr, ender);
2043 regtail(NEXTOPER(NEXTOPER(lastbr)), ender);
2046 regtail(ret, ender);
2050 vFAIL2("Unknown switch condition (?(%.2s", PL_regcomp_parse);
2054 PL_regcomp_parse--; /* for vFAIL to print correctly */
2055 vFAIL("Sequence (? incomplete");
2060 while (*PL_regcomp_parse && strchr("iogcmsx", *PL_regcomp_parse)) {
2061 if (*PL_regcomp_parse != 'o')
2062 pmflag(flagsp, *PL_regcomp_parse);
2065 if (*PL_regcomp_parse == '-') {
2070 PL_regflags |= posflags;
2071 PL_regflags &= ~negflags;
2072 if (*PL_regcomp_parse == ':') {
2078 if (*PL_regcomp_parse != ')') {
2080 vFAIL3("Sequence (%.*s...) not recognized", PL_regcomp_parse-seqstart, seqstart);
2090 ret = reganode(OPEN, parno);
2097 /* Pick up the branches, linking them together. */
2098 br = regbranch(&flags, 1);
2101 if (*PL_regcomp_parse == '|') {
2102 if (!SIZE_ONLY && PL_extralen) {
2103 reginsert(BRANCHJ, br);
2106 reginsert(BRANCH, br);
2109 PL_extralen += 1; /* For BRANCHJ-BRANCH. */
2111 else if (paren == ':') {
2112 *flagp |= flags&SIMPLE;
2114 if (open) { /* Starts with OPEN. */
2115 regtail(ret, br); /* OPEN -> first. */
2117 else if (paren != '?') /* Not Conditional */
2121 *flagp |= flags&SPSTART;
2123 while (*PL_regcomp_parse == '|') {
2124 if (!SIZE_ONLY && PL_extralen) {
2125 ender = reganode(LONGJMP,0);
2126 regtail(NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
2129 PL_extralen += 2; /* Account for LONGJMP. */
2131 br = regbranch(&flags, 0);
2134 regtail(lastbr, br); /* BRANCH -> BRANCH. */
2138 *flagp |= flags&SPSTART;
2141 if (have_branch || paren != ':') {
2142 /* Make a closing node, and hook it on the end. */
2145 ender = reg_node(TAIL);
2148 ender = reganode(CLOSE, parno);
2154 *flagp &= ~HASWIDTH;
2157 ender = reg_node(SUCCEED);
2160 ender = reg_node(END);
2163 regtail(lastbr, ender);
2166 /* Hook the tails of the branches to the closing node. */
2167 for (br = ret; br != NULL; br = regnext(br)) {
2168 regoptail(br, ender);
2175 static char parens[] = "=!<,>";
2177 if (paren && (p = strchr(parens, paren))) {
2178 int node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
2179 int flag = (p - parens) > 1;
2182 node = SUSPEND, flag = 0;
2183 reginsert(node,ret);
2185 regtail(ret, reg_node(TAIL));
2189 /* Check for proper termination. */
2191 PL_regflags = oregflags;
2192 if (PL_regcomp_parse >= PL_regxend || *nextchar() != ')') {
2193 PL_regcomp_parse = oregcomp_parse;
2194 vFAIL("Unmatched (");
2197 else if (!paren && PL_regcomp_parse < PL_regxend) {
2198 if (*PL_regcomp_parse == ')') {
2200 vFAIL("Unmatched )");
2203 FAIL("Junk on end of regexp"); /* "Can't happen". */
2211 - regbranch - one alternative of an | operator
2213 * Implements the concatenation operator.
2216 S_regbranch(pTHX_ I32 *flagp, I32 first)
2219 register regnode *ret;
2220 register regnode *chain = NULL;
2221 register regnode *latest;
2222 I32 flags = 0, c = 0;
2227 if (!SIZE_ONLY && PL_extralen)
2228 ret = reganode(BRANCHJ,0);
2230 ret = reg_node(BRANCH);
2233 if (!first && SIZE_ONLY)
2234 PL_extralen += 1; /* BRANCHJ */
2236 *flagp = WORST; /* Tentatively. */
2240 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != '|' && *PL_regcomp_parse != ')') {
2242 latest = regpiece(&flags);
2243 if (latest == NULL) {
2244 if (flags & TRYAGAIN)
2248 else if (ret == NULL)
2250 *flagp |= flags&HASWIDTH;
2251 if (chain == NULL) /* First piece. */
2252 *flagp |= flags&SPSTART;
2255 regtail(chain, latest);
2260 if (chain == NULL) { /* Loop ran zero times. */
2261 chain = reg_node(NOTHING);
2266 *flagp |= flags&SIMPLE;
2273 - regpiece - something followed by possible [*+?]
2275 * Note that the branching code sequences used for ? and the general cases
2276 * of * and + are somewhat optimized: they use the same NOTHING node as
2277 * both the endmarker for their branch list and the body of the last branch.
2278 * It might seem that this node could be dispensed with entirely, but the
2279 * endmarker role is not redundant.
2282 S_regpiece(pTHX_ I32 *flagp)
2285 register regnode *ret;
2287 register char *next;
2289 char *origparse = PL_regcomp_parse;
2292 I32 max = REG_INFTY;
2294 ret = regatom(&flags);
2296 if (flags & TRYAGAIN)
2301 op = *PL_regcomp_parse;
2303 if (op == '{' && regcurly(PL_regcomp_parse)) {
2304 next = PL_regcomp_parse + 1;
2306 while (isDIGIT(*next) || *next == ',') {
2315 if (*next == '}') { /* got one */
2319 min = atoi(PL_regcomp_parse);
2323 maxpos = PL_regcomp_parse;
2325 if (!max && *maxpos != '0')
2326 max = REG_INFTY; /* meaning "infinity" */
2327 else if (max >= REG_INFTY)
2328 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
2329 PL_regcomp_parse = next;
2333 if ((flags&SIMPLE)) {
2334 PL_regnaughty += 2 + PL_regnaughty / 2;
2335 reginsert(CURLY, ret);
2338 regnode *w = reg_node(WHILEM);
2342 if (!SIZE_ONLY && PL_extralen) {
2343 reginsert(LONGJMP,ret);
2344 reginsert(NOTHING,ret);
2345 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
2347 reginsert(CURLYX,ret);
2348 if (!SIZE_ONLY && PL_extralen)
2349 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
2350 regtail(ret, reg_node(NOTHING));
2352 PL_reg_whilem_seen++, PL_extralen += 3;
2353 PL_regnaughty += 4 + PL_regnaughty; /* compound interest */
2361 if (max && max < min)
2362 vFAIL("Can't do {n,m} with n > m");
2377 #if 0 /* Now runtime fix should be reliable. */
2379 /* if this is reinstated, don't forget to put this back into perldiag:
2381 =item Regexp *+ operand could be empty at {#} in regex m/%s/
2383 (F) The part of the regexp subject to either the * or + quantifier
2384 could match an empty string. The {#} shows in the regular
2385 expression about where the problem was discovered.
2389 if (!(flags&HASWIDTH) && op != '?')
