5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
8 /* This file contains functions for compiling a regular expression. See
9 * also regexec.c which funnily enough, contains functions for executing
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_comp.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
57 **** Alterations to Henry's code are...
59 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
60 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, by Larry Wall and others
62 **** You may distribute under the terms of either the GNU General Public
63 **** License or the Artistic License, as specified in the README file.
66 * Beware that some of this code is subtly aware of the way operator
67 * precedence is structured in regular expressions. Serious changes in
68 * regular-expression syntax might require a total rethink.
71 #define PERL_IN_REGCOMP_C
74 #ifndef PERL_IN_XSUB_RE
79 #ifdef PERL_IN_XSUB_RE
90 # if defined(BUGGY_MSC6)
91 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
92 # pragma optimize("a",off)
93 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
94 # pragma optimize("w",on )
95 # endif /* BUGGY_MSC6 */
102 typedef struct RExC_state_t {
103 U32 flags; /* are we folding, multilining? */
104 char *precomp; /* uncompiled string. */
105 regexp *rx; /* perl core regexp structure */
106 regexp_internal *rxi; /* internal data for regexp object pprivate field */
107 char *start; /* Start of input for compile */
108 char *end; /* End of input for compile */
109 char *parse; /* Input-scan pointer. */
110 I32 whilem_seen; /* number of WHILEM in this expr */
111 regnode *emit_start; /* Start of emitted-code area */
112 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
113 I32 naughty; /* How bad is this pattern? */
114 I32 sawback; /* Did we see \1, ...? */
116 I32 size; /* Code size. */
117 I32 npar; /* Capture buffer count, (OPEN). */
118 I32 cpar; /* Capture buffer count, (CLOSE). */
119 I32 nestroot; /* root parens we are in - used by accept */
123 regnode **open_parens; /* pointers to open parens */
124 regnode **close_parens; /* pointers to close parens */
125 regnode *opend; /* END node in program */
127 HV *charnames; /* cache of named sequences */
128 HV *paren_names; /* Paren names */
129 regnode **recurse; /* Recurse regops */
130 I32 recurse_count; /* Number of recurse regops */
132 char *starttry; /* -Dr: where regtry was called. */
133 #define RExC_starttry (pRExC_state->starttry)
136 const char *lastparse;
138 #define RExC_lastparse (pRExC_state->lastparse)
139 #define RExC_lastnum (pRExC_state->lastnum)
143 #define RExC_flags (pRExC_state->flags)
144 #define RExC_precomp (pRExC_state->precomp)
145 #define RExC_rx (pRExC_state->rx)
146 #define RExC_rxi (pRExC_state->rxi)
147 #define RExC_start (pRExC_state->start)
148 #define RExC_end (pRExC_state->end)
149 #define RExC_parse (pRExC_state->parse)
150 #define RExC_whilem_seen (pRExC_state->whilem_seen)
151 #define RExC_offsets (pRExC_state->rxi->offsets) /* I am not like the others */
152 #define RExC_emit (pRExC_state->emit)
153 #define RExC_emit_start (pRExC_state->emit_start)
154 #define RExC_naughty (pRExC_state->naughty)
155 #define RExC_sawback (pRExC_state->sawback)
156 #define RExC_seen (pRExC_state->seen)
157 #define RExC_size (pRExC_state->size)
158 #define RExC_npar (pRExC_state->npar)
159 #define RExC_nestroot (pRExC_state->nestroot)
160 #define RExC_extralen (pRExC_state->extralen)
161 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
162 #define RExC_seen_evals (pRExC_state->seen_evals)
163 #define RExC_utf8 (pRExC_state->utf8)
164 #define RExC_charnames (pRExC_state->charnames)
165 #define RExC_open_parens (pRExC_state->open_parens)
166 #define RExC_close_parens (pRExC_state->close_parens)
167 #define RExC_opend (pRExC_state->opend)
168 #define RExC_paren_names (pRExC_state->paren_names)
169 #define RExC_recurse (pRExC_state->recurse)
170 #define RExC_recurse_count (pRExC_state->recurse_count)
172 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
173 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
174 ((*s) == '{' && regcurly(s)))
177 #undef SPSTART /* dratted cpp namespace... */
180 * Flags to be passed up and down.
182 #define WORST 0 /* Worst case. */
183 #define HASWIDTH 0x1 /* Known to match non-null strings. */
184 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
185 #define SPSTART 0x4 /* Starts with * or +. */
186 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
188 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
190 /* whether trie related optimizations are enabled */
191 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
192 #define TRIE_STUDY_OPT
193 #define FULL_TRIE_STUDY
199 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
200 #define PBITVAL(paren) (1 << ((paren) & 7))
201 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
202 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
203 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
206 /* About scan_data_t.
208 During optimisation we recurse through the regexp program performing
209 various inplace (keyhole style) optimisations. In addition study_chunk
210 and scan_commit populate this data structure with information about
211 what strings MUST appear in the pattern. We look for the longest
212 string that must appear for at a fixed location, and we look for the
213 longest string that may appear at a floating location. So for instance
218 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
219 strings (because they follow a .* construct). study_chunk will identify
220 both FOO and BAR as being the longest fixed and floating strings respectively.
222 The strings can be composites, for instance
226 will result in a composite fixed substring 'foo'.
228 For each string some basic information is maintained:
230 - offset or min_offset
231 This is the position the string must appear at, or not before.
232 It also implicitly (when combined with minlenp) tells us how many
233 character must match before the string we are searching.
234 Likewise when combined with minlenp and the length of the string
235 tells us how many characters must appear after the string we have
239 Only used for floating strings. This is the rightmost point that
240 the string can appear at. Ifset to I32 max it indicates that the
241 string can occur infinitely far to the right.
244 A pointer to the minimum length of the pattern that the string
245 was found inside. This is important as in the case of positive
246 lookahead or positive lookbehind we can have multiple patterns
251 The minimum length of the pattern overall is 3, the minimum length
252 of the lookahead part is 3, but the minimum length of the part that
253 will actually match is 1. So 'FOO's minimum length is 3, but the
254 minimum length for the F is 1. This is important as the minimum length
255 is used to determine offsets in front of and behind the string being
256 looked for. Since strings can be composites this is the length of the
257 pattern at the time it was commited with a scan_commit. Note that
258 the length is calculated by study_chunk, so that the minimum lengths
259 are not known until the full pattern has been compiled, thus the
260 pointer to the value.
264 In the case of lookbehind the string being searched for can be
265 offset past the start point of the final matching string.
266 If this value was just blithely removed from the min_offset it would
267 invalidate some of the calculations for how many chars must match
268 before or after (as they are derived from min_offset and minlen and
269 the length of the string being searched for).
270 When the final pattern is compiled and the data is moved from the
271 scan_data_t structure into the regexp structure the information
272 about lookbehind is factored in, with the information that would
273 have been lost precalculated in the end_shift field for the
276 The fields pos_min and pos_delta are used to store the minimum offset
277 and the delta to the maximum offset at the current point in the pattern.
281 typedef struct scan_data_t {
282 /*I32 len_min; unused */
283 /*I32 len_delta; unused */
287 I32 last_end; /* min value, <0 unless valid. */
290 SV **longest; /* Either &l_fixed, or &l_float. */
291 SV *longest_fixed; /* longest fixed string found in pattern */
292 I32 offset_fixed; /* offset where it starts */
293 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
294 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
295 SV *longest_float; /* longest floating string found in pattern */
296 I32 offset_float_min; /* earliest point in string it can appear */
297 I32 offset_float_max; /* latest point in string it can appear */
298 I32 *minlen_float; /* pointer to the minlen relevent to the string */
299 I32 lookbehind_float; /* is the position of the string modified by LB */
303 struct regnode_charclass_class *start_class;
307 * Forward declarations for pregcomp()'s friends.
310 static const scan_data_t zero_scan_data =
311 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
313 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
314 #define SF_BEFORE_SEOL 0x0001
315 #define SF_BEFORE_MEOL 0x0002
316 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
317 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
320 # define SF_FIX_SHIFT_EOL (0+2)
321 # define SF_FL_SHIFT_EOL (0+4)
323 # define SF_FIX_SHIFT_EOL (+2)
324 # define SF_FL_SHIFT_EOL (+4)
327 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
328 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
330 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
331 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
332 #define SF_IS_INF 0x0040
333 #define SF_HAS_PAR 0x0080
334 #define SF_IN_PAR 0x0100
335 #define SF_HAS_EVAL 0x0200
336 #define SCF_DO_SUBSTR 0x0400
337 #define SCF_DO_STCLASS_AND 0x0800
338 #define SCF_DO_STCLASS_OR 0x1000
339 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
340 #define SCF_WHILEM_VISITED_POS 0x2000
342 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
343 #define SCF_SEEN_ACCEPT 0x8000
345 #define UTF (RExC_utf8 != 0)
346 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
347 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
349 #define OOB_UNICODE 12345678
350 #define OOB_NAMEDCLASS -1
352 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
353 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
356 /* length of regex to show in messages that don't mark a position within */
357 #define RegexLengthToShowInErrorMessages 127
360 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
361 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
362 * op/pragma/warn/regcomp.
364 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
365 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
367 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
370 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
371 * arg. Show regex, up to a maximum length. If it's too long, chop and add
374 #define _FAIL(code) STMT_START { \
375 const char *ellipses = ""; \
376 IV len = RExC_end - RExC_precomp; \
379 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
380 if (len > RegexLengthToShowInErrorMessages) { \
381 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
382 len = RegexLengthToShowInErrorMessages - 10; \
388 #define FAIL(msg) _FAIL( \
389 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
390 msg, (int)len, RExC_precomp, ellipses))
392 #define FAIL2(msg,arg) _FAIL( \
393 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
394 arg, (int)len, RExC_precomp, ellipses))
397 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
399 #define Simple_vFAIL(m) STMT_START { \
400 const IV offset = RExC_parse - RExC_precomp; \
401 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
402 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
406 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
408 #define vFAIL(m) STMT_START { \
410 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
415 * Like Simple_vFAIL(), but accepts two arguments.
417 #define Simple_vFAIL2(m,a1) STMT_START { \
418 const IV offset = RExC_parse - RExC_precomp; \
419 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
420 (int)offset, RExC_precomp, RExC_precomp + offset); \
424 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
426 #define vFAIL2(m,a1) STMT_START { \
428 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
429 Simple_vFAIL2(m, a1); \
434 * Like Simple_vFAIL(), but accepts three arguments.
436 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
437 const IV offset = RExC_parse - RExC_precomp; \
438 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
439 (int)offset, RExC_precomp, RExC_precomp + offset); \
443 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
445 #define vFAIL3(m,a1,a2) STMT_START { \
447 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
448 Simple_vFAIL3(m, a1, a2); \
452 * Like Simple_vFAIL(), but accepts four arguments.
454 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
455 const IV offset = RExC_parse - RExC_precomp; \
456 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
457 (int)offset, RExC_precomp, RExC_precomp + offset); \
460 #define vWARN(loc,m) STMT_START { \
461 const IV offset = loc - RExC_precomp; \
462 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
463 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
466 #define vWARNdep(loc,m) STMT_START { \
467 const IV offset = loc - RExC_precomp; \
468 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
469 "%s" REPORT_LOCATION, \
470 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
474 #define vWARN2(loc, m, a1) STMT_START { \
475 const IV offset = loc - RExC_precomp; \
476 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
477 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
480 #define vWARN3(loc, m, a1, a2) STMT_START { \
481 const IV offset = loc - RExC_precomp; \
482 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
483 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
486 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
487 const IV offset = loc - RExC_precomp; \
488 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
489 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
492 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
493 const IV offset = loc - RExC_precomp; \
494 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
495 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
499 /* Allow for side effects in s */
500 #define REGC(c,s) STMT_START { \
501 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
504 /* Macros for recording node offsets. 20001227 mjd@plover.com
505 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
506 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
507 * Element 0 holds the number n.
508 * Position is 1 indexed.
511 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
513 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
514 __LINE__, (int)(node), (int)(byte))); \
516 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
518 RExC_offsets[2*(node)-1] = (byte); \
523 #define Set_Node_Offset(node,byte) \
524 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
525 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
527 #define Set_Node_Length_To_R(node,len) STMT_START { \
529 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
530 __LINE__, (int)(node), (int)(len))); \
532 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
534 RExC_offsets[2*(node)] = (len); \
539 #define Set_Node_Length(node,len) \
540 Set_Node_Length_To_R((node)-RExC_emit_start, len)
541 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
542 #define Set_Node_Cur_Length(node) \
543 Set_Node_Length(node, RExC_parse - parse_start)
545 /* Get offsets and lengths */
546 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
547 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
549 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
550 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
551 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
555 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
556 #define EXPERIMENTAL_INPLACESCAN
559 #define DEBUG_STUDYDATA(str,data,depth) \
560 DEBUG_OPTIMISE_MORE_r(if(data){ \
561 PerlIO_printf(Perl_debug_log, \
562 "%*s" str "Pos:%"IVdf"/%"IVdf \
563 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
564 (int)(depth)*2, "", \
565 (IV)((data)->pos_min), \
566 (IV)((data)->pos_delta), \
567 (UV)((data)->flags), \
568 (IV)((data)->whilem_c), \
569 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
570 is_inf ? "INF " : "" \
572 if ((data)->last_found) \
573 PerlIO_printf(Perl_debug_log, \
574 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
575 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
576 SvPVX_const((data)->last_found), \
577 (IV)((data)->last_end), \
578 (IV)((data)->last_start_min), \
579 (IV)((data)->last_start_max), \
580 ((data)->longest && \
581 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
582 SvPVX_const((data)->longest_fixed), \
583 (IV)((data)->offset_fixed), \
584 ((data)->longest && \
585 (data)->longest==&((data)->longest_float)) ? "*" : "", \
586 SvPVX_const((data)->longest_float), \
587 (IV)((data)->offset_float_min), \
588 (IV)((data)->offset_float_max) \
590 PerlIO_printf(Perl_debug_log,"\n"); \
593 static void clear_re(pTHX_ void *r);
595 /* Mark that we cannot extend a found fixed substring at this point.
596 Update the longest found anchored substring and the longest found
597 floating substrings if needed. */
600 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
602 const STRLEN l = CHR_SVLEN(data->last_found);
603 const STRLEN old_l = CHR_SVLEN(*data->longest);
604 GET_RE_DEBUG_FLAGS_DECL;
606 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
607 SvSetMagicSV(*data->longest, data->last_found);
608 if (*data->longest == data->longest_fixed) {
609 data->offset_fixed = l ? data->last_start_min : data->pos_min;
610 if (data->flags & SF_BEFORE_EOL)
612 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
614 data->flags &= ~SF_FIX_BEFORE_EOL;
615 data->minlen_fixed=minlenp;
616 data->lookbehind_fixed=0;
618 else { /* *data->longest == data->longest_float */
619 data->offset_float_min = l ? data->last_start_min : data->pos_min;
620 data->offset_float_max = (l
621 ? data->last_start_max
622 : data->pos_min + data->pos_delta);
623 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
624 data->offset_float_max = I32_MAX;
625 if (data->flags & SF_BEFORE_EOL)
627 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
629 data->flags &= ~SF_FL_BEFORE_EOL;
630 data->minlen_float=minlenp;
631 data->lookbehind_float=0;
634 SvCUR_set(data->last_found, 0);
636 SV * const sv = data->last_found;
637 if (SvUTF8(sv) && SvMAGICAL(sv)) {
638 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
644 data->flags &= ~SF_BEFORE_EOL;
645 DEBUG_STUDYDATA("cl_anything: ",data,0);
648 /* Can match anything (initialization) */
650 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
652 ANYOF_CLASS_ZERO(cl);
653 ANYOF_BITMAP_SETALL(cl);
654 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
656 cl->flags |= ANYOF_LOCALE;
659 /* Can match anything (initialization) */
661 S_cl_is_anything(const struct regnode_charclass_class *cl)
665 for (value = 0; value <= ANYOF_MAX; value += 2)
666 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
668 if (!(cl->flags & ANYOF_UNICODE_ALL))
670 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
675 /* Can match anything (initialization) */
677 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
679 Zero(cl, 1, struct regnode_charclass_class);
681 cl_anything(pRExC_state, cl);
685 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
687 Zero(cl, 1, struct regnode_charclass_class);
689 cl_anything(pRExC_state, cl);
691 cl->flags |= ANYOF_LOCALE;
694 /* 'And' a given class with another one. Can create false positives */
695 /* We assume that cl is not inverted */
697 S_cl_and(struct regnode_charclass_class *cl,
698 const struct regnode_charclass_class *and_with)
701 assert(and_with->type == ANYOF);
702 if (!(and_with->flags & ANYOF_CLASS)
703 && !(cl->flags & ANYOF_CLASS)
704 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
705 && !(and_with->flags & ANYOF_FOLD)
706 && !(cl->flags & ANYOF_FOLD)) {
709 if (and_with->flags & ANYOF_INVERT)
710 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
711 cl->bitmap[i] &= ~and_with->bitmap[i];
713 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
714 cl->bitmap[i] &= and_with->bitmap[i];
715 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
716 if (!(and_with->flags & ANYOF_EOS))
717 cl->flags &= ~ANYOF_EOS;
719 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
720 !(and_with->flags & ANYOF_INVERT)) {
721 cl->flags &= ~ANYOF_UNICODE_ALL;
722 cl->flags |= ANYOF_UNICODE;
723 ARG_SET(cl, ARG(and_with));
725 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
726 !(and_with->flags & ANYOF_INVERT))
727 cl->flags &= ~ANYOF_UNICODE_ALL;
728 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
729 !(and_with->flags & ANYOF_INVERT))
730 cl->flags &= ~ANYOF_UNICODE;
733 /* 'OR' a given class with another one. Can create false positives */
734 /* We assume that cl is not inverted */
736 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
738 if (or_with->flags & ANYOF_INVERT) {
740 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
741 * <= (B1 | !B2) | (CL1 | !CL2)
742 * which is wasteful if CL2 is small, but we ignore CL2:
743 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
744 * XXXX Can we handle case-fold? Unclear:
745 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
746 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
748 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
749 && !(or_with->flags & ANYOF_FOLD)
750 && !(cl->flags & ANYOF_FOLD) ) {
753 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
754 cl->bitmap[i] |= ~or_with->bitmap[i];
755 } /* XXXX: logic is complicated otherwise */
757 cl_anything(pRExC_state, cl);
760 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
761 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
762 && (!(or_with->flags & ANYOF_FOLD)
763 || (cl->flags & ANYOF_FOLD)) ) {
766 /* OR char bitmap and class bitmap separately */
767 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
768 cl->bitmap[i] |= or_with->bitmap[i];
769 if (or_with->flags & ANYOF_CLASS) {
770 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
771 cl->classflags[i] |= or_with->classflags[i];
772 cl->flags |= ANYOF_CLASS;
775 else { /* XXXX: logic is complicated, leave it along for a moment. */
776 cl_anything(pRExC_state, cl);
779 if (or_with->flags & ANYOF_EOS)
780 cl->flags |= ANYOF_EOS;
782 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
783 ARG(cl) != ARG(or_with)) {
784 cl->flags |= ANYOF_UNICODE_ALL;
785 cl->flags &= ~ANYOF_UNICODE;
787 if (or_with->flags & ANYOF_UNICODE_ALL) {
788 cl->flags |= ANYOF_UNICODE_ALL;
789 cl->flags &= ~ANYOF_UNICODE;
793 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
794 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
795 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
796 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
801 dump_trie(trie,widecharmap,revcharmap)
802 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
803 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
805 These routines dump out a trie in a somewhat readable format.
806 The _interim_ variants are used for debugging the interim
807 tables that are used to generate the final compressed
808 representation which is what dump_trie expects.
810 Part of the reason for their existance is to provide a form
811 of documentation as to how the different representations function.
816 Dumps the final compressed table form of the trie to Perl_debug_log.
817 Used for debugging make_trie().
821 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
822 AV *revcharmap, U32 depth)
825 SV *sv=sv_newmortal();
826 int colwidth= widecharmap ? 6 : 4;
827 GET_RE_DEBUG_FLAGS_DECL;
830 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
831 (int)depth * 2 + 2,"",
832 "Match","Base","Ofs" );
834 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
835 SV ** const tmp = av_fetch( revcharmap, state, 0);
837 PerlIO_printf( Perl_debug_log, "%*s",
839 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
840 PL_colors[0], PL_colors[1],
841 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
842 PERL_PV_ESCAPE_FIRSTCHAR
847 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
848 (int)depth * 2 + 2,"");
850 for( state = 0 ; state < trie->uniquecharcount ; state++ )
851 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
852 PerlIO_printf( Perl_debug_log, "\n");
854 for( state = 1 ; state < trie->statecount ; state++ ) {
855 const U32 base = trie->states[ state ].trans.base;
857 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
859 if ( trie->states[ state ].wordnum ) {
860 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
862 PerlIO_printf( Perl_debug_log, "%6s", "" );
865 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
870 while( ( base + ofs < trie->uniquecharcount ) ||
871 ( base + ofs - trie->uniquecharcount < trie->lasttrans
872 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
875 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
877 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
878 if ( ( base + ofs >= trie->uniquecharcount ) &&
879 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
880 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
882 PerlIO_printf( Perl_debug_log, "%*"UVXf,
884 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
886 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
890 PerlIO_printf( Perl_debug_log, "]");
893 PerlIO_printf( Perl_debug_log, "\n" );
897 Dumps a fully constructed but uncompressed trie in list form.
898 List tries normally only are used for construction when the number of
899 possible chars (trie->uniquecharcount) is very high.
900 Used for debugging make_trie().
903 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
904 HV *widecharmap, AV *revcharmap, U32 next_alloc,
908 SV *sv=sv_newmortal();
909 int colwidth= widecharmap ? 6 : 4;
910 GET_RE_DEBUG_FLAGS_DECL;
911 /* print out the table precompression. */
912 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
913 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
914 "------:-----+-----------------\n" );
916 for( state=1 ; state < next_alloc ; state ++ ) {
919 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
920 (int)depth * 2 + 2,"", (UV)state );
921 if ( ! trie->states[ state ].wordnum ) {
922 PerlIO_printf( Perl_debug_log, "%5s| ","");
924 PerlIO_printf( Perl_debug_log, "W%4x| ",
925 trie->states[ state ].wordnum
928 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
929 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
931 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
933 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
934 PL_colors[0], PL_colors[1],
935 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
936 PERL_PV_ESCAPE_FIRSTCHAR
938 TRIE_LIST_ITEM(state,charid).forid,
939 (UV)TRIE_LIST_ITEM(state,charid).newstate
942 PerlIO_printf(Perl_debug_log, "\n%*s| ",
943 (int)((depth * 2) + 14), "");
946 PerlIO_printf( Perl_debug_log, "\n");
951 Dumps a fully constructed but uncompressed trie in table form.
952 This is the normal DFA style state transition table, with a few
953 twists to facilitate compression later.
954 Used for debugging make_trie().
957 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
958 HV *widecharmap, AV *revcharmap, U32 next_alloc,
963 SV *sv=sv_newmortal();
964 int colwidth= widecharmap ? 6 : 4;
965 GET_RE_DEBUG_FLAGS_DECL;
968 print out the table precompression so that we can do a visual check
969 that they are identical.
972 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
974 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
975 SV ** const tmp = av_fetch( revcharmap, charid, 0);
977 PerlIO_printf( Perl_debug_log, "%*s",
979 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
980 PL_colors[0], PL_colors[1],
981 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
982 PERL_PV_ESCAPE_FIRSTCHAR
988 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
990 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
991 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
994 PerlIO_printf( Perl_debug_log, "\n" );
996 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
998 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
999 (int)depth * 2 + 2,"",
1000 (UV)TRIE_NODENUM( state ) );
1002 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1003 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1005 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1007 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1009 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1010 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1012 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1013 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1020 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1021 startbranch: the first branch in the whole branch sequence
1022 first : start branch of sequence of branch-exact nodes.
1023 May be the same as startbranch
1024 last : Thing following the last branch.
1025 May be the same as tail.
1026 tail : item following the branch sequence
1027 count : words in the sequence
1028 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1029 depth : indent depth
1031 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1033 A trie is an N'ary tree where the branches are determined by digital
1034 decomposition of the key. IE, at the root node you look up the 1st character and
1035 follow that branch repeat until you find the end of the branches. Nodes can be
1036 marked as "accepting" meaning they represent a complete word. Eg:
1040 would convert into the following structure. Numbers represent states, letters
1041 following numbers represent valid transitions on the letter from that state, if
1042 the number is in square brackets it represents an accepting state, otherwise it
1043 will be in parenthesis.
1045 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1049 (1) +-i->(6)-+-s->[7]
1051 +-s->(3)-+-h->(4)-+-e->[5]
1053 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1055 This shows that when matching against the string 'hers' we will begin at state 1
1056 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1057 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1058 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1059 single traverse. We store a mapping from accepting to state to which word was
1060 matched, and then when we have multiple possibilities we try to complete the
1061 rest of the regex in the order in which they occured in the alternation.
1063 The only prior NFA like behaviour that would be changed by the TRIE support is
1064 the silent ignoring of duplicate alternations which are of the form:
1066 / (DUPE|DUPE) X? (?{ ... }) Y /x
1068 Thus EVAL blocks follwing a trie may be called a different number of times with
1069 and without the optimisation. With the optimisations dupes will be silently
1070 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1071 the following demonstrates:
1073 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1075 which prints out 'word' three times, but
1077 'words'=~/(word|word|word)(?{ print $1 })S/
1079 which doesnt print it out at all. This is due to other optimisations kicking in.
