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 */
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 /* *These* symbols are masked to allow static link. */
39 # define Perl_pregfree my_regfree
40 # define Perl_regnext my_regnext
41 # define Perl_save_re_context my_save_re_context
42 # define Perl_reginitcolors my_reginitcolors
47 * pregcomp and pregexec -- regsub and regerror are not used in perl
49 * Copyright (c) 1986 by University of Toronto.
50 * Written by Henry Spencer. Not derived from licensed software.
52 * Permission is granted to anyone to use this software for any
53 * purpose on any computer system, and to redistribute it freely,
54 * subject to the following restrictions:
56 * 1. The author is not responsible for the consequences of use of
57 * this software, no matter how awful, even if they arise
60 * 2. The origin of this software must not be misrepresented, either
61 * by explicit claim or by omission.
63 * 3. Altered versions must be plainly marked as such, and must not
64 * be misrepresented as being the original software.
67 **** Alterations to Henry's code are...
69 **** Copyright (c) 1991-1999, Larry Wall
71 **** You may distribute under the terms of either the GNU General Public
72 **** License or the Artistic License, as specified in the README file.
75 * Beware that some of this code is subtly aware of the way operator
76 * precedence is structured in regular expressions. Serious changes in
77 * regular-expression syntax might require a total rethink.
82 #ifndef PERL_IN_XSUB_RE
94 # if defined(BUGGY_MSC6)
95 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
96 # pragma optimize("a",off)
97 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
98 # pragma optimize("w",on )
99 # endif /* BUGGY_MSC6 */
103 #define STATIC static
106 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
107 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
108 ((*s) == '{' && regcurly(s)))
110 #define PERL_META "^$.[()|?+*\\"
112 #define META "^$.[()|?+*\\"
116 #undef SPSTART /* dratted cpp namespace... */
119 * Flags to be passed up and down.
121 #define WORST 0 /* Worst case. */
122 #define HASWIDTH 0x1 /* Known to match non-null strings. */
123 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
124 #define SPSTART 0x4 /* Starts with * or +. */
125 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
128 * Forward declarations for pregcomp()'s friends.
132 static regnode *reg _((I32, I32 *));
133 static regnode *reganode _((U8, U32));
134 static regnode *regatom _((I32 *));
135 static regnode *regbranch _((I32 *, I32));
136 static void regc _((U8, char *));
137 static void reguni _((UV, char *, I32*));
138 static regnode *regclass _((void));
139 static regnode *regclassutf8 _((void));
140 STATIC I32 regcurly _((char *));
141 static regnode *reg_node _((U8));
142 static regnode *regpiece _((I32 *));
143 static void reginsert _((U8, regnode *));
144 static void regoptail _((regnode *, regnode *));
145 static void regtail _((regnode *, regnode *));
146 static char* regwhite _((char *, char *));
147 static char* nextchar _((void));
148 static void re_croak2 _((const char* pat1,const char* pat2,...)) __attribute__((noreturn));
149 static char* regpposixcc _((I32 value));
150 static void clear_re _((void *r));
153 /* Length of a variant. */
159 I32 pos_min; /* CC */
160 I32 pos_delta; /* CC */
162 I32 last_end; /* min value, <0 unless valid. */
163 I32 last_start_min; /* CC */
164 I32 last_start_max; /* CC */
165 SV **longest; /* Either &l_fixed, or &l_float. */
167 I32 offset_fixed; /* CC */
169 I32 offset_float_min; /* CC */
170 I32 offset_float_max; /* CC */
175 static scan_data_t zero_scan_data = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
178 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
179 #define SF_BEFORE_SEOL 0x1
180 #define SF_BEFORE_MEOL 0x2
181 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
182 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
185 # define SF_FIX_SHIFT_EOL (0+2)
186 # define SF_FL_SHIFT_EOL (0+4)
188 # define SF_FIX_SHIFT_EOL (+2)
189 # define SF_FL_SHIFT_EOL (+4)
192 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
193 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
195 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
196 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
197 #define SF_IS_INF 0x40
198 #define SF_HAS_PAR 0x80
199 #define SF_IN_PAR 0x100
200 #define SF_HAS_EVAL 0x200
201 #define SCF_DO_SUBSTR 0x400
204 #define UTF (PL_reg_flags & RF_utf8)
205 #define LOC (PL_regflags & PMf_LOCALE)
206 #define FOLD (PL_regflags & PMf_FOLD)
208 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
209 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
214 ReREFCNT_dec((regexp *)r);
218 scan_commit(scan_data_t *data)
221 STRLEN l = CHR_SVLEN(data->last_found);
222 STRLEN old_l = CHR_SVLEN(*data->longest);
224 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
225 sv_setsv(*data->longest, data->last_found);
226 if (*data->longest == data->longest_fixed) {
227 data->offset_fixed = l ? data->last_start_min : data->pos_min;
228 if (data->flags & SF_BEFORE_EOL)
230 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
232 data->flags &= ~SF_FIX_BEFORE_EOL;
235 data->offset_float_min = l ? data->last_start_min : data->pos_min;
236 data->offset_float_max = (l
237 ? data->last_start_max
238 : data->pos_min + data->pos_delta);
239 if (data->flags & SF_BEFORE_EOL)
241 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
243 data->flags &= ~SF_FL_BEFORE_EOL;
246 SvCUR_set(data->last_found, 0);
248 data->flags &= ~SF_BEFORE_EOL;
251 /* Stops at toplevel WHILEM as well as at `last'. At end *scanp is set
252 to the position after last scanned or to NULL. */
255 study_chunk(regnode **scanp, I32 *deltap, regnode *last, scan_data_t *data, U32 flags)
256 /* scanp: Start here (read-write). */
257 /* deltap: Write maxlen-minlen here. */
258 /* last: Stop before this one. */
261 I32 min = 0, pars = 0, code;
262 regnode *scan = *scanp, *next;
264 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
265 int is_inf_internal = 0; /* The studied chunk is infinite */
266 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
267 scan_data_t data_fake;
269 while (scan && OP(scan) != END && scan < last) {
270 /* Peephole optimizer: */
272 if (PL_regkind[(U8)OP(scan)] == EXACT) {
273 regnode *n = regnext(scan);
276 regnode *stop = scan;
279 next = scan + (*OPERAND(scan) + 2 - 1)/sizeof(regnode) + 2;
280 /* Skip NOTHING, merge EXACT*. */
282 ( PL_regkind[(U8)OP(n)] == NOTHING ||
283 (stringok && (OP(n) == OP(scan))))
285 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
286 if (OP(n) == TAIL || n > next)
288 if (PL_regkind[(U8)OP(n)] == NOTHING) {
289 NEXT_OFF(scan) += NEXT_OFF(n);
290 next = n + NODE_STEP_REGNODE;
298 int oldl = *OPERAND(scan);
299 regnode *nnext = regnext(n);
301 if (oldl + *OPERAND(n) > U8_MAX)
303 NEXT_OFF(scan) += NEXT_OFF(n);
304 *OPERAND(scan) += *OPERAND(n);
305 next = n + (*OPERAND(n) + 2 - 1)/sizeof(regnode) + 2;
306 /* Now we can overwrite *n : */
307 Move(OPERAND(n) + 1, OPERAND(scan) + oldl + 1,
308 *OPERAND(n) + 1, char);
318 n = scan + (*OPERAND(scan) + 2 - 1)/sizeof(regnode) + 2;
320 /* Purify reports a benign UMR here sometimes, because we
321 * don't initialize the OP() slot of a node when that node
322 * is occupied by just the trailing null of the string in
324 if (PL_regkind[(U8)OP(n)] != NOTHING || OP(n) == NOTHING) {
333 if (OP(scan) != CURLYX) {
334 int max = (reg_off_by_arg[OP(scan)]
336 /* I32 may be smaller than U16 on CRAYs! */
337 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
338 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
342 /* Skip NOTHING and LONGJMP. */
343 while ((n = regnext(n))
344 && ((PL_regkind[(U8)OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
345 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
348 if (reg_off_by_arg[OP(scan)])
351 NEXT_OFF(scan) = off;
353 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
354 || OP(scan) == IFTHEN || OP(scan) == SUSPEND) {
355 next = regnext(scan);
358 if (OP(next) == code || code == IFTHEN || code == SUSPEND) {
359 I32 max1 = 0, min1 = I32_MAX, num = 0;
361 if (flags & SCF_DO_SUBSTR)
363 while (OP(scan) == code) {
364 I32 deltanext, minnext;
368 next = regnext(scan);
369 scan = NEXTOPER(scan);
371 scan = NEXTOPER(scan);
372 /* We suppose the run is continuous, last=next...*/
373 minnext = study_chunk(&scan, &deltanext, next,
377 if (max1 < minnext + deltanext)
378 max1 = minnext + deltanext;
379 if (deltanext == I32_MAX)
380 is_inf = is_inf_internal = 1;
382 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
384 if (data && (data_fake.flags & SF_HAS_EVAL))
385 data->flags |= SF_HAS_EVAL;
389 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
391 if (flags & SCF_DO_SUBSTR) {
392 data->pos_min += min1;
393 data->pos_delta += max1 - min1;
394 if (max1 != min1 || is_inf)
395 data->longest = &(data->longest_float);
398 delta += max1 - min1;
400 else if (code == BRANCHJ) /* single branch is optimized. */
401 scan = NEXTOPER(NEXTOPER(scan));
402 else /* single branch is optimized. */
403 scan = NEXTOPER(scan);
406 else if (OP(scan) == EXACT) {
407 I32 l = *OPERAND(scan);
409 unsigned char *s = (unsigned char *)(OPERAND(scan)+1);