2390 vFAIL("Regexp *+ operand could be empty");
2395 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
2397 if (op == '*' && (flags&SIMPLE)) {
2398 reginsert(STAR, ret);
2402 else if (op == '*') {
2406 else if (op == '+' && (flags&SIMPLE)) {
2407 reginsert(PLUS, ret);
2411 else if (op == '+') {
2415 else if (op == '?') {
2420 if (ckWARN(WARN_REGEXP) && !SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3) {
2421 vWARN3(PL_regcomp_parse,
2422 "%.*s matches null string many times",
2423 PL_regcomp_parse - origparse,
2427 if (*PL_regcomp_parse == '?') {
2429 reginsert(MINMOD, ret);
2430 regtail(ret, ret + NODE_STEP_REGNODE);
2432 if (ISMULT2(PL_regcomp_parse)) {
2434 vFAIL("Nested quantifiers");
2441 - regatom - the lowest level
2443 * Optimization: gobbles an entire sequence of ordinary characters so that
2444 * it can turn them into a single node, which is smaller to store and
2445 * faster to run. Backslashed characters are exceptions, each becoming a
2446 * separate node; the code is simpler that way and it's not worth fixing.
2448 * [Yes, it is worth fixing, some scripts can run twice the speed.] */
2450 S_regatom(pTHX_ I32 *flagp)
2453 register regnode *ret = 0;
2456 *flagp = WORST; /* Tentatively. */
2459 switch (*PL_regcomp_parse) {
2463 if (PL_regflags & PMf_MULTILINE)
2464 ret = reg_node(MBOL);
2465 else if (PL_regflags & PMf_SINGLELINE)
2466 ret = reg_node(SBOL);
2468 ret = reg_node(BOL);
2472 if (*PL_regcomp_parse)
2474 if (PL_regflags & PMf_MULTILINE)
2475 ret = reg_node(MEOL);
2476 else if (PL_regflags & PMf_SINGLELINE)
2477 ret = reg_node(SEOL);
2479 ret = reg_node(EOL);
2484 if (PL_regflags & PMf_SINGLELINE)
2485 ret = reg_node(SANYUTF8);
2487 ret = reg_node(ANYUTF8);
2491 if (PL_regflags & PMf_SINGLELINE)
2492 ret = reg_node(SANY);
2494 ret = reg_node(REG_ANY);
2495 *flagp |= HASWIDTH|SIMPLE;
2501 char *oregcomp_parse = ++PL_regcomp_parse;
2502 ret = (UTF ? regclassutf8() : regclass());
2503 if (*PL_regcomp_parse != ']') {
2504 PL_regcomp_parse = oregcomp_parse;
2505 vFAIL("Unmatched [");
2508 *flagp |= HASWIDTH|SIMPLE;
2513 ret = reg(1, &flags);
2515 if (flags & TRYAGAIN) {
2516 if (PL_regcomp_parse == PL_regxend) {
2517 /* Make parent create an empty node if needed. */
2525 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
2529 if (flags & TRYAGAIN) {
2533 vFAIL("Internal urp");
2534 /* Supposed to be caught earlier. */
2537 if (!regcurly(PL_regcomp_parse)) {
2546 vFAIL("Quantifier follows nothing");
2549 switch (*++PL_regcomp_parse) {
2552 ret = reg_node(SBOL);
2557 ret = reg_node(GPOS);
2558 PL_regseen |= REG_SEEN_GPOS;
2563 ret = reg_node(SEOL);
2568 ret = reg_node(EOS);
2570 PL_seen_zerolen++; /* Do not optimize RE away */
2574 ret = reg_node(SANY);
2575 *flagp |= HASWIDTH|SIMPLE;
2579 ret = reg_node(CLUMP);
2582 if (UTF && !PL_utf8_mark)
2583 is_utf8_mark((U8*)"~"); /* preload table */
2588 ? (LOC ? ALNUMLUTF8 : ALNUMUTF8)
2589 : (LOC ? ALNUML : ALNUM));
2590 *flagp |= HASWIDTH|SIMPLE;
2592 if (UTF && !PL_utf8_alnum)
2593 is_utf8_alnum((U8*)"a"); /* preload table */
2598 ? (LOC ? NALNUMLUTF8 : NALNUMUTF8)
2599 : (LOC ? NALNUML : NALNUM));
2600 *flagp |= HASWIDTH|SIMPLE;
2602 if (UTF && !PL_utf8_alnum)
2603 is_utf8_alnum((U8*)"a"); /* preload table */
2607 PL_regseen |= REG_SEEN_LOOKBEHIND;
2610 ? (LOC ? BOUNDLUTF8 : BOUNDUTF8)
2611 : (LOC ? BOUNDL : BOUND));
2614 if (UTF && !PL_utf8_alnum)
2615 is_utf8_alnum((U8*)"a"); /* preload table */
2619 PL_regseen |= REG_SEEN_LOOKBEHIND;
2622 ? (LOC ? NBOUNDLUTF8 : NBOUNDUTF8)
2623 : (LOC ? NBOUNDL : NBOUND));
2626 if (UTF && !PL_utf8_alnum)
2627 is_utf8_alnum((U8*)"a"); /* preload table */
2632 ? (LOC ? SPACELUTF8 : SPACEUTF8)
2633 : (LOC ? SPACEL : SPACE));
2634 *flagp |= HASWIDTH|SIMPLE;
2636 if (UTF && !PL_utf8_space)
2637 is_utf8_space((U8*)" "); /* preload table */
2642 ? (LOC ? NSPACELUTF8 : NSPACEUTF8)
2643 : (LOC ? NSPACEL : NSPACE));
2644 *flagp |= HASWIDTH|SIMPLE;
2646 if (UTF && !PL_utf8_space)
2647 is_utf8_space((U8*)" "); /* preload table */
2650 ret = reg_node(UTF ? DIGITUTF8 : DIGIT);
2651 *flagp |= HASWIDTH|SIMPLE;
2653 if (UTF && !PL_utf8_digit)
2654 is_utf8_digit((U8*)"1"); /* preload table */
2657 ret = reg_node(UTF ? NDIGITUTF8 : NDIGIT);
2658 *flagp |= HASWIDTH|SIMPLE;
2660 if (UTF && !PL_utf8_digit)
2661 is_utf8_digit((U8*)"1"); /* preload table */
2665 { /* a lovely hack--pretend we saw [\pX] instead */
2666 char* oldregxend = PL_regxend;
2668 if (PL_regcomp_parse[1] == '{') {
2669 PL_regxend = strchr(PL_regcomp_parse, '}');
2671 PL_regcomp_parse += 2;
2672 PL_regxend = oldregxend;
2673 vFAIL("Missing right brace on \\p{}");
2678 PL_regxend = PL_regcomp_parse + 2;
2681 ret = regclassutf8();
2683 PL_regxend = oldregxend;
2686 *flagp |= HASWIDTH|SIMPLE;
2699 case '1': case '2': case '3': case '4':
2700 case '5': case '6': case '7': case '8': case '9':
2702 I32 num = atoi(PL_regcomp_parse);
2704 if (num > 9 && num >= PL_regnpar)
2707 while (isDIGIT(*PL_regcomp_parse))
2710 if (!SIZE_ONLY && num > PL_regcomp_rx->nparens)
2711 vFAIL("Reference to nonexistent group");
2714 ? (LOC ? REFFL : REFF)
2723 if (PL_regcomp_parse >= PL_regxend)
2724 FAIL("Trailing \\");
2727 /* Do not generate `unrecognized' warnings here, we fall
2728 back into the quick-grab loop below */
2734 if (PL_regflags & PMf_EXTENDED) {