1081 Example of what happens on a structural level:
1083 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1085 1: CURLYM[1] {1,32767}(18)
1096 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1097 and should turn into:
1099 1: CURLYM[1] {1,32767}(18)
1101 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1109 Cases where tail != last would be like /(?foo|bar)baz/:
1119 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1120 and would end up looking like:
1123 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1130 d = uvuni_to_utf8_flags(d, uv, 0);
1132 is the recommended Unicode-aware way of saying
1137 #define TRIE_STORE_REVCHAR \
1139 SV *tmp = newSVpvs(""); \
1140 if (UTF) SvUTF8_on(tmp); \
1141 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1142 av_push( revcharmap, tmp ); \
1145 #define TRIE_READ_CHAR STMT_START { \
1149 if ( foldlen > 0 ) { \
1150 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1155 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1156 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1157 foldlen -= UNISKIP( uvc ); \
1158 scan = foldbuf + UNISKIP( uvc ); \
1161 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1171 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1172 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1173 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1174 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1176 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1177 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1178 TRIE_LIST_CUR( state )++; \
1181 #define TRIE_LIST_NEW(state) STMT_START { \
1182 Newxz( trie->states[ state ].trans.list, \
1183 4, reg_trie_trans_le ); \
1184 TRIE_LIST_CUR( state ) = 1; \
1185 TRIE_LIST_LEN( state ) = 4; \
1188 #define TRIE_HANDLE_WORD(state) STMT_START { \
1189 U16 dupe= trie->states[ state ].wordnum; \
1190 regnode * const noper_next = regnext( noper ); \
1192 if (trie->wordlen) \
1193 trie->wordlen[ curword ] = wordlen; \
1195 /* store the word for dumping */ \
1197 if (OP(noper) != NOTHING) \
1198 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1200 tmp = newSVpvn( "", 0 ); \
1201 if ( UTF ) SvUTF8_on( tmp ); \
1202 av_push( trie_words, tmp ); \
1207 if ( noper_next < tail ) { \
1209 trie->jump = PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1210 trie->jump[curword] = (U16)(noper_next - convert); \
1212 jumper = noper_next; \
1214 nextbranch= regnext(cur); \
1218 /* So it's a dupe. This means we need to maintain a */\
1219 /* linked-list from the first to the next. */\
1220 /* we only allocate the nextword buffer when there */\
1221 /* a dupe, so first time we have to do the allocation */\
1222 if (!trie->nextword) \
1224 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1225 while ( trie->nextword[dupe] ) \
1226 dupe= trie->nextword[dupe]; \
1227 trie->nextword[dupe]= curword; \
1229 /* we haven't inserted this word yet. */ \
1230 trie->states[ state ].wordnum = curword; \
1235 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1236 ( ( base + charid >= ucharcount \
1237 && base + charid < ubound \
1238 && state == trie->trans[ base - ucharcount + charid ].check \
1239 && trie->trans[ base - ucharcount + charid ].next ) \
1240 ? trie->trans[ base - ucharcount + charid ].next \
1241 : ( state==1 ? special : 0 ) \
1245 #define MADE_JUMP_TRIE 2
1246 #define MADE_EXACT_TRIE 4
1249 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1252 /* first pass, loop through and scan words */
1253 reg_trie_data *trie;
1254 HV *widecharmap = NULL;
1255 AV *revcharmap = newAV();
1257 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1262 regnode *jumper = NULL;
1263 regnode *nextbranch = NULL;
1264 regnode *convert = NULL;
1265 /* we just use folder as a flag in utf8 */
1266 const U8 * const folder = ( flags == EXACTF
1268 : ( flags == EXACTFL
1275 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1276 AV *trie_words = NULL;
1277 /* along with revcharmap, this only used during construction but both are
1278 * useful during debugging so we store them in the struct when debugging.
1281 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1282 STRLEN trie_charcount=0;
1284 SV *re_trie_maxbuff;
1285 GET_RE_DEBUG_FLAGS_DECL;
1287 PERL_UNUSED_ARG(depth);
1290 trie = PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1292 trie->startstate = 1;
1293 trie->wordcount = word_count;
1294 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1295 trie->charmap = PerlMemShared_calloc( 256, sizeof(U16) );
1296 if (!(UTF && folder))
1297 trie->bitmap = PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1299 trie_words = newAV();
1302 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1303 if (!SvIOK(re_trie_maxbuff)) {
1304 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1307 PerlIO_printf( Perl_debug_log,
1308 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1309 (int)depth * 2 + 2, "",
1310 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1311 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1315 /* Find the node we are going to overwrite */
1316 if ( first == startbranch && OP( last ) != BRANCH ) {
1317 /* whole branch chain */
1320 /* branch sub-chain */
1321 convert = NEXTOPER( first );
1324 /* -- First loop and Setup --
1326 We first traverse the branches and scan each word to determine if it
1327 contains widechars, and how many unique chars there are, this is
1328 important as we have to build a table with at least as many columns as we
1331 We use an array of integers to represent the character codes 0..255
1332 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1333 native representation of the character value as the key and IV's for the
1336 *TODO* If we keep track of how many times each character is used we can
1337 remap the columns so that the table compression later on is more
1338 efficient in terms of memory by ensuring most common value is in the
1339 middle and the least common are on the outside. IMO this would be better
1340 than a most to least common mapping as theres a decent chance the most
1341 common letter will share a node with the least common, meaning the node
1342 will not be compressable. With a middle is most common approach the worst
1343 case is when we have the least common nodes twice.
1347 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1348 regnode * const noper = NEXTOPER( cur );
1349 const U8 *uc = (U8*)STRING( noper );
1350 const U8 * const e = uc + STR_LEN( noper );
1352 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1353 const U8 *scan = (U8*)NULL;
1354 U32 wordlen = 0; /* required init */
1357 if (OP(noper) == NOTHING) {
1362 TRIE_BITMAP_SET(trie,*uc);
1363 if ( folder ) TRIE_BITMAP_SET(trie,folder[ *uc ]);
1365 for ( ; uc < e ; uc += len ) {
1366 TRIE_CHARCOUNT(trie)++;
1370 if ( !trie->charmap[ uvc ] ) {
1371 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1373 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1379 widecharmap = newHV();
1381 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1384 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1386 if ( !SvTRUE( *svpp ) ) {
1387 sv_setiv( *svpp, ++trie->uniquecharcount );
1392 if( cur == first ) {
1395 } else if (chars < trie->minlen) {
1397 } else if (chars > trie->maxlen) {
1401 } /* end first pass */
1402 DEBUG_TRIE_COMPILE_r(
1403 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1404 (int)depth * 2 + 2,"",
1405 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1406 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1407 (int)trie->minlen, (int)trie->maxlen )
1409 trie->wordlen = PerlMemShared_calloc( word_count, sizeof(U32) );
1412 We now know what we are dealing with in terms of unique chars and
1413 string sizes so we can calculate how much memory a naive
1414 representation using a flat table will take. If it's over a reasonable
1415 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1416 conservative but potentially much slower representation using an array
1419 At the end we convert both representations into the same compressed
1420 form that will be used in regexec.c for matching with. The latter
1421 is a form that cannot be used to construct with but has memory
1422 properties similar to the list form and access properties similar
1423 to the table form making it both suitable for fast searches and
1424 small enough that its feasable to store for the duration of a program.
1426 See the comment in the code where the compressed table is produced
1427 inplace from the flat tabe representation for an explanation of how
1428 the compression works.
1433 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1435 Second Pass -- Array Of Lists Representation
1437 Each state will be represented by a list of charid:state records
1438 (reg_trie_trans_le) the first such element holds the CUR and LEN
1439 points of the allocated array. (See defines above).
1441 We build the initial structure using the lists, and then convert
1442 it into the compressed table form which allows faster lookups
1443 (but cant be modified once converted).
1446 STRLEN transcount = 1;
1448 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1449 "%*sCompiling trie using list compiler\n",
1450 (int)depth * 2 + 2, ""));
1452 trie->states = PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1453 sizeof(reg_trie_state) );
1457 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1459 regnode * const noper = NEXTOPER( cur );
1460 U8 *uc = (U8*)STRING( noper );
1461 const U8 * const e = uc + STR_LEN( noper );
1462 U32 state = 1; /* required init */
1463 U16 charid = 0; /* sanity init */
1464 U8 *scan = (U8*)NULL; /* sanity init */
1465 STRLEN foldlen = 0; /* required init */
1466 U32 wordlen = 0; /* required init */
1467 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1469 if (OP(noper) != NOTHING) {
1470 for ( ; uc < e ; uc += len ) {
1475 charid = trie->charmap[ uvc ];
1477 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1481 charid=(U16)SvIV( *svpp );
1484 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1491 if ( !trie->states[ state ].trans.list ) {
1492 TRIE_LIST_NEW( state );
1494 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1495 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1496 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1501 newstate = next_alloc++;
1502 TRIE_LIST_PUSH( state, charid, newstate );
1507 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1511 TRIE_HANDLE_WORD(state);
1513 } /* end second pass */
1515 /* next alloc is the NEXT state to be allocated */
1516 trie->statecount = next_alloc;
1517 trie->states = PerlMemShared_realloc( trie->states, next_alloc
1518 * sizeof(reg_trie_state) );
1520 /* and now dump it out before we compress it */
1521 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1522 revcharmap, next_alloc,
1527 = PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1534 for( state=1 ; state < next_alloc ; state ++ ) {
1538 DEBUG_TRIE_COMPILE_MORE_r(
1539 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1543 if (trie->states[state].trans.list) {
1544 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1548 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1549 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1550 if ( forid < minid ) {
1552 } else if ( forid > maxid ) {
1556 if ( transcount < tp + maxid - minid + 1) {
1559 = PerlMemShared_realloc( trie->trans,
1561 * sizeof(reg_trie_trans) );
1562 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1564 base = trie->uniquecharcount + tp - minid;
1565 if ( maxid == minid ) {
1567 for ( ; zp < tp ; zp++ ) {
1568 if ( ! trie->trans[ zp ].next ) {
1569 base = trie->uniquecharcount + zp - minid;
1570 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1571 trie->trans[ zp ].check = state;
1577 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1578 trie->trans[ tp ].check = state;
1583 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1584 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1585 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1586 trie->trans[ tid ].check = state;
1588 tp += ( maxid - minid + 1 );
1590 Safefree(trie->states[ state ].trans.list);
1593 DEBUG_TRIE_COMPILE_MORE_r(
1594 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1597 trie->states[ state ].trans.base=base;
1599 trie->lasttrans = tp + 1;
1603 Second Pass -- Flat Table Representation.
1605 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1606 We know that we will need Charcount+1 trans at most to store the data
1607 (one row per char at worst case) So we preallocate both structures
1608 assuming worst case.
1610 We then construct the trie using only the .next slots of the entry
1613 We use the .check field of the first entry of the node temporarily to
1614 make compression both faster and easier by keeping track of how many non
1615 zero fields are in the node.
1617 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1620 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1621 number representing the first entry of the node, and state as a
1622 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1623 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1624 are 2 entrys per node. eg:
1632 The table is internally in the right hand, idx form. However as we also
1633 have to deal with the states array which is indexed by nodenum we have to
1634 use TRIE_NODENUM() to convert.
1637 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1638 "%*sCompiling trie using table compiler\n",
1639 (int)depth * 2 + 2, ""));
1641 trie->trans = PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1642 * trie->uniquecharcount + 1,
1643 sizeof(reg_trie_trans) );
1644 trie->states = PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1645 sizeof(reg_trie_state) );
1646 next_alloc = trie->uniquecharcount + 1;
1649 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1651 regnode * const noper = NEXTOPER( cur );
1652 const U8 *uc = (U8*)STRING( noper );
1653 const U8 * const e = uc + STR_LEN( noper );
1655 U32 state = 1; /* required init */
1657 U16 charid = 0; /* sanity init */
1658 U32 accept_state = 0; /* sanity init */
1659 U8 *scan = (U8*)NULL; /* sanity init */
1661 STRLEN foldlen = 0; /* required init */
1662 U32 wordlen = 0; /* required init */
1663 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1665 if ( OP(noper) != NOTHING ) {
1666 for ( ; uc < e ; uc += len ) {
1671 charid = trie->charmap[ uvc ];
1673 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1674 charid = svpp ? (U16)SvIV(*svpp) : 0;
1678 if ( !trie->trans[ state + charid ].next ) {
1679 trie->trans[ state + charid ].next = next_alloc;
1680 trie->trans[ state ].check++;
1681 next_alloc += trie->uniquecharcount;
1683 state = trie->trans[ state + charid ].next;
1685 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1687 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1690 accept_state = TRIE_NODENUM( state );
1691 TRIE_HANDLE_WORD(accept_state);
1693 } /* end second pass */
1695 /* and now dump it out before we compress it */
1696 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1698 next_alloc, depth+1));
1702 * Inplace compress the table.*
1704 For sparse data sets the table constructed by the trie algorithm will
1705 be mostly 0/FAIL transitions or to put it another way mostly empty.
1706 (Note that leaf nodes will not contain any transitions.)
1708 This algorithm compresses the tables by eliminating most such
1709 transitions, at the cost of a modest bit of extra work during lookup:
1711 - Each states[] entry contains a .base field which indicates the
1712 index in the state[] array wheres its transition data is stored.
1714 - If .base is 0 there are no valid transitions from that node.
1716 - If .base is nonzero then charid is added to it to find an entry in
1719 -If trans[states[state].base+charid].check!=state then the
1720 transition is taken to be a 0/Fail transition. Thus if there are fail
1721 transitions at the front of the node then the .base offset will point
1722 somewhere inside the previous nodes data (or maybe even into a node
1723 even earlier), but the .check field determines if the transition is
1727 The following process inplace converts the table to the compressed
1728 table: We first do not compress the root node 1,and mark its all its
1729 .check pointers as 1 and set its .base pointer as 1 as well. This
1730 allows to do a DFA construction from the compressed table later, and
1731 ensures that any .base pointers we calculate later are greater than
1734 - We set 'pos' to indicate the first entry of the second node.
1736 - We then iterate over the columns of the node, finding the first and
1737 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1738 and set the .check pointers accordingly, and advance pos
1739 appropriately and repreat for the next node. Note that when we copy
1740 the next pointers we have to convert them from the original
1741 NODEIDX form to NODENUM form as the former is not valid post
1744 - If a node has no transitions used we mark its base as 0 and do not
1745 advance the pos pointer.
1747 - If a node only has one transition we use a second pointer into the
1748 structure to fill in allocated fail transitions from other states.
1749 This pointer is independent of the main pointer and scans forward
1750 looking for null transitions that are allocated to a state. When it
1751 finds one it writes the single transition into the "hole". If the
1752 pointer doesnt find one the single transition is appended as normal.
1754 - Once compressed we can Renew/realloc the structures to release the
1757 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1758 specifically Fig 3.47 and the associated pseudocode.
1762 const U32 laststate = TRIE_NODENUM( next_alloc );
1765 trie->statecount = laststate;
1767 for ( state = 1 ; state < laststate ; state++ ) {
1769 const U32 stateidx = TRIE_NODEIDX( state );
1770 const U32 o_used = trie->trans[ stateidx ].check;
1771 U32 used = trie->trans[ stateidx ].check;
1772 trie->trans[ stateidx ].check = 0;
1774 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1775 if ( flag || trie->trans[ stateidx + charid ].next ) {
1776 if ( trie->trans[ stateidx + charid ].next ) {
1778 for ( ; zp < pos ; zp++ ) {
1779 if ( ! trie->trans[ zp ].next ) {
1783 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1784 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1785 trie->trans[ zp ].check = state;
1786 if ( ++zp > pos ) pos = zp;
1793 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1795 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1796 trie->trans[ pos ].check = state;
1801 trie->lasttrans = pos + 1;
1802 trie->states = PerlMemShared_realloc( trie->states, laststate
1803 * sizeof(reg_trie_state) );
1804 DEBUG_TRIE_COMPILE_MORE_r(
1805 PerlIO_printf( Perl_debug_log,
1806 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1807 (int)depth * 2 + 2,"",
1808 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1811 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1814 } /* end table compress */
1816 DEBUG_TRIE_COMPILE_MORE_r(
1817 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1818 (int)depth * 2 + 2, "",
1819 (UV)trie->statecount,
1820 (UV)trie->lasttrans)
1822 /* resize the trans array to remove unused space */
1823 trie->trans = PerlMemShared_realloc( trie->trans, trie->lasttrans
1824 * sizeof(reg_trie_trans) );
1826 /* and now dump out the compressed format */
1827 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1829 { /* Modify the program and insert the new TRIE node*/
1830 U8 nodetype =(U8)(flags & 0xFF);
1834 regnode *optimize = NULL;
1836 U32 mjd_nodelen = 0;
1839 This means we convert either the first branch or the first Exact,
1840 depending on whether the thing following (in 'last') is a branch
1841 or not and whther first is the startbranch (ie is it a sub part of
1842 the alternation or is it the whole thing.)
1843 Assuming its a sub part we conver the EXACT otherwise we convert
1844 the whole branch sequence, including the first.
1846 /* Find the node we are going to overwrite */
1847 if ( first != startbranch || OP( last ) == BRANCH ) {
1848 /* branch sub-chain */
1849 NEXT_OFF( first ) = (U16)(last - first);
1851 mjd_offset= Node_Offset((convert));
1852 mjd_nodelen= Node_Length((convert));
1854 /* whole branch chain */
1857 const regnode *nop = NEXTOPER( convert );
1858 mjd_offset= Node_Offset((nop));
1859 mjd_nodelen= Node_Length((nop));
1864 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1865 (int)depth * 2 + 2, "",
1866 (UV)mjd_offset, (UV)mjd_nodelen)
1869 /* But first we check to see if there is a common prefix we can
1870 split out as an EXACT and put in front of the TRIE node. */
1871 trie->startstate= 1;
1872 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1874 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1878 const U32 base = trie->states[ state ].trans.base;
1880 if ( trie->states[state].wordnum )
1883 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1884 if ( ( base + ofs >= trie->uniquecharcount ) &&
1885 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1886 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1888 if ( ++count > 1 ) {
1889 SV **tmp = av_fetch( revcharmap, ofs, 0);
1890 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1891 if ( state == 1 ) break;
1893 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1895 PerlIO_printf(Perl_debug_log,
1896 "%*sNew Start State=%"UVuf" Class: [",
1897 (int)depth * 2 + 2, "",
1900 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1901 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1903 TRIE_BITMAP_SET(trie,*ch);
1905 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1907 PerlIO_printf(Perl_debug_log, (char*)ch)
1911 TRIE_BITMAP_SET(trie,*ch);
1913 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1914 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1920 SV **tmp = av_fetch( revcharmap, idx, 0);
1921 char *ch = SvPV_nolen( *tmp );
1923 SV *sv=sv_newmortal();
1924 PerlIO_printf( Perl_debug_log,
1925 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1926 (int)depth * 2 + 2, "",
1928 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1929 PL_colors[0], PL_colors[1],
1930 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1931 PERL_PV_ESCAPE_FIRSTCHAR
1936 OP( convert ) = nodetype;
1937 str=STRING(convert);
1948 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
1954 regnode *n = convert+NODE_SZ_STR(convert);
1955 NEXT_OFF(convert) = NODE_SZ_STR(convert);
1956 trie->startstate = state;
1957 trie->minlen -= (state - 1);
1958 trie->maxlen -= (state - 1);
1960 regnode *fix = convert;
1961 U32 word = trie->wordcount;
1963 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
1964 while( ++fix < n ) {
1965 Set_Node_Offset_Length(fix, 0, 0);
1968 SV ** const tmp = av_fetch( trie_words, word, 0 );
1970 if ( STR_LEN(convert) <= SvCUR(*tmp) )
1971 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
1973 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
1980 NEXT_OFF(convert) = (U16)(tail - convert);
1981 DEBUG_r(optimize= n);
1987 if ( trie->maxlen ) {
1988 NEXT_OFF( convert ) = (U16)(tail - convert);
1989 ARG_SET( convert, data_slot );
1990 /* Store the offset to the first unabsorbed branch in
1991 jump[0], which is otherwise unused by the jump logic.
1992 We use this when dumping a trie and during optimisation. */
1994 trie->jump[0] = (U16)(nextbranch - convert);
1997 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
1998 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2000 OP( convert ) = TRIEC;
2001 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2002 PerlMemShared_free(trie->bitmap);
2005 OP( convert ) = TRIE;
2007 /* store the type in the flags */
2008 convert->flags = nodetype;
2012 + regarglen[ OP( convert ) ];
2014 /* XXX We really should free up the resource in trie now,
2015 as we won't use them - (which resources?) dmq */
2017 /* needed for dumping*/
2018 DEBUG_r(if (optimize) {
2019 regnode *opt = convert;
2020 while ( ++opt < optimize) {
2021 Set_Node_Offset_Length(opt,0,0);
2024 Try to clean up some of the debris left after the
2027 while( optimize < jumper ) {
2028 mjd_nodelen += Node_Length((optimize));
2029 OP( optimize ) = OPTIMIZED;
2030 Set_Node_Offset_Length(optimize,0,0);
2033 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2035 } /* end node insert */
2036 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2038 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2039 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2041 SvREFCNT_dec(revcharmap);
2045 : trie->startstate>1
2051 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2053 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2055 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2056 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2059 We find the fail state for each state in the trie, this state is the longest proper
2060 suffix of the current states 'word' that is also a proper prefix of another word in our
2061 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2062 the DFA not to have to restart after its tried and failed a word at a given point, it
2063 simply continues as though it had been matching the other word in the first place.
2065 'abcdgu'=~/abcdefg|cdgu/
2066 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2067 fail, which would bring use to the state representing 'd' in the second word where we would
2068 try 'g' and succeed, prodceding to match 'cdgu'.
2070 /* add a fail transition */
2071 const U32 trie_offset = ARG(source);
2072 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2074 const U32 ucharcount = trie->uniquecharcount;
2075 const U32 numstates = trie->statecount;
2076 const U32 ubound = trie->lasttrans + ucharcount;
2080 U32 base = trie->states[ 1 ].trans.base;
2083 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2084 GET_RE_DEBUG_FLAGS_DECL;
2086 PERL_UNUSED_ARG(depth);
2090 ARG_SET( stclass, data_slot );
2091 aho = PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2092 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2093 aho->trie=trie_offset;
2094 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2095 Copy( trie->states, aho->states, numstates, reg_trie_state );
2096 Newxz( q, numstates, U32);
2097 aho->fail = PerlMemShared_calloc( numstates, sizeof(U32) );
2100 /* initialize fail[0..1] to be 1 so that we always have
2101 a valid final fail state */
2102 fail[ 0 ] = fail[ 1 ] = 1;
2104 for ( charid = 0; charid < ucharcount ; charid++ ) {
2105 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2107 q[ q_write ] = newstate;
2108 /* set to point at the root */
2109 fail[ q[ q_write++ ] ]=1;
2112 while ( q_read < q_write) {
2113 const U32 cur = q[ q_read++ % numstates ];
2114 base = trie->states[ cur ].trans.base;
2116 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2117 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2119 U32 fail_state = cur;
2122 fail_state = fail[ fail_state ];
2123 fail_base = aho->states[ fail_state ].trans.base;
2124 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2126 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2127 fail[ ch_state ] = fail_state;
2128 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2130 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2132 q[ q_write++ % numstates] = ch_state;
2136 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2137 when we fail in state 1, this allows us to use the
2138 charclass scan to find a valid start char. This is based on the principle
2139 that theres a good chance the string being searched contains lots of stuff
2140 that cant be a start char.
2142 fail[ 0 ] = fail[ 1 ] = 0;
2143 DEBUG_TRIE_COMPILE_r({
2144 PerlIO_printf(Perl_debug_log,
2145 "%*sStclass Failtable (%"UVuf" states): 0",
2146 (int)(depth * 2), "", (UV)numstates
2148 for( q_read=1; q_read<numstates; q_read++ ) {
2149 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2151 PerlIO_printf(Perl_debug_log, "\n");
2154 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2159 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2160 * These need to be revisited when a newer toolchain becomes available.
2162 #if defined(__sparc64__) && defined(__GNUC__)
2163 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2164 # undef SPARC64_GCC_WORKAROUND
2165 # define SPARC64_GCC_WORKAROUND 1
2169 #define DEBUG_PEEP(str,scan,depth) \
2170 DEBUG_OPTIMISE_r({if (scan){ \
2171 SV * const mysv=sv_newmortal(); \
2172 regnode *Next = regnext(scan); \
2173 regprop(RExC_rx, mysv, scan); \
2174 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2175 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2176 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2183 #define JOIN_EXACT(scan,min,flags) \
2184 if (PL_regkind[OP(scan)] == EXACT) \
2185 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2188 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2189 /* Merge several consecutive EXACTish nodes into one. */
2190 regnode *n = regnext(scan);
2192 regnode *next = scan + NODE_SZ_STR(scan);
2196 regnode *stop = scan;
2197 GET_RE_DEBUG_FLAGS_DECL;
2199 PERL_UNUSED_ARG(depth);
2201 #ifndef EXPERIMENTAL_INPLACESCAN
2202 PERL_UNUSED_ARG(flags);
2203 PERL_UNUSED_ARG(val);
2205 DEBUG_PEEP("join",scan,depth);
2207 /* Skip NOTHING, merge EXACT*. */
2209 ( PL_regkind[OP(n)] == NOTHING ||
2210 (stringok && (OP(n) == OP(scan))))
2212 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2214 if (OP(n) == TAIL || n > next)
2216 if (PL_regkind[OP(n)] == NOTHING) {
2217 DEBUG_PEEP("skip:",n,depth);
2218 NEXT_OFF(scan) += NEXT_OFF(n);
2219 next = n + NODE_STEP_REGNODE;
2226 else if (stringok) {
2227 const unsigned int oldl = STR_LEN(scan);
2228 regnode * const nnext = regnext(n);
2230 DEBUG_PEEP("merg",n,depth);
2233 if (oldl + STR_LEN(n) > U8_MAX)
2235 NEXT_OFF(scan) += NEXT_OFF(n);
2236 STR_LEN(scan) += STR_LEN(n);
2237 next = n + NODE_SZ_STR(n);
2238 /* Now we can overwrite *n : */
2239 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2247 #ifdef EXPERIMENTAL_INPLACESCAN
2248 if (flags && !NEXT_OFF(n)) {
2249 DEBUG_PEEP("atch", val, depth);
2250 if (reg_off_by_arg[OP(n)]) {
2251 ARG_SET(n, val - n);
2254 NEXT_OFF(n) = val - n;
2261 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2263 Two problematic code points in Unicode casefolding of EXACT nodes:
2265 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2266 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2272 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2273 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2275 This means that in case-insensitive matching (or "loose matching",
2276 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2277 length of the above casefolded versions) can match a target string
2278 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2279 This would rather mess up the minimum length computation.
2281 What we'll do is to look for the tail four bytes, and then peek
2282 at the preceding two bytes to see whether we need to decrease
2283 the minimum length by four (six minus two).
2285 Thanks to the design of UTF-8, there cannot be false matches:
2286 A sequence of valid UTF-8 bytes cannot be a subsequence of
2287 another valid sequence of UTF-8 bytes.
2290 char * const s0 = STRING(scan), *s, *t;
2291 char * const s1 = s0 + STR_LEN(scan) - 1;
2292 char * const s2 = s1 - 4;
2293 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2294 const char t0[] = "\xaf\x49\xaf\x42";
2296 const char t0[] = "\xcc\x88\xcc\x81";
2298 const char * const t1 = t0 + 3;
2301 s < s2 && (t = ninstr(s, s1, t0, t1));
2304 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2305 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2307 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2308 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2316 n = scan + NODE_SZ_STR(scan);
2318 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2325 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2329 /* REx optimizer. Converts nodes into quickier variants "in place".