410 unsigned char *e = s + l;
419 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
420 /* The code below prefers earlier match for fixed
421 offset, later match for variable offset. */
422 if (data->last_end == -1) { /* Update the start info. */
423 data->last_start_min = data->pos_min;
424 data->last_start_max = is_inf
425 ? I32_MAX : data->pos_min + data->pos_delta;
427 sv_catpvn(data->last_found, (char *)(OPERAND(scan)+1), *OPERAND(scan));
428 data->last_end = data->pos_min + l;
429 data->pos_min += l; /* As in the first entry. */
430 data->flags &= ~SF_BEFORE_EOL;
433 else if (PL_regkind[(U8)OP(scan)] == EXACT) {
434 I32 l = *OPERAND(scan);
435 if (flags & SCF_DO_SUBSTR)
438 unsigned char *s = (unsigned char *)(OPERAND(scan)+1);
439 unsigned char *e = s + l;
448 if (data && (flags & SCF_DO_SUBSTR))
451 else if (strchr(PL_varies,OP(scan))) {
452 I32 mincount, maxcount, minnext, deltanext, pos_before, fl;
453 regnode *oscan = scan;
455 switch (PL_regkind[(U8)OP(scan)]) {
457 scan = NEXTOPER(scan);
460 if (flags & SCF_DO_SUBSTR) {
461 next = NEXTOPER(scan);
462 if (OP(next) == EXACT) {
464 maxcount = REG_INFTY;
465 next = regnext(scan);
466 scan = NEXTOPER(scan);
470 if (flags & SCF_DO_SUBSTR)
475 is_inf = is_inf_internal = 1;
476 scan = regnext(scan);
477 if (flags & SCF_DO_SUBSTR) {
479 data->longest = &(data->longest_float);
481 goto optimize_curly_tail;
483 mincount = ARG1(scan);
484 maxcount = ARG2(scan);
485 next = regnext(scan);
486 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
488 if (flags & SCF_DO_SUBSTR) {
489 if (mincount == 0) scan_commit(data);
490 pos_before = data->pos_min;
494 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
496 data->flags |= SF_IS_INF;
498 /* This will finish on WHILEM, setting scan, or on NULL: */
499 minnext = study_chunk(&scan, &deltanext, last, data,
501 ? (flags & ~SCF_DO_SUBSTR) : flags);
502 if (!scan) /* It was not CURLYX, but CURLY. */
504 if (ckWARN(WARN_UNSAFE) && (minnext + deltanext == 0)
505 && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
506 && maxcount <= REG_INFTY/3) /* Complement check for big count */
507 warner(WARN_UNSAFE, "Strange *+?{} on zero-length expression");
508 min += minnext * mincount;
509 is_inf_internal |= (maxcount == REG_INFTY
510 && (minnext + deltanext) > 0
511 || deltanext == I32_MAX);
512 is_inf |= is_inf_internal;
513 delta += (minnext + deltanext) * maxcount - minnext * mincount;
515 /* Try powerful optimization CURLYX => CURLYN. */
516 if ( OP(oscan) == CURLYX && data
517 && data->flags & SF_IN_PAR
518 && !(data->flags & SF_HAS_EVAL)
519 && !deltanext && minnext == 1 ) {
520 /* Try to optimize to CURLYN. */
521 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
522 regnode *nxt1 = nxt, *nxt2;
526 if (!strchr(PL_simple,OP(nxt))
527 && !(PL_regkind[(U8)OP(nxt)] == EXACT
528 && *OPERAND(nxt) == 1))
532 if (OP(nxt) != CLOSE)
534 /* Now we know that nxt2 is the only contents: */
535 oscan->flags = ARG(nxt);
537 OP(nxt1) = NOTHING; /* was OPEN. */
539 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
540 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
541 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
542 OP(nxt) = OPTIMIZED; /* was CLOSE. */
543 OP(nxt + 1) = OPTIMIZED; /* was count. */
544 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
549 /* Try optimization CURLYX => CURLYM. */
550 if ( OP(oscan) == CURLYX && data
551 && !(data->flags & SF_HAS_PAR)
552 && !(data->flags & SF_HAS_EVAL)
554 /* XXXX How to optimize if data == 0? */
555 /* Optimize to a simpler form. */
556 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
560 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
561 && (OP(nxt2) != WHILEM))
563 OP(nxt2) = SUCCEED; /* Whas WHILEM */
564 /* Need to optimize away parenths. */
565 if (data->flags & SF_IN_PAR) {
566 /* Set the parenth number. */
567 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
569 if (OP(nxt) != CLOSE)
570 FAIL("panic opt close");
571 oscan->flags = ARG(nxt);
572 OP(nxt1) = OPTIMIZED; /* was OPEN. */
573 OP(nxt) = OPTIMIZED; /* was CLOSE. */
575 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
576 OP(nxt + 1) = OPTIMIZED; /* was count. */
577 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
578 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
581 while ( nxt1 && (OP(nxt1) != WHILEM)) {
582 regnode *nnxt = regnext(nxt1);
585 if (reg_off_by_arg[OP(nxt1)])
586 ARG_SET(nxt1, nxt2 - nxt1);
587 else if (nxt2 - nxt1 < U16_MAX)
588 NEXT_OFF(nxt1) = nxt2 - nxt1;
590 OP(nxt) = NOTHING; /* Cannot beautify */
595 /* Optimize again: */
596 study_chunk(&nxt1, &deltanext, nxt, NULL, 0);
601 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
603 if (flags & SCF_DO_SUBSTR) {
604 SV *last_str = Nullsv;
605 int counted = mincount != 0;
607 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
608 I32 b = pos_before >= data->last_start_min
609 ? pos_before : data->last_start_min;
611 char *s = SvPV(data->last_found, l);
612 I32 old = b - data->last_start_min;
615 old = utf8_hop((U8*)s, old) - (U8*)s;
618 /* Get the added string: */
619 last_str = newSVpvn(s + old, l);
620 if (deltanext == 0 && pos_before == b) {
621 /* What was added is a constant string */
623 SvGROW(last_str, (mincount * l) + 1);
624 repeatcpy(SvPVX(last_str) + l,
625 SvPVX(last_str), l, mincount - 1);
626 SvCUR(last_str) *= mincount;
627 /* Add additional parts. */
628 SvCUR_set(data->last_found,
629 SvCUR(data->last_found) - l);
630 sv_catsv(data->last_found, last_str);
631 data->last_end += l * (mincount - 1);
635 /* It is counted once already... */
636 data->pos_min += minnext * (mincount - counted);
637 data->pos_delta += - counted * deltanext +
638 (minnext + deltanext) * maxcount - minnext * mincount;
639 if (mincount != maxcount) {
641 if (mincount && last_str) {
642 sv_setsv(data->last_found, last_str);
643 data->last_end = data->pos_min;
644 data->last_start_min =
645 data->pos_min - CHR_SVLEN(last_str);
646 data->last_start_max = is_inf
648 : data->pos_min + data->pos_delta
649 - CHR_SVLEN(last_str);
651 data->longest = &(data->longest_float);
653 SvREFCNT_dec(last_str);
655 if (data && (fl & SF_HAS_EVAL))
656 data->flags |= SF_HAS_EVAL;
658 if (OP(oscan) != CURLYX) {
659 while (PL_regkind[(U8)OP(next = regnext(oscan))] == NOTHING
661 NEXT_OFF(oscan) += NEXT_OFF(next);
664 default: /* REF only? */
665 if (flags & SCF_DO_SUBSTR) {
667 data->longest = &(data->longest_float);
669 is_inf = is_inf_internal = 1;
673 else if (strchr(PL_simple,OP(scan)) || PL_regkind[(U8)OP(scan)] == ANYUTF8) {
674 if (flags & SCF_DO_SUBSTR) {
680 else if (PL_regkind[(U8)OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
681 data->flags |= (OP(scan) == MEOL
685 else if (PL_regkind[(U8)OP(scan)] == BRANCHJ
686 && (scan->flags || data)
687 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
688 I32 deltanext, minnext;
692 next = regnext(scan);
693 nscan = NEXTOPER(NEXTOPER(scan));
694 minnext = study_chunk(&nscan, &deltanext, last, &data_fake, 0);
697 FAIL("variable length lookbehind not implemented");
699 else if (minnext > U8_MAX) {
700 FAIL2("lookbehind longer than %d not implemented", U8_MAX);
702 scan->flags = minnext;
704 if (data && data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
706 if (data && (data_fake.flags & SF_HAS_EVAL))
707 data->flags |= SF_HAS_EVAL;
709 else if (OP(scan) == OPEN) {
712 else if (OP(scan) == CLOSE && ARG(scan) == is_par) {
713 next = regnext(scan);
715 if ( next && (OP(next) != WHILEM) && next < last)
716 is_par = 0; /* Disable optimization */
718 else if (OP(scan) == EVAL) {
720 data->flags |= SF_HAS_EVAL;
722 else if (OP(scan) == LOGICAL && scan->flags == 2) { /* Embedded */
723 if (flags & SCF_DO_SUBSTR) {
725 data->longest = &(data->longest_float);
727 is_inf = is_inf_internal = 1;
729 /* Else: zero-length, ignore. */
730 scan = regnext(scan);
735 *deltap = is_inf_internal ? I32_MAX : delta;
736 if (flags & SCF_DO_SUBSTR && is_inf)
737 data->pos_delta = I32_MAX - data->pos_min;
740 if (is_par && pars==1 && data) {
741 data->flags |= SF_IN_PAR;
742 data->flags &= ~SF_HAS_PAR;
744 else if (pars && data) {
745 data->flags |= SF_HAS_PAR;
746 data->flags &= ~SF_IN_PAR;
752 add_data(I32 n, char *s)
755 if (PL_regcomp_rx->data) {
756 Renewc(PL_regcomp_rx->data,
757 sizeof(*PL_regcomp_rx->data) + sizeof(void*) * (PL_regcomp_rx->data->count + n - 1),
758 char, struct reg_data);
759 Renew(PL_regcomp_rx->data->what, PL_regcomp_rx->data->count + n, U8);
760 PL_regcomp_rx->data->count += n;
763 Newc(1207, PL_regcomp_rx->data, sizeof(*PL_regcomp_rx->data) + sizeof(void*) * (n - 1),
764 char, struct reg_data);
765 New(1208, PL_regcomp_rx->data->what, n, U8);
766 PL_regcomp_rx->data->count = n;
768 Copy(s, PL_regcomp_rx->data->what + PL_regcomp_rx->data->count - n, n, U8);
769 return PL_regcomp_rx->data->count - n;
777 char *s = PerlEnv_getenv("PERL_RE_COLORS");
780 PL_colors[0] = s = savepv(s);
788 PL_colors[i] = s = "";
798 - pregcomp - compile a regular expression into internal code
800 * We can't allocate space until we know how big the compiled form will be,
801 * but we can't compile it (and thus know how big it is) until we've got a
802 * place to put the code. So we cheat: we compile it twice, once with code
803 * generation turned off and size counting turned on, and once "for real".
804 * This also means that we don't allocate space until we are sure that the
805 * thing really will compile successfully, and we never have to move the
806 * code and thus invalidate pointers into it. (Note that it has to be in
807 * one piece because free() must be able to free it all.) [NB: not true in perl]
809 * Beware that the optimization-preparation code in here knows about some
810 * of the structure of the compiled regexp. [I'll say.]
813 pregcomp(char *exp, char *xend, PMOP *pm)
828 FAIL("NULL regexp argument");
830 if (PL_curcop == &PL_compiling ? (PL_hints & HINT_UTF8) : IN_UTF8)
831 PL_reg_flags |= RF_utf8;
835 PL_regprecomp = savepvn(exp, xend - exp);
836 DEBUG_r(if (!PL_colorset) reginitcolors());
837 DEBUG_r(PerlIO_printf(Perl_debug_log, "%sCompiling%s RE `%s%*s%s'\n",
838 PL_colors[4],PL_colors[5],PL_colors[0],
839 xend - exp, PL_regprecomp, PL_colors[1]));
840 PL_regflags = pm->op_pmflags;
844 PL_seen_zerolen = *exp == '^' ? -1 : 0;
848 /* First pass: determine size, legality. */
849 PL_regcomp_parse = exp;
854 PL_regcode = &PL_regdummy;
855 regc((U8)REG_MAGIC, (char*)PL_regcode);
856 if (reg(0, &flags) == NULL) {
857 Safefree(PL_regprecomp);
858 PL_regprecomp = Nullch;
861 DEBUG_r(PerlIO_printf(Perl_debug_log, "size %d ", PL_regsize));
863 /* Small enough for pointer-storage convention?
864 If extralen==0, this means that we will not need long jumps. */
865 if (PL_regsize >= 0x10000L && PL_extralen)
866 PL_regsize += PL_extralen;
870 /* Allocate space and initialize. */
871 Newc(1001, r, sizeof(regexp) + (unsigned)PL_regsize * sizeof(regnode),
874 FAIL("regexp out of space");
876 r->prelen = xend - exp;
877 r->precomp = PL_regprecomp;
879 r->reganch = pm->op_pmflags & PMf_COMPILETIME;
880 r->nparens = PL_regnpar - 1; /* set early to validate backrefs */
882 r->substrs = 0; /* Useful during FAIL. */
883 r->startp = 0; /* Useful during FAIL. */
884 r->endp = 0; /* Useful during FAIL. */
888 /* Second pass: emit code. */
889 PL_regcomp_parse = exp;
893 PL_regcode = r->program;
894 /* Store the count of eval-groups for security checks: */
895 PL_regcode->next_off = ((PL_seen_evals > U16_MAX) ? U16_MAX : PL_seen_evals);
896 regc((U8)REG_MAGIC, (char*) PL_regcode++);
898 if (reg(0, &flags) == NULL)
901 /* Dig out information for optimizations. */
902 r->reganch = pm->op_pmflags & PMf_COMPILETIME; /* Again? */
903 pm->op_pmflags = PL_regflags;
905 r->reganch |= ROPT_UTF8;
906 r->regstclass = NULL;
907 if (PL_regnaughty >= 10) /* Probably an expensive pattern. */
908 r->reganch |= ROPT_NAUGHTY;
909 scan = r->program + 1; /* First BRANCH. */
911 /* XXXX To minimize changes to RE engine we always allocate
912 3-units-long substrs field. */
913 Newz(1004, r->substrs, 1, struct reg_substr_data);
915 if (OP(scan) != BRANCH) { /* Only one top-level choice. */
918 STRLEN longest_float_length, longest_fixed_length;
920 StructCopy(&zero_scan_data, &data, scan_data_t);
922 /* Skip introductions and multiplicators >= 1. */
923 while ((OP(first) == OPEN && (sawopen = 1)) ||
924 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
925 (OP(first) == PLUS) ||
926 (OP(first) == MINMOD) ||
927 (PL_regkind[(U8)OP(first)] == CURLY && ARG1(first) > 0) ) {
928 if (OP(first) == PLUS)
931 first += regarglen[(U8)OP(first)];
932 first = NEXTOPER(first);
935 /* Starting-point info. */
937 if (OP(first) == EXACT); /* Empty, get anchored substr later. */
938 else if (strchr(PL_simple+4,OP(first)))
939 r->regstclass = first;
940 else if (PL_regkind[(U8)OP(first)] == BOUND ||
941 PL_regkind[(U8)OP(first)] == NBOUND)
942 r->regstclass = first;
943 else if (PL_regkind[(U8)OP(first)] == BOL) {
944 r->reganch |= (OP(first) == MBOL ? ROPT_ANCH_MBOL: ROPT_ANCH_BOL);
945 first = NEXTOPER(first);
948 else if (OP(first) == GPOS) {
949 r->reganch |= ROPT_ANCH_GPOS;
950 first = NEXTOPER(first);
953 else if ((OP(first) == STAR &&
954 PL_regkind[(U8)OP(NEXTOPER(first))] == REG_ANY) &&
955 !(r->reganch & ROPT_ANCH) )
957 /* turn .* into ^.* with an implied $*=1 */
958 r->reganch |= ROPT_ANCH_BOL | ROPT_IMPLICIT;
959 first = NEXTOPER(first);
962 if (sawplus && (!sawopen || !PL_regsawback))
963 r->reganch |= ROPT_SKIP; /* x+ must match 1st of run */
965 /* Scan is after the zeroth branch, first is atomic matcher. */
966 DEBUG_r(PerlIO_printf(Perl_debug_log, "first at %d\n",
969 * If there's something expensive in the r.e., find the
970 * longest literal string that must appear and make it the
971 * regmust. Resolve ties in favor of later strings, since
972 * the regstart check works with the beginning of the r.e.
973 * and avoiding duplication strengthens checking. Not a
974 * strong reason, but sufficient in the absence of others.
975 * [Now we resolve ties in favor of the earlier string if
976 * it happens that c_offset_min has been invalidated, since the
977 * earlier string may buy us something the later one won't.]