2735 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != '\n') PL_regcomp_parse++;
2736 if (PL_regcomp_parse < PL_regxend)
2742 register STRLEN len;
2752 ? (LOC ? EXACTFL : EXACTF)
2755 for (len = 0, p = PL_regcomp_parse - 1;
2756 len < 127 && p < PL_regxend;
2761 if (PL_regflags & PMf_EXTENDED)
2762 p = regwhite(p, PL_regxend);
2824 char* e = strchr(p, '}');
2827 PL_regcomp_parse = p + 1;
2828 vFAIL("Missing right brace on \\x{}");
2831 numlen = 1; /* allow underscores */
2832 ender = (UV)scan_hex(p + 1, e - p - 1, &numlen);
2833 /* numlen is generous */
2834 if (numlen + len >= 127) {
2842 PL_regcomp_parse = e + 1;
2843 vFAIL("Can't use \\x{} without 'use utf8' declaration");
2848 numlen = 0; /* disallow underscores */
2849 ender = (UV)scan_hex(p, 2, &numlen);
2855 ender = UCHARAT(p++);
2856 ender = toCTRL(ender);
2858 case '0': case '1': case '2': case '3':case '4':
2859 case '5': case '6': case '7': case '8':case '9':
2861 (isDIGIT(p[1]) && atoi(p) >= PL_regnpar) ) {
2862 numlen = 0; /* disallow underscores */
2863 ender = (UV)scan_oct(p, 3, &numlen);
2872 if (p >= PL_regxend)
2873 FAIL("Trailing \\");
2876 if (!SIZE_ONLY && ckWARN(WARN_REGEXP) && isALPHA(*p))
2877 vWARN2(p +1, "Unrecognized escape \\%c passed through", *p);
2878 goto normal_default;
2883 if ((*p & 0xc0) == 0xc0 && UTF) {
2884 ender = utf8_to_uv((U8*)p, PL_regxend - p,
2892 if (PL_regflags & PMf_EXTENDED)
2893 p = regwhite(p, PL_regxend);
2896 ender = toLOWER_LC_uni(ender);
2898 ender = toLOWER_uni(ender);
2900 if (ISMULT2(p)) { /* Back off on ?+*. */
2903 else if (ender >= 0x80 && UTF) {
2904 reguni(ender, s, &numlen);
2914 if (ender >= 0x80 && UTF) {
2915 reguni(ender, s, &numlen);
2923 PL_regcomp_parse = p - 1;
2926 vFAIL("Internal disaster");
2934 PL_regsize += STR_SZ(len);
2936 PL_regcode += STR_SZ(len);
2945 S_regwhite(pTHX_ char *p, char *e)
2950 else if (*p == '#') {
2953 } while (p < e && *p != '\n');
2961 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
2962 Character classes ([:foo:]) can also be negated ([:^foo:]).
2963 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
2964 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
2965 but trigger warnings because they are currently unimplemented. */
2967 S_regpposixcc(pTHX_ I32 value)
2971 I32 namedclass = OOB_NAMEDCLASS;
2973 if (value == '[' && PL_regcomp_parse + 1 < PL_regxend &&
2974 /* I smell either [: or [= or [. -- POSIX has been here, right? */
2975 (*PL_regcomp_parse == ':' ||
2976 *PL_regcomp_parse == '=' ||
2977 *PL_regcomp_parse == '.')) {
2978 char c = *PL_regcomp_parse;
2979 char* s = PL_regcomp_parse++;
2981 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != c)
2983 if (PL_regcomp_parse == PL_regxend)
2984 /* Grandfather lone [:, [=, [. */
2985 PL_regcomp_parse = s;
2987 char* t = PL_regcomp_parse++; /* skip over the c */
2989 if (*PL_regcomp_parse == ']') {
2990 PL_regcomp_parse++; /* skip over the ending ] */
2993 I32 complement = *posixcc == '^' ? *posixcc++ : 0;
2994 I32 skip = 5; /* the most common skip */
2998 if (strnEQ(posixcc, "alnum", 5))
3000 complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
3001 else if (strnEQ(posixcc, "alpha", 5))
3003 complement ? ANYOF_NALPHA : ANYOF_ALPHA;
3004 else if (strnEQ(posixcc, "ascii", 5))
3006 complement ? ANYOF_NASCII : ANYOF_ASCII;
3009 if (strnEQ(posixcc, "blank", 5))
3011 complement ? ANYOF_NBLANK : ANYOF_BLANK;
3014 if (strnEQ(posixcc, "cntrl", 5))
3016 complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
3019 if (strnEQ(posixcc, "digit", 5))
3021 complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
3024 if (strnEQ(posixcc, "graph", 5))
3026 complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
3029 if (strnEQ(posixcc, "lower", 5))
3031 complement ? ANYOF_NLOWER : ANYOF_LOWER;
3034 if (strnEQ(posixcc, "print", 5))
3036 complement ? ANYOF_NPRINT : ANYOF_PRINT;
3037 else if (strnEQ(posixcc, "punct", 5))
3039 complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
3042 if (strnEQ(posixcc, "space", 5))
3044 complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
3047 if (strnEQ(posixcc, "upper", 5))
3049 complement ? ANYOF_NUPPER : ANYOF_UPPER;
3051 case 'w': /* this is not POSIX, this is the Perl \w */
3052 if (strnEQ(posixcc, "word", 4)) {
3054 complement ? ANYOF_NALNUM : ANYOF_ALNUM;
3059 if (strnEQ(posixcc, "xdigit", 6)) {
3061 complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
3066 if (namedclass == OOB_NAMEDCLASS ||
3067 posixcc[skip] != ':' ||
3068 posixcc[skip+1] != ']')
3070 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
3073 } else if (!SIZE_ONLY) {
3074 /* [[=foo=]] and [[.foo.]] are still future. */
3076 /* adjust PL_regcomp_parse so the warning shows after
3078 while (*PL_regcomp_parse && *PL_regcomp_parse != ']')
3080 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
3083 /* Maternal grandfather:
3084 * "[:" ending in ":" but not in ":]" */
3085 PL_regcomp_parse = s;
3094 S_checkposixcc(pTHX)
3096 if (!SIZE_ONLY && ckWARN(WARN_REGEXP) &&
3097 (*PL_regcomp_parse == ':' ||
3098 *PL_regcomp_parse == '=' ||
3099 *PL_regcomp_parse == '.')) {
3100 char *s = PL_regcomp_parse;
3103 while(*s && isALNUM(*s))
3105 if (*s && c == *s && s[1] == ']') {
3106 vWARN3(s+2, "POSIX syntax [%c %c] belongs inside character classes", c, c);
3108 /* [[=foo=]] and [[.foo.]] are still future. */
3109 if (c == '=' || c == '.')