2330 Finds fixed substrings. */
2332 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2333 to the position after last scanned or to NULL. */
2335 #define INIT_AND_WITHP \
2336 assert(!and_withp); \
2337 Newx(and_withp,1,struct regnode_charclass_class); \
2338 SAVEFREEPV(and_withp)
2340 /* this is a chain of data about sub patterns we are processing that
2341 need to be handled seperately/specially in study_chunk. Its so
2342 we can simulate recursion without losing state. */
2344 typedef struct scan_frame {
2345 regnode *last; /* last node to process in this frame */
2346 regnode *next; /* next node to process when last is reached */
2347 struct scan_frame *prev; /*previous frame*/
2348 I32 stop; /* what stopparen do we use */
2352 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2355 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2356 I32 *minlenp, I32 *deltap,
2361 struct regnode_charclass_class *and_withp,
2362 U32 flags, U32 depth)
2363 /* scanp: Start here (read-write). */
2364 /* deltap: Write maxlen-minlen here. */
2365 /* last: Stop before this one. */
2366 /* data: string data about the pattern */
2367 /* stopparen: treat close N as END */
2368 /* recursed: which subroutines have we recursed into */
2369 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2372 I32 min = 0, pars = 0, code;
2373 regnode *scan = *scanp, *next;
2375 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2376 int is_inf_internal = 0; /* The studied chunk is infinite */
2377 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2378 scan_data_t data_fake;
2379 SV *re_trie_maxbuff = NULL;
2380 regnode *first_non_open = scan;
2381 I32 stopmin = I32_MAX;
2382 scan_frame *frame = NULL;
2384 GET_RE_DEBUG_FLAGS_DECL;
2387 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2391 while (first_non_open && OP(first_non_open) == OPEN)
2392 first_non_open=regnext(first_non_open);
2397 while ( scan && OP(scan) != END && scan < last ){
2398 /* Peephole optimizer: */
2399 DEBUG_STUDYDATA("Peep:", data,depth);
2400 DEBUG_PEEP("Peep",scan,depth);
2401 JOIN_EXACT(scan,&min,0);
2403 /* Follow the next-chain of the current node and optimize
2404 away all the NOTHINGs from it. */
2405 if (OP(scan) != CURLYX) {
2406 const int max = (reg_off_by_arg[OP(scan)]
2408 /* I32 may be smaller than U16 on CRAYs! */
2409 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2410 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2414 /* Skip NOTHING and LONGJMP. */
2415 while ((n = regnext(n))
2416 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2417 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2418 && off + noff < max)
2420 if (reg_off_by_arg[OP(scan)])
2423 NEXT_OFF(scan) = off;
2428 /* The principal pseudo-switch. Cannot be a switch, since we
2429 look into several different things. */
2430 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2431 || OP(scan) == IFTHEN) {
2432 next = regnext(scan);
2434 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2436 if (OP(next) == code || code == IFTHEN) {
2437 /* NOTE - There is similar code to this block below for handling
2438 TRIE nodes on a re-study. If you change stuff here check there
2440 I32 max1 = 0, min1 = I32_MAX, num = 0;
2441 struct regnode_charclass_class accum;
2442 regnode * const startbranch=scan;
2444 if (flags & SCF_DO_SUBSTR)
2445 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2446 if (flags & SCF_DO_STCLASS)
2447 cl_init_zero(pRExC_state, &accum);
2449 while (OP(scan) == code) {
2450 I32 deltanext, minnext, f = 0, fake;
2451 struct regnode_charclass_class this_class;
2454 data_fake.flags = 0;
2456 data_fake.whilem_c = data->whilem_c;
2457 data_fake.last_closep = data->last_closep;
2460 data_fake.last_closep = &fake;
2462 data_fake.pos_delta = delta;
2463 next = regnext(scan);
2464 scan = NEXTOPER(scan);
2466 scan = NEXTOPER(scan);
2467 if (flags & SCF_DO_STCLASS) {
2468 cl_init(pRExC_state, &this_class);
2469 data_fake.start_class = &this_class;
2470 f = SCF_DO_STCLASS_AND;
2472 if (flags & SCF_WHILEM_VISITED_POS)
2473 f |= SCF_WHILEM_VISITED_POS;
2475 /* we suppose the run is continuous, last=next...*/
2476 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2478 stopparen, recursed, NULL, f,depth+1);
2481 if (max1 < minnext + deltanext)
2482 max1 = minnext + deltanext;
2483 if (deltanext == I32_MAX)
2484 is_inf = is_inf_internal = 1;
2486 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2488 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2489 if ( stopmin > minnext)
2490 stopmin = min + min1;
2491 flags &= ~SCF_DO_SUBSTR;
2493 data->flags |= SCF_SEEN_ACCEPT;
2496 if (data_fake.flags & SF_HAS_EVAL)
2497 data->flags |= SF_HAS_EVAL;
2498 data->whilem_c = data_fake.whilem_c;
2500 if (flags & SCF_DO_STCLASS)
2501 cl_or(pRExC_state, &accum, &this_class);
2503 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2505 if (flags & SCF_DO_SUBSTR) {
2506 data->pos_min += min1;
2507 data->pos_delta += max1 - min1;
2508 if (max1 != min1 || is_inf)
2509 data->longest = &(data->longest_float);
2512 delta += max1 - min1;
2513 if (flags & SCF_DO_STCLASS_OR) {
2514 cl_or(pRExC_state, data->start_class, &accum);
2516 cl_and(data->start_class, and_withp);
2517 flags &= ~SCF_DO_STCLASS;
2520 else if (flags & SCF_DO_STCLASS_AND) {
2522 cl_and(data->start_class, &accum);
2523 flags &= ~SCF_DO_STCLASS;
2526 /* Switch to OR mode: cache the old value of
2527 * data->start_class */
2529 StructCopy(data->start_class, and_withp,
2530 struct regnode_charclass_class);
2531 flags &= ~SCF_DO_STCLASS_AND;
2532 StructCopy(&accum, data->start_class,
2533 struct regnode_charclass_class);
2534 flags |= SCF_DO_STCLASS_OR;
2535 data->start_class->flags |= ANYOF_EOS;
2539 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2542 Assuming this was/is a branch we are dealing with: 'scan' now
2543 points at the item that follows the branch sequence, whatever
2544 it is. We now start at the beginning of the sequence and look
2551 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2553 If we can find such a subseqence we need to turn the first
2554 element into a trie and then add the subsequent branch exact
2555 strings to the trie.
2559 1. patterns where the whole set of branch can be converted.
2561 2. patterns where only a subset can be converted.
2563 In case 1 we can replace the whole set with a single regop
2564 for the trie. In case 2 we need to keep the start and end
2567 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2568 becomes BRANCH TRIE; BRANCH X;
2570 There is an additional case, that being where there is a
2571 common prefix, which gets split out into an EXACT like node
2572 preceding the TRIE node.
2574 If x(1..n)==tail then we can do a simple trie, if not we make
2575 a "jump" trie, such that when we match the appropriate word
2576 we "jump" to the appopriate tail node. Essentailly we turn
2577 a nested if into a case structure of sorts.
2582 if (!re_trie_maxbuff) {
2583 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2584 if (!SvIOK(re_trie_maxbuff))
2585 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2587 if ( SvIV(re_trie_maxbuff)>=0 ) {
2589 regnode *first = (regnode *)NULL;
2590 regnode *last = (regnode *)NULL;
2591 regnode *tail = scan;
2596 SV * const mysv = sv_newmortal(); /* for dumping */
2598 /* var tail is used because there may be a TAIL
2599 regop in the way. Ie, the exacts will point to the
2600 thing following the TAIL, but the last branch will
2601 point at the TAIL. So we advance tail. If we
2602 have nested (?:) we may have to move through several
2606 while ( OP( tail ) == TAIL ) {
2607 /* this is the TAIL generated by (?:) */
2608 tail = regnext( tail );
2613 regprop(RExC_rx, mysv, tail );
2614 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2615 (int)depth * 2 + 2, "",
2616 "Looking for TRIE'able sequences. Tail node is: ",
2617 SvPV_nolen_const( mysv )
2623 step through the branches, cur represents each
2624 branch, noper is the first thing to be matched
2625 as part of that branch and noper_next is the
2626 regnext() of that node. if noper is an EXACT
2627 and noper_next is the same as scan (our current
2628 position in the regex) then the EXACT branch is
2629 a possible optimization target. Once we have
2630 two or more consequetive such branches we can
2631 create a trie of the EXACT's contents and stich
2632 it in place. If the sequence represents all of
2633 the branches we eliminate the whole thing and
2634 replace it with a single TRIE. If it is a
2635 subsequence then we need to stitch it in. This
2636 means the first branch has to remain, and needs
2637 to be repointed at the item on the branch chain
2638 following the last branch optimized. This could
2639 be either a BRANCH, in which case the
2640 subsequence is internal, or it could be the
2641 item following the branch sequence in which
2642 case the subsequence is at the end.
2646 /* dont use tail as the end marker for this traverse */
2647 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2648 regnode * const noper = NEXTOPER( cur );
2649 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2650 regnode * const noper_next = regnext( noper );
2654 regprop(RExC_rx, mysv, cur);
2655 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2656 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2658 regprop(RExC_rx, mysv, noper);
2659 PerlIO_printf( Perl_debug_log, " -> %s",
2660 SvPV_nolen_const(mysv));
2663 regprop(RExC_rx, mysv, noper_next );
2664 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2665 SvPV_nolen_const(mysv));
2667 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2668 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2670 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2671 : PL_regkind[ OP( noper ) ] == EXACT )
2672 || OP(noper) == NOTHING )
2674 && noper_next == tail
2679 if ( !first || optype == NOTHING ) {
2680 if (!first) first = cur;
2681 optype = OP( noper );
2687 make_trie( pRExC_state,
2688 startbranch, first, cur, tail, count,
2691 if ( PL_regkind[ OP( noper ) ] == EXACT
2693 && noper_next == tail
2698 optype = OP( noper );
2708 regprop(RExC_rx, mysv, cur);
2709 PerlIO_printf( Perl_debug_log,
2710 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2711 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2715 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2716 #ifdef TRIE_STUDY_OPT
2717 if ( ((made == MADE_EXACT_TRIE &&
2718 startbranch == first)
2719 || ( first_non_open == first )) &&
2721 flags |= SCF_TRIE_RESTUDY;
2722 if ( startbranch == first
2725 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2735 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2736 scan = NEXTOPER(NEXTOPER(scan));
2737 } else /* single branch is optimized. */
2738 scan = NEXTOPER(scan);
2740 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2741 scan_frame *newframe = NULL;
2746 if (OP(scan) != SUSPEND) {
2747 /* set the pointer */
2748 if (OP(scan) == GOSUB) {
2750 RExC_recurse[ARG2L(scan)] = scan;
2751 start = RExC_open_parens[paren-1];
2752 end = RExC_close_parens[paren-1];
2755 start = RExC_rxi->program + 1;
2759 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2760 SAVEFREEPV(recursed);
2762 if (!PAREN_TEST(recursed,paren+1)) {
2763 PAREN_SET(recursed,paren+1);
2764 Newx(newframe,1,scan_frame);
2766 if (flags & SCF_DO_SUBSTR) {
2767 SCAN_COMMIT(pRExC_state,data,minlenp);
2768 data->longest = &(data->longest_float);
2770 is_inf = is_inf_internal = 1;
2771 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2772 cl_anything(pRExC_state, data->start_class);
2773 flags &= ~SCF_DO_STCLASS;
2776 Newx(newframe,1,scan_frame);
2779 end = regnext(scan);
2784 SAVEFREEPV(newframe);
2785 newframe->next = regnext(scan);
2786 newframe->last = last;
2787 newframe->stop = stopparen;
2788 newframe->prev = frame;
2798 else if (OP(scan) == EXACT) {
2799 I32 l = STR_LEN(scan);
2802 const U8 * const s = (U8*)STRING(scan);
2803 l = utf8_length(s, s + l);
2804 uc = utf8_to_uvchr(s, NULL);
2806 uc = *((U8*)STRING(scan));
2809 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2810 /* The code below prefers earlier match for fixed
2811 offset, later match for variable offset. */
2812 if (data->last_end == -1) { /* Update the start info. */
2813 data->last_start_min = data->pos_min;
2814 data->last_start_max = is_inf
2815 ? I32_MAX : data->pos_min + data->pos_delta;
2817 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2819 SvUTF8_on(data->last_found);
2821 SV * const sv = data->last_found;
2822 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2823 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2824 if (mg && mg->mg_len >= 0)
2825 mg->mg_len += utf8_length((U8*)STRING(scan),
2826 (U8*)STRING(scan)+STR_LEN(scan));
2828 data->last_end = data->pos_min + l;
2829 data->pos_min += l; /* As in the first entry. */
2830 data->flags &= ~SF_BEFORE_EOL;
2832 if (flags & SCF_DO_STCLASS_AND) {
2833 /* Check whether it is compatible with what we know already! */
2837 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2838 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2839 && (!(data->start_class->flags & ANYOF_FOLD)
2840 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2843 ANYOF_CLASS_ZERO(data->start_class);
2844 ANYOF_BITMAP_ZERO(data->start_class);
2846 ANYOF_BITMAP_SET(data->start_class, uc);
2847 data->start_class->flags &= ~ANYOF_EOS;
2849 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2851 else if (flags & SCF_DO_STCLASS_OR) {
2852 /* false positive possible if the class is case-folded */
2854 ANYOF_BITMAP_SET(data->start_class, uc);
2856 data->start_class->flags |= ANYOF_UNICODE_ALL;
2857 data->start_class->flags &= ~ANYOF_EOS;
2858 cl_and(data->start_class, and_withp);
2860 flags &= ~SCF_DO_STCLASS;
2862 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2863 I32 l = STR_LEN(scan);
2864 UV uc = *((U8*)STRING(scan));
2866 /* Search for fixed substrings supports EXACT only. */
2867 if (flags & SCF_DO_SUBSTR) {
2869 SCAN_COMMIT(pRExC_state, data, minlenp);
2872 const U8 * const s = (U8 *)STRING(scan);
2873 l = utf8_length(s, s + l);
2874 uc = utf8_to_uvchr(s, NULL);
2877 if (flags & SCF_DO_SUBSTR)
2879 if (flags & SCF_DO_STCLASS_AND) {
2880 /* Check whether it is compatible with what we know already! */
2884 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2885 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2886 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2888 ANYOF_CLASS_ZERO(data->start_class);
2889 ANYOF_BITMAP_ZERO(data->start_class);
2891 ANYOF_BITMAP_SET(data->start_class, uc);
2892 data->start_class->flags &= ~ANYOF_EOS;
2893 data->start_class->flags |= ANYOF_FOLD;
2894 if (OP(scan) == EXACTFL)
2895 data->start_class->flags |= ANYOF_LOCALE;
2898 else if (flags & SCF_DO_STCLASS_OR) {
2899 if (data->start_class->flags & ANYOF_FOLD) {
2900 /* false positive possible if the class is case-folded.
2901 Assume that the locale settings are the same... */
2903 ANYOF_BITMAP_SET(data->start_class, uc);
2904 data->start_class->flags &= ~ANYOF_EOS;
2906 cl_and(data->start_class, and_withp);
2908 flags &= ~SCF_DO_STCLASS;
2910 else if (strchr((const char*)PL_varies,OP(scan))) {
2911 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2912 I32 f = flags, pos_before = 0;
2913 regnode * const oscan = scan;
2914 struct regnode_charclass_class this_class;
2915 struct regnode_charclass_class *oclass = NULL;
2916 I32 next_is_eval = 0;
2918 switch (PL_regkind[OP(scan)]) {
2919 case WHILEM: /* End of (?:...)* . */
2920 scan = NEXTOPER(scan);
2923 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2924 next = NEXTOPER(scan);
2925 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2927 maxcount = REG_INFTY;
2928 next = regnext(scan);
2929 scan = NEXTOPER(scan);
2933 if (flags & SCF_DO_SUBSTR)
2938 if (flags & SCF_DO_STCLASS) {
2940 maxcount = REG_INFTY;
2941 next = regnext(scan);
2942 scan = NEXTOPER(scan);
2945 is_inf = is_inf_internal = 1;
2946 scan = regnext(scan);
2947 if (flags & SCF_DO_SUBSTR) {
2948 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2949 data->longest = &(data->longest_float);
2951 goto optimize_curly_tail;
2953 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
2954 && (scan->flags == stopparen))
2959 mincount = ARG1(scan);
2960 maxcount = ARG2(scan);
2962 next = regnext(scan);
2963 if (OP(scan) == CURLYX) {
2964 I32 lp = (data ? *(data->last_closep) : 0);
2965 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
2967 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
2968 next_is_eval = (OP(scan) == EVAL);
2970 if (flags & SCF_DO_SUBSTR) {
2971 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
2972 pos_before = data->pos_min;
2976 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
2978 data->flags |= SF_IS_INF;
2980 if (flags & SCF_DO_STCLASS) {
2981 cl_init(pRExC_state, &this_class);
2982 oclass = data->start_class;
2983 data->start_class = &this_class;
2984 f |= SCF_DO_STCLASS_AND;
2985 f &= ~SCF_DO_STCLASS_OR;
2987 /* These are the cases when once a subexpression
2988 fails at a particular position, it cannot succeed
2989 even after backtracking at the enclosing scope.
2991 XXXX what if minimal match and we are at the
2992 initial run of {n,m}? */
2993 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
2994 f &= ~SCF_WHILEM_VISITED_POS;
2996 /* This will finish on WHILEM, setting scan, or on NULL: */
2997 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2998 last, data, stopparen, recursed, NULL,
3000 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3002 if (flags & SCF_DO_STCLASS)
3003 data->start_class = oclass;
3004 if (mincount == 0 || minnext == 0) {
3005 if (flags & SCF_DO_STCLASS_OR) {
3006 cl_or(pRExC_state, data->start_class, &this_class);
3008 else if (flags & SCF_DO_STCLASS_AND) {
3009 /* Switch to OR mode: cache the old value of
3010 * data->start_class */
3012 StructCopy(data->start_class, and_withp,
3013 struct regnode_charclass_class);
3014 flags &= ~SCF_DO_STCLASS_AND;
3015 StructCopy(&this_class, data->start_class,
3016 struct regnode_charclass_class);
3017 flags |= SCF_DO_STCLASS_OR;
3018 data->start_class->flags |= ANYOF_EOS;
3020 } else { /* Non-zero len */
3021 if (flags & SCF_DO_STCLASS_OR) {
3022 cl_or(pRExC_state, data->start_class, &this_class);
3023 cl_and(data->start_class, and_withp);
3025 else if (flags & SCF_DO_STCLASS_AND)
3026 cl_and(data->start_class, &this_class);
3027 flags &= ~SCF_DO_STCLASS;
3029 if (!scan) /* It was not CURLYX, but CURLY. */
3031 if ( /* ? quantifier ok, except for (?{ ... }) */
3032 (next_is_eval || !(mincount == 0 && maxcount == 1))
3033 && (minnext == 0) && (deltanext == 0)
3034 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3035 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3036 && ckWARN(WARN_REGEXP))
3039 "Quantifier unexpected on zero-length expression");
3042 min += minnext * mincount;
3043 is_inf_internal |= ((maxcount == REG_INFTY
3044 && (minnext + deltanext) > 0)
3045 || deltanext == I32_MAX);
3046 is_inf |= is_inf_internal;
3047 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3049 /* Try powerful optimization CURLYX => CURLYN. */
3050 if ( OP(oscan) == CURLYX && data
3051 && data->flags & SF_IN_PAR
3052 && !(data->flags & SF_HAS_EVAL)
3053 && !deltanext && minnext == 1 ) {
3054 /* Try to optimize to CURLYN. */
3055 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3056 regnode * const nxt1 = nxt;
3063 if (!strchr((const char*)PL_simple,OP(nxt))
3064 && !(PL_regkind[OP(nxt)] == EXACT
3065 && STR_LEN(nxt) == 1))
3071 if (OP(nxt) != CLOSE)
3073 if (RExC_open_parens) {
3074 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3075 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3077 /* Now we know that nxt2 is the only contents: */
3078 oscan->flags = (U8)ARG(nxt);
3080 OP(nxt1) = NOTHING; /* was OPEN. */
3083 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3084 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3085 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3086 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3087 OP(nxt + 1) = OPTIMIZED; /* was count. */
3088 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3093 /* Try optimization CURLYX => CURLYM. */
3094 if ( OP(oscan) == CURLYX && data
3095 && !(data->flags & SF_HAS_PAR)
3096 && !(data->flags & SF_HAS_EVAL)
3097 && !deltanext /* atom is fixed width */
3098 && minnext != 0 /* CURLYM can't handle zero width */
3100 /* XXXX How to optimize if data == 0? */
3101 /* Optimize to a simpler form. */
3102 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3106 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3107 && (OP(nxt2) != WHILEM))
3109 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3110 /* Need to optimize away parenths. */
3111 if (data->flags & SF_IN_PAR) {
3112 /* Set the parenth number. */
3113 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3115 if (OP(nxt) != CLOSE)
3116 FAIL("Panic opt close");
3117 oscan->flags = (U8)ARG(nxt);
3118 if (RExC_open_parens) {
3119 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3120 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3122 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3123 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3126 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3127 OP(nxt + 1) = OPTIMIZED; /* was count. */
3128 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3129 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3132 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3133 regnode *nnxt = regnext(nxt1);
3136 if (reg_off_by_arg[OP(nxt1)])
3137 ARG_SET(nxt1, nxt2 - nxt1);
3138 else if (nxt2 - nxt1 < U16_MAX)
3139 NEXT_OFF(nxt1) = nxt2 - nxt1;
3141 OP(nxt) = NOTHING; /* Cannot beautify */
3146 /* Optimize again: */
3147 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3148 NULL, stopparen, recursed, NULL, 0,depth+1);
3153 else if ((OP(oscan) == CURLYX)
3154 && (flags & SCF_WHILEM_VISITED_POS)
3155 /* See the comment on a similar expression above.
3156 However, this time it not a subexpression
3157 we care about, but the expression itself. */
3158 && (maxcount == REG_INFTY)
3159 && data && ++data->whilem_c < 16) {
3160 /* This stays as CURLYX, we can put the count/of pair. */
3161 /* Find WHILEM (as in regexec.c) */
3162 regnode *nxt = oscan + NEXT_OFF(oscan);
3164 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3166 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3167 | (RExC_whilem_seen << 4)); /* On WHILEM */
3169 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3171 if (flags & SCF_DO_SUBSTR) {
3172 SV *last_str = NULL;
3173 int counted = mincount != 0;
3175 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3176 #if defined(SPARC64_GCC_WORKAROUND)
3179 const char *s = NULL;
3182 if (pos_before >= data->last_start_min)
3185 b = data->last_start_min;
3188 s = SvPV_const(data->last_found, l);
3189 old = b - data->last_start_min;
3192 I32 b = pos_before >= data->last_start_min
3193 ? pos_before : data->last_start_min;
3195 const char * const s = SvPV_const(data->last_found, l);
3196 I32 old = b - data->last_start_min;
3200 old = utf8_hop((U8*)s, old) - (U8*)s;
3203 /* Get the added string: */
3204 last_str = newSVpvn(s + old, l);
3206 SvUTF8_on(last_str);
3207 if (deltanext == 0 && pos_before == b) {
3208 /* What was added is a constant string */
3210 SvGROW(last_str, (mincount * l) + 1);
3211 repeatcpy(SvPVX(last_str) + l,
3212 SvPVX_const(last_str), l, mincount - 1);
3213 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3214 /* Add additional parts. */
3215 SvCUR_set(data->last_found,
3216 SvCUR(data->last_found) - l);
3217 sv_catsv(data->last_found, last_str);
3219 SV * sv = data->last_found;
3221 SvUTF8(sv) && SvMAGICAL(sv) ?
3222 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3223 if (mg && mg->mg_len >= 0)
3224 mg->mg_len += CHR_SVLEN(last_str);
3226 data->last_end += l * (mincount - 1);
3229 /* start offset must point into the last copy */
3230 data->last_start_min += minnext * (mincount - 1);
3231 data->last_start_max += is_inf ? I32_MAX
3232 : (maxcount - 1) * (minnext + data->pos_delta);
3235 /* It is counted once already... */
3236 data->pos_min += minnext * (mincount - counted);
3237 data->pos_delta += - counted * deltanext +
3238 (minnext + deltanext) * maxcount - minnext * mincount;
3239 if (mincount != maxcount) {
3240 /* Cannot extend fixed substrings found inside
3242 SCAN_COMMIT(pRExC_state,data,minlenp);
3243 if (mincount && last_str) {
3244 SV * const sv = data->last_found;
3245 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3246 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3250 sv_setsv(sv, last_str);
3251 data->last_end = data->pos_min;
3252 data->last_start_min =
3253 data->pos_min - CHR_SVLEN(last_str);
3254 data->last_start_max = is_inf
3256 : data->pos_min + data->pos_delta
3257 - CHR_SVLEN(last_str);
3259 data->longest = &(data->longest_float);
3261 SvREFCNT_dec(last_str);
3263 if (data && (fl & SF_HAS_EVAL))
3264 data->flags |= SF_HAS_EVAL;
3265 optimize_curly_tail:
3266 if (OP(oscan) != CURLYX) {
3267 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3269 NEXT_OFF(oscan) += NEXT_OFF(next);
3272 default: /* REF and CLUMP only? */
3273 if (flags & SCF_DO_SUBSTR) {
3274 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3275 data->longest = &(data->longest_float);
3277 is_inf = is_inf_internal = 1;
3278 if (flags & SCF_DO_STCLASS_OR)
3279 cl_anything(pRExC_state, data->start_class);
3280 flags &= ~SCF_DO_STCLASS;
3284 else if (strchr((const char*)PL_simple,OP(scan))) {
3287 if (flags & SCF_DO_SUBSTR) {
3288 SCAN_COMMIT(pRExC_state,data,minlenp);
3292 if (flags & SCF_DO_STCLASS) {
3293 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3295 /* Some of the logic below assumes that switching
3296 locale on will only add false positives. */
3297 switch (PL_regkind[OP(scan)]) {
3301 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3302 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3303 cl_anything(pRExC_state, data->start_class);
3306 if (OP(scan) == SANY)
3308 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3309 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3310 || (data->start_class->flags & ANYOF_CLASS));
3311 cl_anything(pRExC_state, data->start_class);
3313 if (flags & SCF_DO_STCLASS_AND || !value)
3314 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3317 if (flags & SCF_DO_STCLASS_AND)
3318 cl_and(data->start_class,
3319 (struct regnode_charclass_class*)scan);
3321 cl_or(pRExC_state, data->start_class,
3322 (struct regnode_charclass_class*)scan);
3325 if (flags & SCF_DO_STCLASS_AND) {
3326 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3327 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3328 for (value = 0; value < 256; value++)
3329 if (!isALNUM(value))
3330 ANYOF_BITMAP_CLEAR(data->start_class, value);
3334 if (data->start_class->flags & ANYOF_LOCALE)
3335 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3337 for (value = 0; value < 256; value++)
3339 ANYOF_BITMAP_SET(data->start_class, value);
3344 if (flags & SCF_DO_STCLASS_AND) {
3345 if (data->start_class->flags & ANYOF_LOCALE)
3346 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3349 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3350 data->start_class->flags |= ANYOF_LOCALE;
3354 if (flags & SCF_DO_STCLASS_AND) {
3355 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3356 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3357 for (value = 0; value < 256; value++)
3359 ANYOF_BITMAP_CLEAR(data->start_class, value);
3363 if (data->start_class->flags & ANYOF_LOCALE)
3364 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3366 for (value = 0; value < 256; value++)
3367 if (!isALNUM(value))
3368 ANYOF_BITMAP_SET(data->start_class, value);
3373 if (flags & SCF_DO_STCLASS_AND) {
3374 if (data->start_class->flags & ANYOF_LOCALE)
3375 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3378 data->start_class->flags |= ANYOF_LOCALE;
3379 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3383 if (flags & SCF_DO_STCLASS_AND) {
3384 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3385 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3386 for (value = 0; value < 256; value++)
3387 if (!isSPACE(value))
3388 ANYOF_BITMAP_CLEAR(data->start_class, value);
3392 if (data->start_class->flags & ANYOF_LOCALE)
3393 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3395 for (value = 0; value < 256; value++)
3397 ANYOF_BITMAP_SET(data->start_class, value);
3402 if (flags & SCF_DO_STCLASS_AND) {
3403 if (data->start_class->flags & ANYOF_LOCALE)
3404 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3407 data->start_class->flags |= ANYOF_LOCALE;
3408 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3412 if (flags & SCF_DO_STCLASS_AND) {
3413 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3414 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3415 for (value = 0; value < 256; value++)
3417 ANYOF_BITMAP_CLEAR(data->start_class, value);
3421 if (data->start_class->flags & ANYOF_LOCALE)
3422 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3424 for (value = 0; value < 256; value++)
3425 if (!isSPACE(value))
3426 ANYOF_BITMAP_SET(data->start_class, value);
3431 if (flags & SCF_DO_STCLASS_AND) {
3432 if (data->start_class->flags & ANYOF_LOCALE) {
3433 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3434 for (value = 0; value < 256; value++)
3435 if (!isSPACE(value))
3436 ANYOF_BITMAP_CLEAR(data->start_class, value);
3440 data->start_class->flags |= ANYOF_LOCALE;
3441 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3445 if (flags & SCF_DO_STCLASS_AND) {
3446 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3447 for (value = 0; value < 256; value++)
3448 if (!isDIGIT(value))
3449 ANYOF_BITMAP_CLEAR(data->start_class, value);
3452 if (data->start_class->flags & ANYOF_LOCALE)
3453 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3455 for (value = 0; value < 256; value++)
3457 ANYOF_BITMAP_SET(data->start_class, value);
3462 if (flags & SCF_DO_STCLASS_AND) {
3463 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3464 for (value = 0; value < 256; value++)
3466 ANYOF_BITMAP_CLEAR(data->start_class, value);
3469 if (data->start_class->flags & ANYOF_LOCALE)
3470 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3472 for (value = 0; value < 256; value++)
3473 if (!isDIGIT(value))
3474 ANYOF_BITMAP_SET(data->start_class, value);
3479 if (flags & SCF_DO_STCLASS_OR)
3480 cl_and(data->start_class, and_withp);
3481 flags &= ~SCF_DO_STCLASS;
3484 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3485 data->flags |= (OP(scan) == MEOL
3489 else if ( PL_regkind[OP(scan)] == BRANCHJ
3490 /* Lookbehind, or need to calculate parens/evals/stclass: */
3491 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3492 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3493 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3494 || OP(scan) == UNLESSM )