981 data.longest_fixed = newSVpvn("",0);
982 data.longest_float = newSVpvn("",0);
983 data.last_found = newSVpvn("",0);
984 data.longest = &(data.longest_fixed);
987 minlen = study_chunk(&first, &fake, scan + PL_regsize, /* Up to end */
988 &data, SCF_DO_SUBSTR);
989 if ( PL_regnpar == 1 && data.longest == &(data.longest_fixed)
990 && data.last_start_min == 0 && data.last_end > 0
992 && (!(PL_regseen & REG_SEEN_GPOS) || (r->reganch & ROPT_ANCH_GPOS)))
993 r->reganch |= ROPT_CHECK_ALL;
995 SvREFCNT_dec(data.last_found);
997 longest_float_length = CHR_SVLEN(data.longest_float);
998 if (longest_float_length
999 || (data.flags & SF_FL_BEFORE_EOL
1000 && (!(data.flags & SF_FL_BEFORE_MEOL)
1001 || (PL_regflags & PMf_MULTILINE)))) {
1004 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
1005 && data.offset_fixed == data.offset_float_min
1006 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
1007 goto remove_float; /* As in (a)+. */
1009 r->float_substr = data.longest_float;
1010 r->float_min_offset = data.offset_float_min;
1011 r->float_max_offset = data.offset_float_max;
1012 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
1013 && (!(data.flags & SF_FL_BEFORE_MEOL)
1014 || (PL_regflags & PMf_MULTILINE)));
1015 fbm_compile(r->float_substr, t ? FBMcf_TAIL : 0);
1019 r->float_substr = Nullsv;
1020 SvREFCNT_dec(data.longest_float);
1021 longest_float_length = 0;
1024 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
1025 if (longest_fixed_length
1026 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
1027 && (!(data.flags & SF_FIX_BEFORE_MEOL)
1028 || (PL_regflags & PMf_MULTILINE)))) {
1031 r->anchored_substr = data.longest_fixed;
1032 r->anchored_offset = data.offset_fixed;
1033 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
1034 && (!(data.flags & SF_FIX_BEFORE_MEOL)
1035 || (PL_regflags & PMf_MULTILINE)));
1036 fbm_compile(r->anchored_substr, t ? FBMcf_TAIL : 0);
1039 r->anchored_substr = Nullsv;
1040 SvREFCNT_dec(data.longest_fixed);
1041 longest_fixed_length = 0;
1044 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
1045 if (longest_fixed_length > longest_float_length) {
1046 r->check_substr = r->anchored_substr;
1047 r->check_offset_min = r->check_offset_max = r->anchored_offset;
1048 if (r->reganch & ROPT_ANCH_SINGLE)
1049 r->reganch |= ROPT_NOSCAN;
1052 r->check_substr = r->float_substr;
1053 r->check_offset_min = data.offset_float_min;
1054 r->check_offset_max = data.offset_float_max;
1058 /* Several toplevels. Best we can is to set minlen. */
1061 DEBUG_r(PerlIO_printf(Perl_debug_log, "\n"));
1062 scan = r->program + 1;
1063 minlen = study_chunk(&scan, &fake, scan + PL_regsize, NULL, 0);
1064 r->check_substr = r->anchored_substr = r->float_substr = Nullsv;
1068 if (PL_regseen & REG_SEEN_GPOS)
1069 r->reganch |= ROPT_GPOS_SEEN;
1070 if (PL_regseen & REG_SEEN_LOOKBEHIND)
1071 r->reganch |= ROPT_LOOKBEHIND_SEEN;
1072 if (PL_regseen & REG_SEEN_EVAL)
1073 r->reganch |= ROPT_EVAL_SEEN;
1074 Newz(1002, r->startp, PL_regnpar, I32);
1075 Newz(1002, r->endp, PL_regnpar, I32);
1076 DEBUG_r(regdump(r));
1081 - reg - regular expression, i.e. main body or parenthesized thing
1083 * Caller must absorb opening parenthesis.
1085 * Combining parenthesis handling with the base level of regular expression
1086 * is a trifle forced, but the need to tie the tails of the branches to what
1087 * follows makes it hard to avoid.
1090 reg(I32 paren, I32 *flagp)
1091 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
1094 register regnode *ret; /* Will be the head of the group. */
1095 register regnode *br;
1096 register regnode *lastbr;
1097 register regnode *ender = 0;
1098 register I32 parno = 0;
1099 I32 flags, oregflags = PL_regflags, have_branch = 0, open = 0;
1102 *flagp = 0; /* Tentatively. */
1104 /* Make an OPEN node, if parenthesized. */
1106 if (*PL_regcomp_parse == '?') {
1107 U16 posflags = 0, negflags = 0;
1108 U16 *flagsp = &posflags;
1112 paren = *PL_regcomp_parse++;
1113 ret = NULL; /* For look-ahead/behind. */
1116 PL_regseen |= REG_SEEN_LOOKBEHIND;
1117 if (*PL_regcomp_parse == '!')
1119 if (*PL_regcomp_parse != '=' && *PL_regcomp_parse != '!')
1130 FAIL2("Sequence (?%c...) not implemented", (int)paren);
1133 while (*PL_regcomp_parse && *PL_regcomp_parse != ')')
1135 if (*PL_regcomp_parse != ')')
1136 FAIL("Sequence (?#... not terminated");
1142 paren = *PL_regcomp_parse++;
1147 I32 count = 1, n = 0;
1149 char *s = PL_regcomp_parse;
1151 OP_4tree *sop, *rop;
1154 PL_regseen |= REG_SEEN_EVAL;
1155 while (count && (c = *PL_regcomp_parse)) {
1156 if (c == '\\' && PL_regcomp_parse[1])
1164 if (*PL_regcomp_parse != ')')
1165 FAIL("Sequence (?{...}) not terminated or not {}-balanced");
1169 if (PL_regcomp_parse - 1 - s)
1170 sv = newSVpvn(s, PL_regcomp_parse - 1 - s);
1172 sv = newSVpvn("", 0);
1174 rop = sv_compile_2op(sv, &sop, "re", &av);
1176 n = add_data(3, "nop");
1177 PL_regcomp_rx->data->data[n] = (void*)rop;
1178 PL_regcomp_rx->data->data[n+1] = (void*)sop;
1179 PL_regcomp_rx->data->data[n+2] = (void*)av;
1182 else { /* First pass */
1183 if (PL_reginterp_cnt < ++PL_seen_evals
1184 && PL_curcop != &PL_compiling)
1185 /* No compiled RE interpolated, has runtime
1186 components ===> unsafe. */
1187 FAIL("Eval-group not allowed at runtime, use re 'eval'");
1189 FAIL("Eval-group in insecure regular expression");
1194 ret = reg_node(LOGICAL);
1197 regtail(ret, reganode(EVAL, n));
1200 return reganode(EVAL, n);
1204 if (PL_regcomp_parse[0] == '?') {
1205 if (PL_regcomp_parse[1] == '=' || PL_regcomp_parse[1] == '!'
1206 || PL_regcomp_parse[1] == '<'
1207 || PL_regcomp_parse[1] == '{') { /* Lookahead or eval. */
1210 ret = reg_node(LOGICAL);
1213 regtail(ret, reg(1, &flag));
1217 else if (PL_regcomp_parse[0] >= '1' && PL_regcomp_parse[0] <= '9' ) {
1218 parno = atoi(PL_regcomp_parse++);
1220 while (isDIGIT(*PL_regcomp_parse))
1222 ret = reganode(GROUPP, parno);
1223 if ((c = *nextchar()) != ')')
1224 FAIL2("Switch (?(number%c not recognized", c);
1226 regtail(ret, reganode(IFTHEN, 0));
1227 br = regbranch(&flags, 1);
1229 br = reganode(LONGJMP, 0);
1231 regtail(br, reganode(LONGJMP, 0));
1236 lastbr = reganode(IFTHEN, 0); /* Fake one for optimizer. */
1237 regbranch(&flags, 1);
1238 regtail(ret, lastbr);
1246 FAIL("Switch (?(condition)... contains too many branches");
1247 ender = reg_node(TAIL);
1250 regtail(lastbr, ender);
1251 regtail(NEXTOPER(NEXTOPER(lastbr)), ender);
1254 regtail(ret, ender);
1258 FAIL2("Unknown condition for (?(%.2s", PL_regcomp_parse);
1262 FAIL("Sequence (? incomplete");
1267 while (*PL_regcomp_parse && strchr("iogcmsx", *PL_regcomp_parse)) {
1268 if (*PL_regcomp_parse != 'o')
1269 pmflag(flagsp, *PL_regcomp_parse);
1272 if (*PL_regcomp_parse == '-') {
1277 PL_regflags |= posflags;
1278 PL_regflags &= ~negflags;
1279 if (*PL_regcomp_parse == ':') {
1285 if (*PL_regcomp_parse != ')')
1286 FAIL2("Sequence (?%c...) not recognized", *PL_regcomp_parse);
1295 ret = reganode(OPEN, parno);
1302 /* Pick up the branches, linking them together. */
1303 br = regbranch(&flags, 1);
1306 if (*PL_regcomp_parse == '|') {
1307 if (!SIZE_ONLY && PL_extralen) {
1308 reginsert(BRANCHJ, br);
1311 reginsert(BRANCH, br);
1314 PL_extralen += 1; /* For BRANCHJ-BRANCH. */
1316 else if (paren == ':') {
1317 *flagp |= flags&SIMPLE;
1319 if (open) { /* Starts with OPEN. */
1320 regtail(ret, br); /* OPEN -> first. */
1322 else if (paren != '?') /* Not Conditional */
1326 *flagp |= flags&SPSTART;
1328 while (*PL_regcomp_parse == '|') {
1329 if (!SIZE_ONLY && PL_extralen) {
1330 ender = reganode(LONGJMP,0);
1331 regtail(NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
1334 PL_extralen += 2; /* Account for LONGJMP. */
1336 br = regbranch(&flags, 0);
1339 regtail(lastbr, br); /* BRANCH -> BRANCH. */
1343 *flagp |= flags&SPSTART;
1346 if (have_branch || paren != ':') {
1347 /* Make a closing node, and hook it on the end. */
1350 ender = reg_node(TAIL);
1353 ender = reganode(CLOSE, parno);
1359 *flagp &= ~HASWIDTH;
1362 ender = reg_node(SUCCEED);
1365 ender = reg_node(END);
1368 regtail(lastbr, ender);
1371 /* Hook the tails of the branches to the closing node. */
1372 for (br = ret; br != NULL; br = regnext(br)) {
1373 regoptail(br, ender);
1380 static char parens[] = "=!<,>";
1382 if (paren && (p = strchr(parens, paren))) {
1383 int node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
1384 int flag = (p - parens) > 1;
1387 node = SUSPEND, flag = 0;
1388 reginsert(node,ret);
1390 regtail(ret, reg_node(TAIL));
1394 /* Check for proper termination. */
1396 PL_regflags = oregflags;
1397 if (PL_regcomp_parse >= PL_regxend || *nextchar() != ')') {
1398 FAIL("unmatched () in regexp");
1401 else if (!paren && PL_regcomp_parse < PL_regxend) {
1402 if (*PL_regcomp_parse == ')') {
1403 FAIL("unmatched () in regexp");
1406 FAIL("junk on end of regexp"); /* "Can't happen". */
1414 - regbranch - one alternative of an | operator
1416 * Implements the concatenation operator.
1419 regbranch(I32 *flagp, I32 first)
1422 register regnode *ret;
1423 register regnode *chain = NULL;
1424 register regnode *latest;
1425 I32 flags = 0, c = 0;
1430 if (!SIZE_ONLY && PL_extralen)
1431 ret = reganode(BRANCHJ,0);
1433 ret = reg_node(BRANCH);
1436 if (!first && SIZE_ONLY)
1437 PL_extralen += 1; /* BRANCHJ */
1439 *flagp = WORST; /* Tentatively. */
1443 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != '|' && *PL_regcomp_parse != ')') {
1445 latest = regpiece(&flags);
1446 if (latest == NULL) {
1447 if (flags & TRYAGAIN)
1451 else if (ret == NULL)
1453 *flagp |= flags&HASWIDTH;
1454 if (chain == NULL) /* First piece. */
1455 *flagp |= flags&SPSTART;
1458 regtail(chain, latest);
1463 if (chain == NULL) { /* Loop ran zero times. */
1464 chain = reg_node(NOTHING);
1469 *flagp |= flags&SIMPLE;
1476 - regpiece - something followed by possible [*+?]
1478 * Note that the branching code sequences used for ? and the general cases
1479 * of * and + are somewhat optimized: they use the same NOTHING node as
1480 * both the endmarker for their branch list and the body of the last branch.
1481 * It might seem that this node could be dispensed with entirely, but the
1482 * endmarker role is not redundant.
1485 regpiece(I32 *flagp)
1488 register regnode *ret;
1490 register char *next;
1492 char *origparse = PL_regcomp_parse;
1495 I32 max = REG_INFTY;
1497 ret = regatom(&flags);
1499 if (flags & TRYAGAIN)
1504 op = *PL_regcomp_parse;
1506 if (op == '{' && regcurly(PL_regcomp_parse)) {
1507 next = PL_regcomp_parse + 1;
1509 while (isDIGIT(*next) || *next == ',') {
1518 if (*next == '}') { /* got one */
1522 min = atoi(PL_regcomp_parse);
1526 maxpos = PL_regcomp_parse;
1528 if (!max && *maxpos != '0')
1529 max = REG_INFTY; /* meaning "infinity" */
1530 else if (max >= REG_INFTY)
1531 FAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
1532 PL_regcomp_parse = next;
1536 if ((flags&SIMPLE)) {
1537 PL_regnaughty += 2 + PL_regnaughty / 2;
1538 reginsert(CURLY, ret);
1541 PL_regnaughty += 4 + PL_regnaughty; /* compound interest */
1542 regtail(ret, reg_node(WHILEM));
1543 if (!SIZE_ONLY && PL_extralen) {
1544 reginsert(LONGJMP,ret);
1545 reginsert(NOTHING,ret);
1546 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
1548 reginsert(CURLYX,ret);
1549 if (!SIZE_ONLY && PL_extralen)
1550 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
1551 regtail(ret, reg_node(NOTHING));
1561 if (max && max < min)
1562 FAIL("Can't do {n,m} with n > m");
1577 #if 0 /* Now runtime fix should be reliable. */
1578 if (!(flags&HASWIDTH) && op != '?')
1579 FAIL("regexp *+ operand could be empty");
1584 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
1586 if (op == '*' && (flags&SIMPLE)) {
1587 reginsert(STAR, ret);
1591 else if (op == '*') {
1595 else if (op == '+' && (flags&SIMPLE)) {
1596 reginsert(PLUS, ret);
1600 else if (op == '+') {
1604 else if (op == '?') {
1609 if (ckWARN(WARN_UNSAFE) && !SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3) {
1610 warner(WARN_UNSAFE, "%.*s matches null string many times",
1611 PL_regcomp_parse - origparse, origparse);
1614 if (*PL_regcomp_parse == '?') {
1616 reginsert(MINMOD, ret);
1617 regtail(ret, ret + NODE_STEP_REGNODE);
1619 if (ISMULT2(PL_regcomp_parse))
1620 FAIL("nested *?+ in regexp");
1626 - regatom - the lowest level
1628 * Optimization: gobbles an entire sequence of ordinary characters so that
1629 * it can turn them into a single node, which is smaller to store and
1630 * faster to run. Backslashed characters are exceptions, each becoming a
1631 * separate node; the code is simpler that way and it's not worth fixing.
1633 * [Yes, it is worth fixing, some scripts can run twice the speed.]
1639 register regnode *ret = 0;
1642 *flagp = WORST; /* Tentatively. */
1645 switch (*PL_regcomp_parse) {
1649 if (PL_regflags & PMf_MULTILINE)
1650 ret = reg_node(MBOL);
1651 else if (PL_regflags & PMf_SINGLELINE)
1652 ret = reg_node(SBOL);
1654 ret = reg_node(BOL);
1657 if (PL_regcomp_parse[1])
1660 if (PL_regflags & PMf_MULTILINE)
1661 ret = reg_node(MEOL);
1662 else if (PL_regflags & PMf_SINGLELINE)
1663 ret = reg_node(SEOL);
1665 ret = reg_node(EOL);
1670 if (PL_regflags & PMf_SINGLELINE)
1671 ret = reg_node(SANYUTF8);
1673 ret = reg_node(ANYUTF8);
1677 if (PL_regflags & PMf_SINGLELINE)
1678 ret = reg_node(SANY);
1680 ret = reg_node(REG_ANY);
1681 *flagp |= HASWIDTH|SIMPLE;
1687 ret = (UTF ? regclassutf8() : regclass());
1688 if (*PL_regcomp_parse != ']')
1689 FAIL("unmatched [] in regexp");
1691 *flagp |= HASWIDTH|SIMPLE;
1695 ret = reg(1, &flags);
1697 if (flags & TRYAGAIN)
1701 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
1705 if (flags & TRYAGAIN) {
1709 FAIL2("internal urp in regexp at /%s/", PL_regcomp_parse);
1710 /* Supposed to be caught earlier. */
1713 if (!regcurly(PL_regcomp_parse)) {
1721 FAIL("?+*{} follows nothing in regexp");
1724 switch (*++PL_regcomp_parse) {
1727 ret = reg_node(SBOL);
1732 ret = reg_node(GPOS);
1733 PL_regseen |= REG_SEEN_GPOS;
1738 ret = reg_node(SEOL);
1743 ret = reg_node(EOS);
1745 PL_seen_zerolen++; /* Do not optimize RE away */
1749 ret = reg_node(SANY);
1750 *flagp |= HASWIDTH|SIMPLE;
1754 ret = reg_node(CLUMP);
1757 if (UTF && !PL_utf8_mark)
1758 is_utf8_mark((U8*)"~"); /* preload table */
1763 ? (LOC ? ALNUMLUTF8 : ALNUMUTF8)
1764 : (LOC ? ALNUML : ALNUM));
1765 *flagp |= HASWIDTH|SIMPLE;
1767 if (UTF && !PL_utf8_alnum)
1768 is_utf8_alnum((U8*)"a"); /* preload table */
1773 ? (LOC ? NALNUMLUTF8 : NALNUMUTF8)
1774 : (LOC ? NALNUML : NALNUM));
1775 *flagp |= HASWIDTH|SIMPLE;
1777 if (UTF && !PL_utf8_alnum)
1778 is_utf8_alnum((U8*)"a"); /* preload table */
1782 PL_regseen |= REG_SEEN_LOOKBEHIND;
1785 ? (LOC ? BOUNDLUTF8 : BOUNDUTF8)
1786 : (LOC ? BOUNDL : BOUND));
1789 if (UTF && !PL_utf8_alnum)
1790 is_utf8_alnum((U8*)"a"); /* preload table */
1794 PL_regseen |= REG_SEEN_LOOKBEHIND;
1797 ? (LOC ? NBOUNDLUTF8 : NBOUNDUTF8)
1798 : (LOC ? NBOUNDL : NBOUND));
1801 if (UTF && !PL_utf8_alnum)
1802 is_utf8_alnum((U8*)"a"); /* preload table */
1807 ? (LOC ? SPACELUTF8 : SPACEUTF8)
1808 : (LOC ? SPACEL : SPACE));
1809 *flagp |= HASWIDTH|SIMPLE;
1811 if (UTF && !PL_utf8_space)
1812 is_utf8_space((U8*)" "); /* preload table */
1817 ? (LOC ? NSPACELUTF8 : NSPACEUTF8)
1818 : (LOC ? NSPACEL : NSPACE));
1819 *flagp |= HASWIDTH|SIMPLE;
1821 if (UTF && !PL_utf8_space)
1822 is_utf8_space((U8*)" "); /* preload table */
1825 ret = reg_node(UTF ? DIGITUTF8 : DIGIT);
1826 *flagp |= HASWIDTH|SIMPLE;
1828 if (UTF && !PL_utf8_digit)
1829 is_utf8_digit((U8*)"1"); /* preload table */
1832 ret = reg_node(UTF ? NDIGITUTF8 : NDIGIT);
1833 *flagp |= HASWIDTH|SIMPLE;
1835 if (UTF && !PL_utf8_digit)
1836 is_utf8_digit((U8*)"1"); /* preload table */
1840 { /* a lovely hack--pretend we saw [\pX] instead */
1841 char* oldregxend = PL_regxend;
1843 if (PL_regcomp_parse[1] == '{') {
1844 PL_regxend = strchr(PL_regcomp_parse, '}');
1846 FAIL("Missing right brace on \\p{}");
1850 PL_regxend = PL_regcomp_parse + 2;
1853 ret = regclassutf8();
1855 PL_regxend = oldregxend;
1858 *flagp |= HASWIDTH|SIMPLE;
1871 case '1': case '2': case '3': case '4':
1872 case '5': case '6': case '7': case '8': case '9':
1874 I32 num = atoi(PL_regcomp_parse);
1876 if (num > 9 && num >= PL_regnpar)
1879 if (!SIZE_ONLY && num > PL_regcomp_rx->nparens)
1880 FAIL("reference to nonexistent group");
1883 ? (LOC ? REFFL : REFF)
1886 while (isDIGIT(*PL_regcomp_parse))
1894 if (PL_regcomp_parse >= PL_regxend)
1895 FAIL("trailing \\ in regexp");
1898 /* Do not generate `unrecognized' warnings here, we fall
1899 back into the quick-grab loop below */
1905 if (PL_regflags & PMf_EXTENDED) {
1906 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != '\n') PL_regcomp_parse++;
1907 if (PL_regcomp_parse < PL_regxend)
1923 ? (LOC ? EXACTFL : EXACTF)
1925 s = (char *) OPERAND(ret);
1926 regc(0, s++); /* save spot for len */
1927 for (len = 0, p = PL_regcomp_parse - 1;
1928 len < 127 && p < PL_regxend;
1933 if (PL_regflags & PMf_EXTENDED)
1934 p = regwhite(p, PL_regxend);
1988 char* e = strchr(p, '}');
1991 FAIL("Missing right brace on \\x{}");
1993 ender = scan_hex(p + 1, e - p, &numlen);
1994 if (numlen + len >= 127) { /* numlen is generous */
2001 FAIL("Can't use \\x{} without 'use utf8' declaration");
2004 ender = scan_hex(p, 2, &numlen);
2010 ender = UCHARAT(p++);
2011 ender = toCTRL(ender);
2013 case '0': case '1': case '2': case '3':case '4':
2014 case '5': case '6': case '7': case '8':case '9':
2016 (isDIGIT(p[1]) && atoi(p) >= PL_regnpar) ) {
2017 ender = scan_oct(p, 3, &numlen);
2026 if (p >= PL_regxend)
2027 FAIL("trailing \\ in regexp");
2030 if (!SIZE_ONLY && ckWARN(WARN_UNSAFE) && isALPHA(*p))
2032 "/%.127s/: Unrecognized escape \\%c passed through",
2035 goto normal_default;
2040 if ((*p & 0xc0) == 0xc0 && UTF) {
2041 ender = utf8_to_uv((U8*)p, &numlen);
2048 if (PL_regflags & PMf_EXTENDED)
2049 p = regwhite(p, PL_regxend);
2052 ender = toLOWER_LC_uni(ender);
2054 ender = toLOWER_uni(ender);
2056 if (ISMULT2(p)) { /* Back off on ?+*. */
2059 else if (ender >= 0x80 && UTF) {
2060 reguni(ender, s, &numlen);
2070 if (ender >= 0x80 && UTF) {
2071 reguni(ender, s, &numlen);
2079 PL_regcomp_parse = p - 1;
2082 FAIL("internal disaster in regexp");
2088 *OPERAND(ret) = len;
2091 PL_regsize += (len + 2 + sizeof(regnode) - 1) / sizeof(regnode);
2094 PL_regcode += (len + 2 + sizeof(regnode) - 1) / sizeof(regnode);
2104 regwhite(char *p, char *e)
2109 else if (*p == '#') {
2112 } while (p < e && *p != '\n');
2120 /* parse POSIX character classes like [[:foo:]] */
2122 regpposixcc(I32 value)
2127 if (value == '[' && PL_regcomp_parse + 1 < PL_regxend &&
2128 /* I smell either [: or [= or [. -- POSIX has been here, right? */
2129 (*PL_regcomp_parse == ':' ||
2130 *PL_regcomp_parse == '=' ||
2131 *PL_regcomp_parse == '.')) {
2132 char c = *PL_regcomp_parse;
2133 char* s = PL_regcomp_parse++;
2135 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != c)
2137 if (PL_regcomp_parse == PL_regxend)
2138 /* Grandfather lone [:, [=, [. */
2139 PL_regcomp_parse = s;
2141 PL_regcomp_parse++; /* skip over the c */
2142 if (*PL_regcomp_parse == ']') {
2143 /* Not Implemented Yet.
2144 * (POSIX Extended Character Classes, that is)
2145 * The text between e.g. [: and :] would start
2146 * at s + 1 and stop at regcomp_parse - 2. */
2147 if (ckWARN(WARN_UNSAFE) && !SIZE_ONLY)
2149 "Character class syntax [%c %c] is reserved for future extensions", c, c);
2150 PL_regcomp_parse++; /* skip over the ending ] */
2154 /* maternal grandfather */
2155 PL_regcomp_parse = s;
2167 register char *opnd, *s;
2169 register I32 lastvalue = 1234;
2170 register I32 range = 0;
2171 register regnode *ret;
2175 s = opnd = (char *) OPERAND(PL_regcode);
2176 ret = reg_node(ANYOF);
2177 for (value = 0; value < 33; value++)
2179 if (*PL_regcomp_parse == '^') { /* Complement of range. */
2183 *opnd |= ANYOF_INVERT;
2186 PL_regcode += ANY_SKIP;
2188 *opnd |= ANYOF_FOLD;
2190 *opnd |= ANYOF_LOCALE;
2193 PL_regsize += ANY_SKIP;
2195 if (*PL_regcomp_parse == ']' || *PL_regcomp_parse == '-')
2196 goto skipcond; /* allow 1st char to be ] or - */
2197 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != ']') {
2199 value = UCHARAT(PL_regcomp_parse++);
2201 (void)regpposixcc(value); /* ignore the return value for now */
2202 else if (value == '\\') {
2203 value = UCHARAT(PL_regcomp_parse++);
2208 *opnd |= ANYOF_ALNUML;
2210 for (value = 0; value < 256; value++)
2212 ANYOF_SET(opnd, value);
2220 *opnd |= ANYOF_NALNUML;
2222 for (value = 0; value < 256; value++)
2223 if (!isALNUM(value))
2224 ANYOF_SET(opnd, value);
2232 *opnd |= ANYOF_SPACEL;
2234 for (value = 0; value < 256; value++)
2236 ANYOF_SET(opnd, value);
2244 *opnd |= ANYOF_NSPACEL;
2246 for (value = 0; value < 256; value++)
2247 if (!isSPACE(value))
2248 ANYOF_SET(opnd, value);
2255 for (value = '0'; value <= '9'; value++)
2256 ANYOF_SET(opnd, value);
2262 for (value = 0; value < '0'; value++)
2263 ANYOF_SET(opnd, value);
2264 for (value = '9' + 1; value < 256; value++)
2265 ANYOF_SET(opnd, value);
2291 value = scan_hex(PL_regcomp_parse, 2, &numlen);
2292 PL_regcomp_parse += numlen;
2295 value = UCHARAT(PL_regcomp_parse++);
2296 value = toCTRL(value);
2298 case '0': case '1': case '2': case '3': case '4':
2299 case '5': case '6': case '7': case '8': case '9':
2300 value = scan_oct(--PL_regcomp_parse, 3, &numlen);
2301 PL_regcomp_parse += numlen;
2306 if (lastvalue > value)
2307 FAIL("invalid [] range in regexp");
2312 if (*PL_regcomp_parse == '-' && PL_regcomp_parse+1 < PL_regxend &&
2313 PL_regcomp_parse[1] != ']') {
2316 continue; /* do it next time */
2321 if ((isLOWER(lastvalue) && isLOWER(value)) ||
2322 (isUPPER(lastvalue) && isUPPER(value)))
2325 if (isLOWER(lastvalue)) {
2326 for (i = lastvalue; i <= value; i++)
2330 for (i = lastvalue; i <= value; i++)
2337 for ( ; lastvalue <= value; lastvalue++)
2338 ANYOF_SET(opnd, lastvalue);
2342 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
2343 if (!SIZE_ONLY && (*opnd & (0xFF ^ ANYOF_INVERT)) == ANYOF_FOLD) {
2344 for (value = 0; value < 256; ++value) {
2345 if (ANYOF_TEST(opnd, value)) {
2346 I32 cf = PL_fold[value];
2347 ANYOF_SET(opnd, cf);
2350 *opnd &= ~ANYOF_FOLD;
2352 /* optimize inverted simple patterns (e.g. [^a-z]) */
2353 if (!SIZE_ONLY && (*opnd & 0xFF) == ANYOF_INVERT) {
2354 for (value = 0; value < 32; ++value)
2355 opnd[1 + value] ^= 0xFF;
2364 register char *opnd, *e;
2366 register U32 lastvalue = 123456;
2367 register I32 range = 0;
2368 register regnode *ret;
2375 if (*PL_regcomp_parse == '^') { /* Complement of range. */