3111 /* adjust PL_regcomp_parse so the error shows after
3113 while (*PL_regcomp_parse && *PL_regcomp_parse++ != ']')
3115 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
3126 register I32 lastvalue = OOB_CHAR8;
3127 register I32 range = 0;
3128 register regnode *ret;
3132 bool need_class = 0;
3134 ret = reg_node(ANYOF);
3136 PL_regsize += ANYOF_SKIP;
3139 ANYOF_BITMAP_ZERO(ret);
3140 PL_regcode += ANYOF_SKIP;
3142 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
3144 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
3146 if (*PL_regcomp_parse == '^') { /* Complement of range. */
3150 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
3153 if (!SIZE_ONLY && ckWARN(WARN_REGEXP))
3156 if (*PL_regcomp_parse == ']' || *PL_regcomp_parse == '-')
3157 goto skipcond; /* allow 1st char to be ] or - */
3158 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != ']') {
3160 namedclass = OOB_NAMEDCLASS;
3162 rangebegin = PL_regcomp_parse;
3163 value = UCHARAT(PL_regcomp_parse++);
3165 namedclass = regpposixcc(value);
3166 else if (value == '\\') {
3167 value = UCHARAT(PL_regcomp_parse++);
3168 /* Some compilers cannot handle switching on 64-bit integer
3169 * values, therefore the 'value' cannot be an UV. --jhi */
3171 case 'w': namedclass = ANYOF_ALNUM; break;
3172 case 'W': namedclass = ANYOF_NALNUM; break;
3173 case 's': namedclass = ANYOF_SPACE; break;
3174 case 'S': namedclass = ANYOF_NSPACE; break;
3175 case 'd': namedclass = ANYOF_DIGIT; break;
3176 case 'D': namedclass = ANYOF_NDIGIT; break;
3177 case 'n': value = '\n'; break;
3178 case 'r': value = '\r'; break;
3179 case 't': value = '\t'; break;
3180 case 'f': value = '\f'; break;
3181 case 'b': value = '\b'; break;
3183 case 'e': value = '\033'; break;
3184 case 'a': value = '\007'; break;
3186 case 'e': value = '\047'; break;
3187 case 'a': value = '\057'; break;
3190 numlen = 0; /* disallow underscores */
3191 value = (UV)scan_hex(PL_regcomp_parse, 2, &numlen);
3192 PL_regcomp_parse += numlen;
3195 value = UCHARAT(PL_regcomp_parse++);
3196 value = toCTRL(value);
3198 case '0': case '1': case '2': case '3': case '4':
3199 case '5': case '6': case '7': case '8': case '9':
3200 numlen = 0; /* disallow underscores */
3201 value = (UV)scan_oct(--PL_regcomp_parse, 3, &numlen);
3202 PL_regcomp_parse += numlen;
3205 if (!SIZE_ONLY && ckWARN(WARN_REGEXP) && isALPHA(value))
3207 vWARN2(PL_regcomp_parse, "Unrecognized escape \\%c in character class passed through", (int)value);
3211 if (namedclass > OOB_NAMEDCLASS) {
3212 if (!need_class && !SIZE_ONLY)
3213 ANYOF_CLASS_ZERO(ret);
3215 if (range) { /* a-\d, a-[:digit:] */
3217 if (ckWARN(WARN_REGEXP))
3218 vWARN4(PL_regcomp_parse,
3219 "False [] range \"%*.*s\"",
3220 PL_regcomp_parse - rangebegin,
3221 PL_regcomp_parse - rangebegin,
3223 ANYOF_BITMAP_SET(ret, lastvalue);
3224 ANYOF_BITMAP_SET(ret, '-');
3226 range = 0; /* this is not a true range */
3229 switch (namedclass) {
3232 ANYOF_CLASS_SET(ret, ANYOF_ALNUM);
3234 for (value = 0; value < 256; value++)
3236 ANYOF_BITMAP_SET(ret, value);
3241 ANYOF_CLASS_SET(ret, ANYOF_NALNUM);
3243 for (value = 0; value < 256; value++)
3244 if (!isALNUM(value))
3245 ANYOF_BITMAP_SET(ret, value);
3250 ANYOF_CLASS_SET(ret, ANYOF_SPACE);
3252 for (value = 0; value < 256; value++)
3254 ANYOF_BITMAP_SET(ret, value);
3259 ANYOF_CLASS_SET(ret, ANYOF_NSPACE);
3261 for (value = 0; value < 256; value++)
3262 if (!isSPACE(value))
3263 ANYOF_BITMAP_SET(ret, value);
3268 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
3270 for (value = '0'; value <= '9'; value++)
3271 ANYOF_BITMAP_SET(ret, value);
3276 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
3278 for (value = 0; value < '0'; value++)
3279 ANYOF_BITMAP_SET(ret, value);
3280 for (value = '9' + 1; value < 256; value++)
3281 ANYOF_BITMAP_SET(ret, value);
3286 ANYOF_CLASS_SET(ret, ANYOF_NALNUMC);
3288 for (value = 0; value < 256; value++)
3289 if (!isALNUMC(value))
3290 ANYOF_BITMAP_SET(ret, value);
3295 ANYOF_CLASS_SET(ret, ANYOF_ALNUMC);
3297 for (value = 0; value < 256; value++)
3298 if (isALNUMC(value))
3299 ANYOF_BITMAP_SET(ret, value);
3304 ANYOF_CLASS_SET(ret, ANYOF_ALPHA);
3306 for (value = 0; value < 256; value++)
3308 ANYOF_BITMAP_SET(ret, value);
3313 ANYOF_CLASS_SET(ret, ANYOF_NALPHA);
3315 for (value = 0; value < 256; value++)
3316 if (!isALPHA(value))
3317 ANYOF_BITMAP_SET(ret, value);
3322 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
3325 for (value = 0; value < 128; value++)
3326 ANYOF_BITMAP_SET(ret, value);
3328 for (value = 0; value < 256; value++)
3330 ANYOF_BITMAP_SET(ret, value);
3336 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
3339 for (value = 128; value < 256; value++)
3340 ANYOF_BITMAP_SET(ret, value);
3342 for (value = 0; value < 256; value++)
3343 if (!isASCII(value))
3344 ANYOF_BITMAP_SET(ret, value);
3350 ANYOF_CLASS_SET(ret, ANYOF_BLANK);
3352 for (value = 0; value < 256; value++)
3354 ANYOF_BITMAP_SET(ret, value);
3359 ANYOF_CLASS_SET(ret, ANYOF_NBLANK);
3361 for (value = 0; value < 256; value++)
3362 if (!isBLANK(value))
3363 ANYOF_BITMAP_SET(ret, value);
3368 ANYOF_CLASS_SET(ret, ANYOF_CNTRL);
3370 for (value = 0; value < 256; value++)
3372 ANYOF_BITMAP_SET(ret, value);
3374 lastvalue = OOB_CHAR8;
3378 ANYOF_CLASS_SET(ret, ANYOF_NCNTRL);
3380 for (value = 0; value < 256; value++)
3381 if (!isCNTRL(value))
3382 ANYOF_BITMAP_SET(ret, value);
3387 ANYOF_CLASS_SET(ret, ANYOF_GRAPH);
3389 for (value = 0; value < 256; value++)
3391 ANYOF_BITMAP_SET(ret, value);
3396 ANYOF_CLASS_SET(ret, ANYOF_NGRAPH);
3398 for (value = 0; value < 256; value++)
3399 if (!isGRAPH(value))
3400 ANYOF_BITMAP_SET(ret, value);
3405 ANYOF_CLASS_SET(ret, ANYOF_LOWER);
3407 for (value = 0; value < 256; value++)