3496 /* Negative Lookahead/lookbehind
3497 In this case we can't do fixed string optimisation.
3500 I32 deltanext, minnext, fake = 0;
3502 struct regnode_charclass_class intrnl;
3505 data_fake.flags = 0;
3507 data_fake.whilem_c = data->whilem_c;
3508 data_fake.last_closep = data->last_closep;
3511 data_fake.last_closep = &fake;
3512 data_fake.pos_delta = delta;
3513 if ( flags & SCF_DO_STCLASS && !scan->flags
3514 && OP(scan) == IFMATCH ) { /* Lookahead */
3515 cl_init(pRExC_state, &intrnl);
3516 data_fake.start_class = &intrnl;
3517 f |= SCF_DO_STCLASS_AND;
3519 if (flags & SCF_WHILEM_VISITED_POS)
3520 f |= SCF_WHILEM_VISITED_POS;
3521 next = regnext(scan);
3522 nscan = NEXTOPER(NEXTOPER(scan));
3523 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3524 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3527 FAIL("Variable length lookbehind not implemented");
3529 else if (minnext > (I32)U8_MAX) {
3530 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3532 scan->flags = (U8)minnext;
3535 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3537 if (data_fake.flags & SF_HAS_EVAL)
3538 data->flags |= SF_HAS_EVAL;
3539 data->whilem_c = data_fake.whilem_c;
3541 if (f & SCF_DO_STCLASS_AND) {
3542 const int was = (data->start_class->flags & ANYOF_EOS);
3544 cl_and(data->start_class, &intrnl);
3546 data->start_class->flags |= ANYOF_EOS;
3549 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3551 /* Positive Lookahead/lookbehind
3552 In this case we can do fixed string optimisation,
3553 but we must be careful about it. Note in the case of
3554 lookbehind the positions will be offset by the minimum
3555 length of the pattern, something we won't know about
3556 until after the recurse.
3558 I32 deltanext, fake = 0;
3560 struct regnode_charclass_class intrnl;
3562 /* We use SAVEFREEPV so that when the full compile
3563 is finished perl will clean up the allocated
3564 minlens when its all done. This was we don't
3565 have to worry about freeing them when we know
3566 they wont be used, which would be a pain.
3569 Newx( minnextp, 1, I32 );
3570 SAVEFREEPV(minnextp);
3573 StructCopy(data, &data_fake, scan_data_t);
3574 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3577 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3578 data_fake.last_found=newSVsv(data->last_found);
3582 data_fake.last_closep = &fake;
3583 data_fake.flags = 0;
3584 data_fake.pos_delta = delta;
3586 data_fake.flags |= SF_IS_INF;
3587 if ( flags & SCF_DO_STCLASS && !scan->flags
3588 && OP(scan) == IFMATCH ) { /* Lookahead */
3589 cl_init(pRExC_state, &intrnl);
3590 data_fake.start_class = &intrnl;
3591 f |= SCF_DO_STCLASS_AND;
3593 if (flags & SCF_WHILEM_VISITED_POS)
3594 f |= SCF_WHILEM_VISITED_POS;
3595 next = regnext(scan);
3596 nscan = NEXTOPER(NEXTOPER(scan));
3598 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3599 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3602 FAIL("Variable length lookbehind not implemented");
3604 else if (*minnextp > (I32)U8_MAX) {
3605 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3607 scan->flags = (U8)*minnextp;
3612 if (f & SCF_DO_STCLASS_AND) {
3613 const int was = (data->start_class->flags & ANYOF_EOS);
3615 cl_and(data->start_class, &intrnl);
3617 data->start_class->flags |= ANYOF_EOS;
3620 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3622 if (data_fake.flags & SF_HAS_EVAL)
3623 data->flags |= SF_HAS_EVAL;
3624 data->whilem_c = data_fake.whilem_c;
3625 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3626 if (RExC_rx->minlen<*minnextp)
3627 RExC_rx->minlen=*minnextp;
3628 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3629 SvREFCNT_dec(data_fake.last_found);
3631 if ( data_fake.minlen_fixed != minlenp )
3633 data->offset_fixed= data_fake.offset_fixed;
3634 data->minlen_fixed= data_fake.minlen_fixed;
3635 data->lookbehind_fixed+= scan->flags;
3637 if ( data_fake.minlen_float != minlenp )
3639 data->minlen_float= data_fake.minlen_float;
3640 data->offset_float_min=data_fake.offset_float_min;
3641 data->offset_float_max=data_fake.offset_float_max;
3642 data->lookbehind_float+= scan->flags;
3651 else if (OP(scan) == OPEN) {
3652 if (stopparen != (I32)ARG(scan))
3655 else if (OP(scan) == CLOSE) {
3656 if (stopparen == (I32)ARG(scan)) {
3659 if ((I32)ARG(scan) == is_par) {
3660 next = regnext(scan);
3662 if ( next && (OP(next) != WHILEM) && next < last)
3663 is_par = 0; /* Disable optimization */
3666 *(data->last_closep) = ARG(scan);
3668 else if (OP(scan) == EVAL) {
3670 data->flags |= SF_HAS_EVAL;
3672 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3673 if (flags & SCF_DO_SUBSTR) {
3674 SCAN_COMMIT(pRExC_state,data,minlenp);
3675 flags &= ~SCF_DO_SUBSTR;
3677 if (data && OP(scan)==ACCEPT) {
3678 data->flags |= SCF_SEEN_ACCEPT;
3683 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3685 if (flags & SCF_DO_SUBSTR) {
3686 SCAN_COMMIT(pRExC_state,data,minlenp);
3687 data->longest = &(data->longest_float);
3689 is_inf = is_inf_internal = 1;
3690 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3691 cl_anything(pRExC_state, data->start_class);
3692 flags &= ~SCF_DO_STCLASS;
3694 else if (OP(scan) == GPOS) {
3695 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3696 !(delta || is_inf || (data && data->pos_delta)))
3698 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3699 RExC_rx->extflags |= RXf_ANCH_GPOS;
3700 if (RExC_rx->gofs < (U32)min)
3701 RExC_rx->gofs = min;
3703 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3707 #ifdef TRIE_STUDY_OPT
3708 #ifdef FULL_TRIE_STUDY
3709 else if (PL_regkind[OP(scan)] == TRIE) {
3710 /* NOTE - There is similar code to this block above for handling
3711 BRANCH nodes on the initial study. If you change stuff here
3713 regnode *trie_node= scan;
3714 regnode *tail= regnext(scan);
3715 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3716 I32 max1 = 0, min1 = I32_MAX;
3717 struct regnode_charclass_class accum;
3719 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3720 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3721 if (flags & SCF_DO_STCLASS)
3722 cl_init_zero(pRExC_state, &accum);
3728 const regnode *nextbranch= NULL;
3731 for ( word=1 ; word <= trie->wordcount ; word++)
3733 I32 deltanext=0, minnext=0, f = 0, fake;
3734 struct regnode_charclass_class this_class;
3736 data_fake.flags = 0;
3738 data_fake.whilem_c = data->whilem_c;
3739 data_fake.last_closep = data->last_closep;
3742 data_fake.last_closep = &fake;
3743 data_fake.pos_delta = delta;
3744 if (flags & SCF_DO_STCLASS) {
3745 cl_init(pRExC_state, &this_class);
3746 data_fake.start_class = &this_class;
3747 f = SCF_DO_STCLASS_AND;
3749 if (flags & SCF_WHILEM_VISITED_POS)
3750 f |= SCF_WHILEM_VISITED_POS;
3752 if (trie->jump[word]) {
3754 nextbranch = trie_node + trie->jump[0];
3755 scan= trie_node + trie->jump[word];
3756 /* We go from the jump point to the branch that follows
3757 it. Note this means we need the vestigal unused branches
3758 even though they arent otherwise used.
3760 minnext = study_chunk(pRExC_state, &scan, minlenp,
3761 &deltanext, (regnode *)nextbranch, &data_fake,
3762 stopparen, recursed, NULL, f,depth+1);
3764 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3765 nextbranch= regnext((regnode*)nextbranch);
3767 if (min1 > (I32)(minnext + trie->minlen))
3768 min1 = minnext + trie->minlen;
3769 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3770 max1 = minnext + deltanext + trie->maxlen;
3771 if (deltanext == I32_MAX)
3772 is_inf = is_inf_internal = 1;
3774 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3776 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3777 if ( stopmin > min + min1)
3778 stopmin = min + min1;
3779 flags &= ~SCF_DO_SUBSTR;
3781 data->flags |= SCF_SEEN_ACCEPT;
3784 if (data_fake.flags & SF_HAS_EVAL)
3785 data->flags |= SF_HAS_EVAL;
3786 data->whilem_c = data_fake.whilem_c;
3788 if (flags & SCF_DO_STCLASS)
3789 cl_or(pRExC_state, &accum, &this_class);
3792 if (flags & SCF_DO_SUBSTR) {
3793 data->pos_min += min1;
3794 data->pos_delta += max1 - min1;
3795 if (max1 != min1 || is_inf)
3796 data->longest = &(data->longest_float);
3799 delta += max1 - min1;
3800 if (flags & SCF_DO_STCLASS_OR) {
3801 cl_or(pRExC_state, data->start_class, &accum);
3803 cl_and(data->start_class, and_withp);
3804 flags &= ~SCF_DO_STCLASS;
3807 else if (flags & SCF_DO_STCLASS_AND) {
3809 cl_and(data->start_class, &accum);
3810 flags &= ~SCF_DO_STCLASS;
3813 /* Switch to OR mode: cache the old value of
3814 * data->start_class */
3816 StructCopy(data->start_class, and_withp,
3817 struct regnode_charclass_class);
3818 flags &= ~SCF_DO_STCLASS_AND;
3819 StructCopy(&accum, data->start_class,
3820 struct regnode_charclass_class);
3821 flags |= SCF_DO_STCLASS_OR;
3822 data->start_class->flags |= ANYOF_EOS;
3829 else if (PL_regkind[OP(scan)] == TRIE) {
3830 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3833 min += trie->minlen;
3834 delta += (trie->maxlen - trie->minlen);
3835 flags &= ~SCF_DO_STCLASS; /* xxx */
3836 if (flags & SCF_DO_SUBSTR) {
3837 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3838 data->pos_min += trie->minlen;
3839 data->pos_delta += (trie->maxlen - trie->minlen);
3840 if (trie->maxlen != trie->minlen)
3841 data->longest = &(data->longest_float);
3843 if (trie->jump) /* no more substrings -- for now /grr*/
3844 flags &= ~SCF_DO_SUBSTR;
3846 #endif /* old or new */
3847 #endif /* TRIE_STUDY_OPT */
3848 /* Else: zero-length, ignore. */
3849 scan = regnext(scan);
3854 stopparen = frame->stop;
3855 frame = frame->prev;
3856 goto fake_study_recurse;
3861 DEBUG_STUDYDATA("pre-fin:",data,depth);
3864 *deltap = is_inf_internal ? I32_MAX : delta;
3865 if (flags & SCF_DO_SUBSTR && is_inf)
3866 data->pos_delta = I32_MAX - data->pos_min;
3867 if (is_par > (I32)U8_MAX)
3869 if (is_par && pars==1 && data) {
3870 data->flags |= SF_IN_PAR;
3871 data->flags &= ~SF_HAS_PAR;
3873 else if (pars && data) {
3874 data->flags |= SF_HAS_PAR;
3875 data->flags &= ~SF_IN_PAR;
3877 if (flags & SCF_DO_STCLASS_OR)
3878 cl_and(data->start_class, and_withp);
3879 if (flags & SCF_TRIE_RESTUDY)
3880 data->flags |= SCF_TRIE_RESTUDY;
3882 DEBUG_STUDYDATA("post-fin:",data,depth);
3884 return min < stopmin ? min : stopmin;
3888 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3890 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3892 Renewc(RExC_rxi->data,
3893 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3894 char, struct reg_data);
3896 Renew(RExC_rxi->data->what, count + n, U8);
3898 Newx(RExC_rxi->data->what, n, U8);
3899 RExC_rxi->data->count = count + n;
3900 Copy(s, RExC_rxi->data->what + count, n, U8);
3904 /*XXX: todo make this not included in a non debugging perl */
3905 #ifndef PERL_IN_XSUB_RE
3907 Perl_reginitcolors(pTHX)
3910 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3912 char *t = savepv(s);
3916 t = strchr(t, '\t');
3922 PL_colors[i] = t = (char *)"";
3927 PL_colors[i++] = (char *)"";
3934 #ifdef TRIE_STUDY_OPT
3935 #define CHECK_RESTUDY_GOTO \
3937 (data.flags & SCF_TRIE_RESTUDY) \
3941 #define CHECK_RESTUDY_GOTO
3945 - pregcomp - compile a regular expression into internal code
3947 * We can't allocate space until we know how big the compiled form will be,
3948 * but we can't compile it (and thus know how big it is) until we've got a
3949 * place to put the code. So we cheat: we compile it twice, once with code
3950 * generation turned off and size counting turned on, and once "for real".
3951 * This also means that we don't allocate space until we are sure that the
3952 * thing really will compile successfully, and we never have to move the
3953 * code and thus invalidate pointers into it. (Note that it has to be in
3954 * one piece because free() must be able to free it all.) [NB: not true in perl]
3956 * Beware that the optimization-preparation code in here knows about some
3957 * of the structure of the compiled regexp. [I'll say.]
3962 #ifndef PERL_IN_XSUB_RE
3963 #define RE_ENGINE_PTR &PL_core_reg_engine
3965 extern const struct regexp_engine my_reg_engine;
3966 #define RE_ENGINE_PTR &my_reg_engine
3968 /* these make a few things look better, to avoid indentation */
3969 #define BEGIN_BLOCK {
3973 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
3976 GET_RE_DEBUG_FLAGS_DECL;
3977 DEBUG_r(if (!PL_colorset) reginitcolors());
3978 #ifndef PERL_IN_XSUB_RE
3980 /* Dispatch a request to compile a regexp to correct
3982 HV * const table = GvHV(PL_hintgv);
3984 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
3985 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
3986 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
3988 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
3991 return CALLREGCOMP_ENG(eng, exp, xend, pm);
3998 register regexp_internal *ri;
4006 RExC_state_t RExC_state;
4007 RExC_state_t * const pRExC_state = &RExC_state;
4008 #ifdef TRIE_STUDY_OPT
4010 RExC_state_t copyRExC_state;
4013 FAIL("NULL regexp argument");
4015 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4019 SV *dsv= sv_newmortal();
4020 RE_PV_QUOTED_DECL(s, RExC_utf8,
4021 dsv, RExC_precomp, (xend - exp), 60);
4022 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4023 PL_colors[4],PL_colors[5],s);
4025 RExC_flags = pm->op_pmflags;
4029 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4030 RExC_seen_evals = 0;
4033 /* First pass: determine size, legality. */
4041 RExC_emit = &PL_regdummy;
4042 RExC_whilem_seen = 0;
4043 RExC_charnames = NULL;
4044 RExC_open_parens = NULL;
4045 RExC_close_parens = NULL;
4047 RExC_paren_names = NULL;
4048 RExC_recurse = NULL;
4049 RExC_recurse_count = 0;
4051 #if 0 /* REGC() is (currently) a NOP at the first pass.
4052 * Clever compilers notice this and complain. --jhi */
4053 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4055 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4056 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4057 RExC_precomp = NULL;
4061 PerlIO_printf(Perl_debug_log,
4062 "Required size %"IVdf" nodes\n"
4063 "Starting second pass (creation)\n",
4066 RExC_lastparse=NULL;
4068 /* Small enough for pointer-storage convention?
4069 If extralen==0, this means that we will not need long jumps. */
4070 if (RExC_size >= 0x10000L && RExC_extralen)
4071 RExC_size += RExC_extralen;
4074 if (RExC_whilem_seen > 15)
4075 RExC_whilem_seen = 15;
4078 /* Make room for a sentinel value at the end of the program */
4082 /* Allocate space and zero-initialize. Note, the two step process
4083 of zeroing when in debug mode, thus anything assigned has to
4084 happen after that */
4085 Newxz(r, 1, regexp);
4086 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4087 char, regexp_internal);
4088 if ( r == NULL || ri == NULL )
4089 FAIL("Regexp out of space");
4091 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4092 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4094 /* bulk initialize base fields with 0. */
4095 Zero(ri, sizeof(regexp_internal), char);
4098 /* non-zero initialization begins here */
4100 r->engine= RE_ENGINE_PTR;
4102 r->prelen = xend - exp;
4103 r->precomp = savepvn(RExC_precomp, r->prelen);
4104 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4106 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4108 if (RExC_seen & REG_SEEN_RECURSE) {
4109 Newxz(RExC_open_parens, RExC_npar,regnode *);
4110 SAVEFREEPV(RExC_open_parens);
4111 Newxz(RExC_close_parens,RExC_npar,regnode *);
4112 SAVEFREEPV(RExC_close_parens);
4115 /* Useful during FAIL. */
4116 Newxz(ri->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4118 ri->offsets[0] = RExC_size;
4120 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4121 "%s %"UVuf" bytes for offset annotations.\n",
4122 ri->offsets ? "Got" : "Couldn't get",
4123 (UV)((2*RExC_size+1) * sizeof(U32))));
4128 /* Second pass: emit code. */
4129 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4134 RExC_emit_start = ri->program;
4135 RExC_emit = ri->program;
4137 /* put a sentinal on the end of the program so we can check for
4139 ri->program[RExC_size].type = 255;
4141 /* Store the count of eval-groups for security checks: */
4142 RExC_rx->seen_evals = RExC_seen_evals;
4143 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4144 if (reg(pRExC_state, 0, &flags,1) == NULL)
4147 /* XXXX To minimize changes to RE engine we always allocate
4148 3-units-long substrs field. */
4149 Newx(r->substrs, 1, struct reg_substr_data);
4150 if (RExC_recurse_count) {
4151 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4152 SAVEFREEPV(RExC_recurse);
4156 r->minlen = minlen = sawplus = sawopen = 0;
4157 Zero(r->substrs, 1, struct reg_substr_data);
4159 #ifdef TRIE_STUDY_OPT
4162 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4164 RExC_state = copyRExC_state;
4165 if (seen & REG_TOP_LEVEL_BRANCHES)
4166 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4168 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4169 if (data.last_found) {
4170 SvREFCNT_dec(data.longest_fixed);
4171 SvREFCNT_dec(data.longest_float);
4172 SvREFCNT_dec(data.last_found);
4174 StructCopy(&zero_scan_data, &data, scan_data_t);
4176 StructCopy(&zero_scan_data, &data, scan_data_t);
4177 copyRExC_state = RExC_state;
4180 StructCopy(&zero_scan_data, &data, scan_data_t);
4183 /* Dig out information for optimizations. */
4184 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4185 pm->op_pmflags = RExC_flags;
4187 r->extflags |= RXf_UTF8; /* Unicode in it? */
4188 ri->regstclass = NULL;
4189 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4190 r->intflags |= PREGf_NAUGHTY;
4191 scan = ri->program + 1; /* First BRANCH. */
4193 /* testing for BRANCH here tells us whether there is "must appear"
4194 data in the pattern. If there is then we can use it for optimisations */
4195 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4197 STRLEN longest_float_length, longest_fixed_length;
4198 struct regnode_charclass_class ch_class; /* pointed to by data */
4200 I32 last_close = 0; /* pointed to by data */
4203 /* Skip introductions and multiplicators >= 1. */
4204 while ((OP(first) == OPEN && (sawopen = 1)) ||
4205 /* An OR of *one* alternative - should not happen now. */
4206 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4207 /* for now we can't handle lookbehind IFMATCH*/
4208 (OP(first) == IFMATCH && !first->flags) ||
4209 (OP(first) == PLUS) ||
4210 (OP(first) == MINMOD) ||
4211 /* An {n,m} with n>0 */
4212 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4215 if (OP(first) == PLUS)
4218 first += regarglen[OP(first)];
4219 if (OP(first) == IFMATCH) {
4220 first = NEXTOPER(first);
4221 first += EXTRA_STEP_2ARGS;
4222 } else /* XXX possible optimisation for /(?=)/ */
4223 first = NEXTOPER(first);
4226 /* Starting-point info. */
4228 DEBUG_PEEP("first:",first,0);
4229 /* Ignore EXACT as we deal with it later. */
4230 if (PL_regkind[OP(first)] == EXACT) {
4231 if (OP(first) == EXACT)
4232 NOOP; /* Empty, get anchored substr later. */
4233 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4234 ri->regstclass = first;
4237 else if (PL_regkind[OP(first)] == TRIE &&
4238 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4241 /* this can happen only on restudy */
4242 if ( OP(first) == TRIE ) {
4243 struct regnode_1 *trieop =
4244 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4245 StructCopy(first,trieop,struct regnode_1);
4246 trie_op=(regnode *)trieop;
4248 struct regnode_charclass *trieop =
4249 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4250 StructCopy(first,trieop,struct regnode_charclass);
4251 trie_op=(regnode *)trieop;
4254 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4255 ri->regstclass = trie_op;
4258 else if (strchr((const char*)PL_simple,OP(first)))
4259 ri->regstclass = first;
4260 else if (PL_regkind[OP(first)] == BOUND ||
4261 PL_regkind[OP(first)] == NBOUND)
4262 ri->regstclass = first;
4263 else if (PL_regkind[OP(first)] == BOL) {
4264 r->extflags |= (OP(first) == MBOL
4266 : (OP(first) == SBOL
4269 first = NEXTOPER(first);
4272 else if (OP(first) == GPOS) {
4273 r->extflags |= RXf_ANCH_GPOS;
4274 first = NEXTOPER(first);
4277 else if ((!sawopen || !RExC_sawback) &&
4278 (OP(first) == STAR &&
4279 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4280 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4282 /* turn .* into ^.* with an implied $*=1 */
4284 (OP(NEXTOPER(first)) == REG_ANY)
4287 r->extflags |= type;
4288 r->intflags |= PREGf_IMPLICIT;
4289 first = NEXTOPER(first);
4292 if (sawplus && (!sawopen || !RExC_sawback)
4293 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4294 /* x+ must match at the 1st pos of run of x's */
4295 r->intflags |= PREGf_SKIP;
4297 /* Scan is after the zeroth branch, first is atomic matcher. */
4298 #ifdef TRIE_STUDY_OPT
4301 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4302 (IV)(first - scan + 1))
4306 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4307 (IV)(first - scan + 1))
4313 * If there's something expensive in the r.e., find the
4314 * longest literal string that must appear and make it the
4315 * regmust. Resolve ties in favor of later strings, since
4316 * the regstart check works with the beginning of the r.e.
4317 * and avoiding duplication strengthens checking. Not a
4318 * strong reason, but sufficient in the absence of others.
4319 * [Now we resolve ties in favor of the earlier string if
4320 * it happens that c_offset_min has been invalidated, since the
4321 * earlier string may buy us something the later one won't.]
4324 data.longest_fixed = newSVpvs("");
4325 data.longest_float = newSVpvs("");
4326 data.last_found = newSVpvs("");
4327 data.longest = &(data.longest_fixed);
4329 if (!ri->regstclass) {
4330 cl_init(pRExC_state, &ch_class);
4331 data.start_class = &ch_class;
4332 stclass_flag = SCF_DO_STCLASS_AND;
4333 } else /* XXXX Check for BOUND? */
4335 data.last_closep = &last_close;
4337 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4338 &data, -1, NULL, NULL,
4339 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4345 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4346 && data.last_start_min == 0 && data.last_end > 0
4347 && !RExC_seen_zerolen
4348 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4349 r->extflags |= RXf_CHECK_ALL;
4350 scan_commit(pRExC_state, &data,&minlen,0);
4351 SvREFCNT_dec(data.last_found);
4353 /* Note that code very similar to this but for anchored string
4354 follows immediately below, changes may need to be made to both.