2379 flags |= ANYOF_INVERT;
2383 flags |= ANYOF_FOLD;
2385 flags |= ANYOF_LOCALE;
2386 listsv = newSVpvn("# comment\n",10);
2389 if (*PL_regcomp_parse == ']' || *PL_regcomp_parse == '-')
2390 goto skipcond; /* allow 1st char to be ] or - */
2392 while (PL_regcomp_parse < PL_regxend && *PL_regcomp_parse != ']') {
2394 value = utf8_to_uv((U8*)PL_regcomp_parse, &numlen);
2395 PL_regcomp_parse += numlen;
2398 (void)regpposixcc(value); /* ignore the return value for now */
2399 else if (value == '\\') {
2400 value = utf8_to_uv((U8*)PL_regcomp_parse, &numlen);
2401 PL_regcomp_parse += numlen;
2406 flags |= ANYOF_ALNUML;
2408 sv_catpvf(listsv, "+utf8::IsAlnum\n");
2415 flags |= ANYOF_NALNUML;
2418 "-utf8::IsAlpha\n-utf8::IsDigit\n0000\t%04x\n%04x\tffff\n",
2427 flags |= ANYOF_SPACEL;
2428 sv_catpvf(listsv, "+utf8::IsSpace\n");
2430 is_utf8_space((U8*)" ");
2437 flags |= ANYOF_NSPACEL;
2439 "!utf8::IsSpace\n");
2441 is_utf8_space((U8*)" ");
2447 sv_catpvf(listsv, "+utf8::IsDigit\n");
2454 "!utf8::IsDigit\n");
2460 if (*PL_regcomp_parse == '{') {
2461 e = strchr(PL_regcomp_parse++, '}');
2463 FAIL("Missing right brace on \\p{}");
2464 n = e - PL_regcomp_parse;
2467 e = PL_regcomp_parse;
2472 sv_catpvf(listsv, "+utf8::%.*s\n", n, PL_regcomp_parse);
2475 "!utf8::%.*s\n", n, PL_regcomp_parse);
2477 PL_regcomp_parse = e + 1;
2502 if (*PL_regcomp_parse == '{') {
2503 e = strchr(PL_regcomp_parse++, '}');
2505 FAIL("Missing right brace on \\x{}");
2506 value = scan_hex(PL_regcomp_parse, e - PL_regcomp_parse, &numlen);
2507 PL_regcomp_parse = e + 1;
2510 value = scan_hex(PL_regcomp_parse, 2, &numlen);
2511 PL_regcomp_parse += numlen;
2515 value = UCHARAT(PL_regcomp_parse++);
2516 value = toCTRL(value);
2518 case '0': case '1': case '2': case '3': case '4':
2519 case '5': case '6': case '7': case '8': case '9':
2520 value = scan_oct(--PL_regcomp_parse, 3, &numlen);
2521 PL_regcomp_parse += numlen;
2526 if (lastvalue > value)
2527 FAIL("invalid [] range in regexp");
2529 sv_catpvf(listsv, "%04x\t%04x\n", lastvalue, value);
2535 if (*PL_regcomp_parse == '-' && PL_regcomp_parse+1 < PL_regxend &&
2536 PL_regcomp_parse[1] != ']') {
2539 continue; /* do it next time */
2542 sv_catpvf(listsv, "%04x\n", value);
2546 ret = reganode(ANYOFUTF8, 0);
2549 SV *rv = swash_init("utf8", "", listsv, 1, 0);
2550 SvREFCNT_dec(listsv);
2551 n = add_data(1,"s");
2552 PL_regcomp_rx->data->data[n] = (void*)rv;
2553 ARG1_SET(ret, flags);
2564 char* retval = PL_regcomp_parse++;
2567 if (*PL_regcomp_parse == '(' && PL_regcomp_parse[1] == '?' &&
2568 PL_regcomp_parse[2] == '#') {
2569 while (*PL_regcomp_parse && *PL_regcomp_parse != ')')
2574 if (PL_regflags & PMf_EXTENDED) {
2575 if (isSPACE(*PL_regcomp_parse)) {
2579 else if (*PL_regcomp_parse == '#') {
2580 while (*PL_regcomp_parse && *PL_regcomp_parse != '\n')
2591 - reg_node - emit a node
2593 STATIC regnode * /* Location. */
2597 register regnode *ret;
2598 register regnode *ptr;
2602 SIZE_ALIGN(PL_regsize);
2607 NODE_ALIGN_FILL(ret);
2609 FILL_ADVANCE_NODE(ptr, op);
2616 - reganode - emit a node with an argument
2618 STATIC regnode * /* Location. */
2619 reganode(U8 op, U32 arg)
2622 register regnode *ret;
2623 register regnode *ptr;
2627 SIZE_ALIGN(PL_regsize);
2632 NODE_ALIGN_FILL(ret);
2634 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
2641 - regc - emit (if appropriate) a Unicode character
2644 reguni(UV uv, char* s, I32* lenp)
2649 *lenp = uv_to_utf8(tmpbuf, uv) - tmpbuf;
2652 *lenp = uv_to_utf8((U8*)s, uv) - (U8*)s;
2657 - regc - emit (if appropriate) a byte of code
2668 - reginsert - insert an operator in front of already-emitted operand
2670 * Means relocating the operand.
2673 reginsert(U8 op, regnode *opnd)
2676 register regnode *src;
2677 register regnode *dst;
2678 register regnode *place;
2679 register int offset = regarglen[(U8)op];
2681 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
2684 PL_regsize += NODE_STEP_REGNODE + offset;
2689 PL_regcode += NODE_STEP_REGNODE + offset;
2692 StructCopy(--src, --dst, regnode);
2694 place = opnd; /* Op node, where operand used to be. */
2695 src = NEXTOPER(place);
2696 FILL_ADVANCE_NODE(place, op);
2697 Zero(src, offset, regnode);
2701 - regtail - set the next-pointer at the end of a node chain of p to val.
2704 regtail(regnode *p, regnode *val)
2707 register regnode *scan;
2708 register regnode *temp;
2709 register I32 offset;
2714 /* Find last node. */
2717 temp = regnext(scan);
2723 if (reg_off_by_arg[OP(scan)]) {
2724 ARG_SET(scan, val - scan);
2727 NEXT_OFF(scan) = val - scan;
2732 - regoptail - regtail on operand of first argument; nop if operandless
2735 regoptail(regnode *p, regnode *val)
2738 /* "Operandless" and "op != BRANCH" are synonymous in practice. */
2739 if (p == NULL || SIZE_ONLY)
2741 if (PL_regkind[(U8)OP(p)] == BRANCH) {
2742 regtail(NEXTOPER(p), val);
2744 else if ( PL_regkind[(U8)OP(p)] == BRANCHJ) {
2745 regtail(NEXTOPER(NEXTOPER(p)), val);
2752 - regcurly - a little FSA that accepts {\d+,?\d*}
2755 regcurly(register char *s)
2774 dumpuntil(regnode *start, regnode *node, regnode *last, SV* sv, I32 l)
2777 register char op = EXACT; /* Arbitrary non-END op. */
2778 register regnode *next, *onode;
2780 while (op != END && (!last || node < last)) {
2781 /* While that wasn't END last time... */
2787 next = regnext(node);
2789 if (OP(node) == OPTIMIZED)
2792 PerlIO_printf(Perl_debug_log, "%4d:%*s%s", node - start,
2793 2*l + 1, "", SvPVX(sv));
2794 if (next == NULL) /* Next ptr. */
2795 PerlIO_printf(Perl_debug_log, "(0)");
2797 PerlIO_printf(Perl_debug_log, "(%d)", next - start);
2798 (void)PerlIO_putc(Perl_debug_log, '\n');
2800 if (PL_regkind[(U8)op] == BRANCHJ) {
2801 register regnode *nnode = (OP(next) == LONGJMP
2804 if (last && nnode > last)
2806 node = dumpuntil(start, NEXTOPER(NEXTOPER(node)), nnode, sv, l + 1);
2808 else if (PL_regkind[(U8)op] == BRANCH) {
2809 node = dumpuntil(start, NEXTOPER(node), next, sv, l + 1);
2811 else if ( op == CURLY) { /* `next' might be very big: optimizer */
2812 node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
2813 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1, sv, l + 1);
2815 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
2816 node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
2819 else if ( op == PLUS || op == STAR) {
2820 node = dumpuntil(start, NEXTOPER(node), NEXTOPER(node) + 1, sv, l + 1);
2822 else if (op == ANYOF) {
2823 node = NEXTOPER(node);
2826 else if (PL_regkind[(U8)op] == EXACT) {
2827 /* Literal string, where present. */
2828 node += ((*OPERAND(node)) + 2 + sizeof(regnode) - 1) / sizeof(regnode);
2829 node = NEXTOPER(node);
2832 node = NEXTOPER(node);
2833 node += regarglen[(U8)op];
2835 if (op == CURLYX || op == OPEN)
2837 else if (op == WHILEM)
2840 #endif /* DEBUGGING */
2845 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
2852 SV *sv = sv_newmortal();
2854 (void)dumpuntil(r->program, r->program + 1, NULL, sv, 0);
2856 /* Header fields of interest. */
2857 if (r->anchored_substr)
2858 PerlIO_printf(Perl_debug_log, "anchored `%s%s%s'%s at %d ",
2860 SvPVX(r->anchored_substr),
2862 SvTAIL(r->anchored_substr) ? "$" : "",
2863 r->anchored_offset);
2864 if (r->float_substr)
2865 PerlIO_printf(Perl_debug_log, "floating `%s%s%s'%s at %d..%u ",
2867 SvPVX(r->float_substr),
2869 SvTAIL(r->float_substr) ? "$" : "",
2870 r->float_min_offset, r->float_max_offset);
2871 if (r->check_substr)
2872 PerlIO_printf(Perl_debug_log,
2873 r->check_substr == r->float_substr
2874 ? "(checking floating" : "(checking anchored");
2875 if (r->reganch & ROPT_NOSCAN)
2876 PerlIO_printf(Perl_debug_log, " noscan");
2877 if (r->reganch & ROPT_CHECK_ALL)
2878 PerlIO_printf(Perl_debug_log, " isall");
2879 if (r->check_substr)
2880 PerlIO_printf(Perl_debug_log, ") ");
2882 if (r->regstclass) {
2883 regprop(sv, r->regstclass);
2884 PerlIO_printf(Perl_debug_log, "stclass `%s' ", SvPVX(sv));
2886 if (r->reganch & ROPT_ANCH) {
2887 PerlIO_printf(Perl_debug_log, "anchored");
2888 if (r->reganch & ROPT_ANCH_BOL)
2889 PerlIO_printf(Perl_debug_log, "(BOL)");
2890 if (r->reganch & ROPT_ANCH_MBOL)
2891 PerlIO_printf(Perl_debug_log, "(MBOL)");
2892 if (r->reganch & ROPT_ANCH_GPOS)
2893 PerlIO_printf(Perl_debug_log, "(GPOS)");
2894 PerlIO_putc(Perl_debug_log, ' ');
2896 if (r->reganch & ROPT_GPOS_SEEN)
2897 PerlIO_printf(Perl_debug_log, "GPOS ");
2898 if (r->reganch & ROPT_SKIP)
2899 PerlIO_printf(Perl_debug_log, "plus ");
2900 if (r->reganch & ROPT_IMPLICIT)
2901 PerlIO_printf(Perl_debug_log, "implicit ");
2902 PerlIO_printf(Perl_debug_log, "minlen %ld ", (long) r->minlen);
2903 if (r->reganch & ROPT_EVAL_SEEN)
2904 PerlIO_printf(Perl_debug_log, "with eval ");
2905 PerlIO_printf(Perl_debug_log, "\n");
2906 #endif /* DEBUGGING */
2910 - regprop - printable representation of opcode
2913 regprop(SV *sv, regnode *o)
2919 sv_setpvn(sv, "", 0);
2920 if (OP(o) >= reg_num) /* regnode.type is unsigned */
2921 FAIL("corrupted regexp opcode");
2922 sv_catpv(sv, (char*)reg_name[OP(o)]); /* Take off const! */
2924 k = PL_regkind[(U8)OP(o)];
2927 sv_catpvf(sv, " <%s%s%s>", PL_colors[0], OPERAND(o) + 1, PL_colors[1]);
2928 else if (k == CURLY) {
2929 if (OP(o) == CURLYM || OP(o) == CURLYN)
2930 sv_catpvf(sv, "[%d]", o->flags); /* Parenth number */
2931 sv_catpvf(sv, " {%d,%d}", ARG1(o), ARG2(o));
2933 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP )
2934 sv_catpvf(sv, "%d", ARG(o)); /* Parenth number */
2935 else if (k == LOGICAL)
2936 sv_catpvf(sv, "[%d]", ARG(o)); /* 2: embedded, otherwise 1 */
2937 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
2938 sv_catpvf(sv, "[-%d]", o->flags);
2939 #endif /* DEBUGGING */
2943 pregfree(struct regexp *r)
2946 if (!r || (--r->refcnt > 0))
2949 Safefree(r->precomp);
2950 if (RX_MATCH_COPIED(r))
2951 Safefree(r->subbeg);
2953 if (r->anchored_substr)
2954 SvREFCNT_dec(r->anchored_substr);
2955 if (r->float_substr)
2956 SvREFCNT_dec(r->float_substr);
2957 Safefree(r->substrs);
2960 int n = r->data->count;
2961 AV* new_comppad = NULL;
2966 switch (r->data->what[n]) {
2968 SvREFCNT_dec((SV*)r->data->data[n]);
2971 new_comppad = (AV*)r->data->data[n];
2974 if (new_comppad == NULL)
2975 croak("panic: pregfree comppad");
2976 old_comppad = PL_comppad;
2977 old_curpad = PL_curpad;
2978 PL_comppad = new_comppad;
2979 PL_curpad = AvARRAY(new_comppad);
2980 op_free((OP_4tree*)r->data->data[n]);
2981 PL_comppad = old_comppad;
2982 PL_curpad = old_curpad;
2983 SvREFCNT_dec((SV*)new_comppad);
2989 FAIL2("panic: regfree data code '%c'", r->data->what[n]);
2992 Safefree(r->data->what);
2995 Safefree(r->startp);
3001 - regnext - dig the "next" pointer out of a node
3003 * [Note, when REGALIGN is defined there are two places in regmatch()
3004 * that bypass this code for speed.]
3007 regnext(register regnode *p)
3010 register I32 offset;
3012 if (p == &PL_regdummy)
3015 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
3023 re_croak2(const char* pat1,const char* pat2,...)
3026 STRLEN l1 = strlen(pat1);
3027 STRLEN l2 = strlen(pat2);
3036 Copy(pat1, buf, l1 , char);
3037 Copy(pat2, buf + l1, l2 , char);
3038 buf[l1 + l2] = '\n';
3039 buf[l1 + l2 + 1] = '\0';
3041 /* ANSI variant takes additional second argument */
3042 va_start(args, pat2);
3046 msv = mess(buf, &args);
3048 message = SvPV(msv,l1);
3051 Copy(message, buf, l1 , char);
3052 buf[l1] = '\0'; /* Overwrite \n */
3056 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
3059 save_re_context(void)
3063 SAVEPPTR(PL_regprecomp); /* uncompiled string. */
3064 SAVEI32(PL_regnpar); /* () count. */
3065 SAVEI32(PL_regsize); /* Code size. */
3066 SAVEI16(PL_regflags); /* are we folding, multilining? */
3067 SAVEPPTR(PL_reginput); /* String-input pointer. */
3068 SAVEPPTR(PL_regbol); /* Beginning of input, for ^ check. */
3069 SAVEPPTR(PL_regeol); /* End of input, for $ check. */
3070 SAVESPTR(PL_regstartp); /* Pointer to startp array. */
3071 SAVESPTR(PL_regendp); /* Ditto for endp. */
3072 SAVESPTR(PL_reglastparen); /* Similarly for lastparen. */
3073 SAVEPPTR(PL_regtill); /* How far we are required to go. */
3074 SAVEI32(PL_regprev); /* char before regbol, \n if none */
3075 SAVESPTR(PL_reg_start_tmp); /* from regexec.c */
3076 PL_reg_start_tmp = 0;
3077 SAVEFREEPV(PL_reg_start_tmp);
3078 SAVEI32(PL_reg_start_tmpl); /* from regexec.c */
3079 PL_reg_start_tmpl = 0;
3080 SAVESPTR(PL_regdata);
3081 SAVEI32(PL_reg_flags); /* from regexec.c */
3082 SAVEI32(PL_reg_eval_set); /* from regexec.c */
3083 SAVEI32(PL_regnarrate); /* from regexec.c */
3084 SAVESPTR(PL_regprogram); /* from regexec.c */
3085 SAVEINT(PL_regindent); /* from regexec.c */
3086 SAVESPTR(PL_regcc); /* from regexec.c */
3087 SAVESPTR(PL_curcop);
3088 SAVESPTR(PL_regcomp_rx); /* from regcomp.c */
3089 SAVEI32(PL_regseen); /* from regcomp.c */
3090 SAVEI32(PL_regsawback); /* Did we see \1, ...? */
3091 SAVEI32(PL_regnaughty); /* How bad is this pattern? */
3092 SAVESPTR(PL_regcode); /* Code-emit pointer; ®dummy = don't */
3093 SAVEPPTR(PL_regxend); /* End of input for compile */
3094 SAVEPPTR(PL_regcomp_parse); /* Input-scan pointer. */
3095 SAVESPTR(PL_reg_call_cc); /* from regexec.c */
3096 SAVESPTR(PL_reg_re); /* from regexec.c */
3097 SAVEPPTR(PL_reg_ganch); /* from regexec.c */
3098 SAVESPTR(PL_reg_sv); /* from regexec.c */
3099 SAVESPTR(PL_reg_magic); /* from regexec.c */
3100 SAVEI32(PL_reg_oldpos); /* from regexec.c */
3101 SAVESPTR(PL_reg_oldcurpm); /* from regexec.c */
3102 SAVESPTR(PL_reg_curpm); /* from regexec.c */
3104 SAVEPPTR(PL_reg_starttry); /* from regexec.c */