3409 ANYOF_BITMAP_SET(ret, value);
3414 ANYOF_CLASS_SET(ret, ANYOF_NLOWER);
3416 for (value = 0; value < 256; value++)
3417 if (!isLOWER(value))
3418 ANYOF_BITMAP_SET(ret, value);
3423 ANYOF_CLASS_SET(ret, ANYOF_PRINT);
3425 for (value = 0; value < 256; value++)
3427 ANYOF_BITMAP_SET(ret, value);
3432 ANYOF_CLASS_SET(ret, ANYOF_NPRINT);
3434 for (value = 0; value < 256; value++)
3435 if (!isPRINT(value))
3436 ANYOF_BITMAP_SET(ret, value);
3441 ANYOF_CLASS_SET(ret, ANYOF_PSXSPC);
3443 for (value = 0; value < 256; value++)
3444 if (isPSXSPC(value))
3445 ANYOF_BITMAP_SET(ret, value);
3450 ANYOF_CLASS_SET(ret, ANYOF_NPSXSPC);
3452 for (value = 0; value < 256; value++)
3453 if (!isPSXSPC(value))
3454 ANYOF_BITMAP_SET(ret, value);
3459 ANYOF_CLASS_SET(ret, ANYOF_PUNCT);
3461 for (value = 0; value < 256; value++)
3463 ANYOF_BITMAP_SET(ret, value);
3468 ANYOF_CLASS_SET(ret, ANYOF_NPUNCT);
3470 for (value = 0; value < 256; value++)
3471 if (!isPUNCT(value))
3472 ANYOF_BITMAP_SET(ret, value);
3477 ANYOF_CLASS_SET(ret, ANYOF_UPPER);
3479 for (value = 0; value < 256; value++)
3481 ANYOF_BITMAP_SET(ret, value);
3486 ANYOF_CLASS_SET(ret, ANYOF_NUPPER);
3488 for (value = 0; value < 256; value++)
3489 if (!isUPPER(value))
3490 ANYOF_BITMAP_SET(ret, value);
3495 ANYOF_CLASS_SET(ret, ANYOF_XDIGIT);
3497 for (value = 0; value < 256; value++)
3498 if (isXDIGIT(value))
3499 ANYOF_BITMAP_SET(ret, value);
3504 ANYOF_CLASS_SET(ret, ANYOF_NXDIGIT);
3506 for (value = 0; value < 256; value++)
3507 if (!isXDIGIT(value))
3508 ANYOF_BITMAP_SET(ret, value);
3512 vFAIL("Invalid [::] class");
3516 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
3521 if (lastvalue > value) /* b-a */ {
3522 Simple_vFAIL4("Invalid [] range \"%*.*s\"",
3523 PL_regcomp_parse - rangebegin,
3524 PL_regcomp_parse - rangebegin,
3531 if (*PL_regcomp_parse == '-' && PL_regcomp_parse+1 < PL_regxend &&
3532 PL_regcomp_parse[1] != ']') {
3534 if (namedclass > OOB_NAMEDCLASS) { /* \w-, [:word:]- */
3535 if (ckWARN(WARN_REGEXP))
3536 vWARN4(PL_regcomp_parse,
3537 "False [] range \"%*.*s\"",
3538 PL_regcomp_parse - rangebegin,
3539 PL_regcomp_parse - rangebegin,
3542 ANYOF_BITMAP_SET(ret, '-');
3545 continue; /* do it next time */
3548 /* now is the next time */
3550 #ifndef ASCIIish /* EBCDIC, for example. */
3551 if ((isLOWER(lastvalue) && isLOWER(value)) ||
3552 (isUPPER(lastvalue) && isUPPER(value)))
3555 if (isLOWER(lastvalue)) {
3556 for (i = lastvalue; i <= value; i++)
3558 ANYOF_BITMAP_SET(ret, i);
3560 for (i = lastvalue; i <= value; i++)
3562 ANYOF_BITMAP_SET(ret, i);
3567 for ( ; lastvalue <= value; lastvalue++)
3568 ANYOF_BITMAP_SET(ret, lastvalue);
3574 PL_regsize += ANYOF_CLASS_ADD_SKIP;
3576 PL_regcode += ANYOF_CLASS_ADD_SKIP;
3578 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
3580 (ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD) {
3581 for (value = 0; value < 256; ++value) {
3582 if (ANYOF_BITMAP_TEST(ret, value)) {
3583 I32 cf = PL_fold[value];
3584 ANYOF_BITMAP_SET(ret, cf);
3587 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
3589 /* optimize inverted simple patterns (e.g. [^a-z]) */
3590 if (!SIZE_ONLY && (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
3591 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
3592 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
3593 ANYOF_FLAGS(ret) = 0;
3599 S_regclassutf8(pTHX)
3604 register U32 lastvalue = OOB_UTF8;
3605 register I32 range = 0;
3606 register regnode *ret;
3614 if (*PL_regcomp_parse == '^') { /* Complement of range. */
3618 flags |= ANYOF_INVERT;
3622 flags |= ANYOF_FOLD;
3624 flags |= ANYOF_LOCALE;
3625 listsv = newSVpvn("# comment\n",10);
3628 if (!SIZE_ONLY && ckWARN(WARN_REGEXP))
3631 if (*PL_regcomp_parse == ']' || *PL_regcomp_parse == '-')
3632 goto skipcond; /* allow 1st char to be ] or - */
3634 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != ']') {
3636 namedclass = OOB_NAMEDCLASS;
3638 rangebegin = PL_regcomp_parse;
3639 value = utf8_to_uv((U8*)PL_regcomp_parse,
3640 PL_regxend - PL_regcomp_parse,
3642 PL_regcomp_parse += numlen;
3644 namedclass = regpposixcc(value);
3645 else if (value == '\\') {
3646 value = (U32)utf8_to_uv((U8*)PL_regcomp_parse,
3647 PL_regxend - PL_regcomp_parse,
3649 PL_regcomp_parse += numlen;
3650 /* Some compilers cannot handle switching on 64-bit integer
3651 * values, therefore value cannot be an UV. Yes, this will
3652 * be a problem later if we want switch on Unicode. --jhi */
3654 case 'w': namedclass = ANYOF_ALNUM; break;
3655 case 'W': namedclass = ANYOF_NALNUM; break;
3656 case 's': namedclass = ANYOF_SPACE; break;
3657 case 'S': namedclass = ANYOF_NSPACE; break;
3658 case 'd': namedclass = ANYOF_DIGIT; break;
3659 case 'D': namedclass = ANYOF_NDIGIT; break;
3662 if (*PL_regcomp_parse == '{') {
3663 e = strchr(PL_regcomp_parse++, '}');
3665 vFAIL("Missing right brace on \\p{}");
3666 n = e - PL_regcomp_parse;
3669 e = PL_regcomp_parse;
3674 Perl_sv_catpvf(aTHX_ listsv,
3675 "+utf8::%.*s\n", (int)n, PL_regcomp_parse);
3677 Perl_sv_catpvf(aTHX_ listsv,
3678 "!utf8::%.*s\n", (int)n, PL_regcomp_parse);
3680 PL_regcomp_parse = e + 1;
3681 lastvalue = OOB_UTF8;
3683 case 'n': value = '\n'; break;
3684 case 'r': value = '\r'; break;
3685 case 't': value = '\t'; break;
3686 case 'f': value = '\f'; break;
3687 case 'b': value = '\b'; break;
3689 case 'e': value = '\033'; break;
3690 case 'a': value = '\007'; break;
3692 case 'e': value = '\047'; break;
3693 case 'a': value = '\057'; break;
3696 if (*PL_regcomp_parse == '{') {
3697 e = strchr(PL_regcomp_parse++, '}');
3699 vFAIL("Missing right brace on \\x{}");
3700 numlen = 1; /* allow underscores */
3701 value = (UV)scan_hex(PL_regcomp_parse,
3702 e - PL_regcomp_parse,
3704 PL_regcomp_parse = e + 1;