4357 longest_float_length = CHR_SVLEN(data.longest_float);
4358 if (longest_float_length
4359 || (data.flags & SF_FL_BEFORE_EOL
4360 && (!(data.flags & SF_FL_BEFORE_MEOL)
4361 || (RExC_flags & RXf_PMf_MULTILINE))))
4365 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4366 && data.offset_fixed == data.offset_float_min
4367 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4368 goto remove_float; /* As in (a)+. */
4370 /* copy the information about the longest float from the reg_scan_data
4371 over to the program. */
4372 if (SvUTF8(data.longest_float)) {
4373 r->float_utf8 = data.longest_float;
4374 r->float_substr = NULL;
4376 r->float_substr = data.longest_float;
4377 r->float_utf8 = NULL;
4379 /* float_end_shift is how many chars that must be matched that
4380 follow this item. We calculate it ahead of time as once the
4381 lookbehind offset is added in we lose the ability to correctly
4383 ml = data.minlen_float ? *(data.minlen_float)
4384 : (I32)longest_float_length;
4385 r->float_end_shift = ml - data.offset_float_min
4386 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4387 + data.lookbehind_float;
4388 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4389 r->float_max_offset = data.offset_float_max;
4390 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4391 r->float_max_offset -= data.lookbehind_float;
4393 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4394 && (!(data.flags & SF_FL_BEFORE_MEOL)
4395 || (RExC_flags & RXf_PMf_MULTILINE)));
4396 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4400 r->float_substr = r->float_utf8 = NULL;
4401 SvREFCNT_dec(data.longest_float);
4402 longest_float_length = 0;
4405 /* Note that code very similar to this but for floating string
4406 is immediately above, changes may need to be made to both.
4409 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4410 if (longest_fixed_length
4411 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4412 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4413 || (RExC_flags & RXf_PMf_MULTILINE))))
4417 /* copy the information about the longest fixed
4418 from the reg_scan_data over to the program. */
4419 if (SvUTF8(data.longest_fixed)) {
4420 r->anchored_utf8 = data.longest_fixed;
4421 r->anchored_substr = NULL;
4423 r->anchored_substr = data.longest_fixed;
4424 r->anchored_utf8 = NULL;
4426 /* fixed_end_shift is how many chars that must be matched that
4427 follow this item. We calculate it ahead of time as once the
4428 lookbehind offset is added in we lose the ability to correctly
4430 ml = data.minlen_fixed ? *(data.minlen_fixed)
4431 : (I32)longest_fixed_length;
4432 r->anchored_end_shift = ml - data.offset_fixed
4433 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4434 + data.lookbehind_fixed;
4435 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4437 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4438 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4439 || (RExC_flags & RXf_PMf_MULTILINE)));
4440 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4443 r->anchored_substr = r->anchored_utf8 = NULL;
4444 SvREFCNT_dec(data.longest_fixed);
4445 longest_fixed_length = 0;
4448 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4449 ri->regstclass = NULL;
4450 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4452 && !(data.start_class->flags & ANYOF_EOS)
4453 && !cl_is_anything(data.start_class))
4455 const U32 n = add_data(pRExC_state, 1, "f");
4457 Newx(RExC_rxi->data->data[n], 1,
4458 struct regnode_charclass_class);
4459 StructCopy(data.start_class,
4460 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4461 struct regnode_charclass_class);
4462 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4463 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4464 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4465 regprop(r, sv, (regnode*)data.start_class);
4466 PerlIO_printf(Perl_debug_log,
4467 "synthetic stclass \"%s\".\n",
4468 SvPVX_const(sv));});
4471 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4472 if (longest_fixed_length > longest_float_length) {
4473 r->check_end_shift = r->anchored_end_shift;
4474 r->check_substr = r->anchored_substr;
4475 r->check_utf8 = r->anchored_utf8;
4476 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4477 if (r->extflags & RXf_ANCH_SINGLE)
4478 r->extflags |= RXf_NOSCAN;
4481 r->check_end_shift = r->float_end_shift;
4482 r->check_substr = r->float_substr;
4483 r->check_utf8 = r->float_utf8;
4484 r->check_offset_min = r->float_min_offset;
4485 r->check_offset_max = r->float_max_offset;
4487 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4488 This should be changed ASAP! */
4489 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4490 r->extflags |= RXf_USE_INTUIT;
4491 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4492 r->extflags |= RXf_INTUIT_TAIL;
4494 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4495 if ( (STRLEN)minlen < longest_float_length )
4496 minlen= longest_float_length;
4497 if ( (STRLEN)minlen < longest_fixed_length )
4498 minlen= longest_fixed_length;
4502 /* Several toplevels. Best we can is to set minlen. */
4504 struct regnode_charclass_class ch_class;
4507 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4509 scan = ri->program + 1;
4510 cl_init(pRExC_state, &ch_class);
4511 data.start_class = &ch_class;
4512 data.last_closep = &last_close;
4515 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4516 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4520 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4521 = r->float_substr = r->float_utf8 = NULL;
4522 if (!(data.start_class->flags & ANYOF_EOS)
4523 && !cl_is_anything(data.start_class))
4525 const U32 n = add_data(pRExC_state, 1, "f");
4527 Newx(RExC_rxi->data->data[n], 1,
4528 struct regnode_charclass_class);
4529 StructCopy(data.start_class,
4530 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4531 struct regnode_charclass_class);
4532 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4533 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4534 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4535 regprop(r, sv, (regnode*)data.start_class);
4536 PerlIO_printf(Perl_debug_log,
4537 "synthetic stclass \"%s\".\n",
4538 SvPVX_const(sv));});
4542 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4543 the "real" pattern. */
4545 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4546 (IV)minlen, (IV)r->minlen);
4548 r->minlenret = minlen;
4549 if (r->minlen < minlen)
4552 if (RExC_seen & REG_SEEN_GPOS)
4553 r->extflags |= RXf_GPOS_SEEN;
4554 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4555 r->extflags |= RXf_LOOKBEHIND_SEEN;
4556 if (RExC_seen & REG_SEEN_EVAL)
4557 r->extflags |= RXf_EVAL_SEEN;
4558 if (RExC_seen & REG_SEEN_CANY)
4559 r->extflags |= RXf_CANY_SEEN;
4560 if (RExC_seen & REG_SEEN_VERBARG)
4561 r->intflags |= PREGf_VERBARG_SEEN;
4562 if (RExC_seen & REG_SEEN_CUTGROUP)
4563 r->intflags |= PREGf_CUTGROUP_SEEN;
4564 if (RExC_paren_names)
4565 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4567 r->paren_names = NULL;
4569 if (RExC_recurse_count) {
4570 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4571 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4572 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4575 Newxz(r->startp, RExC_npar, I32);
4576 Newxz(r->endp, RExC_npar, I32);
4577 /* assume we don't need to swap parens around before we match */
4580 PerlIO_printf(Perl_debug_log,"Final program:\n");
4583 DEBUG_OFFSETS_r(if (ri->offsets) {
4584 const U32 len = ri->offsets[0];
4586 GET_RE_DEBUG_FLAGS_DECL;
4587 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->offsets[0]);
4588 for (i = 1; i <= len; i++) {
4589 if (ri->offsets[i*2-1] || ri->offsets[i*2])
4590 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4591 (UV)i, (UV)ri->offsets[i*2-1], (UV)ri->offsets[i*2]);
4593 PerlIO_printf(Perl_debug_log, "\n");
4599 #undef CORE_ONLY_BLOCK
4601 #undef RE_ENGINE_PTR
4603 #ifndef PERL_IN_XSUB_RE
4605 Perl_reg_named_buff_sv(pTHX_ SV* namesv)
4607 I32 parno = 0; /* no match */
4609 const REGEXP * const rx = PM_GETRE(PL_curpm);
4610 if (rx && rx->paren_names) {
4611 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4614 SV* sv_dat=HeVAL(he_str);
4615 I32 *nums=(I32*)SvPVX(sv_dat);
4616 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4617 if ((I32)(rx->lastparen) >= nums[i] &&
4618 rx->endp[nums[i]] != -1)
4631 SV *sv= sv_newmortal();
4632 Perl_sv_setpvf(aTHX_ sv, "%"IVdf,(IV)parno);
4633 gv_paren= Perl_gv_fetchsv(aTHX_ sv, GV_ADD, SVt_PVGV);
4634 return GvSVn(gv_paren);
4639 /* Scans the name of a named buffer from the pattern.
4640 * If flags is REG_RSN_RETURN_NULL returns null.
4641 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4642 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4643 * to the parsed name as looked up in the RExC_paren_names hash.
4644 * If there is an error throws a vFAIL().. type exception.
4647 #define REG_RSN_RETURN_NULL 0
4648 #define REG_RSN_RETURN_NAME 1
4649 #define REG_RSN_RETURN_DATA 2
4652 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4653 char *name_start = RExC_parse;
4656 while( isIDFIRST_uni(utf8n_to_uvchr((U8*)RExC_parse,
4657 RExC_end - RExC_parse, &numlen, UTF8_ALLOW_DEFAULT)))
4659 RExC_parse += numlen;
4662 while( isIDFIRST(*RExC_parse) )
4666 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4667 (int)(RExC_parse - name_start)));
4670 if ( flags == REG_RSN_RETURN_NAME)
4672 else if (flags==REG_RSN_RETURN_DATA) {
4675 if ( ! sv_name ) /* should not happen*/
4676 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4677 if (RExC_paren_names)
4678 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4680 sv_dat = HeVAL(he_str);
4682 vFAIL("Reference to nonexistent named group");
4686 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4693 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4694 int rem=(int)(RExC_end - RExC_parse); \
4703 if (RExC_lastparse!=RExC_parse) \
4704 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4707 iscut ? "..." : "<" \
4710 PerlIO_printf(Perl_debug_log,"%16s",""); \
4715 num=REG_NODE_NUM(RExC_emit); \
4716 if (RExC_lastnum!=num) \
4717 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4719 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4720 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4721 (int)((depth*2)), "", \
4725 RExC_lastparse=RExC_parse; \
4730 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4731 DEBUG_PARSE_MSG((funcname)); \
4732 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4734 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4735 DEBUG_PARSE_MSG((funcname)); \
4736 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4739 - reg - regular expression, i.e. main body or parenthesized thing
4741 * Caller must absorb opening parenthesis.
4743 * Combining parenthesis handling with the base level of regular expression
4744 * is a trifle forced, but the need to tie the tails of the branches to what
4745 * follows makes it hard to avoid.
4747 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4749 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4751 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4754 /* this idea is borrowed from STR_WITH_LEN in handy.h */
4755 #define CHECK_WORD(s,v,l) \
4756 (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
4759 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4760 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4763 register regnode *ret; /* Will be the head of the group. */
4764 register regnode *br;
4765 register regnode *lastbr;
4766 register regnode *ender = NULL;
4767 register I32 parno = 0;
4769 const I32 oregflags = RExC_flags;
4770 bool have_branch = 0;
4773 /* for (?g), (?gc), and (?o) warnings; warning
4774 about (?c) will warn about (?g) -- japhy */
4776 #define WASTED_O 0x01
4777 #define WASTED_G 0x02
4778 #define WASTED_C 0x04
4779 #define WASTED_GC (0x02|0x04)
4780 I32 wastedflags = 0x00;
4782 char * parse_start = RExC_parse; /* MJD */
4783 char * const oregcomp_parse = RExC_parse;
4785 GET_RE_DEBUG_FLAGS_DECL;
4786 DEBUG_PARSE("reg ");
4789 *flagp = 0; /* Tentatively. */
4792 /* Make an OPEN node, if parenthesized. */
4794 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4795 char *start_verb = RExC_parse;
4796 STRLEN verb_len = 0;
4797 char *start_arg = NULL;
4798 unsigned char op = 0;
4800 int internal_argval = 0; /* internal_argval is only useful if !argok */
4801 while ( *RExC_parse && *RExC_parse != ')' ) {
4802 if ( *RExC_parse == ':' ) {
4803 start_arg = RExC_parse + 1;
4809 verb_len = RExC_parse - start_verb;
4812 while ( *RExC_parse && *RExC_parse != ')' )
4814 if ( *RExC_parse != ')' )
4815 vFAIL("Unterminated verb pattern argument");
4816 if ( RExC_parse == start_arg )
4819 if ( *RExC_parse != ')' )
4820 vFAIL("Unterminated verb pattern");
4823 switch ( *start_verb ) {
4824 case 'A': /* (*ACCEPT) */
4825 if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
4827 internal_argval = RExC_nestroot;
4830 case 'C': /* (*COMMIT) */
4831 if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
4834 case 'F': /* (*FAIL) */
4835 if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
4840 case ':': /* (*:NAME) */
4841 case 'M': /* (*MARK:NAME) */
4842 if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
4847 case 'P': /* (*PRUNE) */
4848 if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
4851 case 'S': /* (*SKIP) */
4852 if ( CHECK_WORD("SKIP",start_verb,verb_len) )
4855 case 'T': /* (*THEN) */
4856 /* [19:06] <TimToady> :: is then */
4857 if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
4859 RExC_seen |= REG_SEEN_CUTGROUP;
4865 vFAIL3("Unknown verb pattern '%.*s'",
4866 verb_len, start_verb);
4869 if ( start_arg && internal_argval ) {
4870 vFAIL3("Verb pattern '%.*s' may not have an argument",
4871 verb_len, start_verb);
4872 } else if ( argok < 0 && !start_arg ) {
4873 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
4874 verb_len, start_verb);
4876 ret = reganode(pRExC_state, op, internal_argval);
4877 if ( ! internal_argval && ! SIZE_ONLY ) {
4879 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
4880 ARG(ret) = add_data( pRExC_state, 1, "S" );
4881 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
4888 if (!internal_argval)
4889 RExC_seen |= REG_SEEN_VERBARG;
4890 } else if ( start_arg ) {
4891 vFAIL3("Verb pattern '%.*s' may not have an argument",
4892 verb_len, start_verb);
4894 ret = reg_node(pRExC_state, op);
4896 nextchar(pRExC_state);
4899 if (*RExC_parse == '?') { /* (?...) */
4900 U32 posflags = 0, negflags = 0;
4901 U32 *flagsp = &posflags;
4902 bool is_logical = 0;
4903 const char * const seqstart = RExC_parse;
4906 paren = *RExC_parse++;
4907 ret = NULL; /* For look-ahead/behind. */
4910 case '<': /* (?<...) */
4911 if (*RExC_parse == '!')
4913 else if (*RExC_parse != '=')
4918 case '\'': /* (?'...') */
4919 name_start= RExC_parse;
4920 svname = reg_scan_name(pRExC_state,
4921 SIZE_ONLY ? /* reverse test from the others */
4922 REG_RSN_RETURN_NAME :
4923 REG_RSN_RETURN_NULL);
4924 if (RExC_parse == name_start)
4926 if (*RExC_parse != paren)
4927 vFAIL2("Sequence (?%c... not terminated",
4928 paren=='>' ? '<' : paren);
4932 if (!svname) /* shouldnt happen */
4934 "panic: reg_scan_name returned NULL");
4935 if (!RExC_paren_names) {
4936 RExC_paren_names= newHV();
4937 sv_2mortal((SV*)RExC_paren_names);
4939 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
4941 sv_dat = HeVAL(he_str);
4943 /* croak baby croak */
4945 "panic: paren_name hash element allocation failed");
4946 } else if ( SvPOK(sv_dat) ) {
4947 IV count=SvIV(sv_dat);
4948 I32 *pv=(I32*)SvGROW(sv_dat,SvCUR(sv_dat)+sizeof(I32)+1);
4949 SvCUR_set(sv_dat,SvCUR(sv_dat)+sizeof(I32));
4950 pv[count]=RExC_npar;
4953 (void)SvUPGRADE(sv_dat,SVt_PVNV);
4954 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
4959 /*sv_dump(sv_dat);*/
4961 nextchar(pRExC_state);
4963 goto capturing_parens;
4965 RExC_seen |= REG_SEEN_LOOKBEHIND;
4967 case '=': /* (?=...) */
4968 case '!': /* (?!...) */
4969 RExC_seen_zerolen++;
4970 if (*RExC_parse == ')') {
4971 ret=reg_node(pRExC_state, OPFAIL);
4972 nextchar(pRExC_state);
4975 case ':': /* (?:...) */
4976 case '>': /* (?>...) */
4978 case '$': /* (?$...) */
4979 case '@': /* (?@...) */
4980 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
4982 case '#': /* (?#...) */
4983 while (*RExC_parse && *RExC_parse != ')')
4985 if (*RExC_parse != ')')
4986 FAIL("Sequence (?#... not terminated");
4987 nextchar(pRExC_state);
4990 case '0' : /* (?0) */
4991 case 'R' : /* (?R) */
4992 if (*RExC_parse != ')')
4993 FAIL("Sequence (?R) not terminated");
4994 ret = reg_node(pRExC_state, GOSTART);
4995 nextchar(pRExC_state);
4998 { /* named and numeric backreferences */
5001 case '&': /* (?&NAME) */
5002 parse_start = RExC_parse - 1;
5004 SV *sv_dat = reg_scan_name(pRExC_state,
5005 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5006 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5008 goto gen_recurse_regop;
5011 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5013 vFAIL("Illegal pattern");
5015 goto parse_recursion;
5017 case '-': /* (?-1) */
5018 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5019 RExC_parse--; /* rewind to let it be handled later */
5023 case '1': case '2': case '3': case '4': /* (?1) */
5024 case '5': case '6': case '7': case '8': case '9':
5027 num = atoi(RExC_parse);
5028 parse_start = RExC_parse - 1; /* MJD */
5029 if (*RExC_parse == '-')
5031 while (isDIGIT(*RExC_parse))
5033 if (*RExC_parse!=')')
5034 vFAIL("Expecting close bracket");
5037 if ( paren == '-' ) {
5039 Diagram of capture buffer numbering.
5040 Top line is the normal capture buffer numbers
5041 Botton line is the negative indexing as from
5045 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5049 num = RExC_npar + num;
5052 vFAIL("Reference to nonexistent group");
5054 } else if ( paren == '+' ) {
5055 num = RExC_npar + num - 1;
5058 ret = reganode(pRExC_state, GOSUB, num);
5060 if (num > (I32)RExC_rx->nparens) {
5062 vFAIL("Reference to nonexistent group");
5064 ARG2L_SET( ret, RExC_recurse_count++);
5066 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5067 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5071 RExC_seen |= REG_SEEN_RECURSE;
5072 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5073 Set_Node_Offset(ret, parse_start); /* MJD */
5075 nextchar(pRExC_state);
5077 } /* named and numeric backreferences */
5080 case 'p': /* (?p...) */
5081 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
5082 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
5084 case '?': /* (??...) */
5086 if (*RExC_parse != '{')
5088 paren = *RExC_parse++;
5090 case '{': /* (?{...}) */
5095 char *s = RExC_parse;
5097 RExC_seen_zerolen++;
5098 RExC_seen |= REG_SEEN_EVAL;
5099 while (count && (c = *RExC_parse)) {
5110 if (*RExC_parse != ')') {
5112 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5116 OP_4tree *sop, *rop;
5117 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5120 Perl_save_re_context(aTHX);
5121 rop = sv_compile_2op(sv, &sop, "re", &pad);
5122 sop->op_private |= OPpREFCOUNTED;
5123 /* re_dup will OpREFCNT_inc */
5124 OpREFCNT_set(sop, 1);
5127 n = add_data(pRExC_state, 3, "nop");
5128 RExC_rxi->data->data[n] = (void*)rop;
5129 RExC_rxi->data->data[n+1] = (void*)sop;
5130 RExC_rxi->data->data[n+2] = (void*)pad;
5133 else { /* First pass */
5134 if (PL_reginterp_cnt < ++RExC_seen_evals
5136 /* No compiled RE interpolated, has runtime
5137 components ===> unsafe. */
5138 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5139 if (PL_tainting && PL_tainted)
5140 FAIL("Eval-group in insecure regular expression");
5141 #if PERL_VERSION > 8
5142 if (IN_PERL_COMPILETIME)
5147 nextchar(pRExC_state);
5149 ret = reg_node(pRExC_state, LOGICAL);
5152 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5153 /* deal with the length of this later - MJD */
5156 ret = reganode(pRExC_state, EVAL, n);
5157 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5158 Set_Node_Offset(ret, parse_start);
5161 case '(': /* (?(?{...})...) and (?(?=...)...) */
5164 if (RExC_parse[0] == '?') { /* (?(?...)) */
5165 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5166 || RExC_parse[1] == '<'
5167 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5170 ret = reg_node(pRExC_state, LOGICAL);
5173 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5177 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5178 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5180 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5181 char *name_start= RExC_parse++;
5183 SV *sv_dat=reg_scan_name(pRExC_state,
5184 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5185 if (RExC_parse == name_start || *RExC_parse != ch)
5186 vFAIL2("Sequence (?(%c... not terminated",
5187 (ch == '>' ? '<' : ch));
5190 num = add_data( pRExC_state, 1, "S" );
5191 RExC_rxi->data->data[num]=(void*)sv_dat;
5192 SvREFCNT_inc(sv_dat);
5194 ret = reganode(pRExC_state,NGROUPP,num);
5195 goto insert_if_check_paren;
5197 else if (RExC_parse[0] == 'D' &&
5198 RExC_parse[1] == 'E' &&
5199 RExC_parse[2] == 'F' &&
5200 RExC_parse[3] == 'I' &&
5201 RExC_parse[4] == 'N' &&
5202 RExC_parse[5] == 'E')
5204 ret = reganode(pRExC_state,DEFINEP,0);
5207 goto insert_if_check_paren;
5209 else if (RExC_parse[0] == 'R') {
5212 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5213 parno = atoi(RExC_parse++);
5214 while (isDIGIT(*RExC_parse))
5216 } else if (RExC_parse[0] == '&') {
5219 sv_dat = reg_scan_name(pRExC_state,
5220 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5221 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5223 ret = reganode(pRExC_state,INSUBP,parno);
5224 goto insert_if_check_paren;
5226 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5229 parno = atoi(RExC_parse++);
5231 while (isDIGIT(*RExC_parse))
5233 ret = reganode(pRExC_state, GROUPP, parno);
5235 insert_if_check_paren:
5236 if ((c = *nextchar(pRExC_state)) != ')')
5237 vFAIL("Switch condition not recognized");
5239 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5240 br = regbranch(pRExC_state, &flags, 1,depth+1);
5242 br = reganode(pRExC_state, LONGJMP, 0);
5244 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5245 c = *nextchar(pRExC_state);
5250 vFAIL("(?(DEFINE)....) does not allow branches");
5251 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5252 regbranch(pRExC_state, &flags, 1,depth+1);
5253 REGTAIL(pRExC_state, ret, lastbr);
5256 c = *nextchar(pRExC_state);
5261 vFAIL("Switch (?(condition)... contains too many branches");
5262 ender = reg_node(pRExC_state, TAIL);
5263 REGTAIL(pRExC_state, br, ender);
5265 REGTAIL(pRExC_state, lastbr, ender);
5266 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5269 REGTAIL(pRExC_state, ret, ender);
5273 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5277 RExC_parse--; /* for vFAIL to print correctly */
5278 vFAIL("Sequence (? incomplete");
5282 parse_flags: /* (?i) */
5283 while (*RExC_parse && strchr("iogcmsx", *RExC_parse)) {
5284 /* (?g), (?gc) and (?o) are useless here
5285 and must be globally applied -- japhy */
5287 if (*RExC_parse == 'o' || *RExC_parse == 'g') {
5288 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5289 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5290 if (! (wastedflags & wflagbit) ) {
5291 wastedflags |= wflagbit;
5294 "Useless (%s%c) - %suse /%c modifier",
5295 flagsp == &negflags ? "?-" : "?",
5297 flagsp == &negflags ? "don't " : "",
5303 else if (*RExC_parse == 'c') {
5304 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5305 if (! (wastedflags & WASTED_C) ) {
5306 wastedflags |= WASTED_GC;
5309 "Useless (%sc) - %suse /gc modifier",
5310 flagsp == &negflags ? "?-" : "?",
5311 flagsp == &negflags ? "don't " : ""
5316 else { pmflag(flagsp, *RExC_parse); }
5320 if (*RExC_parse == '-') {
5322 wastedflags = 0; /* reset so (?g-c) warns twice */
5326 RExC_flags |= posflags;
5327 RExC_flags &= ~negflags;
5328 if (*RExC_parse == ':') {
5334 if (*RExC_parse != ')') {
5336 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5338 nextchar(pRExC_state);
5348 ret = reganode(pRExC_state, OPEN, parno);
5351 RExC_nestroot = parno;
5352 if (RExC_seen & REG_SEEN_RECURSE) {
5353 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5354 "Setting open paren #%"IVdf" to %d\n",
5355 (IV)parno, REG_NODE_NUM(ret)));
5356 RExC_open_parens[parno-1]= ret;
5359 Set_Node_Length(ret, 1); /* MJD */
5360 Set_Node_Offset(ret, RExC_parse); /* MJD */
5367 /* Pick up the branches, linking them together. */
5368 parse_start = RExC_parse; /* MJD */
5369 br = regbranch(pRExC_state, &flags, 1,depth+1);
5370 /* branch_len = (paren != 0); */
5374 if (*RExC_parse == '|') {
5375 if (!SIZE_ONLY && RExC_extralen) {
5376 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5379 reginsert(pRExC_state, BRANCH, br, depth+1);
5380 Set_Node_Length(br, paren != 0);
5381 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5385 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5387 else if (paren == ':') {
5388 *flagp |= flags&SIMPLE;
5390 if (is_open) { /* Starts with OPEN. */
5391 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5393 else if (paren != '?') /* Not Conditional */
5395 *flagp |= flags & (SPSTART | HASWIDTH);
5397 while (*RExC_parse == '|') {
5398 if (!SIZE_ONLY && RExC_extralen) {
5399 ender = reganode(pRExC_state, LONGJMP,0);
5400 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5403 RExC_extralen += 2; /* Account for LONGJMP. */
5404 nextchar(pRExC_state);
5405 br = regbranch(pRExC_state, &flags, 0, depth+1);
5409 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5413 *flagp |= flags&SPSTART;
5416 if (have_branch || paren != ':') {
5417 /* Make a closing node, and hook it on the end. */
5420 ender = reg_node(pRExC_state, TAIL);
5423 ender = reganode(pRExC_state, CLOSE, parno);
5424 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5425 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5426 "Setting close paren #%"IVdf" to %d\n",
5427 (IV)parno, REG_NODE_NUM(ender)));
5428 RExC_close_parens[parno-1]= ender;
5429 if (RExC_nestroot == parno)
5432 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5433 Set_Node_Length(ender,1); /* MJD */
5439 *flagp &= ~HASWIDTH;
5442 ender = reg_node(pRExC_state, SUCCEED);
5445 ender = reg_node(pRExC_state, END);
5447 assert(!RExC_opend); /* there can only be one! */
5452 REGTAIL(pRExC_state, lastbr, ender);
5454 if (have_branch && !SIZE_ONLY) {
5456 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5458 /* Hook the tails of the branches to the closing node. */
5459 for (br = ret; br; br = regnext(br)) {
5460 const U8 op = PL_regkind[OP(br)];
5462 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5464 else if (op == BRANCHJ) {
5465 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5473 static const char parens[] = "=!<,>";
5475 if (paren && (p = strchr(parens, paren))) {
5476 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5477 int flag = (p - parens) > 1;
5480 node = SUSPEND, flag = 0;
5481 reginsert(pRExC_state, node,ret, depth+1);
5482 Set_Node_Cur_Length(ret);
5483 Set_Node_Offset(ret, parse_start + 1);
5485 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5489 /* Check for proper termination. */
5491 RExC_flags = oregflags;
5492 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5493 RExC_parse = oregcomp_parse;
5494 vFAIL("Unmatched (");
5497 else if (!paren && RExC_parse < RExC_end) {
5498 if (*RExC_parse == ')') {
5500 vFAIL("Unmatched )");
5503 FAIL("Junk on end of regexp"); /* "Can't happen". */
5511 - regbranch - one alternative of an | operator
5513 * Implements the concatenation operator.