3707 numlen = 0; /* disallow underscores */
3708 value = (UV)scan_hex(PL_regcomp_parse, 2, &numlen);
3709 PL_regcomp_parse += numlen;
3713 value = UCHARAT(PL_regcomp_parse++);
3714 value = toCTRL(value);
3716 case '0': case '1': case '2': case '3': case '4':
3717 case '5': case '6': case '7': case '8': case '9':
3718 numlen = 0; /* disallow underscores */
3719 value = (UV)scan_oct(--PL_regcomp_parse, 3, &numlen);
3720 PL_regcomp_parse += numlen;
3723 if (!SIZE_ONLY && ckWARN(WARN_REGEXP) && isALPHA(value))
3724 vWARN2(PL_regcomp_parse,
3725 "Unrecognized escape \\%c in character class passed through",
3730 if (namedclass > OOB_NAMEDCLASS) {
3731 if (range) { /* a-\d, a-[:digit:] */
3733 if (ckWARN(WARN_REGEXP))
3734 vWARN4(PL_regcomp_parse,
3735 "False [] range \"%*.*s\"",
3736 PL_regcomp_parse - rangebegin,
3737 PL_regcomp_parse - rangebegin,
3739 Perl_sv_catpvf(aTHX_ listsv,
3740 /* 0x002D is Unicode for '-' */
3741 "%04"UVxf"\n002D\n", (UV)lastvalue);
3746 switch (namedclass) {
3748 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsWord\n"); break;
3750 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsWord\n"); break;
3752 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsAlnum\n"); break;
3754 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsAlnum\n"); break;
3756 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsAlpha\n"); break;
3758 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsAlpha\n"); break;
3760 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsASCII\n"); break;
3762 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsASCII\n"); break;
3764 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsCntrl\n"); break;
3766 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsCntrl\n"); break;
3768 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsGraph\n"); break;
3770 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsGraph\n"); break;
3772 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsDigit\n"); break;
3774 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsDigit\n"); break;
3776 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsLower\n"); break;
3778 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsLower\n"); break;
3780 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsPrint\n"); break;
3782 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsPrint\n"); break;
3784 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsPunct\n"); break;
3786 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsPunct\n"); break;
3790 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsSpace\n"); break;
3794 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsSpace\n"); break;
3796 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsUpper\n"); break;
3798 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsUpper\n"); break;
3800 Perl_sv_catpvf(aTHX_ listsv, "+utf8::IsXDigit\n"); break;
3802 Perl_sv_catpvf(aTHX_ listsv, "!utf8::IsXDigit\n"); break;
3808 if (lastvalue > value) { /* b-a */
3809 Simple_vFAIL4("invalid [] range \"%*.*s\"",
3810 PL_regcomp_parse - rangebegin,
3811 PL_regcomp_parse - rangebegin,
3818 if (*PL_regcomp_parse == '-' && PL_regcomp_parse+1 < PL_regxend &&
3819 PL_regcomp_parse[1] != ']') {
3821 if (namedclass > OOB_NAMEDCLASS) { /* \w-, [:word:]- */
3822 if (ckWARN(WARN_REGEXP))
3823 vWARN4(PL_regcomp_parse,
3824 "False [] range \"%*.*s\"",
3825 PL_regcomp_parse - rangebegin,
3826 PL_regcomp_parse - rangebegin,
3829 Perl_sv_catpvf(aTHX_ listsv,
3830 /* 0x002D is Unicode for '-' */
3834 continue; /* do it next time */
3837 /* now is the next time */
3839 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
3840 (UV)lastvalue, (UV)value);
3844 ret = reganode(ANYOFUTF8, 0);
3847 SV *rv = swash_init("utf8", "", listsv, 1, 0);
3848 SvREFCNT_dec(listsv);
3849 n = add_data(1,"s");
3850 PL_regcomp_rx->data->data[n] = (void*)rv;
3851 ARG1_SET(ret, flags);
3862 char* retval = PL_regcomp_parse++;
3865 if (*PL_regcomp_parse == '(' && PL_regcomp_parse[1] == '?' &&
3866 PL_regcomp_parse[2] == '#') {
3867 while (*PL_regcomp_parse && *PL_regcomp_parse != ')')
3872 if (PL_regflags & PMf_EXTENDED) {
3873 if (isSPACE(*PL_regcomp_parse)) {
3877 else if (*PL_regcomp_parse == '#') {
3878 while (*PL_regcomp_parse && *PL_regcomp_parse != '\n')
3889 - reg_node - emit a node
3891 STATIC regnode * /* Location. */
3892 S_reg_node(pTHX_ U8 op)
3895 register regnode *ret;
3896 register regnode *ptr;
3900 SIZE_ALIGN(PL_regsize);
3905 NODE_ALIGN_FILL(ret);
3907 FILL_ADVANCE_NODE(ptr, op);
3914 - reganode - emit a node with an argument
3916 STATIC regnode * /* Location. */
3917 S_reganode(pTHX_ U8 op, U32 arg)
3920 register regnode *ret;
3921 register regnode *ptr;
3925 SIZE_ALIGN(PL_regsize);
3930 NODE_ALIGN_FILL(ret);
3932 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
3939 - reguni - emit (if appropriate) a Unicode character
3942 S_reguni(pTHX_ UV uv, char* s, STRLEN* lenp)
3946 U8 tmpbuf[UTF8_MAXLEN];
3947 *lenp = uv_to_utf8(tmpbuf, uv) - tmpbuf;
3950 *lenp = uv_to_utf8((U8*)s, uv) - (U8*)s;
3955 - reginsert - insert an operator in front of already-emitted operand
3957 * Means relocating the operand.
3960 S_reginsert(pTHX_ U8 op, regnode *opnd)
3963 register regnode *src;
3964 register regnode *dst;
3965 register regnode *place;
3966 register int offset = regarglen[(U8)op];
3968 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
3971 PL_regsize += NODE_STEP_REGNODE + offset;
3976 PL_regcode += NODE_STEP_REGNODE + offset;
3979 StructCopy(--src, --dst, regnode);
3981 place = opnd; /* Op node, where operand used to be. */
3982 src = NEXTOPER(place);
3983 FILL_ADVANCE_NODE(place, op);
3984 Zero(src, offset, regnode);
3988 - regtail - set the next-pointer at the end of a node chain of p to val.
3991 S_regtail(pTHX_ regnode *p, regnode *val)
3994 register regnode *scan;
3995 register regnode *temp;
4000 /* Find last node. */
4003 temp = regnext(scan);
4009 if (reg_off_by_arg[OP(scan)]) {
4010 ARG_SET(scan, val - scan);
4013 NEXT_OFF(scan) = val - scan;
4018 - regoptail - regtail on operand of first argument; nop if operandless
4021 S_regoptail(pTHX_ regnode *p, regnode *val)
4024 /* "Operandless" and "op != BRANCH" are synonymous in practice. */
4025 if (p == NULL || SIZE_ONLY)
4027 if (PL_regkind[(U8)OP(p)] == BRANCH) {
4028 regtail(NEXTOPER(p), val);
4030 else if ( PL_regkind[(U8)OP(p)] == BRANCHJ) {
4031 regtail(NEXTOPER(NEXTOPER(p)), val);
4038 - regcurly - a little FSA that accepts {\d+,?\d*}
4041 S_regcurly(pTHX_ register char *s)
4060 S_dumpuntil(pTHX_ regnode *start, regnode *node, regnode *last, SV* sv, I32 l)
4063 register U8 op = EXACT; /* Arbitrary non-END op. */
4064 register regnode *next;
4066 while (op != END && (!last || node < last)) {
4067 /* While that wasn't END last time... */
4073 next = regnext(node);
4075 if (OP(node) == OPTIMIZED)
4078 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
4079 (int)(2*l + 1), "", SvPVX(sv));
4080 if (next == NULL) /* Next ptr. */
4081 PerlIO_printf(Perl_debug_log, "(0)");
4083 PerlIO_printf(Perl_debug_log, "(%"IVdf")", (IV)(next - start));
4084 (void)PerlIO_putc(Perl_debug_log, '\n');
4086 if (PL_regkind[(U8)op] == BRANCHJ) {
4087 register regnode *nnode = (OP(next) == LONGJMP
4090 if (last && nnode > last)
4092 node = dumpuntil(start, NEXTOPER(NEXTOPER(node)), nnode, sv, l + 1);
4094 else if (PL_regkind[(U8)op] == BRANCH) {
4095 node = dumpuntil(start, NEXTOPER(node), next, sv, l + 1);
4097 else if ( op == CURLY) { /* `next' might be very big: optimizer */
4098 node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
4099 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1, sv, l + 1);
4101 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
4102 node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
4105 else if ( op == PLUS || op == STAR) {
4106 node = dumpuntil(start, NEXTOPER(node), NEXTOPER(node) + 1, sv, l + 1);
4108 else if (op == ANYOF) {
4109 node = NEXTOPER(node);
4112 else if (PL_regkind[(U8)op] == EXACT) {
4113 /* Literal string, where present. */
4114 node += NODE_SZ_STR(node) - 1;
4115 node = NEXTOPER(node);
4118 node = NEXTOPER(node);
4119 node += regarglen[(U8)op];
4121 if (op == CURLYX || op == OPEN)
4123 else if (op == WHILEM)
4126 #endif /* DEBUGGING */
4131 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
4134 Perl_regdump(pTHX_ regexp *r)
4138 SV *sv = sv_newmortal();
4140 (void)dumpuntil(r->program, r->program + 1, NULL, sv, 0);
4142 /* Header fields of interest. */
4143 if (r->anchored_substr)
4144 PerlIO_printf(Perl_debug_log,
4145 "anchored `%s%.*s%s'%s at %"IVdf" ",
4147 (int)(SvCUR(r->anchored_substr) - (SvTAIL(r->anchored_substr)!=0)),
4148 SvPVX(r->anchored_substr),
4150 SvTAIL(r->anchored_substr) ? "$" : "",
4151 (IV)r->anchored_offset);
4152 if (r->float_substr)
4153 PerlIO_printf(Perl_debug_log,
4154 "floating `%s%.*s%s'%s at %"IVdf"..%"UVuf" ",
4156 (int)(SvCUR(r->float_substr) - (SvTAIL(r->float_substr)!=0)),
4157 SvPVX(r->float_substr),
4159 SvTAIL(r->float_substr) ? "$" : "",
4160 (IV)r->float_min_offset, (UV)r->float_max_offset);
4161 if (r->check_substr)
4162 PerlIO_printf(Perl_debug_log,
4163 r->check_substr == r->float_substr
4164 ? "(checking floating" : "(checking anchored");
4165 if (r->reganch & ROPT_NOSCAN)
4166 PerlIO_printf(Perl_debug_log, " noscan");
4167 if (r->reganch & ROPT_CHECK_ALL)
4168 PerlIO_printf(Perl_debug_log, " isall");
4169 if (r->check_substr)
4170 PerlIO_printf(Perl_debug_log, ") ");
4172 if (r->regstclass) {
4173 regprop(sv, r->regstclass);
4174 PerlIO_printf(Perl_debug_log, "stclass `%s' ", SvPVX(sv));
4176 if (r->reganch & ROPT_ANCH) {
4177 PerlIO_printf(Perl_debug_log, "anchored");
4178 if (r->reganch & ROPT_ANCH_BOL)
4179 PerlIO_printf(Perl_debug_log, "(BOL)");
4180 if (r->reganch & ROPT_ANCH_MBOL)
4181 PerlIO_printf(Perl_debug_log, "(MBOL)");
4182 if (r->reganch & ROPT_ANCH_SBOL)
4183 PerlIO_printf(Perl_debug_log, "(SBOL)");
4184 if (r->reganch & ROPT_ANCH_GPOS)
4185 PerlIO_printf(Perl_debug_log, "(GPOS)");
4186 PerlIO_putc(Perl_debug_log, ' ');
4188 if (r->reganch & ROPT_GPOS_SEEN)
4189 PerlIO_printf(Perl_debug_log, "GPOS ");
4190 if (r->reganch & ROPT_SKIP)
4191 PerlIO_printf(Perl_debug_log, "plus ");
4192 if (r->reganch & ROPT_IMPLICIT)
4193 PerlIO_printf(Perl_debug_log, "implicit ");
4194 PerlIO_printf(Perl_debug_log, "minlen %ld ", (long) r->minlen);
4195 if (r->reganch & ROPT_EVAL_SEEN)
4196 PerlIO_printf(Perl_debug_log, "with eval ");
4197 PerlIO_printf(Perl_debug_log, "\n");
4198 #endif /* DEBUGGING */
4202 S_put_byte(pTHX_ SV *sv, int c)
4204 if (c <= ' ' || c == 127 || c == 255)
4205 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
4206 else if (c == '-' || c == ']' || c == '\\' || c == '^')
4207 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
4209 Perl_sv_catpvf(aTHX_ sv, "%c", c);
4213 - regprop - printable representation of opcode
4216 Perl_regprop(pTHX_ SV *sv, regnode *o)
4222 sv_setpvn(sv, "", 0);
4223 if (OP(o) >= reg_num) /* regnode.type is unsigned */
4224 FAIL("Corrupted regexp opcode");
4225 sv_catpv(sv, (char*)reg_name[OP(o)]); /* Take off const! */
4227 k = PL_regkind[(U8)OP(o)];
4230 Perl_sv_catpvf(aTHX_ sv, " <%s%.*s%s>", PL_colors[0],
4231 STR_LEN(o), STRING(o), PL_colors[1]);
4232 else if (k == CURLY) {
4233 if (OP(o) == CURLYM || OP(o) == CURLYN)
4234 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
4235 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
4237 else if (k == WHILEM && o->flags) /* Ordinal/of */
4238 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
4239 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP )
4240 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