5516 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5519 register regnode *ret;
5520 register regnode *chain = NULL;
5521 register regnode *latest;
5522 I32 flags = 0, c = 0;
5523 GET_RE_DEBUG_FLAGS_DECL;
5524 DEBUG_PARSE("brnc");
5528 if (!SIZE_ONLY && RExC_extralen)
5529 ret = reganode(pRExC_state, BRANCHJ,0);
5531 ret = reg_node(pRExC_state, BRANCH);
5532 Set_Node_Length(ret, 1);
5536 if (!first && SIZE_ONLY)
5537 RExC_extralen += 1; /* BRANCHJ */
5539 *flagp = WORST; /* Tentatively. */
5542 nextchar(pRExC_state);
5543 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5545 latest = regpiece(pRExC_state, &flags,depth+1);
5546 if (latest == NULL) {
5547 if (flags & TRYAGAIN)
5551 else if (ret == NULL)
5553 *flagp |= flags&HASWIDTH;
5554 if (chain == NULL) /* First piece. */
5555 *flagp |= flags&SPSTART;
5558 REGTAIL(pRExC_state, chain, latest);
5563 if (chain == NULL) { /* Loop ran zero times. */
5564 chain = reg_node(pRExC_state, NOTHING);
5569 *flagp |= flags&SIMPLE;
5576 - regpiece - something followed by possible [*+?]
5578 * Note that the branching code sequences used for ? and the general cases
5579 * of * and + are somewhat optimized: they use the same NOTHING node as
5580 * both the endmarker for their branch list and the body of the last branch.
5581 * It might seem that this node could be dispensed with entirely, but the
5582 * endmarker role is not redundant.
5585 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5588 register regnode *ret;
5590 register char *next;
5592 const char * const origparse = RExC_parse;
5594 I32 max = REG_INFTY;
5596 const char *maxpos = NULL;
5597 GET_RE_DEBUG_FLAGS_DECL;
5598 DEBUG_PARSE("piec");
5600 ret = regatom(pRExC_state, &flags,depth+1);
5602 if (flags & TRYAGAIN)
5609 if (op == '{' && regcurly(RExC_parse)) {
5611 parse_start = RExC_parse; /* MJD */
5612 next = RExC_parse + 1;
5613 while (isDIGIT(*next) || *next == ',') {
5622 if (*next == '}') { /* got one */
5626 min = atoi(RExC_parse);
5630 maxpos = RExC_parse;
5632 if (!max && *maxpos != '0')
5633 max = REG_INFTY; /* meaning "infinity" */
5634 else if (max >= REG_INFTY)
5635 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5637 nextchar(pRExC_state);
5640 if ((flags&SIMPLE)) {
5641 RExC_naughty += 2 + RExC_naughty / 2;
5642 reginsert(pRExC_state, CURLY, ret, depth+1);
5643 Set_Node_Offset(ret, parse_start+1); /* MJD */
5644 Set_Node_Cur_Length(ret);
5647 regnode * const w = reg_node(pRExC_state, WHILEM);
5650 REGTAIL(pRExC_state, ret, w);
5651 if (!SIZE_ONLY && RExC_extralen) {
5652 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5653 reginsert(pRExC_state, NOTHING,ret, depth+1);
5654 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5656 reginsert(pRExC_state, CURLYX,ret, depth+1);
5658 Set_Node_Offset(ret, parse_start+1);
5659 Set_Node_Length(ret,
5660 op == '{' ? (RExC_parse - parse_start) : 1);
5662 if (!SIZE_ONLY && RExC_extralen)
5663 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5664 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5666 RExC_whilem_seen++, RExC_extralen += 3;
5667 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5675 if (max && max < min)
5676 vFAIL("Can't do {n,m} with n > m");
5678 ARG1_SET(ret, (U16)min);
5679 ARG2_SET(ret, (U16)max);
5691 #if 0 /* Now runtime fix should be reliable. */
5693 /* if this is reinstated, don't forget to put this back into perldiag:
5695 =item Regexp *+ operand could be empty at {#} in regex m/%s/
5697 (F) The part of the regexp subject to either the * or + quantifier
5698 could match an empty string. The {#} shows in the regular
5699 expression about where the problem was discovered.
5703 if (!(flags&HASWIDTH) && op != '?')
5704 vFAIL("Regexp *+ operand could be empty");
5707 parse_start = RExC_parse;
5708 nextchar(pRExC_state);
5710 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
5712 if (op == '*' && (flags&SIMPLE)) {
5713 reginsert(pRExC_state, STAR, ret, depth+1);
5717 else if (op == '*') {
5721 else if (op == '+' && (flags&SIMPLE)) {
5722 reginsert(pRExC_state, PLUS, ret, depth+1);
5726 else if (op == '+') {
5730 else if (op == '?') {
5735 if (!SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
5737 "%.*s matches null string many times",
5738 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
5742 if (RExC_parse < RExC_end && *RExC_parse == '?') {
5743 nextchar(pRExC_state);
5744 reginsert(pRExC_state, MINMOD, ret, depth+1);
5745 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
5747 #ifndef REG_ALLOW_MINMOD_SUSPEND
5750 if (RExC_parse < RExC_end && *RExC_parse == '+') {
5752 nextchar(pRExC_state);
5753 ender = reg_node(pRExC_state, SUCCEED);
5754 REGTAIL(pRExC_state, ret, ender);
5755 reginsert(pRExC_state, SUSPEND, ret, depth+1);
5757 ender = reg_node(pRExC_state, TAIL);
5758 REGTAIL(pRExC_state, ret, ender);
5762 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
5764 vFAIL("Nested quantifiers");
5771 /* reg_namedseq(pRExC_state,UVp)
5773 This is expected to be called by a parser routine that has
5774 recognized'\N' and needs to handle the rest. RExC_parse is
5775 expected to point at the first char following the N at the time
5778 If valuep is non-null then it is assumed that we are parsing inside
5779 of a charclass definition and the first codepoint in the resolved
5780 string is returned via *valuep and the routine will return NULL.
5781 In this mode if a multichar string is returned from the charnames
5782 handler a warning will be issued, and only the first char in the
5783 sequence will be examined. If the string returned is zero length
5784 then the value of *valuep is undefined and NON-NULL will
5785 be returned to indicate failure. (This will NOT be a valid pointer
5788 If value is null then it is assumed that we are parsing normal text
5789 and inserts a new EXACT node into the program containing the resolved
5790 string and returns a pointer to the new node. If the string is
5791 zerolength a NOTHING node is emitted.
5793 On success RExC_parse is set to the char following the endbrace.
5794 Parsing failures will generate a fatal errorvia vFAIL(...)
5796 NOTE: We cache all results from the charnames handler locally in
5797 the RExC_charnames hash (created on first use) to prevent a charnames
5798 handler from playing silly-buggers and returning a short string and
5799 then a long string for a given pattern. Since the regexp program
5800 size is calculated during an initial parse this would result
5801 in a buffer overrun so we cache to prevent the charname result from
5802 changing during the course of the parse.
5806 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
5808 char * name; /* start of the content of the name */
5809 char * endbrace; /* endbrace following the name */
5812 STRLEN len; /* this has various purposes throughout the code */
5813 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
5814 regnode *ret = NULL;
5816 if (*RExC_parse != '{') {
5817 vFAIL("Missing braces on \\N{}");
5819 name = RExC_parse+1;
5820 endbrace = strchr(RExC_parse, '}');
5823 vFAIL("Missing right brace on \\N{}");
5825 RExC_parse = endbrace + 1;
5828 /* RExC_parse points at the beginning brace,
5829 endbrace points at the last */
5830 if ( name[0]=='U' && name[1]=='+' ) {
5831 /* its a "unicode hex" notation {U+89AB} */
5832 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
5833 | PERL_SCAN_DISALLOW_PREFIX
5834 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
5836 len = (STRLEN)(endbrace - name - 2);
5837 cp = grok_hex(name + 2, &len, &fl, NULL);
5838 if ( len != (STRLEN)(endbrace - name - 2) ) {
5847 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
5849 /* fetch the charnames handler for this scope */
5850 HV * const table = GvHV(PL_hintgv);
5852 hv_fetchs(table, "charnames", FALSE) :
5854 SV *cv= cvp ? *cvp : NULL;
5857 /* create an SV with the name as argument */
5858 sv_name = newSVpvn(name, endbrace - name);
5860 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
5861 vFAIL2("Constant(\\N{%s}) unknown: "
5862 "(possibly a missing \"use charnames ...\")",
5865 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
5866 vFAIL2("Constant(\\N{%s}): "
5867 "$^H{charnames} is not defined",SvPVX(sv_name));
5872 if (!RExC_charnames) {
5873 /* make sure our cache is allocated */
5874 RExC_charnames = newHV();
5875 sv_2mortal((SV*)RExC_charnames);
5877 /* see if we have looked this one up before */
5878 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
5880 sv_str = HeVAL(he_str);
5893 count= call_sv(cv, G_SCALAR);
5895 if (count == 1) { /* XXXX is this right? dmq */
5897 SvREFCNT_inc_simple_void(sv_str);
5905 if ( !sv_str || !SvOK(sv_str) ) {
5906 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
5907 "did not return a defined value",SvPVX(sv_name));
5909 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
5914 char *p = SvPV(sv_str, len);
5917 if ( SvUTF8(sv_str) ) {
5918 *valuep = utf8_to_uvchr((U8*)p, &numlen);
5922 We have to turn on utf8 for high bit chars otherwise
5923 we get failures with
5925 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5926 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5928 This is different from what \x{} would do with the same
5929 codepoint, where the condition is > 0xFF.
5936 /* warn if we havent used the whole string? */
5938 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5940 "Ignoring excess chars from \\N{%s} in character class",
5944 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5946 "Ignoring zero length \\N{%s} in character class",
5951 SvREFCNT_dec(sv_name);
5953 SvREFCNT_dec(sv_str);
5954 return len ? NULL : (regnode *)&len;
5955 } else if(SvCUR(sv_str)) {
5960 char * parse_start = name-3; /* needed for the offsets */
5961 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
5963 ret = reg_node(pRExC_state,
5964 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
5967 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
5968 sv_utf8_upgrade(sv_str);
5969 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
5973 p = SvPV(sv_str, len);
5975 /* len is the length written, charlen is the size the char read */
5976 for ( len = 0; p < pend; p += charlen ) {
5978 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
5980 STRLEN foldlen,numlen;
5981 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
5982 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
5983 /* Emit all the Unicode characters. */
5985 for (foldbuf = tmpbuf;
5989 uvc = utf8_to_uvchr(foldbuf, &numlen);
5991 const STRLEN unilen = reguni(pRExC_state, uvc, s);
5994 /* In EBCDIC the numlen
5995 * and unilen can differ. */
5997 if (numlen >= foldlen)
6001 break; /* "Can't happen." */
6004 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6016 RExC_size += STR_SZ(len);
6019 RExC_emit += STR_SZ(len);
6021 Set_Node_Cur_Length(ret); /* MJD */
6023 nextchar(pRExC_state);
6025 ret = reg_node(pRExC_state,NOTHING);
6028 SvREFCNT_dec(sv_str);
6031 SvREFCNT_dec(sv_name);
6041 * It returns the code point in utf8 for the value in *encp.
6042 * value: a code value in the source encoding
6043 * encp: a pointer to an Encode object
6045 * If the result from Encode is not a single character,
6046 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6049 S_reg_recode(pTHX_ const char value, SV **encp)
6052 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6053 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6055 const STRLEN newlen = SvCUR(sv);
6056 UV uv = UNICODE_REPLACEMENT;
6060 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6063 if (!newlen || numlen != newlen) {
6064 uv = UNICODE_REPLACEMENT;
6073 - regatom - the lowest level
6075 * Optimization: gobbles an entire sequence of ordinary characters so that
6076 * it can turn them into a single node, which is smaller to store and
6077 * faster to run. Backslashed characters are exceptions, each becoming a
6078 * separate node; the code is simpler that way and it's not worth fixing.
6080 * [Yes, it is worth fixing, some scripts can run twice the speed.]
6081 * [It looks like its ok, as in S_study_chunk we merge adjacent EXACT nodes]
6084 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6087 register regnode *ret = NULL;
6089 char *parse_start = RExC_parse;
6090 GET_RE_DEBUG_FLAGS_DECL;
6091 DEBUG_PARSE("atom");
6092 *flagp = WORST; /* Tentatively. */
6095 switch (*RExC_parse) {
6097 RExC_seen_zerolen++;
6098 nextchar(pRExC_state);
6099 if (RExC_flags & RXf_PMf_MULTILINE)
6100 ret = reg_node(pRExC_state, MBOL);
6101 else if (RExC_flags & RXf_PMf_SINGLELINE)
6102 ret = reg_node(pRExC_state, SBOL);
6104 ret = reg_node(pRExC_state, BOL);
6105 Set_Node_Length(ret, 1); /* MJD */
6108 nextchar(pRExC_state);
6110 RExC_seen_zerolen++;
6111 if (RExC_flags & RXf_PMf_MULTILINE)
6112 ret = reg_node(pRExC_state, MEOL);
6113 else if (RExC_flags & RXf_PMf_SINGLELINE)
6114 ret = reg_node(pRExC_state, SEOL);
6116 ret = reg_node(pRExC_state, EOL);
6117 Set_Node_Length(ret, 1); /* MJD */
6120 nextchar(pRExC_state);
6121 if (RExC_flags & RXf_PMf_SINGLELINE)
6122 ret = reg_node(pRExC_state, SANY);
6124 ret = reg_node(pRExC_state, REG_ANY);
6125 *flagp |= HASWIDTH|SIMPLE;
6127 Set_Node_Length(ret, 1); /* MJD */
6131 char * const oregcomp_parse = ++RExC_parse;
6132 ret = regclass(pRExC_state,depth+1);
6133 if (*RExC_parse != ']') {
6134 RExC_parse = oregcomp_parse;
6135 vFAIL("Unmatched [");
6137 nextchar(pRExC_state);
6138 *flagp |= HASWIDTH|SIMPLE;
6139 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6143 nextchar(pRExC_state);
6144 ret = reg(pRExC_state, 1, &flags,depth+1);
6146 if (flags & TRYAGAIN) {
6147 if (RExC_parse == RExC_end) {
6148 /* Make parent create an empty node if needed. */
6156 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
6160 if (flags & TRYAGAIN) {
6164 vFAIL("Internal urp");
6165 /* Supposed to be caught earlier. */
6168 if (!regcurly(RExC_parse)) {
6177 vFAIL("Quantifier follows nothing");
6180 switch (*++RExC_parse) {
6182 RExC_seen_zerolen++;
6183 ret = reg_node(pRExC_state, SBOL);
6185 nextchar(pRExC_state);
6186 Set_Node_Length(ret, 2); /* MJD */
6189 ret = reg_node(pRExC_state, GPOS);
6190 RExC_seen |= REG_SEEN_GPOS;
6192 nextchar(pRExC_state);
6193 Set_Node_Length(ret, 2); /* MJD */
6196 ret = reg_node(pRExC_state, SEOL);
6198 RExC_seen_zerolen++; /* Do not optimize RE away */
6199 nextchar(pRExC_state);
6202 ret = reg_node(pRExC_state, EOS);
6204 RExC_seen_zerolen++; /* Do not optimize RE away */
6205 nextchar(pRExC_state);
6206 Set_Node_Length(ret, 2); /* MJD */
6209 ret = reg_node(pRExC_state, CANY);
6210 RExC_seen |= REG_SEEN_CANY;
6211 *flagp |= HASWIDTH|SIMPLE;
6212 nextchar(pRExC_state);
6213 Set_Node_Length(ret, 2); /* MJD */
6216 ret = reg_node(pRExC_state, CLUMP);
6218 nextchar(pRExC_state);
6219 Set_Node_Length(ret, 2); /* MJD */
6222 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6223 *flagp |= HASWIDTH|SIMPLE;
6224 nextchar(pRExC_state);
6225 Set_Node_Length(ret, 2); /* MJD */
6228 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6229 *flagp |= HASWIDTH|SIMPLE;
6230 nextchar(pRExC_state);
6231 Set_Node_Length(ret, 2); /* MJD */
6234 RExC_seen_zerolen++;
6235 RExC_seen |= REG_SEEN_LOOKBEHIND;
6236 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6238 nextchar(pRExC_state);
6239 Set_Node_Length(ret, 2); /* MJD */
6242 RExC_seen_zerolen++;
6243 RExC_seen |= REG_SEEN_LOOKBEHIND;
6244 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6246 nextchar(pRExC_state);
6247 Set_Node_Length(ret, 2); /* MJD */
6250 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6251 *flagp |= HASWIDTH|SIMPLE;
6252 nextchar(pRExC_state);
6253 Set_Node_Length(ret, 2); /* MJD */
6256 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6257 *flagp |= HASWIDTH|SIMPLE;
6258 nextchar(pRExC_state);
6259 Set_Node_Length(ret, 2); /* MJD */
6262 ret = reg_node(pRExC_state, DIGIT);
6263 *flagp |= HASWIDTH|SIMPLE;
6264 nextchar(pRExC_state);
6265 Set_Node_Length(ret, 2); /* MJD */
6268 ret = reg_node(pRExC_state, NDIGIT);
6269 *flagp |= HASWIDTH|SIMPLE;
6270 nextchar(pRExC_state);
6271 Set_Node_Length(ret, 2); /* MJD */
6276 char* const oldregxend = RExC_end;
6277 char* parse_start = RExC_parse - 2;
6279 if (RExC_parse[1] == '{') {
6280 /* a lovely hack--pretend we saw [\pX] instead */
6281 RExC_end = strchr(RExC_parse, '}');
6283 const U8 c = (U8)*RExC_parse;
6285 RExC_end = oldregxend;
6286 vFAIL2("Missing right brace on \\%c{}", c);
6291 RExC_end = RExC_parse + 2;
6292 if (RExC_end > oldregxend)
6293 RExC_end = oldregxend;
6297 ret = regclass(pRExC_state,depth+1);
6299 RExC_end = oldregxend;
6302 Set_Node_Offset(ret, parse_start + 2);
6303 Set_Node_Cur_Length(ret);
6304 nextchar(pRExC_state);
6305 *flagp |= HASWIDTH|SIMPLE;
6309 /* Handle \N{NAME} here and not below because it can be
6310 multicharacter. join_exact() will join them up later on.
6311 Also this makes sure that things like /\N{BLAH}+/ and
6312 \N{BLAH} being multi char Just Happen. dmq*/
6314 ret= reg_namedseq(pRExC_state, NULL);
6316 case 'k': /* Handle \k<NAME> and \k'NAME' */
6318 char ch= RExC_parse[1];
6319 if (ch != '<' && ch != '\'') {
6321 vWARN( RExC_parse + 1,
6322 "Possible broken named back reference treated as literal k");
6326 char* name_start = (RExC_parse += 2);
6328 SV *sv_dat = reg_scan_name(pRExC_state,
6329 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6330 ch= (ch == '<') ? '>' : '\'';
6332 if (RExC_parse == name_start || *RExC_parse != ch)
6333 vFAIL2("Sequence \\k%c... not terminated",
6334 (ch == '>' ? '<' : ch));
6337 ret = reganode(pRExC_state,
6338 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6344 num = add_data( pRExC_state, 1, "S" );
6346 RExC_rxi->data->data[num]=(void*)sv_dat;
6347 SvREFCNT_inc(sv_dat);
6349 /* override incorrect value set in reganode MJD */
6350 Set_Node_Offset(ret, parse_start+1);
6351 Set_Node_Cur_Length(ret); /* MJD */
6352 nextchar(pRExC_state);
6368 case '1': case '2': case '3': case '4':
6369 case '5': case '6': case '7': case '8': case '9':
6372 bool isrel=(*RExC_parse=='R');
6375 num = atoi(RExC_parse);
6377 num = RExC_npar - num;
6379 vFAIL("Reference to nonexistent or unclosed group");
6381 if (num > 9 && num >= RExC_npar)
6384 char * const parse_start = RExC_parse - 1; /* MJD */
6385 while (isDIGIT(*RExC_parse))
6389 if (num > (I32)RExC_rx->nparens)
6390 vFAIL("Reference to nonexistent group");
6393 ret = reganode(pRExC_state,
6394 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6398 /* override incorrect value set in reganode MJD */
6399 Set_Node_Offset(ret, parse_start+1);
6400 Set_Node_Cur_Length(ret); /* MJD */
6402 nextchar(pRExC_state);
6407 if (RExC_parse >= RExC_end)
6408 FAIL("Trailing \\");
6411 /* Do not generate "unrecognized" warnings here, we fall
6412 back into the quick-grab loop below */
6419 if (RExC_flags & RXf_PMf_EXTENDED) {
6420 while (RExC_parse < RExC_end && *RExC_parse != '\n')
6422 if (RExC_parse < RExC_end)
6428 register STRLEN len;
6433 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6435 parse_start = RExC_parse - 1;
6441 ret = reg_node(pRExC_state,
6442 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6444 for (len = 0, p = RExC_parse - 1;
6445 len < 127 && p < RExC_end;
6448 char * const oldp = p;
6450 if (RExC_flags & RXf_PMf_EXTENDED)
6451 p = regwhite(p, RExC_end);
6501 ender = ASCII_TO_NATIVE('\033');
6505 ender = ASCII_TO_NATIVE('\007');
6510 char* const e = strchr(p, '}');
6514 vFAIL("Missing right brace on \\x{}");
6517 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6518 | PERL_SCAN_DISALLOW_PREFIX;
6519 STRLEN numlen = e - p - 1;
6520 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6527 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6529 ender = grok_hex(p, &numlen, &flags, NULL);
6532 if (PL_encoding && ender < 0x100)
6533 goto recode_encoding;
6537 ender = UCHARAT(p++);
6538 ender = toCTRL(ender);
6540 case '0': case '1': case '2': case '3':case '4':
6541 case '5': case '6': case '7': case '8':case '9':
6543 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6546 ender = grok_oct(p, &numlen, &flags, NULL);
6553 if (PL_encoding && ender < 0x100)
6554 goto recode_encoding;
6558 SV* enc = PL_encoding;
6559 ender = reg_recode((const char)(U8)ender, &enc);
6560 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6561 vWARN(p, "Invalid escape in the specified encoding");
6567 FAIL("Trailing \\");
6570 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6571 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6572 goto normal_default;
6577 if (UTF8_IS_START(*p) && UTF) {
6579 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6580 &numlen, UTF8_ALLOW_DEFAULT);
6587 if (RExC_flags & RXf_PMf_EXTENDED)
6588 p = regwhite(p, RExC_end);
6590 /* Prime the casefolded buffer. */
6591 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6593 if (ISMULT2(p)) { /* Back off on ?+*. */
6598 /* Emit all the Unicode characters. */
6600 for (foldbuf = tmpbuf;
6602 foldlen -= numlen) {
6603 ender = utf8_to_uvchr(foldbuf, &numlen);
6605 const STRLEN unilen = reguni(pRExC_state, ender, s);
6608 /* In EBCDIC the numlen
6609 * and unilen can differ. */
6611 if (numlen >= foldlen)
6615 break; /* "Can't happen." */
6619 const STRLEN unilen = reguni(pRExC_state, ender, s);
6628 REGC((char)ender, s++);
6634 /* Emit all the Unicode characters. */
6636 for (foldbuf = tmpbuf;
6638 foldlen -= numlen) {
6639 ender = utf8_to_uvchr(foldbuf, &numlen);
6641 const STRLEN unilen = reguni(pRExC_state, ender, s);
6644 /* In EBCDIC the numlen
6645 * and unilen can differ. */
6647 if (numlen >= foldlen)
6655 const STRLEN unilen = reguni(pRExC_state, ender, s);
6664 REGC((char)ender, s++);
6668 Set_Node_Cur_Length(ret); /* MJD */
6669 nextchar(pRExC_state);
6671 /* len is STRLEN which is unsigned, need to copy to signed */
6674 vFAIL("Internal disaster");
6678 if (len == 1 && UNI_IS_INVARIANT(ender))
6682 RExC_size += STR_SZ(len);
6685 RExC_emit += STR_SZ(len);
6695 S_regwhite(char *p, const char *e)
6700 else if (*p == '#') {
6703 } while (p < e && *p != '\n');
6711 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
6712 Character classes ([:foo:]) can also be negated ([:^foo:]).
6713 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
6714 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
6715 but trigger failures because they are currently unimplemented. */
6717 #define POSIXCC_DONE(c) ((c) == ':')
6718 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
6719 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
6722 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
6725 I32 namedclass = OOB_NAMEDCLASS;
6727 if (value == '[' && RExC_parse + 1 < RExC_end &&
6728 /* I smell either [: or [= or [. -- POSIX has been here, right? */
6729 POSIXCC(UCHARAT(RExC_parse))) {
6730 const char c = UCHARAT(RExC_parse);
6731 char* const s = RExC_parse++;
6733 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
6735 if (RExC_parse == RExC_end)
6736 /* Grandfather lone [:, [=, [. */
6739 const char* const t = RExC_parse++; /* skip over the c */
6742 if (UCHARAT(RExC_parse) == ']') {
6743 const char *posixcc = s + 1;
6744 RExC_parse++; /* skip over the ending ] */
6747 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
6748 const I32 skip = t - posixcc;
6750 /* Initially switch on the length of the name. */
6753 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
6754 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
6757 /* Names all of length 5. */
6758 /* alnum alpha ascii blank cntrl digit graph lower
6759 print punct space upper */
6760 /* Offset 4 gives the best switch position. */
6761 switch (posixcc[4]) {
6763 if (memEQ(posixcc, "alph", 4)) /* alpha */
6764 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
6767 if (memEQ(posixcc, "spac", 4)) /* space */
6768 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
6771 if (memEQ(posixcc, "grap", 4)) /* graph */
6772 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
6775 if (memEQ(posixcc, "asci", 4)) /* ascii */
6776 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
6779 if (memEQ(posixcc, "blan", 4)) /* blank */
6780 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
6783 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
6784 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
6787 if (memEQ(posixcc, "alnu", 4)) /* alnum */
6788 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
6791 if (memEQ(posixcc, "lowe", 4)) /* lower */
6792 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
6793 else if (memEQ(posixcc, "uppe", 4)) /* upper */
6794 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
6797 if (memEQ(posixcc, "digi", 4)) /* digit */
6798 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
6799 else if (memEQ(posixcc, "prin", 4)) /* print */
6800 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
6801 else if (memEQ(posixcc, "punc", 4)) /* punct */
6802 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
6807 if (memEQ(posixcc, "xdigit", 6))
6808 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
6812 if (namedclass == OOB_NAMEDCLASS)
6813 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
6815 assert (posixcc[skip] == ':');
6816 assert (posixcc[skip+1] == ']');
6817 } else if (!SIZE_ONLY) {
6818 /* [[=foo=]] and [[.foo.]] are still future. */
6820 /* adjust RExC_parse so the warning shows after
6822 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
6824 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6827 /* Maternal grandfather:
6828 * "[:" ending in ":" but not in ":]" */
6838 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
6841 if (POSIXCC(UCHARAT(RExC_parse))) {
6842 const char *s = RExC_parse;
6843 const char c = *s++;
6847 if (*s && c == *s && s[1] == ']') {
6848 if (ckWARN(WARN_REGEXP))
6850 "POSIX syntax [%c %c] belongs inside character classes",
6853 /* [[=foo=]] and [[.foo.]] are still future. */
6854 if (POSIXCC_NOTYET(c)) {
6855 /* adjust RExC_parse so the error shows after
6857 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
6859 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6867 parse a class specification and produce either an ANYOF node that
6868 matches the pattern. If the pattern matches a single char only and
6869 that char is < 256 then we produce an EXACT node instead.