4241 else if (k == LOGICAL)
4242 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
4243 else if (k == ANYOF) {
4244 int i, rangestart = -1;
4245 const char * const out[] = { /* Should be syncronized with
4246 ANYOF_ #xdefines in regcomp.h */
4279 if (o->flags & ANYOF_LOCALE)
4280 sv_catpv(sv, "{loc}");
4281 if (o->flags & ANYOF_FOLD)
4282 sv_catpv(sv, "{i}");
4283 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
4284 if (o->flags & ANYOF_INVERT)
4286 for (i = 0; i <= 256; i++) {
4287 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
4288 if (rangestart == -1)
4290 } else if (rangestart != -1) {
4291 if (i <= rangestart + 3)
4292 for (; rangestart < i; rangestart++)
4293 put_byte(sv, rangestart);
4295 put_byte(sv, rangestart);
4297 put_byte(sv, i - 1);
4302 if (o->flags & ANYOF_CLASS)
4303 for (i = 0; i < sizeof(out)/sizeof(char*); i++)
4304 if (ANYOF_CLASS_TEST(o,i))
4305 sv_catpv(sv, out[i]);
4306 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
4308 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
4309 Perl_sv_catpvf(aTHX_ sv, "[-%d]", o->flags);
4310 #endif /* DEBUGGING */
4314 Perl_re_intuit_string(pTHX_ regexp *prog)
4315 { /* Assume that RE_INTUIT is set */
4318 char *s = SvPV(prog->check_substr,n_a);
4320 if (!PL_colorset) reginitcolors();
4321 PerlIO_printf(Perl_debug_log,
4322 "%sUsing REx substr:%s `%s%.60s%s%s'\n",
4323 PL_colors[4],PL_colors[5],PL_colors[0],
4326 (strlen(s) > 60 ? "..." : ""));
4329 return prog->check_substr;
4333 Perl_pregfree(pTHX_ struct regexp *r)
4336 DEBUG_r(if (!PL_colorset) reginitcolors());
4338 if (!r || (--r->refcnt > 0))
4340 DEBUG_r(PerlIO_printf(Perl_debug_log,
4341 "%sFreeing REx:%s `%s%.60s%s%s'\n",
4342 PL_colors[4],PL_colors[5],PL_colors[0],
4345 (strlen(r->precomp) > 60 ? "..." : "")));
4348 Safefree(r->precomp);
4349 if (RX_MATCH_COPIED(r))
4350 Safefree(r->subbeg);
4352 if (r->anchored_substr)
4353 SvREFCNT_dec(r->anchored_substr);
4354 if (r->float_substr)
4355 SvREFCNT_dec(r->float_substr);
4356 Safefree(r->substrs);
4359 int n = r->data->count;
4360 AV* new_comppad = NULL;
4365 switch (r->data->what[n]) {
4367 SvREFCNT_dec((SV*)r->data->data[n]);
4370 Safefree(r->data->data[n]);
4373 new_comppad = (AV*)r->data->data[n];
4376 if (new_comppad == NULL)
4377 Perl_croak(aTHX_ "panic: pregfree comppad");
4378 old_comppad = PL_comppad;
4379 old_curpad = PL_curpad;
4380 /* Watch out for global destruction's random ordering. */
4381 if (SvTYPE(new_comppad) == SVt_PVAV) {
4382 PL_comppad = new_comppad;
4383 PL_curpad = AvARRAY(new_comppad);
4387 op_free((OP_4tree*)r->data->data[n]);
4388 PL_comppad = old_comppad;
4389 PL_curpad = old_curpad;
4390 SvREFCNT_dec((SV*)new_comppad);
4396 FAIL2("panic: regfree data code '%c'", r->data->what[n]);
4399 Safefree(r->data->what);
4402 Safefree(r->startp);
4408 - regnext - dig the "next" pointer out of a node
4410 * [Note, when REGALIGN is defined there are two places in regmatch()
4411 * that bypass this code for speed.]
4414 Perl_regnext(pTHX_ register regnode *p)
4417 register I32 offset;
4419 if (p == &PL_regdummy)
4422 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
4430 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
4433 STRLEN l1 = strlen(pat1);
4434 STRLEN l2 = strlen(pat2);
4443 Copy(pat1, buf, l1 , char);
4444 Copy(pat2, buf + l1, l2 , char);
4445 buf[l1 + l2] = '\n';
4446 buf[l1 + l2 + 1] = '\0';
4448 /* ANSI variant takes additional second argument */
4449 va_start(args, pat2);
4453 msv = vmess(buf, &args);
4455 message = SvPV(msv,l1);
4458 Copy(message, buf, l1 , char);
4459 buf[l1] = '\0'; /* Overwrite \n */
4460 Perl_croak(aTHX_ "%s", buf);
4463 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
4466 Perl_save_re_context(pTHX)
4470 SAVEPPTR(PL_regprecomp); /* uncompiled string. */
4471 SAVEI32(PL_regnpar); /* () count. */
4472 SAVEI32(PL_regsize); /* Code size. */
4473 SAVEI16(PL_regflags); /* are we folding, multilining? */
4474 SAVEPPTR(PL_reginput); /* String-input pointer. */
4475 SAVEPPTR(PL_regbol); /* Beginning of input, for ^ check. */
4476 SAVEPPTR(PL_regeol); /* End of input, for $ check. */
4477 SAVEVPTR(PL_regstartp); /* Pointer to startp array. */
4478 SAVEVPTR(PL_regendp); /* Ditto for endp. */
4479 SAVEVPTR(PL_reglastparen); /* Similarly for lastparen. */
4480 SAVEPPTR(PL_regtill); /* How far we are required to go. */
4481 SAVEI8(PL_regprev); /* char before regbol, \n if none */
4482 SAVEVPTR(PL_reg_start_tmp); /* from regexec.c */
4483 PL_reg_start_tmp = 0;
4484 SAVEFREEPV(PL_reg_start_tmp);
4485 SAVEI32(PL_reg_start_tmpl); /* from regexec.c */
4486 PL_reg_start_tmpl = 0;
4487 SAVEVPTR(PL_regdata);
4488 SAVEI32(PL_reg_flags); /* from regexec.c */
4489 SAVEI32(PL_reg_eval_set); /* from regexec.c */
4490 SAVEI32(PL_regnarrate); /* from regexec.c */
4491 SAVEVPTR(PL_regprogram); /* from regexec.c */
4492 SAVEINT(PL_regindent); /* from regexec.c */
4493 SAVEVPTR(PL_regcc); /* from regexec.c */
4494 SAVEVPTR(PL_curcop);
4495 SAVEVPTR(PL_regcomp_rx); /* from regcomp.c */
4496 SAVEI32(PL_regseen); /* from regcomp.c */
4497 SAVEI32(PL_regsawback); /* Did we see \1, ...? */
4498 SAVEI32(PL_regnaughty); /* How bad is this pattern? */
4499 SAVEVPTR(PL_regcode); /* Code-emit pointer; ®dummy = don't */
4500 SAVEPPTR(PL_regxend); /* End of input for compile */
4501 SAVEPPTR(PL_regcomp_parse); /* Input-scan pointer. */
4502 SAVEVPTR(PL_reg_call_cc); /* from regexec.c */
4503 SAVEVPTR(PL_reg_re); /* from regexec.c */
4504 SAVEPPTR(PL_reg_ganch); /* from regexec.c */
4505 SAVESPTR(PL_reg_sv); /* from regexec.c */
4506 SAVEVPTR(PL_reg_magic); /* from regexec.c */
4507 SAVEI32(PL_reg_oldpos); /* from regexec.c */
4508 SAVEVPTR(PL_reg_oldcurpm); /* from regexec.c */
4509 SAVEVPTR(PL_reg_curpm); /* from regexec.c */
4511 SAVEPPTR(PL_reg_starttry); /* from regexec.c */
4522 clear_re(pTHXo_ void *r)
4524 ReREFCNT_dec((regexp *)r);