6872 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
6875 register UV value = 0;
6876 register UV nextvalue;
6877 register IV prevvalue = OOB_UNICODE;
6878 register IV range = 0;
6879 register regnode *ret;
6882 char *rangebegin = NULL;
6883 bool need_class = 0;
6886 bool optimize_invert = TRUE;
6887 AV* unicode_alternate = NULL;
6889 UV literal_endpoint = 0;
6891 UV stored = 0; /* number of chars stored in the class */
6893 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
6894 case we need to change the emitted regop to an EXACT. */
6895 const char * orig_parse = RExC_parse;
6896 GET_RE_DEBUG_FLAGS_DECL;
6898 PERL_UNUSED_ARG(depth);
6901 DEBUG_PARSE("clas");
6903 /* Assume we are going to generate an ANYOF node. */
6904 ret = reganode(pRExC_state, ANYOF, 0);
6907 ANYOF_FLAGS(ret) = 0;
6909 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
6913 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
6917 RExC_size += ANYOF_SKIP;
6918 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
6921 RExC_emit += ANYOF_SKIP;
6923 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
6925 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
6926 ANYOF_BITMAP_ZERO(ret);
6927 listsv = newSVpvs("# comment\n");
6930 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6932 if (!SIZE_ONLY && POSIXCC(nextvalue))
6933 checkposixcc(pRExC_state);
6935 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
6936 if (UCHARAT(RExC_parse) == ']')
6940 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
6944 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
6947 rangebegin = RExC_parse;
6949 value = utf8n_to_uvchr((U8*)RExC_parse,
6950 RExC_end - RExC_parse,
6951 &numlen, UTF8_ALLOW_DEFAULT);
6952 RExC_parse += numlen;
6955 value = UCHARAT(RExC_parse++);
6957 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6958 if (value == '[' && POSIXCC(nextvalue))
6959 namedclass = regpposixcc(pRExC_state, value);
6960 else if (value == '\\') {
6962 value = utf8n_to_uvchr((U8*)RExC_parse,
6963 RExC_end - RExC_parse,
6964 &numlen, UTF8_ALLOW_DEFAULT);
6965 RExC_parse += numlen;
6968 value = UCHARAT(RExC_parse++);
6969 /* Some compilers cannot handle switching on 64-bit integer
6970 * values, therefore value cannot be an UV. Yes, this will
6971 * be a problem later if we want switch on Unicode.
6972 * A similar issue a little bit later when switching on
6973 * namedclass. --jhi */
6974 switch ((I32)value) {
6975 case 'w': namedclass = ANYOF_ALNUM; break;
6976 case 'W': namedclass = ANYOF_NALNUM; break;
6977 case 's': namedclass = ANYOF_SPACE; break;
6978 case 'S': namedclass = ANYOF_NSPACE; break;
6979 case 'd': namedclass = ANYOF_DIGIT; break;
6980 case 'D': namedclass = ANYOF_NDIGIT; break;
6981 case 'N': /* Handle \N{NAME} in class */
6983 /* We only pay attention to the first char of
6984 multichar strings being returned. I kinda wonder
6985 if this makes sense as it does change the behaviour
6986 from earlier versions, OTOH that behaviour was broken
6988 UV v; /* value is register so we cant & it /grrr */
6989 if (reg_namedseq(pRExC_state, &v)) {
6999 if (RExC_parse >= RExC_end)
7000 vFAIL2("Empty \\%c{}", (U8)value);
7001 if (*RExC_parse == '{') {
7002 const U8 c = (U8)value;
7003 e = strchr(RExC_parse++, '}');
7005 vFAIL2("Missing right brace on \\%c{}", c);
7006 while (isSPACE(UCHARAT(RExC_parse)))
7008 if (e == RExC_parse)
7009 vFAIL2("Empty \\%c{}", c);
7011 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7019 if (UCHARAT(RExC_parse) == '^') {
7022 value = value == 'p' ? 'P' : 'p'; /* toggle */
7023 while (isSPACE(UCHARAT(RExC_parse))) {
7028 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7029 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7032 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7033 namedclass = ANYOF_MAX; /* no official name, but it's named */
7036 case 'n': value = '\n'; break;
7037 case 'r': value = '\r'; break;
7038 case 't': value = '\t'; break;
7039 case 'f': value = '\f'; break;
7040 case 'b': value = '\b'; break;
7041 case 'e': value = ASCII_TO_NATIVE('\033');break;
7042 case 'a': value = ASCII_TO_NATIVE('\007');break;
7044 if (*RExC_parse == '{') {
7045 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7046 | PERL_SCAN_DISALLOW_PREFIX;
7047 char * const e = strchr(RExC_parse++, '}');
7049 vFAIL("Missing right brace on \\x{}");
7051 numlen = e - RExC_parse;
7052 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7056 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7058 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7059 RExC_parse += numlen;
7061 if (PL_encoding && value < 0x100)
7062 goto recode_encoding;
7065 value = UCHARAT(RExC_parse++);
7066 value = toCTRL(value);
7068 case '0': case '1': case '2': case '3': case '4':
7069 case '5': case '6': case '7': case '8': case '9':
7073 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7074 RExC_parse += numlen;
7075 if (PL_encoding && value < 0x100)
7076 goto recode_encoding;
7081 SV* enc = PL_encoding;
7082 value = reg_recode((const char)(U8)value, &enc);
7083 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7085 "Invalid escape in the specified encoding");
7089 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7091 "Unrecognized escape \\%c in character class passed through",
7095 } /* end of \blah */
7101 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7103 if (!SIZE_ONLY && !need_class)
7104 ANYOF_CLASS_ZERO(ret);
7108 /* a bad range like a-\d, a-[:digit:] ? */
7111 if (ckWARN(WARN_REGEXP)) {
7113 RExC_parse >= rangebegin ?
7114 RExC_parse - rangebegin : 0;
7116 "False [] range \"%*.*s\"",
7119 if (prevvalue < 256) {
7120 ANYOF_BITMAP_SET(ret, prevvalue);
7121 ANYOF_BITMAP_SET(ret, '-');
7124 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7125 Perl_sv_catpvf(aTHX_ listsv,
7126 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7130 range = 0; /* this was not a true range */
7134 const char *what = NULL;
7137 if (namedclass > OOB_NAMEDCLASS)
7138 optimize_invert = FALSE;
7139 /* Possible truncation here but in some 64-bit environments
7140 * the compiler gets heartburn about switch on 64-bit values.
7141 * A similar issue a little earlier when switching on value.
7143 switch ((I32)namedclass) {
7146 ANYOF_CLASS_SET(ret, ANYOF_ALNUM);
7148 for (value = 0; value < 256; value++)
7150 ANYOF_BITMAP_SET(ret, value);
7157 ANYOF_CLASS_SET(ret, ANYOF_NALNUM);
7159 for (value = 0; value < 256; value++)
7160 if (!isALNUM(value))
7161 ANYOF_BITMAP_SET(ret, value);
7168 ANYOF_CLASS_SET(ret, ANYOF_ALNUMC);
7170 for (value = 0; value < 256; value++)
7171 if (isALNUMC(value))
7172 ANYOF_BITMAP_SET(ret, value);
7179 ANYOF_CLASS_SET(ret, ANYOF_NALNUMC);
7181 for (value = 0; value < 256; value++)
7182 if (!isALNUMC(value))
7183 ANYOF_BITMAP_SET(ret, value);
7190 ANYOF_CLASS_SET(ret, ANYOF_ALPHA);
7192 for (value = 0; value < 256; value++)
7194 ANYOF_BITMAP_SET(ret, value);
7201 ANYOF_CLASS_SET(ret, ANYOF_NALPHA);
7203 for (value = 0; value < 256; value++)
7204 if (!isALPHA(value))
7205 ANYOF_BITMAP_SET(ret, value);
7212 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7215 for (value = 0; value < 128; value++)
7216 ANYOF_BITMAP_SET(ret, value);
7218 for (value = 0; value < 256; value++) {
7220 ANYOF_BITMAP_SET(ret, value);
7229 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7232 for (value = 128; value < 256; value++)
7233 ANYOF_BITMAP_SET(ret, value);
7235 for (value = 0; value < 256; value++) {
7236 if (!isASCII(value))
7237 ANYOF_BITMAP_SET(ret, value);
7246 ANYOF_CLASS_SET(ret, ANYOF_BLANK);
7248 for (value = 0; value < 256; value++)
7250 ANYOF_BITMAP_SET(ret, value);
7257 ANYOF_CLASS_SET(ret, ANYOF_NBLANK);
7259 for (value = 0; value < 256; value++)
7260 if (!isBLANK(value))
7261 ANYOF_BITMAP_SET(ret, value);
7268 ANYOF_CLASS_SET(ret, ANYOF_CNTRL);
7270 for (value = 0; value < 256; value++)
7272 ANYOF_BITMAP_SET(ret, value);
7279 ANYOF_CLASS_SET(ret, ANYOF_NCNTRL);
7281 for (value = 0; value < 256; value++)
7282 if (!isCNTRL(value))
7283 ANYOF_BITMAP_SET(ret, value);
7290 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7292 /* consecutive digits assumed */
7293 for (value = '0'; value <= '9'; value++)
7294 ANYOF_BITMAP_SET(ret, value);
7301 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7303 /* consecutive digits assumed */
7304 for (value = 0; value < '0'; value++)
7305 ANYOF_BITMAP_SET(ret, value);
7306 for (value = '9' + 1; value < 256; value++)
7307 ANYOF_BITMAP_SET(ret, value);
7314 ANYOF_CLASS_SET(ret, ANYOF_GRAPH);
7316 for (value = 0; value < 256; value++)
7318 ANYOF_BITMAP_SET(ret, value);
7325 ANYOF_CLASS_SET(ret, ANYOF_NGRAPH);
7327 for (value = 0; value < 256; value++)
7328 if (!isGRAPH(value))
7329 ANYOF_BITMAP_SET(ret, value);
7336 ANYOF_CLASS_SET(ret, ANYOF_LOWER);
7338 for (value = 0; value < 256; value++)
7340 ANYOF_BITMAP_SET(ret, value);
7347 ANYOF_CLASS_SET(ret, ANYOF_NLOWER);
7349 for (value = 0; value < 256; value++)
7350 if (!isLOWER(value))
7351 ANYOF_BITMAP_SET(ret, value);
7358 ANYOF_CLASS_SET(ret, ANYOF_PRINT);
7360 for (value = 0; value < 256; value++)
7362 ANYOF_BITMAP_SET(ret, value);
7369 ANYOF_CLASS_SET(ret, ANYOF_NPRINT);
7371 for (value = 0; value < 256; value++)
7372 if (!isPRINT(value))
7373 ANYOF_BITMAP_SET(ret, value);
7380 ANYOF_CLASS_SET(ret, ANYOF_PSXSPC);
7382 for (value = 0; value < 256; value++)
7383 if (isPSXSPC(value))
7384 ANYOF_BITMAP_SET(ret, value);
7391 ANYOF_CLASS_SET(ret, ANYOF_NPSXSPC);
7393 for (value = 0; value < 256; value++)
7394 if (!isPSXSPC(value))
7395 ANYOF_BITMAP_SET(ret, value);
7402 ANYOF_CLASS_SET(ret, ANYOF_PUNCT);
7404 for (value = 0; value < 256; value++)
7406 ANYOF_BITMAP_SET(ret, value);
7413 ANYOF_CLASS_SET(ret, ANYOF_NPUNCT);
7415 for (value = 0; value < 256; value++)
7416 if (!isPUNCT(value))
7417 ANYOF_BITMAP_SET(ret, value);
7424 ANYOF_CLASS_SET(ret, ANYOF_SPACE);
7426 for (value = 0; value < 256; value++)
7428 ANYOF_BITMAP_SET(ret, value);
7435 ANYOF_CLASS_SET(ret, ANYOF_NSPACE);
7437 for (value = 0; value < 256; value++)
7438 if (!isSPACE(value))
7439 ANYOF_BITMAP_SET(ret, value);
7446 ANYOF_CLASS_SET(ret, ANYOF_UPPER);
7448 for (value = 0; value < 256; value++)
7450 ANYOF_BITMAP_SET(ret, value);
7457 ANYOF_CLASS_SET(ret, ANYOF_NUPPER);
7459 for (value = 0; value < 256; value++)
7460 if (!isUPPER(value))
7461 ANYOF_BITMAP_SET(ret, value);
7468 ANYOF_CLASS_SET(ret, ANYOF_XDIGIT);
7470 for (value = 0; value < 256; value++)
7471 if (isXDIGIT(value))
7472 ANYOF_BITMAP_SET(ret, value);
7479 ANYOF_CLASS_SET(ret, ANYOF_NXDIGIT);
7481 for (value = 0; value < 256; value++)
7482 if (!isXDIGIT(value))
7483 ANYOF_BITMAP_SET(ret, value);
7489 /* this is to handle \p and \P */
7492 vFAIL("Invalid [::] class");
7496 /* Strings such as "+utf8::isWord\n" */
7497 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7500 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7503 } /* end of namedclass \blah */
7506 if (prevvalue > (IV)value) /* b-a */ {
7507 const int w = RExC_parse - rangebegin;
7508 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7509 range = 0; /* not a valid range */
7513 prevvalue = value; /* save the beginning of the range */
7514 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7515 RExC_parse[1] != ']') {
7518 /* a bad range like \w-, [:word:]- ? */
7519 if (namedclass > OOB_NAMEDCLASS) {
7520 if (ckWARN(WARN_REGEXP)) {
7522 RExC_parse >= rangebegin ?
7523 RExC_parse - rangebegin : 0;
7525 "False [] range \"%*.*s\"",
7529 ANYOF_BITMAP_SET(ret, '-');
7531 range = 1; /* yeah, it's a range! */
7532 continue; /* but do it the next time */
7536 /* now is the next time */
7537 /*stored += (value - prevvalue + 1);*/
7539 if (prevvalue < 256) {
7540 const IV ceilvalue = value < 256 ? value : 255;
7543 /* In EBCDIC [\x89-\x91] should include
7544 * the \x8e but [i-j] should not. */
7545 if (literal_endpoint == 2 &&
7546 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7547 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7549 if (isLOWER(prevvalue)) {
7550 for (i = prevvalue; i <= ceilvalue; i++)
7552 ANYOF_BITMAP_SET(ret, i);
7554 for (i = prevvalue; i <= ceilvalue; i++)
7556 ANYOF_BITMAP_SET(ret, i);
7561 for (i = prevvalue; i <= ceilvalue; i++) {
7562 if (!ANYOF_BITMAP_TEST(ret,i)) {
7564 ANYOF_BITMAP_SET(ret, i);
7568 if (value > 255 || UTF) {
7569 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7570 const UV natvalue = NATIVE_TO_UNI(value);
7571 stored+=2; /* can't optimize this class */
7572 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7573 if (prevnatvalue < natvalue) { /* what about > ? */
7574 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7575 prevnatvalue, natvalue);
7577 else if (prevnatvalue == natvalue) {
7578 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7580 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7582 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7584 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7585 if (RExC_precomp[0] == ':' &&
7586 RExC_precomp[1] == '[' &&
7587 (f == 0xDF || f == 0x92)) {
7588 f = NATIVE_TO_UNI(f);
7591 /* If folding and foldable and a single
7592 * character, insert also the folded version
7593 * to the charclass. */
7595 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7596 if ((RExC_precomp[0] == ':' &&
7597 RExC_precomp[1] == '[' &&
7599 (value == 0xFB05 || value == 0xFB06))) ?
7600 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7601 foldlen == (STRLEN)UNISKIP(f) )
7603 if (foldlen == (STRLEN)UNISKIP(f))
7605 Perl_sv_catpvf(aTHX_ listsv,
7608 /* Any multicharacter foldings
7609 * require the following transform:
7610 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7611 * where E folds into "pq" and F folds
7612 * into "rst", all other characters
7613 * fold to single characters. We save
7614 * away these multicharacter foldings,
7615 * to be later saved as part of the
7616 * additional "s" data. */
7619 if (!unicode_alternate)
7620 unicode_alternate = newAV();
7621 sv = newSVpvn((char*)foldbuf, foldlen);
7623 av_push(unicode_alternate, sv);
7627 /* If folding and the value is one of the Greek
7628 * sigmas insert a few more sigmas to make the
7629 * folding rules of the sigmas to work right.
7630 * Note that not all the possible combinations
7631 * are handled here: some of them are handled
7632 * by the standard folding rules, and some of
7633 * them (literal or EXACTF cases) are handled
7634 * during runtime in regexec.c:S_find_byclass(). */
7635 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7636 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7637 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7638 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7639 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7641 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7642 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7643 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7648 literal_endpoint = 0;
7652 range = 0; /* this range (if it was one) is done now */
7656 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7658 RExC_size += ANYOF_CLASS_ADD_SKIP;
7660 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7666 /****** !SIZE_ONLY AFTER HERE *********/
7668 if( stored == 1 && value < 256
7669 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7671 /* optimize single char class to an EXACT node
7672 but *only* when its not a UTF/high char */
7673 const char * cur_parse= RExC_parse;
7674 RExC_emit = (regnode *)orig_emit;
7675 RExC_parse = (char *)orig_parse;
7676 ret = reg_node(pRExC_state,
7677 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7678 RExC_parse = (char *)cur_parse;
7679 *STRING(ret)= (char)value;
7681 RExC_emit += STR_SZ(1);
7684 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7685 if ( /* If the only flag is folding (plus possibly inversion). */
7686 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7688 for (value = 0; value < 256; ++value) {
7689 if (ANYOF_BITMAP_TEST(ret, value)) {
7690 UV fold = PL_fold[value];
7693 ANYOF_BITMAP_SET(ret, fold);
7696 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7699 /* optimize inverted simple patterns (e.g. [^a-z]) */
7700 if (optimize_invert &&
7701 /* If the only flag is inversion. */
7702 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7703 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7704 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7705 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7708 AV * const av = newAV();
7710 /* The 0th element stores the character class description
7711 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7712 * to initialize the appropriate swash (which gets stored in
7713 * the 1st element), and also useful for dumping the regnode.
7714 * The 2nd element stores the multicharacter foldings,
7715 * used later (regexec.c:S_reginclass()). */
7716 av_store(av, 0, listsv);
7717 av_store(av, 1, NULL);
7718 av_store(av, 2, (SV*)unicode_alternate);
7719 rv = newRV_noinc((SV*)av);
7720 n = add_data(pRExC_state, 1, "s");
7721 RExC_rxi->data->data[n] = (void*)rv;
7728 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7730 char* const retval = RExC_parse++;
7733 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7734 RExC_parse[2] == '#') {
7735 while (*RExC_parse != ')') {
7736 if (RExC_parse == RExC_end)
7737 FAIL("Sequence (?#... not terminated");
7743 if (RExC_flags & RXf_PMf_EXTENDED) {
7744 if (isSPACE(*RExC_parse)) {
7748 else if (*RExC_parse == '#') {
7749 while (RExC_parse < RExC_end)
7750 if (*RExC_parse++ == '\n') break;
7759 - reg_node - emit a node
7761 STATIC regnode * /* Location. */
7762 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7765 register regnode *ptr;
7766 regnode * const ret = RExC_emit;
7767 GET_RE_DEBUG_FLAGS_DECL;
7770 SIZE_ALIGN(RExC_size);
7775 if (OP(RExC_emit) == 255)
7776 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %s: %d ",
7777 reg_name[op], OP(RExC_emit));
7779 NODE_ALIGN_FILL(ret);
7781 FILL_ADVANCE_NODE(ptr, op);
7782 if (RExC_offsets) { /* MJD */
7783 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7784 "reg_node", __LINE__,
7786 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7787 ? "Overwriting end of array!\n" : "OK",
7788 (UV)(RExC_emit - RExC_emit_start),
7789 (UV)(RExC_parse - RExC_start),
7790 (UV)RExC_offsets[0]));
7791 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7799 - reganode - emit a node with an argument
7801 STATIC regnode * /* Location. */
7802 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7805 register regnode *ptr;
7806 regnode * const ret = RExC_emit;
7807 GET_RE_DEBUG_FLAGS_DECL;
7810 SIZE_ALIGN(RExC_size);
7815 assert(2==regarglen[op]+1);
7817 Anything larger than this has to allocate the extra amount.
7818 If we changed this to be:
7820 RExC_size += (1 + regarglen[op]);
7822 then it wouldn't matter. Its not clear what side effect
7823 might come from that so its not done so far.
7829 if (OP(RExC_emit) == 255)
7830 Perl_croak(aTHX_ "panic: reganode overwriting end of allocated program space");
7832 NODE_ALIGN_FILL(ret);
7834 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7835 if (RExC_offsets) { /* MJD */
7836 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7840 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7841 "Overwriting end of array!\n" : "OK",
7842 (UV)(RExC_emit - RExC_emit_start),
7843 (UV)(RExC_parse - RExC_start),
7844 (UV)RExC_offsets[0]));
7845 Set_Cur_Node_Offset;
7853 - reguni - emit (if appropriate) a Unicode character
7856 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
7859 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
7863 - reginsert - insert an operator in front of already-emitted operand
7865 * Means relocating the operand.
7868 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
7871 register regnode *src;
7872 register regnode *dst;
7873 register regnode *place;
7874 const int offset = regarglen[(U8)op];
7875 const int size = NODE_STEP_REGNODE + offset;
7876 GET_RE_DEBUG_FLAGS_DECL;
7877 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
7878 DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
7887 if (RExC_open_parens) {
7889 DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);
7890 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
7891 if ( RExC_open_parens[paren] >= opnd ) {
7892 DEBUG_PARSE_FMT("open"," - %d",size);
7893 RExC_open_parens[paren] += size;
7895 DEBUG_PARSE_FMT("open"," - %s","ok");
7897 if ( RExC_close_parens[paren] >= opnd ) {
7898 DEBUG_PARSE_FMT("close"," - %d",size);
7899 RExC_close_parens[paren] += size;
7901 DEBUG_PARSE_FMT("close"," - %s","ok");
7906 while (src > opnd) {
7907 StructCopy(--src, --dst, regnode);
7908 if (RExC_offsets) { /* MJD 20010112 */
7909 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
7913 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
7914 ? "Overwriting end of array!\n" : "OK",
7915 (UV)(src - RExC_emit_start),
7916 (UV)(dst - RExC_emit_start),
7917 (UV)RExC_offsets[0]));
7918 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
7919 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
7924 place = opnd; /* Op node, where operand used to be. */
7925 if (RExC_offsets) { /* MJD */
7926 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7930 (UV)(place - RExC_emit_start) > RExC_offsets[0]
7931 ? "Overwriting end of array!\n" : "OK",
7932 (UV)(place - RExC_emit_start),
7933 (UV)(RExC_parse - RExC_start),
7934 (UV)RExC_offsets[0]));
7935 Set_Node_Offset(place, RExC_parse);
7936 Set_Node_Length(place, 1);
7938 src = NEXTOPER(place);
7939 FILL_ADVANCE_NODE(place, op);
7940 Zero(src, offset, regnode);
7944 - regtail - set the next-pointer at the end of a node chain of p to val.
7945 - SEE ALSO: regtail_study
7947 /* TODO: All three parms should be const */
7949 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7952 register regnode *scan;
7953 GET_RE_DEBUG_FLAGS_DECL;
7955 PERL_UNUSED_ARG(depth);
7961 /* Find last node. */
7964 regnode * const temp = regnext(scan);
7966 SV * const mysv=sv_newmortal();
7967 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
7968 regprop(RExC_rx, mysv, scan);
7969 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
7970 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
7971 (temp == NULL ? "->" : ""),
7972 (temp == NULL ? reg_name[OP(val)] : "")
7980 if (reg_off_by_arg[OP(scan)]) {
7981 ARG_SET(scan, val - scan);
7984 NEXT_OFF(scan) = val - scan;
7990 - regtail_study - set the next-pointer at the end of a node chain of p to val.
7991 - Look for optimizable sequences at the same time.
7992 - currently only looks for EXACT chains.
7994 This is expermental code. The idea is to use this routine to perform
7995 in place optimizations on branches and groups as they are constructed,
7996 with the long term intention of removing optimization from study_chunk so
7997 that it is purely analytical.
7999 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8000 to control which is which.
8003 /* TODO: All four parms should be const */
8006 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8009 register regnode *scan;
8011 #ifdef EXPERIMENTAL_INPLACESCAN
8015 GET_RE_DEBUG_FLAGS_DECL;
8021 /* Find last node. */
8025 regnode * const temp = regnext(scan);
8026 #ifdef EXPERIMENTAL_INPLACESCAN
8027 if (PL_regkind[OP(scan)] == EXACT)
8028 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8036 if( exact == PSEUDO )
8038 else if ( exact != OP(scan) )
8047 SV * const mysv=sv_newmortal();
8048 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8049 regprop(RExC_rx, mysv, scan);
8050 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8051 SvPV_nolen_const(mysv),
8060 SV * const mysv_val=sv_newmortal();
8061 DEBUG_PARSE_MSG("");
8062 regprop(RExC_rx, mysv_val, val);
8063 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8064 SvPV_nolen_const(mysv_val),
8065 (IV)REG_NODE_NUM(val),
8069 if (reg_off_by_arg[OP(scan)]) {
8070 ARG_SET(scan, val - scan);
8073 NEXT_OFF(scan) = val - scan;
8081 - regcurly - a little FSA that accepts {\d+,?\d*}
8084 S_regcurly(register const char *s)
8103 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8106 Perl_regdump(pTHX_ const regexp *r)
8110 SV * const sv = sv_newmortal();
8111 SV *dsv= sv_newmortal();
8114 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8116 /* Header fields of interest. */
8117 if (r->anchored_substr) {
8118 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8119 RE_SV_DUMPLEN(r->anchored_substr), 30);
8120 PerlIO_printf(Perl_debug_log,
8121 "anchored %s%s at %"IVdf" ",
8122 s, RE_SV_TAIL(r->anchored_substr),
8123 (IV)r->anchored_offset);
8124 } else if (r->anchored_utf8) {
8125 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8126 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8127 PerlIO_printf(Perl_debug_log,
8128 "anchored utf8 %s%s at %"IVdf" ",
8129 s, RE_SV_TAIL(r->anchored_utf8),
8130 (IV)r->anchored_offset);
8132 if (r->float_substr) {
8133 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8134 RE_SV_DUMPLEN(r->float_substr), 30);
8135 PerlIO_printf(Perl_debug_log,
8136 "floating %s%s at %"IVdf"..%"UVuf" ",
8137 s, RE_SV_TAIL(r->float_substr),
8138 (IV)r->float_min_offset, (UV)r->float_max_offset);
8139 } else if (r->float_utf8) {
8140 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8141 RE_SV_DUMPLEN(r->float_utf8), 30);
8142 PerlIO_printf(Perl_debug_log,
8143 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8144 s, RE_SV_TAIL(r->float_utf8),
8145 (IV)r->float_min_offset, (UV)r->float_max_offset);
8147 if (r->check_substr || r->check_utf8)
8148 PerlIO_printf(Perl_debug_log,
8150 (r->check_substr == r->float_substr
8151 && r->check_utf8 == r->float_utf8
8152 ? "(checking floating" : "(checking anchored"));
8153 if (r->extflags & RXf_NOSCAN)
8154 PerlIO_printf(Perl_debug_log, " noscan");
8155 if (r->extflags & RXf_CHECK_ALL)
8156 PerlIO_printf(Perl_debug_log, " isall");
8157 if (r->check_substr || r->check_utf8)
8158 PerlIO_printf(Perl_debug_log, ") ");
8160 if (ri->regstclass) {
8161 regprop(r, sv, ri->regstclass);
8162 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8164 if (r->extflags & RXf_ANCH) {
8165 PerlIO_printf(Perl_debug_log, "anchored");
8166 if (r->extflags & RXf_ANCH_BOL)
8167 PerlIO_printf(Perl_debug_log, "(BOL)");
8168 if (r->extflags & RXf_ANCH_MBOL)
8169 PerlIO_printf(Perl_debug_log, "(MBOL)");
8170 if (r->extflags & RXf_ANCH_SBOL)
8171 PerlIO_printf(Perl_debug_log, "(SBOL)");
8172 if (r->extflags & RXf_ANCH_GPOS)
8173 PerlIO_printf(Perl_debug_log, "(GPOS)");
8174 PerlIO_putc(Perl_debug_log, ' ');
8176 if (r->extflags & RXf_GPOS_SEEN)
8177 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8178 if (r->intflags & PREGf_SKIP)
8179 PerlIO_printf(Perl_debug_log, "plus ");
8180 if (r->intflags & PREGf_IMPLICIT)
8181 PerlIO_printf(Perl_debug_log, "implicit ");
8182 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8183 if (r->extflags & RXf_EVAL_SEEN)
8184 PerlIO_printf(Perl_debug_log, "with eval ");
8185 PerlIO_printf(Perl_debug_log, "\n");
8187 PERL_UNUSED_CONTEXT;
8189 #endif /* DEBUGGING */
8193 - regprop - printable representation of opcode
8196 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8201 RXi_GET_DECL(prog,progi);
8202 GET_RE_DEBUG_FLAGS_DECL;
8205 sv_setpvn(sv, "", 0);
8207 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8208 /* It would be nice to FAIL() here, but this may be called from
8209 regexec.c, and it would be hard to supply pRExC_state. */
8210 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8211 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
8213 k = PL_regkind[OP(o)];
8216 SV * const dsv = sv_2mortal(newSVpvs(""));
8217 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8218 * is a crude hack but it may be the best for now since
8219 * we have no flag "this EXACTish node was UTF-8"
8221 const char * const s =
8222 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8223 PL_colors[0], PL_colors[1],
8224 PERL_PV_ESCAPE_UNI_DETECT |
8225 PERL_PV_PRETTY_ELIPSES |
8228 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8229 } else if (k == TRIE) {
8230 /* print the details of the trie in dumpuntil instead, as
8231 * progi->data isn't available here */
8232 const char op = OP(o);
8233 const I32 n = ARG(o);
8234 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8235 (reg_ac_data *)progi->data->data[n] :
8237 const reg_trie_data * const trie
8238 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8240 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
8241 DEBUG_TRIE_COMPILE_r(
8242 Perl_sv_catpvf(aTHX_ sv,
8243 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8244 (UV)trie->startstate,
8245 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8246 (UV)trie->wordcount,
8249 (UV)TRIE_CHARCOUNT(trie),
8250 (UV)trie->uniquecharcount
8253 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8255 int rangestart = -1;
8256 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8257 Perl_sv_catpvf(aTHX_ sv, "[");
8258 for (i = 0; i <= 256; i++) {
8259 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8260 if (rangestart == -1)
8262 } else if (rangestart != -1) {
8263 if (i <= rangestart + 3)
8264 for (; rangestart < i; rangestart++)
8265 put_byte(sv, rangestart);
8267 put_byte(sv, rangestart);
8269 put_byte(sv, i - 1);
8274 Perl_sv_catpvf(aTHX_ sv, "]");
8277 } else if (k == CURLY) {
8278 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8279 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8280 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8282 else if (k == WHILEM && o->flags) /* Ordinal/of */
8283 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8284 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT)
8285 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8286 else if (k == GOSUB)
8287 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8288 else if (k == VERB) {
8290 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8291 (SV*)progi->data->data[ ARG( o ) ]);
8292 } else if (k == LOGICAL)
8293 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8294 else if (k == ANYOF) {
8295 int i, rangestart = -1;
8296 const U8 flags = ANYOF_FLAGS(o);
8298 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8299 static const char * const anyofs[] = {
8332 if (flags & ANYOF_LOCALE)
8333 sv_catpvs(sv, "{loc}");
8334 if (flags & ANYOF_FOLD)
8335 sv_catpvs(sv, "{i}");
8336 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8337 if (flags & ANYOF_INVERT)
8339 for (i = 0; i <= 256; i++) {
8340 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8341 if (rangestart == -1)
8343 } else if (rangestart != -1) {
8344 if (i <= rangestart + 3)
8345 for (; rangestart < i; rangestart++)
8346 put_byte(sv, rangestart);
8348 put_byte(sv, rangestart);
8350 put_byte(sv, i - 1);
8356 if (o->flags & ANYOF_CLASS)
8357 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8358 if (ANYOF_CLASS_TEST(o,i))
8359 sv_catpv(sv, anyofs[i]);
8361 if (flags & ANYOF_UNICODE)
8362 sv_catpvs(sv, "{unicode}");
8363 else if (flags & ANYOF_UNICODE_ALL)
8364 sv_catpvs(sv, "{unicode_all}");
8368 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8372 U8 s[UTF8_MAXBYTES_CASE+1];
8374 for (i = 0; i <= 256; i++) { /* just the first 256 */
8375 uvchr_to_utf8(s, i);
8377 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8378 if (rangestart == -1)
8380 } else if (rangestart != -1) {
8381 if (i <= rangestart + 3)
8382 for (; rangestart < i; rangestart++) {
8383 const U8 * const e = uvchr_to_utf8(s,rangestart);
8385 for(p = s; p < e; p++)
8389 const U8 *e = uvchr_to_utf8(s,rangestart);
8391 for (p = s; p < e; p++)
8394 e = uvchr_to_utf8(s, i-1);
8395 for (p = s; p < e; p++)
8402 sv_catpvs(sv, "..."); /* et cetera */
8406 char *s = savesvpv(lv);
8407 char * const origs = s;
8409 while (*s && *s != '\n')
8413 const char * const t = ++s;
8431 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8433 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8434 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8436 PERL_UNUSED_CONTEXT;
8437 PERL_UNUSED_ARG(sv);
8439 PERL_UNUSED_ARG(prog);
8440 #endif /* DEBUGGING */
8444 Perl_re_intuit_string(pTHX_ regexp *prog)
8445 { /* Assume that RE_INTUIT is set */
8447 GET_RE_DEBUG_FLAGS_DECL;
8448 PERL_UNUSED_CONTEXT;
8452 const char * const s = SvPV_nolen_const(prog->check_substr
8453 ? prog->check_substr : prog->check_utf8);
8455 if (!PL_colorset) reginitcolors();
8456 PerlIO_printf(Perl_debug_log,
8457 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8459 prog->check_substr ? "" : "utf8 ",
8460 PL_colors[5],PL_colors[0],
8463 (strlen(s) > 60 ? "..." : ""));
8466 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8472 handles refcounting and freeing the perl core regexp structure. When
8473 it is necessary to actually free the structure the first thing it
8474 does is call the 'free' method of the regexp_engine associated to to
8475 the regexp, allowing the handling of the void *pprivate; member
8476 first. (This routine is not overridable by extensions, which is why
8477 the extensions free is called first.)
8479 See regdupe and regdupe_internal if you change anything here.
8481 #ifndef PERL_IN_XSUB_RE
8483 Perl_pregfree(pTHX_ struct regexp *r)
8486 GET_RE_DEBUG_FLAGS_DECL;
8488 if (!r || (--r->refcnt > 0))
8491 CALLREGFREE_PVT(r); /* free the private data */
8493 /* gcov results gave these as non-null 100% of the time, so there's no
8494 optimisation in checking them before calling Safefree */
8495 Safefree(r->precomp);
8496 RX_MATCH_COPY_FREE(r);
8497 #ifdef PERL_OLD_COPY_ON_WRITE
8499 SvREFCNT_dec(r->saved_copy);
8502 if (r->anchored_substr)
8503 SvREFCNT_dec(r->anchored_substr);
8504 if (r->anchored_utf8)
8505 SvREFCNT_dec(r->anchored_utf8);
8506 if (r->float_substr)
8507 SvREFCNT_dec(r->float_substr);
8509 SvREFCNT_dec(r->float_utf8);
8510 Safefree(r->substrs);
8513 SvREFCNT_dec(r->paren_names);
8515 Safefree(r->startp);
8521 /* regfree_internal()
8523 Free the private data in a regexp. This is overloadable by
8524 extensions. Perl takes care of the regexp structure in pregfree(),
8525 this covers the *pprivate pointer which technically perldoesnt
8526 know about, however of course we have to handle the
8527 regexp_internal structure when no extension is in use.
8529 Note this is called before freeing anything in the regexp
8534 Perl_regfree_internal(pTHX_ struct regexp *r)
8538 GET_RE_DEBUG_FLAGS_DECL;
8544 SV *dsv= sv_newmortal();
8545 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8546 dsv, r->precomp, r->prelen, 60);
8547 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8548 PL_colors[4],PL_colors[5],s);
8552 Safefree(ri->offsets); /* 20010421 MJD */
8554 int n = ri->data->count;
8555 PAD* new_comppad = NULL;
8560 /* If you add a ->what type here, update the comment in regcomp.h */
8561 switch (ri->data->what[n]) {
8565 SvREFCNT_dec((SV*)ri->data->data[n]);
8568 Safefree(ri->data->data[n]);
8571 new_comppad = (AV*)ri->data->data[n];
8574 if (new_comppad == NULL)
8575 Perl_croak(aTHX_ "panic: pregfree comppad");
8576 PAD_SAVE_LOCAL(old_comppad,
8577 /* Watch out for global destruction's random ordering. */
8578 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8581 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8584 op_free((OP_4tree*)ri->data->data[n]);
8586 PAD_RESTORE_LOCAL(old_comppad);
8587 SvREFCNT_dec((SV*)new_comppad);
8593 { /* Aho Corasick add-on structure for a trie node.
8594 Used in stclass optimization only */
8596 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8598 refcount = --aho->refcount;
8601 PerlMemShared_free(aho->states);
8602 PerlMemShared_free(aho->fail);
8603 /* do this last!!!! */
8604 PerlMemShared_free(ri->data->data[n]);
8605 PerlMemShared_free(ri->regstclass);
8611 /* trie structure. */
8613 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8615 refcount = --trie->refcount;
8618 PerlMemShared_free(trie->charmap);
8619 PerlMemShared_free(trie->states);
8620 PerlMemShared_free(trie->trans);
8622 PerlMemShared_free(trie->bitmap);
8624 PerlMemShared_free(trie->wordlen);
8626 PerlMemShared_free(trie->jump);
8628 PerlMemShared_free(trie->nextword);
8629 /* do this last!!!! */
8630 PerlMemShared_free(ri->data->data[n]);
8635 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8638 Safefree(ri->data->what);
8642 Safefree(ri->swap->startp);
8643 Safefree(ri->swap->endp);
8649 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8650 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8651 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8652 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8655 regdupe - duplicate a regexp.
8657 This routine is called by sv.c's re_dup and is expected to clone a
8658 given regexp structure. It is a no-op when not under USE_ITHREADS.
8659 (Originally this *was* re_dup() for change history see sv.c)
8661 After all of the core data stored in struct regexp is duplicated
8662 the regexp_engine.dupe method is used to copy any private data
8663 stored in the *pprivate pointer. This allows extensions to handle
8664 any duplication it needs to do.
8666 See pregfree() and regfree_internal() if you change anything here.
8668 #if defined(USE_ITHREADS)
8669 #ifndef PERL_IN_XSUB_RE
8671 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
8676 struct reg_substr_datum *s;
8679 return (REGEXP *)NULL;
8681 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8685 npar = r->nparens+1;
8686 Newxz(ret, 1, regexp);
8687 Newx(ret->startp, npar, I32);
8688 Copy(r->startp, ret->startp, npar, I32);
8689 Newx(ret->endp, npar, I32);
8690 Copy(r->endp, ret->endp, npar, I32);
8692 Newx(ret->substrs, 1, struct reg_substr_data);
8693 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8694 s->min_offset = r->substrs->data[i].min_offset;
8695 s->max_offset = r->substrs->data[i].max_offset;
8696 s->end_shift = r->substrs->data[i].end_shift;
8697 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8698 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8702 ret->precomp = SAVEPVN(r->precomp, r->prelen);
8703 ret->refcnt = r->refcnt;
8704 ret->minlen = r->minlen;
8705 ret->minlenret = r->minlenret;
8706 ret->prelen = r->prelen;
8707 ret->nparens = r->nparens;
8708 ret->lastparen = r->lastparen;
8709 ret->lastcloseparen = r->lastcloseparen;
8710 ret->intflags = r->intflags;
8711 ret->extflags = r->extflags;
8713 ret->sublen = r->sublen;
8715 ret->engine = r->engine;
8717 ret->paren_names = hv_dup_inc(r->paren_names, param);
8719 if (RX_MATCH_COPIED(ret))
8720 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8723 #ifdef PERL_OLD_COPY_ON_WRITE
8724 ret->saved_copy = NULL;
8727 ret->pprivate = r->pprivate;
8728 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
8730 ptr_table_store(PL_ptr_table, r, ret);
8733 #endif /* PERL_IN_XSUB_RE */
8738 This is the internal complement to regdupe() which is used to copy
8739 the structure pointed to by the *pprivate pointer in the regexp.
8740 This is the core version of the extension overridable cloning hook.
8741 The regexp structure being duplicated will be copied by perl prior
8742 to this and will be provided as the regexp *r argument, however
8743 with the /old/ structures pprivate pointer value. Thus this routine
8744 may override any copying normally done by perl.
8746 It returns a pointer to the new regexp_internal structure.
8750 Perl_regdupe_internal(pTHX_ const regexp *r, CLONE_PARAMS *param)
8753 regexp_internal *reti;
8757 npar = r->nparens+1;
8758 len = ri->offsets[0];
8760 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8761 Copy(ri->program, reti->program, len+1, regnode);
8764 Newx(reti->swap, 1, regexp_paren_ofs);
8765 /* no need to copy these */
8766 Newx(reti->swap->startp, npar, I32);
8767 Newx(reti->swap->endp, npar, I32);
8773 reti->regstclass = NULL;
8776 const int count = ri->data->count;
8779 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8780 char, struct reg_data);
8781 Newx(d->what, count, U8);
8784 for (i = 0; i < count; i++) {
8785 d->what[i] = ri->data->what[i];
8786 switch (d->what[i]) {
8787 /* legal options are one of: sSfpontTu
8788 see also regcomp.h and pregfree() */
8791 case 'p': /* actually an AV, but the dup function is identical. */
8792 case 'u': /* actually an HV, but the dup function is identical. */
8793 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
8796 /* This is cheating. */
8797 Newx(d->data[i], 1, struct regnode_charclass_class);
8798 StructCopy(ri->data->data[i], d->data[i],
8799 struct regnode_charclass_class);
8800 reti->regstclass = (regnode*)d->data[i];
8803 /* Compiled op trees are readonly and in shared memory,
8804 and can thus be shared without duplication. */
8806 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
8810 /* Trie stclasses are readonly and can thus be shared
8811 * without duplication. We free the stclass in pregfree
8812 * when the corresponding reg_ac_data struct is freed.
8814 reti->regstclass= ri->regstclass;
8818 ((reg_trie_data*)ri->data->data[i])->refcount++;
8822 d->data[i] = ri->data->data[i];
8825 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
8834 Newx(reti->offsets, 2*len+1, U32);
8835 Copy(ri->offsets, reti->offsets, 2*len+1, U32);
8840 #endif /* USE_ITHREADS */
8845 converts a regexp embedded in a MAGIC struct to its stringified form,
8846 caching the converted form in the struct and returns the cached
8849 If lp is nonnull then it is used to return the length of the
8852 If flags is nonnull and the returned string contains UTF8 then
8853 (*flags & 1) will be true.
8855 If haseval is nonnull then it is used to return whether the pattern
8858 Normally called via macro:
8860 CALLREG_STRINGIFY(mg,&len,&utf8);
8864 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
8866 See sv_2pv_flags() in sv.c for an example of internal usage.
8869 #ifndef PERL_IN_XSUB_RE
8871 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
8873 const regexp * const re = (regexp *)mg->mg_obj;
8876 const char *fptr = "msix";
8881 bool need_newline = 0;
8882 U16 reganch = (U16)((re->extflags & RXf_PMf_COMPILETIME) >> 12);
8884 while((ch = *fptr++)) {
8886 reflags[left++] = ch;
8889 reflags[right--] = ch;
8894 reflags[left] = '-';
8898 mg->mg_len = re->prelen + 4 + left;
8900 * If /x was used, we have to worry about a regex ending with a
8901 * comment later being embedded within another regex. If so, we don't
8902 * want this regex's "commentization" to leak out to the right part of
8903 * the enclosing regex, we must cap it with a newline.
8905 * So, if /x was used, we scan backwards from the end of the regex. If
8906 * we find a '#' before we find a newline, we need to add a newline
8907 * ourself. If we find a '\n' first (or if we don't find '#' or '\n'),
8908 * we don't need to add anything. -jfriedl
8910 if (PMf_EXTENDED & re->extflags) {
8911 const char *endptr = re->precomp + re->prelen;
8912 while (endptr >= re->precomp) {
8913 const char c = *(endptr--);
8915 break; /* don't need another */
8917 /* we end while in a comment, so we need a newline */
8918 mg->mg_len++; /* save space for it */
8919 need_newline = 1; /* note to add it */
8925 Newx(mg->mg_ptr, mg->mg_len + 1 + left, char);
8926 mg->mg_ptr[0] = '(';
8927 mg->mg_ptr[1] = '?';
8928 Copy(reflags, mg->mg_ptr+2, left, char);
8929 *(mg->mg_ptr+left+2) = ':';
8930 Copy(re->precomp, mg->mg_ptr+3+left, re->prelen, char);
8932 mg->mg_ptr[mg->mg_len - 2] = '\n';
8933 mg->mg_ptr[mg->mg_len - 1] = ')';
8934 mg->mg_ptr[mg->mg_len] = 0;
8937 *haseval = re->seen_evals;
8939 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
8947 - regnext - dig the "next" pointer out of a node
8950 Perl_regnext(pTHX_ register regnode *p)
8953 register I32 offset;
8958 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
8967 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
8970 STRLEN l1 = strlen(pat1);
8971 STRLEN l2 = strlen(pat2);
8974 const char *message;
8980 Copy(pat1, buf, l1 , char);
8981 Copy(pat2, buf + l1, l2 , char);
8982 buf[l1 + l2] = '\n';
8983 buf[l1 + l2 + 1] = '\0';
8985 /* ANSI variant takes additional second argument */
8986 va_start(args, pat2);
8990 msv = vmess(buf, &args);
8992 message = SvPV_const(msv,l1);
8995 Copy(message, buf, l1 , char);
8996 buf[l1-1] = '\0'; /* Overwrite \n */
8997 Perl_croak(aTHX_ "%s", buf);
9000 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9002 #ifndef PERL_IN_XSUB_RE
9004 Perl_save_re_context(pTHX)
9008 struct re_save_state *state;
9010 SAVEVPTR(PL_curcop);
9011 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9013 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9014 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9015 SSPUSHINT(SAVEt_RE_STATE);
9017 Copy(&PL_reg_state, state, 1, struct re_save_state);
9019 PL_reg_start_tmp = 0;
9020 PL_reg_start_tmpl = 0;
9021 PL_reg_oldsaved = NULL;
9022 PL_reg_oldsavedlen = 0;
9024 PL_reg_leftiter = 0;
9025 PL_reg_poscache = NULL;
9026 PL_reg_poscache_size = 0;
9027 #ifdef PERL_OLD_COPY_ON_WRITE
9031 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9033 const REGEXP * const rx = PM_GETRE(PL_curpm);
9036 for (i = 1; i <= rx->nparens; i++) {
9037 char digits[TYPE_CHARS(long)];
9038 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9039 GV *const *const gvp
9040 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9043 GV * const gv = *gvp;
9044 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9054 clear_re(pTHX_ void *r)
9057 ReREFCNT_dec((regexp *)r);
9063 S_put_byte(pTHX_ SV *sv, int c)
9065 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9066 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9067 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9068 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9070 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9074 #define CLEAR_OPTSTART \
9075 if (optstart) STMT_START { \
9076 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9080 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9082 STATIC const regnode *
9083 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9084 const regnode *last, const regnode *plast,
9085 SV* sv, I32 indent, U32 depth)
9088 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9089 register const regnode *next;
9090 const regnode *optstart= NULL;
9092 GET_RE_DEBUG_FLAGS_DECL;
9094 #ifdef DEBUG_DUMPUNTIL
9095 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9096 last ? last-start : 0,plast ? plast-start : 0);
9099 if (plast && plast < last)
9102 while (PL_regkind[op] != END && (!last || node < last)) {
9103 /* While that wasn't END last time... */
9107 if (op == CLOSE || op == WHILEM)
9109 next = regnext((regnode *)node);
9112 if (OP(node) == OPTIMIZED) {
9113 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9120 regprop(r, sv, node);
9121 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9122 (int)(2*indent + 1), "", SvPVX_const(sv));
9124 if (OP(node) != OPTIMIZED) {
9125 if (next == NULL) /* Next ptr. */
9126 PerlIO_printf(Perl_debug_log, "(0)");
9127 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9128 PerlIO_printf(Perl_debug_log, "(FAIL)");
9130 PerlIO_printf(Perl_debug_log, "(%"IVdf")", (IV)(next - start));
9132 /*if (PL_regkind[(U8)op] != TRIE)*/
9133 (void)PerlIO_putc(Perl_debug_log, '\n');
9137 if (PL_regkind[(U8)op] == BRANCHJ) {
9140 register const regnode *nnode = (OP(next) == LONGJMP
9141 ? regnext((regnode *)next)
9143 if (last && nnode > last)
9145 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9148 else if (PL_regkind[(U8)op] == BRANCH) {
9150 DUMPUNTIL(NEXTOPER(node), next);
9152 else if ( PL_regkind[(U8)op] == TRIE ) {
9153 const regnode *this_trie = node;
9154 const char op = OP(node);
9155 const I32 n = ARG(node);
9156 const reg_ac_data * const ac = op>=AHOCORASICK ?
9157 (reg_ac_data *)ri->data->data[n] :
9159 const reg_trie_data * const trie =
9160 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9162 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9164 const regnode *nextbranch= NULL;
9166 sv_setpvn(sv, "", 0);
9167 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9168 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9170 PerlIO_printf(Perl_debug_log, "%*s%s ",
9171 (int)(2*(indent+3)), "",
9172 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9173 PL_colors[0], PL_colors[1],
9174 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9175 PERL_PV_PRETTY_ELIPSES |
9181 U16 dist= trie->jump[word_idx+1];
9182 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9183 (UV)((dist ? this_trie + dist : next) - start));
9186 nextbranch= this_trie + trie->jump[0];
9187 DUMPUNTIL(this_trie + dist, nextbranch);
9189 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9190 nextbranch= regnext((regnode *)nextbranch);
9192 PerlIO_printf(Perl_debug_log, "\n");
9195 if (last && next > last)
9200 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9201 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9202 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9204 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9206 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9208 else if ( op == PLUS || op == STAR) {
9209 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9211 else if (op == ANYOF) {
9212 /* arglen 1 + class block */
9213 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9214 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9215 node = NEXTOPER(node);
9217 else if (PL_regkind[(U8)op] == EXACT) {
9218 /* Literal string, where present. */
9219 node += NODE_SZ_STR(node) - 1;
9220 node = NEXTOPER(node);
9223 node = NEXTOPER(node);
9224 node += regarglen[(U8)op];
9226 if (op == CURLYX || op == OPEN)
9230 #ifdef DEBUG_DUMPUNTIL
9231 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9236 #endif /* DEBUGGING */
9240 * c-indentation-style: bsd
9242 * indent-tabs-mode: t
9245 * ex: set ts=8 sts=4 sw=4 noet: