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, 2007 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 */
130 regnode **recurse; /* Recurse regops */
131 I32 recurse_count; /* Number of recurse regops */
133 char *starttry; /* -Dr: where regtry was called. */
134 #define RExC_starttry (pRExC_state->starttry)
137 const char *lastparse;
139 AV *paren_name_list; /* idx -> name */
140 #define RExC_lastparse (pRExC_state->lastparse)
141 #define RExC_lastnum (pRExC_state->lastnum)
142 #define RExC_paren_name_list (pRExC_state->paren_name_list)
146 #define RExC_flags (pRExC_state->flags)
147 #define RExC_precomp (pRExC_state->precomp)
148 #define RExC_rx (pRExC_state->rx)
149 #define RExC_rxi (pRExC_state->rxi)
150 #define RExC_start (pRExC_state->start)
151 #define RExC_end (pRExC_state->end)
152 #define RExC_parse (pRExC_state->parse)
153 #define RExC_whilem_seen (pRExC_state->whilem_seen)
154 #define RExC_offsets (pRExC_state->rxi->offsets) /* I am not like the others */
155 #define RExC_emit (pRExC_state->emit)
156 #define RExC_emit_start (pRExC_state->emit_start)
157 #define RExC_naughty (pRExC_state->naughty)
158 #define RExC_sawback (pRExC_state->sawback)
159 #define RExC_seen (pRExC_state->seen)
160 #define RExC_size (pRExC_state->size)
161 #define RExC_npar (pRExC_state->npar)
162 #define RExC_nestroot (pRExC_state->nestroot)
163 #define RExC_extralen (pRExC_state->extralen)
164 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
165 #define RExC_seen_evals (pRExC_state->seen_evals)
166 #define RExC_utf8 (pRExC_state->utf8)
167 #define RExC_charnames (pRExC_state->charnames)
168 #define RExC_open_parens (pRExC_state->open_parens)
169 #define RExC_close_parens (pRExC_state->close_parens)
170 #define RExC_opend (pRExC_state->opend)
171 #define RExC_paren_names (pRExC_state->paren_names)
172 #define RExC_recurse (pRExC_state->recurse)
173 #define RExC_recurse_count (pRExC_state->recurse_count)
175 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
176 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
177 ((*s) == '{' && regcurly(s)))
180 #undef SPSTART /* dratted cpp namespace... */
183 * Flags to be passed up and down.
185 #define WORST 0 /* Worst case. */
186 #define HASWIDTH 0x1 /* Known to match non-null strings. */
187 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
188 #define SPSTART 0x4 /* Starts with * or +. */
189 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
191 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
193 /* whether trie related optimizations are enabled */
194 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
195 #define TRIE_STUDY_OPT
196 #define FULL_TRIE_STUDY
202 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
203 #define PBITVAL(paren) (1 << ((paren) & 7))
204 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
205 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
206 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
209 /* About scan_data_t.
211 During optimisation we recurse through the regexp program performing
212 various inplace (keyhole style) optimisations. In addition study_chunk
213 and scan_commit populate this data structure with information about
214 what strings MUST appear in the pattern. We look for the longest
215 string that must appear for at a fixed location, and we look for the
216 longest string that may appear at a floating location. So for instance
221 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
222 strings (because they follow a .* construct). study_chunk will identify
223 both FOO and BAR as being the longest fixed and floating strings respectively.
225 The strings can be composites, for instance
229 will result in a composite fixed substring 'foo'.
231 For each string some basic information is maintained:
233 - offset or min_offset
234 This is the position the string must appear at, or not before.
235 It also implicitly (when combined with minlenp) tells us how many
236 character must match before the string we are searching.
237 Likewise when combined with minlenp and the length of the string
238 tells us how many characters must appear after the string we have
242 Only used for floating strings. This is the rightmost point that
243 the string can appear at. Ifset to I32 max it indicates that the
244 string can occur infinitely far to the right.
247 A pointer to the minimum length of the pattern that the string
248 was found inside. This is important as in the case of positive
249 lookahead or positive lookbehind we can have multiple patterns
254 The minimum length of the pattern overall is 3, the minimum length
255 of the lookahead part is 3, but the minimum length of the part that
256 will actually match is 1. So 'FOO's minimum length is 3, but the
257 minimum length for the F is 1. This is important as the minimum length
258 is used to determine offsets in front of and behind the string being
259 looked for. Since strings can be composites this is the length of the
260 pattern at the time it was commited with a scan_commit. Note that
261 the length is calculated by study_chunk, so that the minimum lengths
262 are not known until the full pattern has been compiled, thus the
263 pointer to the value.
267 In the case of lookbehind the string being searched for can be
268 offset past the start point of the final matching string.
269 If this value was just blithely removed from the min_offset it would
270 invalidate some of the calculations for how many chars must match
271 before or after (as they are derived from min_offset and minlen and
272 the length of the string being searched for).
273 When the final pattern is compiled and the data is moved from the
274 scan_data_t structure into the regexp structure the information
275 about lookbehind is factored in, with the information that would
276 have been lost precalculated in the end_shift field for the
279 The fields pos_min and pos_delta are used to store the minimum offset
280 and the delta to the maximum offset at the current point in the pattern.
284 typedef struct scan_data_t {
285 /*I32 len_min; unused */
286 /*I32 len_delta; unused */
290 I32 last_end; /* min value, <0 unless valid. */
293 SV **longest; /* Either &l_fixed, or &l_float. */
294 SV *longest_fixed; /* longest fixed string found in pattern */
295 I32 offset_fixed; /* offset where it starts */
296 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
297 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
298 SV *longest_float; /* longest floating string found in pattern */
299 I32 offset_float_min; /* earliest point in string it can appear */
300 I32 offset_float_max; /* latest point in string it can appear */
301 I32 *minlen_float; /* pointer to the minlen relevent to the string */
302 I32 lookbehind_float; /* is the position of the string modified by LB */
306 struct regnode_charclass_class *start_class;
310 * Forward declarations for pregcomp()'s friends.
313 static const scan_data_t zero_scan_data =
314 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
316 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
317 #define SF_BEFORE_SEOL 0x0001
318 #define SF_BEFORE_MEOL 0x0002
319 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
320 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
323 # define SF_FIX_SHIFT_EOL (0+2)
324 # define SF_FL_SHIFT_EOL (0+4)
326 # define SF_FIX_SHIFT_EOL (+2)
327 # define SF_FL_SHIFT_EOL (+4)
330 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
331 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
333 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
334 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
335 #define SF_IS_INF 0x0040
336 #define SF_HAS_PAR 0x0080
337 #define SF_IN_PAR 0x0100
338 #define SF_HAS_EVAL 0x0200
339 #define SCF_DO_SUBSTR 0x0400
340 #define SCF_DO_STCLASS_AND 0x0800
341 #define SCF_DO_STCLASS_OR 0x1000
342 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
343 #define SCF_WHILEM_VISITED_POS 0x2000
345 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
346 #define SCF_SEEN_ACCEPT 0x8000
348 #define UTF (RExC_utf8 != 0)
349 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
350 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
352 #define OOB_UNICODE 12345678
353 #define OOB_NAMEDCLASS -1
355 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
356 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
359 /* length of regex to show in messages that don't mark a position within */
360 #define RegexLengthToShowInErrorMessages 127
363 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
364 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
365 * op/pragma/warn/regcomp.
367 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
368 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
370 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
373 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
374 * arg. Show regex, up to a maximum length. If it's too long, chop and add
377 #define _FAIL(code) STMT_START { \
378 const char *ellipses = ""; \
379 IV len = RExC_end - RExC_precomp; \
382 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
383 if (len > RegexLengthToShowInErrorMessages) { \
384 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
385 len = RegexLengthToShowInErrorMessages - 10; \
391 #define FAIL(msg) _FAIL( \
392 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
393 msg, (int)len, RExC_precomp, ellipses))
395 #define FAIL2(msg,arg) _FAIL( \
396 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
397 arg, (int)len, RExC_precomp, ellipses))
400 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
402 #define Simple_vFAIL(m) STMT_START { \
403 const IV offset = RExC_parse - RExC_precomp; \
404 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
405 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
409 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
411 #define vFAIL(m) STMT_START { \
413 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
418 * Like Simple_vFAIL(), but accepts two arguments.
420 #define Simple_vFAIL2(m,a1) STMT_START { \
421 const IV offset = RExC_parse - RExC_precomp; \
422 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
423 (int)offset, RExC_precomp, RExC_precomp + offset); \
427 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
429 #define vFAIL2(m,a1) STMT_START { \
431 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
432 Simple_vFAIL2(m, a1); \
437 * Like Simple_vFAIL(), but accepts three arguments.
439 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
440 const IV offset = RExC_parse - RExC_precomp; \
441 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
442 (int)offset, RExC_precomp, RExC_precomp + offset); \
446 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
448 #define vFAIL3(m,a1,a2) STMT_START { \
450 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
451 Simple_vFAIL3(m, a1, a2); \
455 * Like Simple_vFAIL(), but accepts four arguments.
457 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
458 const IV offset = RExC_parse - RExC_precomp; \
459 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
460 (int)offset, RExC_precomp, RExC_precomp + offset); \
463 #define vWARN(loc,m) STMT_START { \
464 const IV offset = loc - RExC_precomp; \
465 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
466 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
469 #define vWARNdep(loc,m) STMT_START { \
470 const IV offset = loc - RExC_precomp; \
471 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
472 "%s" REPORT_LOCATION, \
473 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
477 #define vWARN2(loc, m, a1) STMT_START { \
478 const IV offset = loc - RExC_precomp; \
479 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
480 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
483 #define vWARN3(loc, m, a1, a2) STMT_START { \
484 const IV offset = loc - RExC_precomp; \
485 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
486 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
489 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
490 const IV offset = loc - RExC_precomp; \
491 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
492 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
495 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
496 const IV offset = loc - RExC_precomp; \
497 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
498 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
502 /* Allow for side effects in s */
503 #define REGC(c,s) STMT_START { \
504 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
507 /* Macros for recording node offsets. 20001227 mjd@plover.com
508 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
509 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
510 * Element 0 holds the number n.
511 * Position is 1 indexed.
514 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
516 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
517 __LINE__, (int)(node), (int)(byte))); \
519 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
521 RExC_offsets[2*(node)-1] = (byte); \
526 #define Set_Node_Offset(node,byte) \
527 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
528 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
530 #define Set_Node_Length_To_R(node,len) STMT_START { \
532 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
533 __LINE__, (int)(node), (int)(len))); \
535 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
537 RExC_offsets[2*(node)] = (len); \
542 #define Set_Node_Length(node,len) \
543 Set_Node_Length_To_R((node)-RExC_emit_start, len)
544 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
545 #define Set_Node_Cur_Length(node) \
546 Set_Node_Length(node, RExC_parse - parse_start)
548 /* Get offsets and lengths */
549 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
550 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
552 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
553 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
554 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
558 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
559 #define EXPERIMENTAL_INPLACESCAN
562 #define DEBUG_STUDYDATA(str,data,depth) \
563 DEBUG_OPTIMISE_MORE_r(if(data){ \
564 PerlIO_printf(Perl_debug_log, \
565 "%*s" str "Pos:%"IVdf"/%"IVdf \
566 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
567 (int)(depth)*2, "", \
568 (IV)((data)->pos_min), \
569 (IV)((data)->pos_delta), \
570 (UV)((data)->flags), \
571 (IV)((data)->whilem_c), \
572 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
573 is_inf ? "INF " : "" \
575 if ((data)->last_found) \
576 PerlIO_printf(Perl_debug_log, \
577 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
578 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
579 SvPVX_const((data)->last_found), \
580 (IV)((data)->last_end), \
581 (IV)((data)->last_start_min), \
582 (IV)((data)->last_start_max), \
583 ((data)->longest && \
584 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
585 SvPVX_const((data)->longest_fixed), \
586 (IV)((data)->offset_fixed), \
587 ((data)->longest && \
588 (data)->longest==&((data)->longest_float)) ? "*" : "", \
589 SvPVX_const((data)->longest_float), \
590 (IV)((data)->offset_float_min), \
591 (IV)((data)->offset_float_max) \
593 PerlIO_printf(Perl_debug_log,"\n"); \
596 static void clear_re(pTHX_ void *r);
598 /* Mark that we cannot extend a found fixed substring at this point.
599 Update the longest found anchored substring and the longest found
600 floating substrings if needed. */
603 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
605 const STRLEN l = CHR_SVLEN(data->last_found);
606 const STRLEN old_l = CHR_SVLEN(*data->longest);
607 GET_RE_DEBUG_FLAGS_DECL;
609 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
610 SvSetMagicSV(*data->longest, data->last_found);
611 if (*data->longest == data->longest_fixed) {
612 data->offset_fixed = l ? data->last_start_min : data->pos_min;
613 if (data->flags & SF_BEFORE_EOL)
615 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
617 data->flags &= ~SF_FIX_BEFORE_EOL;
618 data->minlen_fixed=minlenp;
619 data->lookbehind_fixed=0;
621 else { /* *data->longest == data->longest_float */
622 data->offset_float_min = l ? data->last_start_min : data->pos_min;
623 data->offset_float_max = (l
624 ? data->last_start_max
625 : data->pos_min + data->pos_delta);
626 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
627 data->offset_float_max = I32_MAX;
628 if (data->flags & SF_BEFORE_EOL)
630 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
632 data->flags &= ~SF_FL_BEFORE_EOL;
633 data->minlen_float=minlenp;
634 data->lookbehind_float=0;
637 SvCUR_set(data->last_found, 0);
639 SV * const sv = data->last_found;
640 if (SvUTF8(sv) && SvMAGICAL(sv)) {
641 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
647 data->flags &= ~SF_BEFORE_EOL;
648 DEBUG_STUDYDATA("cl_anything: ",data,0);
651 /* Can match anything (initialization) */
653 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
655 ANYOF_CLASS_ZERO(cl);
656 ANYOF_BITMAP_SETALL(cl);
657 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
659 cl->flags |= ANYOF_LOCALE;
662 /* Can match anything (initialization) */
664 S_cl_is_anything(const struct regnode_charclass_class *cl)
668 for (value = 0; value <= ANYOF_MAX; value += 2)
669 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
671 if (!(cl->flags & ANYOF_UNICODE_ALL))
673 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
678 /* Can match anything (initialization) */
680 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
682 Zero(cl, 1, struct regnode_charclass_class);
684 cl_anything(pRExC_state, cl);
688 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
690 Zero(cl, 1, struct regnode_charclass_class);
692 cl_anything(pRExC_state, cl);
694 cl->flags |= ANYOF_LOCALE;
697 /* 'And' a given class with another one. Can create false positives */
698 /* We assume that cl is not inverted */
700 S_cl_and(struct regnode_charclass_class *cl,
701 const struct regnode_charclass_class *and_with)
704 assert(and_with->type == ANYOF);
705 if (!(and_with->flags & ANYOF_CLASS)
706 && !(cl->flags & ANYOF_CLASS)
707 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
708 && !(and_with->flags & ANYOF_FOLD)
709 && !(cl->flags & ANYOF_FOLD)) {
712 if (and_with->flags & ANYOF_INVERT)
713 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
714 cl->bitmap[i] &= ~and_with->bitmap[i];
716 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
717 cl->bitmap[i] &= and_with->bitmap[i];
718 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
719 if (!(and_with->flags & ANYOF_EOS))
720 cl->flags &= ~ANYOF_EOS;
722 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
723 !(and_with->flags & ANYOF_INVERT)) {
724 cl->flags &= ~ANYOF_UNICODE_ALL;
725 cl->flags |= ANYOF_UNICODE;
726 ARG_SET(cl, ARG(and_with));
728 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
729 !(and_with->flags & ANYOF_INVERT))
730 cl->flags &= ~ANYOF_UNICODE_ALL;
731 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
732 !(and_with->flags & ANYOF_INVERT))
733 cl->flags &= ~ANYOF_UNICODE;
736 /* 'OR' a given class with another one. Can create false positives */
737 /* We assume that cl is not inverted */
739 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
741 if (or_with->flags & ANYOF_INVERT) {
743 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
744 * <= (B1 | !B2) | (CL1 | !CL2)
745 * which is wasteful if CL2 is small, but we ignore CL2:
746 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
747 * XXXX Can we handle case-fold? Unclear:
748 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
749 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
751 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
752 && !(or_with->flags & ANYOF_FOLD)
753 && !(cl->flags & ANYOF_FOLD) ) {
756 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
757 cl->bitmap[i] |= ~or_with->bitmap[i];
758 } /* XXXX: logic is complicated otherwise */
760 cl_anything(pRExC_state, cl);
763 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
764 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
765 && (!(or_with->flags & ANYOF_FOLD)
766 || (cl->flags & ANYOF_FOLD)) ) {
769 /* OR char bitmap and class bitmap separately */
770 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
771 cl->bitmap[i] |= or_with->bitmap[i];
772 if (or_with->flags & ANYOF_CLASS) {
773 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
774 cl->classflags[i] |= or_with->classflags[i];
775 cl->flags |= ANYOF_CLASS;
778 else { /* XXXX: logic is complicated, leave it along for a moment. */
779 cl_anything(pRExC_state, cl);
782 if (or_with->flags & ANYOF_EOS)
783 cl->flags |= ANYOF_EOS;
785 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
786 ARG(cl) != ARG(or_with)) {
787 cl->flags |= ANYOF_UNICODE_ALL;
788 cl->flags &= ~ANYOF_UNICODE;
790 if (or_with->flags & ANYOF_UNICODE_ALL) {
791 cl->flags |= ANYOF_UNICODE_ALL;
792 cl->flags &= ~ANYOF_UNICODE;
796 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
797 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
798 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
799 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
804 dump_trie(trie,widecharmap,revcharmap)
805 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
806 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
808 These routines dump out a trie in a somewhat readable format.
809 The _interim_ variants are used for debugging the interim
810 tables that are used to generate the final compressed
811 representation which is what dump_trie expects.
813 Part of the reason for their existance is to provide a form
814 of documentation as to how the different representations function.
819 Dumps the final compressed table form of the trie to Perl_debug_log.
820 Used for debugging make_trie().
824 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
825 AV *revcharmap, U32 depth)
828 SV *sv=sv_newmortal();
829 int colwidth= widecharmap ? 6 : 4;
830 GET_RE_DEBUG_FLAGS_DECL;
833 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
834 (int)depth * 2 + 2,"",
835 "Match","Base","Ofs" );
837 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
838 SV ** const tmp = av_fetch( revcharmap, state, 0);
840 PerlIO_printf( Perl_debug_log, "%*s",
842 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
843 PL_colors[0], PL_colors[1],
844 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
845 PERL_PV_ESCAPE_FIRSTCHAR
850 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
851 (int)depth * 2 + 2,"");
853 for( state = 0 ; state < trie->uniquecharcount ; state++ )
854 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
855 PerlIO_printf( Perl_debug_log, "\n");
857 for( state = 1 ; state < trie->statecount ; state++ ) {
858 const U32 base = trie->states[ state ].trans.base;
860 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
862 if ( trie->states[ state ].wordnum ) {
863 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
865 PerlIO_printf( Perl_debug_log, "%6s", "" );
868 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
873 while( ( base + ofs < trie->uniquecharcount ) ||
874 ( base + ofs - trie->uniquecharcount < trie->lasttrans
875 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
878 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
880 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
881 if ( ( base + ofs >= trie->uniquecharcount ) &&
882 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
883 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
885 PerlIO_printf( Perl_debug_log, "%*"UVXf,
887 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
889 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
893 PerlIO_printf( Perl_debug_log, "]");
896 PerlIO_printf( Perl_debug_log, "\n" );
900 Dumps a fully constructed but uncompressed trie in list form.
901 List tries normally only are used for construction when the number of
902 possible chars (trie->uniquecharcount) is very high.
903 Used for debugging make_trie().
906 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
907 HV *widecharmap, AV *revcharmap, U32 next_alloc,
911 SV *sv=sv_newmortal();
912 int colwidth= widecharmap ? 6 : 4;
913 GET_RE_DEBUG_FLAGS_DECL;
914 /* print out the table precompression. */
915 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
916 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
917 "------:-----+-----------------\n" );
919 for( state=1 ; state < next_alloc ; state ++ ) {
922 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
923 (int)depth * 2 + 2,"", (UV)state );
924 if ( ! trie->states[ state ].wordnum ) {
925 PerlIO_printf( Perl_debug_log, "%5s| ","");
927 PerlIO_printf( Perl_debug_log, "W%4x| ",
928 trie->states[ state ].wordnum
931 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
932 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
934 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
936 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
937 PL_colors[0], PL_colors[1],
938 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
939 PERL_PV_ESCAPE_FIRSTCHAR
941 TRIE_LIST_ITEM(state,charid).forid,
942 (UV)TRIE_LIST_ITEM(state,charid).newstate
945 PerlIO_printf(Perl_debug_log, "\n%*s| ",
946 (int)((depth * 2) + 14), "");
949 PerlIO_printf( Perl_debug_log, "\n");
954 Dumps a fully constructed but uncompressed trie in table form.
955 This is the normal DFA style state transition table, with a few
956 twists to facilitate compression later.
957 Used for debugging make_trie().
960 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
961 HV *widecharmap, AV *revcharmap, U32 next_alloc,
966 SV *sv=sv_newmortal();
967 int colwidth= widecharmap ? 6 : 4;
968 GET_RE_DEBUG_FLAGS_DECL;
971 print out the table precompression so that we can do a visual check
972 that they are identical.
975 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
977 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
978 SV ** const tmp = av_fetch( revcharmap, charid, 0);
980 PerlIO_printf( Perl_debug_log, "%*s",
982 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
983 PL_colors[0], PL_colors[1],
984 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
985 PERL_PV_ESCAPE_FIRSTCHAR
991 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
993 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
994 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
997 PerlIO_printf( Perl_debug_log, "\n" );
999 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1001 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1002 (int)depth * 2 + 2,"",
1003 (UV)TRIE_NODENUM( state ) );
1005 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1006 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1008 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1010 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1012 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1013 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1015 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1016 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1023 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1024 startbranch: the first branch in the whole branch sequence
1025 first : start branch of sequence of branch-exact nodes.
1026 May be the same as startbranch
1027 last : Thing following the last branch.
1028 May be the same as tail.
1029 tail : item following the branch sequence
1030 count : words in the sequence
1031 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1032 depth : indent depth
1034 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1036 A trie is an N'ary tree where the branches are determined by digital
1037 decomposition of the key. IE, at the root node you look up the 1st character and
1038 follow that branch repeat until you find the end of the branches. Nodes can be
1039 marked as "accepting" meaning they represent a complete word. Eg:
1043 would convert into the following structure. Numbers represent states, letters
1044 following numbers represent valid transitions on the letter from that state, if
1045 the number is in square brackets it represents an accepting state, otherwise it
1046 will be in parenthesis.
1048 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1052 (1) +-i->(6)-+-s->[7]
1054 +-s->(3)-+-h->(4)-+-e->[5]
1056 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1058 This shows that when matching against the string 'hers' we will begin at state 1
1059 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1060 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1061 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1062 single traverse. We store a mapping from accepting to state to which word was
1063 matched, and then when we have multiple possibilities we try to complete the
1064 rest of the regex in the order in which they occured in the alternation.
1066 The only prior NFA like behaviour that would be changed by the TRIE support is
1067 the silent ignoring of duplicate alternations which are of the form:
1069 / (DUPE|DUPE) X? (?{ ... }) Y /x
1071 Thus EVAL blocks follwing a trie may be called a different number of times with
1072 and without the optimisation. With the optimisations dupes will be silently
1073 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1074 the following demonstrates:
1076 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1078 which prints out 'word' three times, but
1080 'words'=~/(word|word|word)(?{ print $1 })S/
1082 which doesnt print it out at all. This is due to other optimisations kicking in.
1084 Example of what happens on a structural level:
1086 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1088 1: CURLYM[1] {1,32767}(18)
1099 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1100 and should turn into:
1102 1: CURLYM[1] {1,32767}(18)
1104 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1112 Cases where tail != last would be like /(?foo|bar)baz/:
1122 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1123 and would end up looking like:
1126 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1133 d = uvuni_to_utf8_flags(d, uv, 0);
1135 is the recommended Unicode-aware way of saying
1140 #define TRIE_STORE_REVCHAR \
1142 SV *tmp = newSVpvs(""); \
1143 if (UTF) SvUTF8_on(tmp); \
1144 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1145 av_push( revcharmap, tmp ); \
1148 #define TRIE_READ_CHAR STMT_START { \
1152 if ( foldlen > 0 ) { \
1153 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1158 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1159 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1160 foldlen -= UNISKIP( uvc ); \
1161 scan = foldbuf + UNISKIP( uvc ); \
1164 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1174 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1175 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1176 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1177 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1179 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1180 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1181 TRIE_LIST_CUR( state )++; \
1184 #define TRIE_LIST_NEW(state) STMT_START { \
1185 Newxz( trie->states[ state ].trans.list, \
1186 4, reg_trie_trans_le ); \
1187 TRIE_LIST_CUR( state ) = 1; \
1188 TRIE_LIST_LEN( state ) = 4; \
1191 #define TRIE_HANDLE_WORD(state) STMT_START { \
1192 U16 dupe= trie->states[ state ].wordnum; \
1193 regnode * const noper_next = regnext( noper ); \
1195 if (trie->wordlen) \
1196 trie->wordlen[ curword ] = wordlen; \
1198 /* store the word for dumping */ \
1200 if (OP(noper) != NOTHING) \
1201 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1203 tmp = newSVpvn( "", 0 ); \
1204 if ( UTF ) SvUTF8_on( tmp ); \
1205 av_push( trie_words, tmp ); \
1210 if ( noper_next < tail ) { \
1212 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1213 trie->jump[curword] = (U16)(noper_next - convert); \
1215 jumper = noper_next; \
1217 nextbranch= regnext(cur); \
1221 /* So it's a dupe. This means we need to maintain a */\
1222 /* linked-list from the first to the next. */\
1223 /* we only allocate the nextword buffer when there */\
1224 /* a dupe, so first time we have to do the allocation */\
1225 if (!trie->nextword) \
1226 trie->nextword = (U16 *) \
1227 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1228 while ( trie->nextword[dupe] ) \
1229 dupe= trie->nextword[dupe]; \
1230 trie->nextword[dupe]= curword; \
1232 /* we haven't inserted this word yet. */ \
1233 trie->states[ state ].wordnum = curword; \
1238 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1239 ( ( base + charid >= ucharcount \
1240 && base + charid < ubound \
1241 && state == trie->trans[ base - ucharcount + charid ].check \
1242 && trie->trans[ base - ucharcount + charid ].next ) \
1243 ? trie->trans[ base - ucharcount + charid ].next \
1244 : ( state==1 ? special : 0 ) \
1248 #define MADE_JUMP_TRIE 2
1249 #define MADE_EXACT_TRIE 4
1252 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1255 /* first pass, loop through and scan words */
1256 reg_trie_data *trie;
1257 HV *widecharmap = NULL;
1258 AV *revcharmap = newAV();
1260 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1265 regnode *jumper = NULL;
1266 regnode *nextbranch = NULL;
1267 regnode *convert = NULL;
1268 /* we just use folder as a flag in utf8 */
1269 const U8 * const folder = ( flags == EXACTF
1271 : ( flags == EXACTFL
1278 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1279 AV *trie_words = NULL;
1280 /* along with revcharmap, this only used during construction but both are
1281 * useful during debugging so we store them in the struct when debugging.
1284 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1285 STRLEN trie_charcount=0;
1287 SV *re_trie_maxbuff;
1288 GET_RE_DEBUG_FLAGS_DECL;
1290 PERL_UNUSED_ARG(depth);
1293 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1295 trie->startstate = 1;
1296 trie->wordcount = word_count;
1297 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1298 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1299 if (!(UTF && folder))
1300 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1302 trie_words = newAV();
1305 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1306 if (!SvIOK(re_trie_maxbuff)) {
1307 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1310 PerlIO_printf( Perl_debug_log,
1311 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1312 (int)depth * 2 + 2, "",
1313 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1314 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1318 /* Find the node we are going to overwrite */
1319 if ( first == startbranch && OP( last ) != BRANCH ) {
1320 /* whole branch chain */
1323 /* branch sub-chain */
1324 convert = NEXTOPER( first );
1327 /* -- First loop and Setup --
1329 We first traverse the branches and scan each word to determine if it
1330 contains widechars, and how many unique chars there are, this is
1331 important as we have to build a table with at least as many columns as we
1334 We use an array of integers to represent the character codes 0..255
1335 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1336 native representation of the character value as the key and IV's for the
1339 *TODO* If we keep track of how many times each character is used we can
1340 remap the columns so that the table compression later on is more
1341 efficient in terms of memory by ensuring most common value is in the
1342 middle and the least common are on the outside. IMO this would be better
1343 than a most to least common mapping as theres a decent chance the most
1344 common letter will share a node with the least common, meaning the node
1345 will not be compressable. With a middle is most common approach the worst
1346 case is when we have the least common nodes twice.
1350 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1351 regnode * const noper = NEXTOPER( cur );
1352 const U8 *uc = (U8*)STRING( noper );
1353 const U8 * const e = uc + STR_LEN( noper );
1355 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1356 const U8 *scan = (U8*)NULL;
1357 U32 wordlen = 0; /* required init */
1360 if (OP(noper) == NOTHING) {
1365 TRIE_BITMAP_SET(trie,*uc);
1366 if ( folder ) TRIE_BITMAP_SET(trie,folder[ *uc ]);
1368 for ( ; uc < e ; uc += len ) {
1369 TRIE_CHARCOUNT(trie)++;
1373 if ( !trie->charmap[ uvc ] ) {
1374 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1376 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1382 widecharmap = newHV();
1384 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1387 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1389 if ( !SvTRUE( *svpp ) ) {
1390 sv_setiv( *svpp, ++trie->uniquecharcount );
1395 if( cur == first ) {
1398 } else if (chars < trie->minlen) {
1400 } else if (chars > trie->maxlen) {
1404 } /* end first pass */
1405 DEBUG_TRIE_COMPILE_r(
1406 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1407 (int)depth * 2 + 2,"",
1408 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1409 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1410 (int)trie->minlen, (int)trie->maxlen )
1412 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1415 We now know what we are dealing with in terms of unique chars and
1416 string sizes so we can calculate how much memory a naive
1417 representation using a flat table will take. If it's over a reasonable
1418 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1419 conservative but potentially much slower representation using an array
1422 At the end we convert both representations into the same compressed
1423 form that will be used in regexec.c for matching with. The latter
1424 is a form that cannot be used to construct with but has memory
1425 properties similar to the list form and access properties similar
1426 to the table form making it both suitable for fast searches and
1427 small enough that its feasable to store for the duration of a program.
1429 See the comment in the code where the compressed table is produced
1430 inplace from the flat tabe representation for an explanation of how
1431 the compression works.
1436 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1438 Second Pass -- Array Of Lists Representation
1440 Each state will be represented by a list of charid:state records
1441 (reg_trie_trans_le) the first such element holds the CUR and LEN
1442 points of the allocated array. (See defines above).
1444 We build the initial structure using the lists, and then convert
1445 it into the compressed table form which allows faster lookups
1446 (but cant be modified once converted).
1449 STRLEN transcount = 1;
1451 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1452 "%*sCompiling trie using list compiler\n",
1453 (int)depth * 2 + 2, ""));
1455 trie->states = (reg_trie_state *)
1456 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1457 sizeof(reg_trie_state) );
1461 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1463 regnode * const noper = NEXTOPER( cur );
1464 U8 *uc = (U8*)STRING( noper );
1465 const U8 * const e = uc + STR_LEN( noper );
1466 U32 state = 1; /* required init */
1467 U16 charid = 0; /* sanity init */
1468 U8 *scan = (U8*)NULL; /* sanity init */
1469 STRLEN foldlen = 0; /* required init */
1470 U32 wordlen = 0; /* required init */
1471 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1473 if (OP(noper) != NOTHING) {
1474 for ( ; uc < e ; uc += len ) {
1479 charid = trie->charmap[ uvc ];
1481 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1485 charid=(U16)SvIV( *svpp );
1488 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1495 if ( !trie->states[ state ].trans.list ) {
1496 TRIE_LIST_NEW( state );
1498 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1499 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1500 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1505 newstate = next_alloc++;
1506 TRIE_LIST_PUSH( state, charid, newstate );
1511 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1515 TRIE_HANDLE_WORD(state);
1517 } /* end second pass */
1519 /* next alloc is the NEXT state to be allocated */
1520 trie->statecount = next_alloc;
1521 trie->states = (reg_trie_state *)
1522 PerlMemShared_realloc( trie->states,
1524 * sizeof(reg_trie_state) );
1526 /* and now dump it out before we compress it */
1527 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1528 revcharmap, next_alloc,
1532 trie->trans = (reg_trie_trans *)
1533 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1540 for( state=1 ; state < next_alloc ; state ++ ) {
1544 DEBUG_TRIE_COMPILE_MORE_r(
1545 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1549 if (trie->states[state].trans.list) {
1550 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1554 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1555 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1556 if ( forid < minid ) {
1558 } else if ( forid > maxid ) {
1562 if ( transcount < tp + maxid - minid + 1) {
1564 trie->trans = (reg_trie_trans *)
1565 PerlMemShared_realloc( trie->trans,
1567 * sizeof(reg_trie_trans) );
1568 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1570 base = trie->uniquecharcount + tp - minid;
1571 if ( maxid == minid ) {
1573 for ( ; zp < tp ; zp++ ) {
1574 if ( ! trie->trans[ zp ].next ) {
1575 base = trie->uniquecharcount + zp - minid;
1576 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1577 trie->trans[ zp ].check = state;
1583 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1584 trie->trans[ tp ].check = state;
1589 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1590 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1591 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1592 trie->trans[ tid ].check = state;
1594 tp += ( maxid - minid + 1 );
1596 Safefree(trie->states[ state ].trans.list);
1599 DEBUG_TRIE_COMPILE_MORE_r(
1600 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1603 trie->states[ state ].trans.base=base;
1605 trie->lasttrans = tp + 1;
1609 Second Pass -- Flat Table Representation.
1611 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1612 We know that we will need Charcount+1 trans at most to store the data
1613 (one row per char at worst case) So we preallocate both structures
1614 assuming worst case.
1616 We then construct the trie using only the .next slots of the entry
1619 We use the .check field of the first entry of the node temporarily to
1620 make compression both faster and easier by keeping track of how many non
1621 zero fields are in the node.
1623 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1626 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1627 number representing the first entry of the node, and state as a
1628 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1629 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1630 are 2 entrys per node. eg:
1638 The table is internally in the right hand, idx form. However as we also
1639 have to deal with the states array which is indexed by nodenum we have to
1640 use TRIE_NODENUM() to convert.
1643 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1644 "%*sCompiling trie using table compiler\n",
1645 (int)depth * 2 + 2, ""));
1647 trie->trans = (reg_trie_trans *)
1648 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1649 * trie->uniquecharcount + 1,
1650 sizeof(reg_trie_trans) );
1651 trie->states = (reg_trie_state *)
1652 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1653 sizeof(reg_trie_state) );
1654 next_alloc = trie->uniquecharcount + 1;
1657 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1659 regnode * const noper = NEXTOPER( cur );
1660 const U8 *uc = (U8*)STRING( noper );
1661 const U8 * const e = uc + STR_LEN( noper );
1663 U32 state = 1; /* required init */
1665 U16 charid = 0; /* sanity init */
1666 U32 accept_state = 0; /* sanity init */
1667 U8 *scan = (U8*)NULL; /* sanity init */
1669 STRLEN foldlen = 0; /* required init */
1670 U32 wordlen = 0; /* required init */
1671 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1673 if ( OP(noper) != NOTHING ) {
1674 for ( ; uc < e ; uc += len ) {
1679 charid = trie->charmap[ uvc ];
1681 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1682 charid = svpp ? (U16)SvIV(*svpp) : 0;
1686 if ( !trie->trans[ state + charid ].next ) {
1687 trie->trans[ state + charid ].next = next_alloc;
1688 trie->trans[ state ].check++;
1689 next_alloc += trie->uniquecharcount;
1691 state = trie->trans[ state + charid ].next;
1693 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1695 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1698 accept_state = TRIE_NODENUM( state );
1699 TRIE_HANDLE_WORD(accept_state);
1701 } /* end second pass */
1703 /* and now dump it out before we compress it */
1704 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1706 next_alloc, depth+1));
1710 * Inplace compress the table.*
1712 For sparse data sets the table constructed by the trie algorithm will
1713 be mostly 0/FAIL transitions or to put it another way mostly empty.
1714 (Note that leaf nodes will not contain any transitions.)
1716 This algorithm compresses the tables by eliminating most such
1717 transitions, at the cost of a modest bit of extra work during lookup:
1719 - Each states[] entry contains a .base field which indicates the
1720 index in the state[] array wheres its transition data is stored.
1722 - If .base is 0 there are no valid transitions from that node.
1724 - If .base is nonzero then charid is added to it to find an entry in
1727 -If trans[states[state].base+charid].check!=state then the
1728 transition is taken to be a 0/Fail transition. Thus if there are fail
1729 transitions at the front of the node then the .base offset will point
1730 somewhere inside the previous nodes data (or maybe even into a node
1731 even earlier), but the .check field determines if the transition is
1735 The following process inplace converts the table to the compressed
1736 table: We first do not compress the root node 1,and mark its all its
1737 .check pointers as 1 and set its .base pointer as 1 as well. This
1738 allows to do a DFA construction from the compressed table later, and
1739 ensures that any .base pointers we calculate later are greater than
1742 - We set 'pos' to indicate the first entry of the second node.
1744 - We then iterate over the columns of the node, finding the first and
1745 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1746 and set the .check pointers accordingly, and advance pos
1747 appropriately and repreat for the next node. Note that when we copy
1748 the next pointers we have to convert them from the original
1749 NODEIDX form to NODENUM form as the former is not valid post
1752 - If a node has no transitions used we mark its base as 0 and do not
1753 advance the pos pointer.
1755 - If a node only has one transition we use a second pointer into the
1756 structure to fill in allocated fail transitions from other states.
1757 This pointer is independent of the main pointer and scans forward
1758 looking for null transitions that are allocated to a state. When it
1759 finds one it writes the single transition into the "hole". If the
1760 pointer doesnt find one the single transition is appended as normal.
1762 - Once compressed we can Renew/realloc the structures to release the
1765 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1766 specifically Fig 3.47 and the associated pseudocode.
1770 const U32 laststate = TRIE_NODENUM( next_alloc );
1773 trie->statecount = laststate;
1775 for ( state = 1 ; state < laststate ; state++ ) {
1777 const U32 stateidx = TRIE_NODEIDX( state );
1778 const U32 o_used = trie->trans[ stateidx ].check;
1779 U32 used = trie->trans[ stateidx ].check;
1780 trie->trans[ stateidx ].check = 0;
1782 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1783 if ( flag || trie->trans[ stateidx + charid ].next ) {
1784 if ( trie->trans[ stateidx + charid ].next ) {
1786 for ( ; zp < pos ; zp++ ) {
1787 if ( ! trie->trans[ zp ].next ) {
1791 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1792 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1793 trie->trans[ zp ].check = state;
1794 if ( ++zp > pos ) pos = zp;
1801 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1803 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1804 trie->trans[ pos ].check = state;
1809 trie->lasttrans = pos + 1;
1810 trie->states = (reg_trie_state *)
1811 PerlMemShared_realloc( trie->states, laststate
1812 * sizeof(reg_trie_state) );
1813 DEBUG_TRIE_COMPILE_MORE_r(
1814 PerlIO_printf( Perl_debug_log,
1815 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1816 (int)depth * 2 + 2,"",
1817 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1820 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1823 } /* end table compress */
1825 DEBUG_TRIE_COMPILE_MORE_r(
1826 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1827 (int)depth * 2 + 2, "",
1828 (UV)trie->statecount,
1829 (UV)trie->lasttrans)
1831 /* resize the trans array to remove unused space */
1832 trie->trans = (reg_trie_trans *)
1833 PerlMemShared_realloc( trie->trans, trie->lasttrans
1834 * sizeof(reg_trie_trans) );
1836 /* and now dump out the compressed format */
1837 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1839 { /* Modify the program and insert the new TRIE node*/
1840 U8 nodetype =(U8)(flags & 0xFF);
1844 regnode *optimize = NULL;
1846 U32 mjd_nodelen = 0;
1849 This means we convert either the first branch or the first Exact,
1850 depending on whether the thing following (in 'last') is a branch
1851 or not and whther first is the startbranch (ie is it a sub part of
1852 the alternation or is it the whole thing.)
1853 Assuming its a sub part we conver the EXACT otherwise we convert
1854 the whole branch sequence, including the first.
1856 /* Find the node we are going to overwrite */
1857 if ( first != startbranch || OP( last ) == BRANCH ) {
1858 /* branch sub-chain */
1859 NEXT_OFF( first ) = (U16)(last - first);
1861 mjd_offset= Node_Offset((convert));
1862 mjd_nodelen= Node_Length((convert));
1864 /* whole branch chain */
1867 const regnode *nop = NEXTOPER( convert );
1868 mjd_offset= Node_Offset((nop));
1869 mjd_nodelen= Node_Length((nop));
1874 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1875 (int)depth * 2 + 2, "",
1876 (UV)mjd_offset, (UV)mjd_nodelen)
1879 /* But first we check to see if there is a common prefix we can
1880 split out as an EXACT and put in front of the TRIE node. */
1881 trie->startstate= 1;
1882 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1884 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1888 const U32 base = trie->states[ state ].trans.base;
1890 if ( trie->states[state].wordnum )
1893 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1894 if ( ( base + ofs >= trie->uniquecharcount ) &&
1895 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1896 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1898 if ( ++count > 1 ) {
1899 SV **tmp = av_fetch( revcharmap, ofs, 0);
1900 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1901 if ( state == 1 ) break;
1903 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1905 PerlIO_printf(Perl_debug_log,
1906 "%*sNew Start State=%"UVuf" Class: [",
1907 (int)depth * 2 + 2, "",
1910 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1911 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1913 TRIE_BITMAP_SET(trie,*ch);
1915 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1917 PerlIO_printf(Perl_debug_log, (char*)ch)
1921 TRIE_BITMAP_SET(trie,*ch);
1923 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1924 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1930 SV **tmp = av_fetch( revcharmap, idx, 0);
1931 char *ch = SvPV_nolen( *tmp );
1933 SV *sv=sv_newmortal();
1934 PerlIO_printf( Perl_debug_log,
1935 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1936 (int)depth * 2 + 2, "",
1938 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1939 PL_colors[0], PL_colors[1],
1940 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1941 PERL_PV_ESCAPE_FIRSTCHAR
1946 OP( convert ) = nodetype;
1947 str=STRING(convert);
1958 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
1964 regnode *n = convert+NODE_SZ_STR(convert);
1965 NEXT_OFF(convert) = NODE_SZ_STR(convert);
1966 trie->startstate = state;
1967 trie->minlen -= (state - 1);
1968 trie->maxlen -= (state - 1);
1970 regnode *fix = convert;
1971 U32 word = trie->wordcount;
1973 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
1974 while( ++fix < n ) {
1975 Set_Node_Offset_Length(fix, 0, 0);
1978 SV ** const tmp = av_fetch( trie_words, word, 0 );
1980 if ( STR_LEN(convert) <= SvCUR(*tmp) )
1981 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
1983 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
1990 NEXT_OFF(convert) = (U16)(tail - convert);
1991 DEBUG_r(optimize= n);
1997 if ( trie->maxlen ) {
1998 NEXT_OFF( convert ) = (U16)(tail - convert);
1999 ARG_SET( convert, data_slot );
2000 /* Store the offset to the first unabsorbed branch in
2001 jump[0], which is otherwise unused by the jump logic.
2002 We use this when dumping a trie and during optimisation. */
2004 trie->jump[0] = (U16)(nextbranch - convert);
2007 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2008 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2010 OP( convert ) = TRIEC;
2011 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2012 PerlMemShared_free(trie->bitmap);
2015 OP( convert ) = TRIE;
2017 /* store the type in the flags */
2018 convert->flags = nodetype;
2022 + regarglen[ OP( convert ) ];
2024 /* XXX We really should free up the resource in trie now,
2025 as we won't use them - (which resources?) dmq */
2027 /* needed for dumping*/
2028 DEBUG_r(if (optimize) {
2029 regnode *opt = convert;
2030 while ( ++opt < optimize) {
2031 Set_Node_Offset_Length(opt,0,0);
2034 Try to clean up some of the debris left after the
2037 while( optimize < jumper ) {
2038 mjd_nodelen += Node_Length((optimize));
2039 OP( optimize ) = OPTIMIZED;
2040 Set_Node_Offset_Length(optimize,0,0);
2043 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2045 } /* end node insert */
2046 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2048 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2049 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2051 SvREFCNT_dec(revcharmap);
2055 : trie->startstate>1
2061 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2063 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2065 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2066 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2069 We find the fail state for each state in the trie, this state is the longest proper
2070 suffix of the current states 'word' that is also a proper prefix of another word in our
2071 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2072 the DFA not to have to restart after its tried and failed a word at a given point, it
2073 simply continues as though it had been matching the other word in the first place.
2075 'abcdgu'=~/abcdefg|cdgu/
2076 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2077 fail, which would bring use to the state representing 'd' in the second word where we would
2078 try 'g' and succeed, prodceding to match 'cdgu'.
2080 /* add a fail transition */
2081 const U32 trie_offset = ARG(source);
2082 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2084 const U32 ucharcount = trie->uniquecharcount;
2085 const U32 numstates = trie->statecount;
2086 const U32 ubound = trie->lasttrans + ucharcount;
2090 U32 base = trie->states[ 1 ].trans.base;
2093 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2094 GET_RE_DEBUG_FLAGS_DECL;
2096 PERL_UNUSED_ARG(depth);
2100 ARG_SET( stclass, data_slot );
2101 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2102 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2103 aho->trie=trie_offset;
2104 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2105 Copy( trie->states, aho->states, numstates, reg_trie_state );
2106 Newxz( q, numstates, U32);
2107 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2110 /* initialize fail[0..1] to be 1 so that we always have
2111 a valid final fail state */
2112 fail[ 0 ] = fail[ 1 ] = 1;
2114 for ( charid = 0; charid < ucharcount ; charid++ ) {
2115 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2117 q[ q_write ] = newstate;
2118 /* set to point at the root */
2119 fail[ q[ q_write++ ] ]=1;
2122 while ( q_read < q_write) {
2123 const U32 cur = q[ q_read++ % numstates ];
2124 base = trie->states[ cur ].trans.base;
2126 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2127 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2129 U32 fail_state = cur;
2132 fail_state = fail[ fail_state ];
2133 fail_base = aho->states[ fail_state ].trans.base;
2134 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2136 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2137 fail[ ch_state ] = fail_state;
2138 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2140 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2142 q[ q_write++ % numstates] = ch_state;
2146 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2147 when we fail in state 1, this allows us to use the
2148 charclass scan to find a valid start char. This is based on the principle
2149 that theres a good chance the string being searched contains lots of stuff
2150 that cant be a start char.
2152 fail[ 0 ] = fail[ 1 ] = 0;
2153 DEBUG_TRIE_COMPILE_r({
2154 PerlIO_printf(Perl_debug_log,
2155 "%*sStclass Failtable (%"UVuf" states): 0",
2156 (int)(depth * 2), "", (UV)numstates
2158 for( q_read=1; q_read<numstates; q_read++ ) {
2159 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2161 PerlIO_printf(Perl_debug_log, "\n");
2164 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2169 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2170 * These need to be revisited when a newer toolchain becomes available.
2172 #if defined(__sparc64__) && defined(__GNUC__)
2173 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2174 # undef SPARC64_GCC_WORKAROUND
2175 # define SPARC64_GCC_WORKAROUND 1
2179 #define DEBUG_PEEP(str,scan,depth) \
2180 DEBUG_OPTIMISE_r({if (scan){ \
2181 SV * const mysv=sv_newmortal(); \
2182 regnode *Next = regnext(scan); \
2183 regprop(RExC_rx, mysv, scan); \
2184 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2185 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2186 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2193 #define JOIN_EXACT(scan,min,flags) \
2194 if (PL_regkind[OP(scan)] == EXACT) \
2195 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2198 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2199 /* Merge several consecutive EXACTish nodes into one. */
2200 regnode *n = regnext(scan);
2202 regnode *next = scan + NODE_SZ_STR(scan);
2206 regnode *stop = scan;
2207 GET_RE_DEBUG_FLAGS_DECL;
2209 PERL_UNUSED_ARG(depth);
2211 #ifndef EXPERIMENTAL_INPLACESCAN
2212 PERL_UNUSED_ARG(flags);
2213 PERL_UNUSED_ARG(val);
2215 DEBUG_PEEP("join",scan,depth);
2217 /* Skip NOTHING, merge EXACT*. */
2219 ( PL_regkind[OP(n)] == NOTHING ||
2220 (stringok && (OP(n) == OP(scan))))
2222 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2224 if (OP(n) == TAIL || n > next)
2226 if (PL_regkind[OP(n)] == NOTHING) {
2227 DEBUG_PEEP("skip:",n,depth);
2228 NEXT_OFF(scan) += NEXT_OFF(n);
2229 next = n + NODE_STEP_REGNODE;
2236 else if (stringok) {
2237 const unsigned int oldl = STR_LEN(scan);
2238 regnode * const nnext = regnext(n);
2240 DEBUG_PEEP("merg",n,depth);
2243 if (oldl + STR_LEN(n) > U8_MAX)
2245 NEXT_OFF(scan) += NEXT_OFF(n);
2246 STR_LEN(scan) += STR_LEN(n);
2247 next = n + NODE_SZ_STR(n);
2248 /* Now we can overwrite *n : */
2249 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2257 #ifdef EXPERIMENTAL_INPLACESCAN
2258 if (flags && !NEXT_OFF(n)) {
2259 DEBUG_PEEP("atch", val, depth);
2260 if (reg_off_by_arg[OP(n)]) {
2261 ARG_SET(n, val - n);
2264 NEXT_OFF(n) = val - n;
2271 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2273 Two problematic code points in Unicode casefolding of EXACT nodes:
2275 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2276 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2282 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2283 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2285 This means that in case-insensitive matching (or "loose matching",
2286 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2287 length of the above casefolded versions) can match a target string
2288 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2289 This would rather mess up the minimum length computation.
2291 What we'll do is to look for the tail four bytes, and then peek
2292 at the preceding two bytes to see whether we need to decrease
2293 the minimum length by four (six minus two).
2295 Thanks to the design of UTF-8, there cannot be false matches:
2296 A sequence of valid UTF-8 bytes cannot be a subsequence of
2297 another valid sequence of UTF-8 bytes.
2300 char * const s0 = STRING(scan), *s, *t;
2301 char * const s1 = s0 + STR_LEN(scan) - 1;
2302 char * const s2 = s1 - 4;
2303 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2304 const char t0[] = "\xaf\x49\xaf\x42";
2306 const char t0[] = "\xcc\x88\xcc\x81";
2308 const char * const t1 = t0 + 3;
2311 s < s2 && (t = ninstr(s, s1, t0, t1));
2314 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2315 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2317 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2318 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2326 n = scan + NODE_SZ_STR(scan);
2328 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2335 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2339 /* REx optimizer. Converts nodes into quickier variants "in place".
2340 Finds fixed substrings. */
2342 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2343 to the position after last scanned or to NULL. */
2345 #define INIT_AND_WITHP \
2346 assert(!and_withp); \
2347 Newx(and_withp,1,struct regnode_charclass_class); \
2348 SAVEFREEPV(and_withp)
2350 /* this is a chain of data about sub patterns we are processing that
2351 need to be handled seperately/specially in study_chunk. Its so
2352 we can simulate recursion without losing state. */
2354 typedef struct scan_frame {
2355 regnode *last; /* last node to process in this frame */
2356 regnode *next; /* next node to process when last is reached */
2357 struct scan_frame *prev; /*previous frame*/
2358 I32 stop; /* what stopparen do we use */
2362 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2365 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2366 I32 *minlenp, I32 *deltap,
2371 struct regnode_charclass_class *and_withp,
2372 U32 flags, U32 depth)
2373 /* scanp: Start here (read-write). */
2374 /* deltap: Write maxlen-minlen here. */
2375 /* last: Stop before this one. */
2376 /* data: string data about the pattern */
2377 /* stopparen: treat close N as END */
2378 /* recursed: which subroutines have we recursed into */
2379 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2382 I32 min = 0, pars = 0, code;
2383 regnode *scan = *scanp, *next;
2385 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2386 int is_inf_internal = 0; /* The studied chunk is infinite */
2387 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2388 scan_data_t data_fake;
2389 SV *re_trie_maxbuff = NULL;
2390 regnode *first_non_open = scan;
2391 I32 stopmin = I32_MAX;
2392 scan_frame *frame = NULL;
2394 GET_RE_DEBUG_FLAGS_DECL;
2397 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2401 while (first_non_open && OP(first_non_open) == OPEN)
2402 first_non_open=regnext(first_non_open);
2407 while ( scan && OP(scan) != END && scan < last ){
2408 /* Peephole optimizer: */
2409 DEBUG_STUDYDATA("Peep:", data,depth);
2410 DEBUG_PEEP("Peep",scan,depth);
2411 JOIN_EXACT(scan,&min,0);
2413 /* Follow the next-chain of the current node and optimize
2414 away all the NOTHINGs from it. */
2415 if (OP(scan) != CURLYX) {
2416 const int max = (reg_off_by_arg[OP(scan)]
2418 /* I32 may be smaller than U16 on CRAYs! */
2419 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2420 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2424 /* Skip NOTHING and LONGJMP. */
2425 while ((n = regnext(n))
2426 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2427 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2428 && off + noff < max)
2430 if (reg_off_by_arg[OP(scan)])
2433 NEXT_OFF(scan) = off;
2438 /* The principal pseudo-switch. Cannot be a switch, since we
2439 look into several different things. */
2440 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2441 || OP(scan) == IFTHEN) {
2442 next = regnext(scan);
2444 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2446 if (OP(next) == code || code == IFTHEN) {
2447 /* NOTE - There is similar code to this block below for handling
2448 TRIE nodes on a re-study. If you change stuff here check there
2450 I32 max1 = 0, min1 = I32_MAX, num = 0;
2451 struct regnode_charclass_class accum;
2452 regnode * const startbranch=scan;
2454 if (flags & SCF_DO_SUBSTR)
2455 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2456 if (flags & SCF_DO_STCLASS)
2457 cl_init_zero(pRExC_state, &accum);
2459 while (OP(scan) == code) {
2460 I32 deltanext, minnext, f = 0, fake;
2461 struct regnode_charclass_class this_class;
2464 data_fake.flags = 0;
2466 data_fake.whilem_c = data->whilem_c;
2467 data_fake.last_closep = data->last_closep;
2470 data_fake.last_closep = &fake;
2472 data_fake.pos_delta = delta;
2473 next = regnext(scan);
2474 scan = NEXTOPER(scan);
2476 scan = NEXTOPER(scan);
2477 if (flags & SCF_DO_STCLASS) {
2478 cl_init(pRExC_state, &this_class);
2479 data_fake.start_class = &this_class;
2480 f = SCF_DO_STCLASS_AND;
2482 if (flags & SCF_WHILEM_VISITED_POS)
2483 f |= SCF_WHILEM_VISITED_POS;
2485 /* we suppose the run is continuous, last=next...*/
2486 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2488 stopparen, recursed, NULL, f,depth+1);
2491 if (max1 < minnext + deltanext)
2492 max1 = minnext + deltanext;
2493 if (deltanext == I32_MAX)
2494 is_inf = is_inf_internal = 1;
2496 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2498 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2499 if ( stopmin > minnext)
2500 stopmin = min + min1;
2501 flags &= ~SCF_DO_SUBSTR;
2503 data->flags |= SCF_SEEN_ACCEPT;
2506 if (data_fake.flags & SF_HAS_EVAL)
2507 data->flags |= SF_HAS_EVAL;
2508 data->whilem_c = data_fake.whilem_c;
2510 if (flags & SCF_DO_STCLASS)
2511 cl_or(pRExC_state, &accum, &this_class);
2513 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2515 if (flags & SCF_DO_SUBSTR) {
2516 data->pos_min += min1;
2517 data->pos_delta += max1 - min1;
2518 if (max1 != min1 || is_inf)
2519 data->longest = &(data->longest_float);
2522 delta += max1 - min1;
2523 if (flags & SCF_DO_STCLASS_OR) {
2524 cl_or(pRExC_state, data->start_class, &accum);
2526 cl_and(data->start_class, and_withp);
2527 flags &= ~SCF_DO_STCLASS;
2530 else if (flags & SCF_DO_STCLASS_AND) {
2532 cl_and(data->start_class, &accum);
2533 flags &= ~SCF_DO_STCLASS;
2536 /* Switch to OR mode: cache the old value of
2537 * data->start_class */
2539 StructCopy(data->start_class, and_withp,
2540 struct regnode_charclass_class);
2541 flags &= ~SCF_DO_STCLASS_AND;
2542 StructCopy(&accum, data->start_class,
2543 struct regnode_charclass_class);
2544 flags |= SCF_DO_STCLASS_OR;
2545 data->start_class->flags |= ANYOF_EOS;
2549 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2552 Assuming this was/is a branch we are dealing with: 'scan' now
2553 points at the item that follows the branch sequence, whatever
2554 it is. We now start at the beginning of the sequence and look
2561 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2563 If we can find such a subseqence we need to turn the first
2564 element into a trie and then add the subsequent branch exact
2565 strings to the trie.
2569 1. patterns where the whole set of branch can be converted.
2571 2. patterns where only a subset can be converted.
2573 In case 1 we can replace the whole set with a single regop
2574 for the trie. In case 2 we need to keep the start and end
2577 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2578 becomes BRANCH TRIE; BRANCH X;
2580 There is an additional case, that being where there is a
2581 common prefix, which gets split out into an EXACT like node
2582 preceding the TRIE node.
2584 If x(1..n)==tail then we can do a simple trie, if not we make
2585 a "jump" trie, such that when we match the appropriate word
2586 we "jump" to the appopriate tail node. Essentailly we turn
2587 a nested if into a case structure of sorts.
2592 if (!re_trie_maxbuff) {
2593 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2594 if (!SvIOK(re_trie_maxbuff))
2595 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2597 if ( SvIV(re_trie_maxbuff)>=0 ) {
2599 regnode *first = (regnode *)NULL;
2600 regnode *last = (regnode *)NULL;
2601 regnode *tail = scan;
2606 SV * const mysv = sv_newmortal(); /* for dumping */
2608 /* var tail is used because there may be a TAIL
2609 regop in the way. Ie, the exacts will point to the
2610 thing following the TAIL, but the last branch will
2611 point at the TAIL. So we advance tail. If we
2612 have nested (?:) we may have to move through several
2616 while ( OP( tail ) == TAIL ) {
2617 /* this is the TAIL generated by (?:) */
2618 tail = regnext( tail );
2623 regprop(RExC_rx, mysv, tail );
2624 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2625 (int)depth * 2 + 2, "",
2626 "Looking for TRIE'able sequences. Tail node is: ",
2627 SvPV_nolen_const( mysv )
2633 step through the branches, cur represents each
2634 branch, noper is the first thing to be matched
2635 as part of that branch and noper_next is the
2636 regnext() of that node. if noper is an EXACT
2637 and noper_next is the same as scan (our current
2638 position in the regex) then the EXACT branch is
2639 a possible optimization target. Once we have
2640 two or more consequetive such branches we can
2641 create a trie of the EXACT's contents and stich
2642 it in place. If the sequence represents all of
2643 the branches we eliminate the whole thing and
2644 replace it with a single TRIE. If it is a
2645 subsequence then we need to stitch it in. This
2646 means the first branch has to remain, and needs
2647 to be repointed at the item on the branch chain
2648 following the last branch optimized. This could
2649 be either a BRANCH, in which case the
2650 subsequence is internal, or it could be the
2651 item following the branch sequence in which
2652 case the subsequence is at the end.
2656 /* dont use tail as the end marker for this traverse */
2657 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2658 regnode * const noper = NEXTOPER( cur );
2659 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2660 regnode * const noper_next = regnext( noper );
2664 regprop(RExC_rx, mysv, cur);
2665 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2666 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2668 regprop(RExC_rx, mysv, noper);
2669 PerlIO_printf( Perl_debug_log, " -> %s",
2670 SvPV_nolen_const(mysv));
2673 regprop(RExC_rx, mysv, noper_next );
2674 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2675 SvPV_nolen_const(mysv));
2677 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2678 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2680 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2681 : PL_regkind[ OP( noper ) ] == EXACT )
2682 || OP(noper) == NOTHING )
2684 && noper_next == tail
2689 if ( !first || optype == NOTHING ) {
2690 if (!first) first = cur;
2691 optype = OP( noper );
2697 make_trie( pRExC_state,
2698 startbranch, first, cur, tail, count,
2701 if ( PL_regkind[ OP( noper ) ] == EXACT
2703 && noper_next == tail
2708 optype = OP( noper );
2718 regprop(RExC_rx, mysv, cur);
2719 PerlIO_printf( Perl_debug_log,
2720 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2721 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2725 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2726 #ifdef TRIE_STUDY_OPT
2727 if ( ((made == MADE_EXACT_TRIE &&
2728 startbranch == first)
2729 || ( first_non_open == first )) &&
2731 flags |= SCF_TRIE_RESTUDY;
2732 if ( startbranch == first
2735 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2745 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2746 scan = NEXTOPER(NEXTOPER(scan));
2747 } else /* single branch is optimized. */
2748 scan = NEXTOPER(scan);
2750 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2751 scan_frame *newframe = NULL;
2756 if (OP(scan) != SUSPEND) {
2757 /* set the pointer */
2758 if (OP(scan) == GOSUB) {
2760 RExC_recurse[ARG2L(scan)] = scan;
2761 start = RExC_open_parens[paren-1];
2762 end = RExC_close_parens[paren-1];
2765 start = RExC_rxi->program + 1;
2769 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2770 SAVEFREEPV(recursed);
2772 if (!PAREN_TEST(recursed,paren+1)) {
2773 PAREN_SET(recursed,paren+1);
2774 Newx(newframe,1,scan_frame);
2776 if (flags & SCF_DO_SUBSTR) {
2777 SCAN_COMMIT(pRExC_state,data,minlenp);
2778 data->longest = &(data->longest_float);
2780 is_inf = is_inf_internal = 1;
2781 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2782 cl_anything(pRExC_state, data->start_class);
2783 flags &= ~SCF_DO_STCLASS;
2786 Newx(newframe,1,scan_frame);
2789 end = regnext(scan);
2794 SAVEFREEPV(newframe);
2795 newframe->next = regnext(scan);
2796 newframe->last = last;
2797 newframe->stop = stopparen;
2798 newframe->prev = frame;
2808 else if (OP(scan) == EXACT) {
2809 I32 l = STR_LEN(scan);
2812 const U8 * const s = (U8*)STRING(scan);
2813 l = utf8_length(s, s + l);
2814 uc = utf8_to_uvchr(s, NULL);
2816 uc = *((U8*)STRING(scan));
2819 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2820 /* The code below prefers earlier match for fixed
2821 offset, later match for variable offset. */
2822 if (data->last_end == -1) { /* Update the start info. */
2823 data->last_start_min = data->pos_min;
2824 data->last_start_max = is_inf
2825 ? I32_MAX : data->pos_min + data->pos_delta;
2827 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2829 SvUTF8_on(data->last_found);
2831 SV * const sv = data->last_found;
2832 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2833 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2834 if (mg && mg->mg_len >= 0)
2835 mg->mg_len += utf8_length((U8*)STRING(scan),
2836 (U8*)STRING(scan)+STR_LEN(scan));
2838 data->last_end = data->pos_min + l;
2839 data->pos_min += l; /* As in the first entry. */
2840 data->flags &= ~SF_BEFORE_EOL;
2842 if (flags & SCF_DO_STCLASS_AND) {
2843 /* Check whether it is compatible with what we know already! */
2847 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2848 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2849 && (!(data->start_class->flags & ANYOF_FOLD)
2850 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2853 ANYOF_CLASS_ZERO(data->start_class);
2854 ANYOF_BITMAP_ZERO(data->start_class);
2856 ANYOF_BITMAP_SET(data->start_class, uc);
2857 data->start_class->flags &= ~ANYOF_EOS;
2859 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2861 else if (flags & SCF_DO_STCLASS_OR) {
2862 /* false positive possible if the class is case-folded */
2864 ANYOF_BITMAP_SET(data->start_class, uc);
2866 data->start_class->flags |= ANYOF_UNICODE_ALL;
2867 data->start_class->flags &= ~ANYOF_EOS;
2868 cl_and(data->start_class, and_withp);
2870 flags &= ~SCF_DO_STCLASS;
2872 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2873 I32 l = STR_LEN(scan);
2874 UV uc = *((U8*)STRING(scan));
2876 /* Search for fixed substrings supports EXACT only. */
2877 if (flags & SCF_DO_SUBSTR) {
2879 SCAN_COMMIT(pRExC_state, data, minlenp);
2882 const U8 * const s = (U8 *)STRING(scan);
2883 l = utf8_length(s, s + l);
2884 uc = utf8_to_uvchr(s, NULL);
2887 if (flags & SCF_DO_SUBSTR)
2889 if (flags & SCF_DO_STCLASS_AND) {
2890 /* Check whether it is compatible with what we know already! */
2894 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2895 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2896 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2898 ANYOF_CLASS_ZERO(data->start_class);
2899 ANYOF_BITMAP_ZERO(data->start_class);
2901 ANYOF_BITMAP_SET(data->start_class, uc);
2902 data->start_class->flags &= ~ANYOF_EOS;
2903 data->start_class->flags |= ANYOF_FOLD;
2904 if (OP(scan) == EXACTFL)
2905 data->start_class->flags |= ANYOF_LOCALE;
2908 else if (flags & SCF_DO_STCLASS_OR) {
2909 if (data->start_class->flags & ANYOF_FOLD) {
2910 /* false positive possible if the class is case-folded.
2911 Assume that the locale settings are the same... */
2913 ANYOF_BITMAP_SET(data->start_class, uc);
2914 data->start_class->flags &= ~ANYOF_EOS;
2916 cl_and(data->start_class, and_withp);
2918 flags &= ~SCF_DO_STCLASS;
2920 else if (strchr((const char*)PL_varies,OP(scan))) {
2921 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2922 I32 f = flags, pos_before = 0;
2923 regnode * const oscan = scan;
2924 struct regnode_charclass_class this_class;
2925 struct regnode_charclass_class *oclass = NULL;
2926 I32 next_is_eval = 0;
2928 switch (PL_regkind[OP(scan)]) {
2929 case WHILEM: /* End of (?:...)* . */
2930 scan = NEXTOPER(scan);
2933 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2934 next = NEXTOPER(scan);
2935 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2937 maxcount = REG_INFTY;
2938 next = regnext(scan);
2939 scan = NEXTOPER(scan);
2943 if (flags & SCF_DO_SUBSTR)
2948 if (flags & SCF_DO_STCLASS) {
2950 maxcount = REG_INFTY;
2951 next = regnext(scan);
2952 scan = NEXTOPER(scan);
2955 is_inf = is_inf_internal = 1;
2956 scan = regnext(scan);
2957 if (flags & SCF_DO_SUBSTR) {
2958 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2959 data->longest = &(data->longest_float);
2961 goto optimize_curly_tail;
2963 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
2964 && (scan->flags == stopparen))
2969 mincount = ARG1(scan);
2970 maxcount = ARG2(scan);
2972 next = regnext(scan);
2973 if (OP(scan) == CURLYX) {
2974 I32 lp = (data ? *(data->last_closep) : 0);
2975 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
2977 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
2978 next_is_eval = (OP(scan) == EVAL);
2980 if (flags & SCF_DO_SUBSTR) {
2981 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
2982 pos_before = data->pos_min;
2986 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
2988 data->flags |= SF_IS_INF;
2990 if (flags & SCF_DO_STCLASS) {
2991 cl_init(pRExC_state, &this_class);
2992 oclass = data->start_class;
2993 data->start_class = &this_class;
2994 f |= SCF_DO_STCLASS_AND;
2995 f &= ~SCF_DO_STCLASS_OR;
2997 /* These are the cases when once a subexpression
2998 fails at a particular position, it cannot succeed
2999 even after backtracking at the enclosing scope.
3001 XXXX what if minimal match and we are at the
3002 initial run of {n,m}? */
3003 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3004 f &= ~SCF_WHILEM_VISITED_POS;
3006 /* This will finish on WHILEM, setting scan, or on NULL: */
3007 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3008 last, data, stopparen, recursed, NULL,
3010 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3012 if (flags & SCF_DO_STCLASS)
3013 data->start_class = oclass;
3014 if (mincount == 0 || minnext == 0) {
3015 if (flags & SCF_DO_STCLASS_OR) {
3016 cl_or(pRExC_state, data->start_class, &this_class);
3018 else if (flags & SCF_DO_STCLASS_AND) {
3019 /* Switch to OR mode: cache the old value of
3020 * data->start_class */
3022 StructCopy(data->start_class, and_withp,
3023 struct regnode_charclass_class);
3024 flags &= ~SCF_DO_STCLASS_AND;
3025 StructCopy(&this_class, data->start_class,
3026 struct regnode_charclass_class);
3027 flags |= SCF_DO_STCLASS_OR;
3028 data->start_class->flags |= ANYOF_EOS;
3030 } else { /* Non-zero len */
3031 if (flags & SCF_DO_STCLASS_OR) {
3032 cl_or(pRExC_state, data->start_class, &this_class);
3033 cl_and(data->start_class, and_withp);
3035 else if (flags & SCF_DO_STCLASS_AND)
3036 cl_and(data->start_class, &this_class);
3037 flags &= ~SCF_DO_STCLASS;
3039 if (!scan) /* It was not CURLYX, but CURLY. */
3041 if ( /* ? quantifier ok, except for (?{ ... }) */
3042 (next_is_eval || !(mincount == 0 && maxcount == 1))
3043 && (minnext == 0) && (deltanext == 0)
3044 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3045 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3046 && ckWARN(WARN_REGEXP))
3049 "Quantifier unexpected on zero-length expression");
3052 min += minnext * mincount;
3053 is_inf_internal |= ((maxcount == REG_INFTY
3054 && (minnext + deltanext) > 0)
3055 || deltanext == I32_MAX);
3056 is_inf |= is_inf_internal;
3057 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3059 /* Try powerful optimization CURLYX => CURLYN. */
3060 if ( OP(oscan) == CURLYX && data
3061 && data->flags & SF_IN_PAR
3062 && !(data->flags & SF_HAS_EVAL)
3063 && !deltanext && minnext == 1 ) {
3064 /* Try to optimize to CURLYN. */
3065 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3066 regnode * const nxt1 = nxt;
3073 if (!strchr((const char*)PL_simple,OP(nxt))
3074 && !(PL_regkind[OP(nxt)] == EXACT
3075 && STR_LEN(nxt) == 1))
3081 if (OP(nxt) != CLOSE)
3083 if (RExC_open_parens) {
3084 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3085 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3087 /* Now we know that nxt2 is the only contents: */
3088 oscan->flags = (U8)ARG(nxt);
3090 OP(nxt1) = NOTHING; /* was OPEN. */
3093 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3094 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3095 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3096 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3097 OP(nxt + 1) = OPTIMIZED; /* was count. */
3098 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3103 /* Try optimization CURLYX => CURLYM. */
3104 if ( OP(oscan) == CURLYX && data
3105 && !(data->flags & SF_HAS_PAR)
3106 && !(data->flags & SF_HAS_EVAL)
3107 && !deltanext /* atom is fixed width */
3108 && minnext != 0 /* CURLYM can't handle zero width */
3110 /* XXXX How to optimize if data == 0? */
3111 /* Optimize to a simpler form. */
3112 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3116 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3117 && (OP(nxt2) != WHILEM))
3119 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3120 /* Need to optimize away parenths. */
3121 if (data->flags & SF_IN_PAR) {
3122 /* Set the parenth number. */
3123 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3125 if (OP(nxt) != CLOSE)
3126 FAIL("Panic opt close");
3127 oscan->flags = (U8)ARG(nxt);
3128 if (RExC_open_parens) {
3129 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3130 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3132 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3133 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3136 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3137 OP(nxt + 1) = OPTIMIZED; /* was count. */
3138 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3139 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3142 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3143 regnode *nnxt = regnext(nxt1);
3146 if (reg_off_by_arg[OP(nxt1)])
3147 ARG_SET(nxt1, nxt2 - nxt1);
3148 else if (nxt2 - nxt1 < U16_MAX)
3149 NEXT_OFF(nxt1) = nxt2 - nxt1;
3151 OP(nxt) = NOTHING; /* Cannot beautify */
3156 /* Optimize again: */
3157 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3158 NULL, stopparen, recursed, NULL, 0,depth+1);
3163 else if ((OP(oscan) == CURLYX)
3164 && (flags & SCF_WHILEM_VISITED_POS)
3165 /* See the comment on a similar expression above.
3166 However, this time it not a subexpression
3167 we care about, but the expression itself. */
3168 && (maxcount == REG_INFTY)
3169 && data && ++data->whilem_c < 16) {
3170 /* This stays as CURLYX, we can put the count/of pair. */
3171 /* Find WHILEM (as in regexec.c) */
3172 regnode *nxt = oscan + NEXT_OFF(oscan);
3174 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3176 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3177 | (RExC_whilem_seen << 4)); /* On WHILEM */
3179 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3181 if (flags & SCF_DO_SUBSTR) {
3182 SV *last_str = NULL;
3183 int counted = mincount != 0;
3185 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3186 #if defined(SPARC64_GCC_WORKAROUND)
3189 const char *s = NULL;
3192 if (pos_before >= data->last_start_min)
3195 b = data->last_start_min;
3198 s = SvPV_const(data->last_found, l);
3199 old = b - data->last_start_min;
3202 I32 b = pos_before >= data->last_start_min
3203 ? pos_before : data->last_start_min;
3205 const char * const s = SvPV_const(data->last_found, l);
3206 I32 old = b - data->last_start_min;
3210 old = utf8_hop((U8*)s, old) - (U8*)s;
3213 /* Get the added string: */
3214 last_str = newSVpvn(s + old, l);
3216 SvUTF8_on(last_str);
3217 if (deltanext == 0 && pos_before == b) {
3218 /* What was added is a constant string */
3220 SvGROW(last_str, (mincount * l) + 1);
3221 repeatcpy(SvPVX(last_str) + l,
3222 SvPVX_const(last_str), l, mincount - 1);
3223 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3224 /* Add additional parts. */
3225 SvCUR_set(data->last_found,
3226 SvCUR(data->last_found) - l);
3227 sv_catsv(data->last_found, last_str);
3229 SV * sv = data->last_found;
3231 SvUTF8(sv) && SvMAGICAL(sv) ?
3232 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3233 if (mg && mg->mg_len >= 0)
3234 mg->mg_len += CHR_SVLEN(last_str);
3236 data->last_end += l * (mincount - 1);
3239 /* start offset must point into the last copy */
3240 data->last_start_min += minnext * (mincount - 1);
3241 data->last_start_max += is_inf ? I32_MAX
3242 : (maxcount - 1) * (minnext + data->pos_delta);
3245 /* It is counted once already... */
3246 data->pos_min += minnext * (mincount - counted);
3247 data->pos_delta += - counted * deltanext +
3248 (minnext + deltanext) * maxcount - minnext * mincount;
3249 if (mincount != maxcount) {
3250 /* Cannot extend fixed substrings found inside
3252 SCAN_COMMIT(pRExC_state,data,minlenp);
3253 if (mincount && last_str) {
3254 SV * const sv = data->last_found;
3255 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3256 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3260 sv_setsv(sv, last_str);
3261 data->last_end = data->pos_min;
3262 data->last_start_min =
3263 data->pos_min - CHR_SVLEN(last_str);
3264 data->last_start_max = is_inf
3266 : data->pos_min + data->pos_delta
3267 - CHR_SVLEN(last_str);
3269 data->longest = &(data->longest_float);
3271 SvREFCNT_dec(last_str);
3273 if (data && (fl & SF_HAS_EVAL))
3274 data->flags |= SF_HAS_EVAL;
3275 optimize_curly_tail:
3276 if (OP(oscan) != CURLYX) {
3277 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3279 NEXT_OFF(oscan) += NEXT_OFF(next);
3282 default: /* REF and CLUMP only? */
3283 if (flags & SCF_DO_SUBSTR) {
3284 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3285 data->longest = &(data->longest_float);
3287 is_inf = is_inf_internal = 1;
3288 if (flags & SCF_DO_STCLASS_OR)
3289 cl_anything(pRExC_state, data->start_class);
3290 flags &= ~SCF_DO_STCLASS;
3294 else if (strchr((const char*)PL_simple,OP(scan))) {
3297 if (flags & SCF_DO_SUBSTR) {
3298 SCAN_COMMIT(pRExC_state,data,minlenp);
3302 if (flags & SCF_DO_STCLASS) {
3303 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3305 /* Some of the logic below assumes that switching
3306 locale on will only add false positives. */
3307 switch (PL_regkind[OP(scan)]) {
3311 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3312 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3313 cl_anything(pRExC_state, data->start_class);
3316 if (OP(scan) == SANY)
3318 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3319 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3320 || (data->start_class->flags & ANYOF_CLASS));
3321 cl_anything(pRExC_state, data->start_class);
3323 if (flags & SCF_DO_STCLASS_AND || !value)
3324 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3327 if (flags & SCF_DO_STCLASS_AND)
3328 cl_and(data->start_class,
3329 (struct regnode_charclass_class*)scan);
3331 cl_or(pRExC_state, data->start_class,
3332 (struct regnode_charclass_class*)scan);
3335 if (flags & SCF_DO_STCLASS_AND) {
3336 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3337 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3338 for (value = 0; value < 256; value++)
3339 if (!isALNUM(value))
3340 ANYOF_BITMAP_CLEAR(data->start_class, value);
3344 if (data->start_class->flags & ANYOF_LOCALE)
3345 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3347 for (value = 0; value < 256; value++)
3349 ANYOF_BITMAP_SET(data->start_class, value);
3354 if (flags & SCF_DO_STCLASS_AND) {
3355 if (data->start_class->flags & ANYOF_LOCALE)
3356 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3359 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3360 data->start_class->flags |= ANYOF_LOCALE;
3364 if (flags & SCF_DO_STCLASS_AND) {
3365 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3366 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3367 for (value = 0; value < 256; value++)
3369 ANYOF_BITMAP_CLEAR(data->start_class, value);
3373 if (data->start_class->flags & ANYOF_LOCALE)
3374 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3376 for (value = 0; value < 256; value++)
3377 if (!isALNUM(value))
3378 ANYOF_BITMAP_SET(data->start_class, value);
3383 if (flags & SCF_DO_STCLASS_AND) {
3384 if (data->start_class->flags & ANYOF_LOCALE)
3385 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3388 data->start_class->flags |= ANYOF_LOCALE;
3389 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3393 if (flags & SCF_DO_STCLASS_AND) {
3394 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3395 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3396 for (value = 0; value < 256; value++)
3397 if (!isSPACE(value))
3398 ANYOF_BITMAP_CLEAR(data->start_class, value);
3402 if (data->start_class->flags & ANYOF_LOCALE)
3403 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3405 for (value = 0; value < 256; value++)
3407 ANYOF_BITMAP_SET(data->start_class, value);
3412 if (flags & SCF_DO_STCLASS_AND) {
3413 if (data->start_class->flags & ANYOF_LOCALE)
3414 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3417 data->start_class->flags |= ANYOF_LOCALE;
3418 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3422 if (flags & SCF_DO_STCLASS_AND) {
3423 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3424 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3425 for (value = 0; value < 256; value++)
3427 ANYOF_BITMAP_CLEAR(data->start_class, value);
3431 if (data->start_class->flags & ANYOF_LOCALE)
3432 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3434 for (value = 0; value < 256; value++)
3435 if (!isSPACE(value))
3436 ANYOF_BITMAP_SET(data->start_class, value);
3441 if (flags & SCF_DO_STCLASS_AND) {
3442 if (data->start_class->flags & ANYOF_LOCALE) {
3443 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3444 for (value = 0; value < 256; value++)
3445 if (!isSPACE(value))
3446 ANYOF_BITMAP_CLEAR(data->start_class, value);
3450 data->start_class->flags |= ANYOF_LOCALE;
3451 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3455 if (flags & SCF_DO_STCLASS_AND) {
3456 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3457 for (value = 0; value < 256; value++)
3458 if (!isDIGIT(value))
3459 ANYOF_BITMAP_CLEAR(data->start_class, value);
3462 if (data->start_class->flags & ANYOF_LOCALE)
3463 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3465 for (value = 0; value < 256; value++)
3467 ANYOF_BITMAP_SET(data->start_class, value);
3472 if (flags & SCF_DO_STCLASS_AND) {
3473 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3474 for (value = 0; value < 256; value++)
3476 ANYOF_BITMAP_CLEAR(data->start_class, value);
3479 if (data->start_class->flags & ANYOF_LOCALE)
3480 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3482 for (value = 0; value < 256; value++)
3483 if (!isDIGIT(value))
3484 ANYOF_BITMAP_SET(data->start_class, value);
3489 if (flags & SCF_DO_STCLASS_OR)
3490 cl_and(data->start_class, and_withp);
3491 flags &= ~SCF_DO_STCLASS;
3494 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3495 data->flags |= (OP(scan) == MEOL
3499 else if ( PL_regkind[OP(scan)] == BRANCHJ
3500 /* Lookbehind, or need to calculate parens/evals/stclass: */
3501 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3502 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3503 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3504 || OP(scan) == UNLESSM )
3506 /* Negative Lookahead/lookbehind
3507 In this case we can't do fixed string optimisation.
3510 I32 deltanext, minnext, fake = 0;
3512 struct regnode_charclass_class intrnl;
3515 data_fake.flags = 0;
3517 data_fake.whilem_c = data->whilem_c;
3518 data_fake.last_closep = data->last_closep;
3521 data_fake.last_closep = &fake;
3522 data_fake.pos_delta = delta;
3523 if ( flags & SCF_DO_STCLASS && !scan->flags
3524 && OP(scan) == IFMATCH ) { /* Lookahead */
3525 cl_init(pRExC_state, &intrnl);
3526 data_fake.start_class = &intrnl;
3527 f |= SCF_DO_STCLASS_AND;
3529 if (flags & SCF_WHILEM_VISITED_POS)
3530 f |= SCF_WHILEM_VISITED_POS;
3531 next = regnext(scan);
3532 nscan = NEXTOPER(NEXTOPER(scan));
3533 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3534 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3537 FAIL("Variable length lookbehind not implemented");
3539 else if (minnext > (I32)U8_MAX) {
3540 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3542 scan->flags = (U8)minnext;
3545 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3547 if (data_fake.flags & SF_HAS_EVAL)
3548 data->flags |= SF_HAS_EVAL;
3549 data->whilem_c = data_fake.whilem_c;
3551 if (f & SCF_DO_STCLASS_AND) {
3552 const int was = (data->start_class->flags & ANYOF_EOS);
3554 cl_and(data->start_class, &intrnl);
3556 data->start_class->flags |= ANYOF_EOS;
3559 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3561 /* Positive Lookahead/lookbehind
3562 In this case we can do fixed string optimisation,
3563 but we must be careful about it. Note in the case of
3564 lookbehind the positions will be offset by the minimum
3565 length of the pattern, something we won't know about
3566 until after the recurse.
3568 I32 deltanext, fake = 0;
3570 struct regnode_charclass_class intrnl;
3572 /* We use SAVEFREEPV so that when the full compile
3573 is finished perl will clean up the allocated
3574 minlens when its all done. This was we don't
3575 have to worry about freeing them when we know
3576 they wont be used, which would be a pain.
3579 Newx( minnextp, 1, I32 );
3580 SAVEFREEPV(minnextp);
3583 StructCopy(data, &data_fake, scan_data_t);
3584 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3587 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3588 data_fake.last_found=newSVsv(data->last_found);
3592 data_fake.last_closep = &fake;
3593 data_fake.flags = 0;
3594 data_fake.pos_delta = delta;
3596 data_fake.flags |= SF_IS_INF;
3597 if ( flags & SCF_DO_STCLASS && !scan->flags
3598 && OP(scan) == IFMATCH ) { /* Lookahead */
3599 cl_init(pRExC_state, &intrnl);
3600 data_fake.start_class = &intrnl;
3601 f |= SCF_DO_STCLASS_AND;
3603 if (flags & SCF_WHILEM_VISITED_POS)
3604 f |= SCF_WHILEM_VISITED_POS;
3605 next = regnext(scan);
3606 nscan = NEXTOPER(NEXTOPER(scan));
3608 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3609 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3612 FAIL("Variable length lookbehind not implemented");
3614 else if (*minnextp > (I32)U8_MAX) {
3615 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3617 scan->flags = (U8)*minnextp;
3622 if (f & SCF_DO_STCLASS_AND) {
3623 const int was = (data->start_class->flags & ANYOF_EOS);
3625 cl_and(data->start_class, &intrnl);
3627 data->start_class->flags |= ANYOF_EOS;
3630 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3632 if (data_fake.flags & SF_HAS_EVAL)
3633 data->flags |= SF_HAS_EVAL;
3634 data->whilem_c = data_fake.whilem_c;
3635 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3636 if (RExC_rx->minlen<*minnextp)
3637 RExC_rx->minlen=*minnextp;
3638 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3639 SvREFCNT_dec(data_fake.last_found);
3641 if ( data_fake.minlen_fixed != minlenp )
3643 data->offset_fixed= data_fake.offset_fixed;
3644 data->minlen_fixed= data_fake.minlen_fixed;
3645 data->lookbehind_fixed+= scan->flags;
3647 if ( data_fake.minlen_float != minlenp )
3649 data->minlen_float= data_fake.minlen_float;
3650 data->offset_float_min=data_fake.offset_float_min;
3651 data->offset_float_max=data_fake.offset_float_max;
3652 data->lookbehind_float+= scan->flags;
3661 else if (OP(scan) == OPEN) {
3662 if (stopparen != (I32)ARG(scan))
3665 else if (OP(scan) == CLOSE) {
3666 if (stopparen == (I32)ARG(scan)) {
3669 if ((I32)ARG(scan) == is_par) {
3670 next = regnext(scan);
3672 if ( next && (OP(next) != WHILEM) && next < last)
3673 is_par = 0; /* Disable optimization */
3676 *(data->last_closep) = ARG(scan);
3678 else if (OP(scan) == EVAL) {
3680 data->flags |= SF_HAS_EVAL;
3682 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3683 if (flags & SCF_DO_SUBSTR) {
3684 SCAN_COMMIT(pRExC_state,data,minlenp);
3685 flags &= ~SCF_DO_SUBSTR;
3687 if (data && OP(scan)==ACCEPT) {
3688 data->flags |= SCF_SEEN_ACCEPT;
3693 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3695 if (flags & SCF_DO_SUBSTR) {
3696 SCAN_COMMIT(pRExC_state,data,minlenp);
3697 data->longest = &(data->longest_float);
3699 is_inf = is_inf_internal = 1;
3700 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3701 cl_anything(pRExC_state, data->start_class);
3702 flags &= ~SCF_DO_STCLASS;
3704 else if (OP(scan) == GPOS) {
3705 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3706 !(delta || is_inf || (data && data->pos_delta)))
3708 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3709 RExC_rx->extflags |= RXf_ANCH_GPOS;
3710 if (RExC_rx->gofs < (U32)min)
3711 RExC_rx->gofs = min;
3713 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3717 #ifdef TRIE_STUDY_OPT
3718 #ifdef FULL_TRIE_STUDY
3719 else if (PL_regkind[OP(scan)] == TRIE) {
3720 /* NOTE - There is similar code to this block above for handling
3721 BRANCH nodes on the initial study. If you change stuff here
3723 regnode *trie_node= scan;
3724 regnode *tail= regnext(scan);
3725 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3726 I32 max1 = 0, min1 = I32_MAX;
3727 struct regnode_charclass_class accum;
3729 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3730 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3731 if (flags & SCF_DO_STCLASS)
3732 cl_init_zero(pRExC_state, &accum);
3738 const regnode *nextbranch= NULL;
3741 for ( word=1 ; word <= trie->wordcount ; word++)
3743 I32 deltanext=0, minnext=0, f = 0, fake;
3744 struct regnode_charclass_class this_class;
3746 data_fake.flags = 0;
3748 data_fake.whilem_c = data->whilem_c;
3749 data_fake.last_closep = data->last_closep;
3752 data_fake.last_closep = &fake;
3753 data_fake.pos_delta = delta;
3754 if (flags & SCF_DO_STCLASS) {
3755 cl_init(pRExC_state, &this_class);
3756 data_fake.start_class = &this_class;
3757 f = SCF_DO_STCLASS_AND;
3759 if (flags & SCF_WHILEM_VISITED_POS)
3760 f |= SCF_WHILEM_VISITED_POS;
3762 if (trie->jump[word]) {
3764 nextbranch = trie_node + trie->jump[0];
3765 scan= trie_node + trie->jump[word];
3766 /* We go from the jump point to the branch that follows
3767 it. Note this means we need the vestigal unused branches
3768 even though they arent otherwise used.
3770 minnext = study_chunk(pRExC_state, &scan, minlenp,
3771 &deltanext, (regnode *)nextbranch, &data_fake,
3772 stopparen, recursed, NULL, f,depth+1);
3774 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3775 nextbranch= regnext((regnode*)nextbranch);
3777 if (min1 > (I32)(minnext + trie->minlen))
3778 min1 = minnext + trie->minlen;
3779 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3780 max1 = minnext + deltanext + trie->maxlen;
3781 if (deltanext == I32_MAX)
3782 is_inf = is_inf_internal = 1;
3784 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3786 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3787 if ( stopmin > min + min1)
3788 stopmin = min + min1;
3789 flags &= ~SCF_DO_SUBSTR;
3791 data->flags |= SCF_SEEN_ACCEPT;
3794 if (data_fake.flags & SF_HAS_EVAL)
3795 data->flags |= SF_HAS_EVAL;
3796 data->whilem_c = data_fake.whilem_c;
3798 if (flags & SCF_DO_STCLASS)
3799 cl_or(pRExC_state, &accum, &this_class);
3802 if (flags & SCF_DO_SUBSTR) {
3803 data->pos_min += min1;
3804 data->pos_delta += max1 - min1;
3805 if (max1 != min1 || is_inf)
3806 data->longest = &(data->longest_float);
3809 delta += max1 - min1;
3810 if (flags & SCF_DO_STCLASS_OR) {
3811 cl_or(pRExC_state, data->start_class, &accum);
3813 cl_and(data->start_class, and_withp);
3814 flags &= ~SCF_DO_STCLASS;
3817 else if (flags & SCF_DO_STCLASS_AND) {
3819 cl_and(data->start_class, &accum);
3820 flags &= ~SCF_DO_STCLASS;
3823 /* Switch to OR mode: cache the old value of
3824 * data->start_class */
3826 StructCopy(data->start_class, and_withp,
3827 struct regnode_charclass_class);
3828 flags &= ~SCF_DO_STCLASS_AND;
3829 StructCopy(&accum, data->start_class,
3830 struct regnode_charclass_class);
3831 flags |= SCF_DO_STCLASS_OR;
3832 data->start_class->flags |= ANYOF_EOS;
3839 else if (PL_regkind[OP(scan)] == TRIE) {
3840 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3843 min += trie->minlen;
3844 delta += (trie->maxlen - trie->minlen);
3845 flags &= ~SCF_DO_STCLASS; /* xxx */
3846 if (flags & SCF_DO_SUBSTR) {
3847 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3848 data->pos_min += trie->minlen;
3849 data->pos_delta += (trie->maxlen - trie->minlen);
3850 if (trie->maxlen != trie->minlen)
3851 data->longest = &(data->longest_float);
3853 if (trie->jump) /* no more substrings -- for now /grr*/
3854 flags &= ~SCF_DO_SUBSTR;
3856 #endif /* old or new */
3857 #endif /* TRIE_STUDY_OPT */
3858 /* Else: zero-length, ignore. */
3859 scan = regnext(scan);
3864 stopparen = frame->stop;
3865 frame = frame->prev;
3866 goto fake_study_recurse;
3871 DEBUG_STUDYDATA("pre-fin:",data,depth);
3874 *deltap = is_inf_internal ? I32_MAX : delta;
3875 if (flags & SCF_DO_SUBSTR && is_inf)
3876 data->pos_delta = I32_MAX - data->pos_min;
3877 if (is_par > (I32)U8_MAX)
3879 if (is_par && pars==1 && data) {
3880 data->flags |= SF_IN_PAR;
3881 data->flags &= ~SF_HAS_PAR;
3883 else if (pars && data) {
3884 data->flags |= SF_HAS_PAR;
3885 data->flags &= ~SF_IN_PAR;
3887 if (flags & SCF_DO_STCLASS_OR)
3888 cl_and(data->start_class, and_withp);
3889 if (flags & SCF_TRIE_RESTUDY)
3890 data->flags |= SCF_TRIE_RESTUDY;
3892 DEBUG_STUDYDATA("post-fin:",data,depth);
3894 return min < stopmin ? min : stopmin;
3898 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3900 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3902 Renewc(RExC_rxi->data,
3903 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3904 char, struct reg_data);
3906 Renew(RExC_rxi->data->what, count + n, U8);
3908 Newx(RExC_rxi->data->what, n, U8);
3909 RExC_rxi->data->count = count + n;
3910 Copy(s, RExC_rxi->data->what + count, n, U8);
3914 /*XXX: todo make this not included in a non debugging perl */
3915 #ifndef PERL_IN_XSUB_RE
3917 Perl_reginitcolors(pTHX)
3920 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3922 char *t = savepv(s);
3926 t = strchr(t, '\t');
3932 PL_colors[i] = t = (char *)"";
3937 PL_colors[i++] = (char *)"";
3944 #ifdef TRIE_STUDY_OPT
3945 #define CHECK_RESTUDY_GOTO \
3947 (data.flags & SCF_TRIE_RESTUDY) \
3951 #define CHECK_RESTUDY_GOTO
3955 - pregcomp - compile a regular expression into internal code
3957 * We can't allocate space until we know how big the compiled form will be,
3958 * but we can't compile it (and thus know how big it is) until we've got a
3959 * place to put the code. So we cheat: we compile it twice, once with code
3960 * generation turned off and size counting turned on, and once "for real".
3961 * This also means that we don't allocate space until we are sure that the
3962 * thing really will compile successfully, and we never have to move the
3963 * code and thus invalidate pointers into it. (Note that it has to be in
3964 * one piece because free() must be able to free it all.) [NB: not true in perl]
3966 * Beware that the optimization-preparation code in here knows about some
3967 * of the structure of the compiled regexp. [I'll say.]
3972 #ifndef PERL_IN_XSUB_RE
3973 #define RE_ENGINE_PTR &PL_core_reg_engine
3975 extern const struct regexp_engine my_reg_engine;
3976 #define RE_ENGINE_PTR &my_reg_engine
3979 #ifndef PERL_IN_XSUB_RE
3981 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
3984 HV * const table = GvHV(PL_hintgv);
3985 /* Dispatch a request to compile a regexp to correct
3988 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
3989 GET_RE_DEBUG_FLAGS_DECL;
3990 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
3991 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
3993 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
3996 return CALLREGCOMP_ENG(eng, exp, xend, pm);
3999 return Perl_re_compile(aTHX_ exp, xend, pm);
4004 Perl_re_compile(pTHX_ char *exp, char *xend, PMOP *pm)
4008 register regexp_internal *ri;
4016 RExC_state_t RExC_state;
4017 RExC_state_t * const pRExC_state = &RExC_state;
4018 #ifdef TRIE_STUDY_OPT
4020 RExC_state_t copyRExC_state;
4022 GET_RE_DEBUG_FLAGS_DECL;
4023 DEBUG_r(if (!PL_colorset) reginitcolors());
4026 FAIL("NULL regexp argument");
4028 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4032 SV *dsv= sv_newmortal();
4033 RE_PV_QUOTED_DECL(s, RExC_utf8,
4034 dsv, RExC_precomp, (xend - exp), 60);
4035 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4036 PL_colors[4],PL_colors[5],s);
4038 RExC_flags = pm->op_pmflags;
4042 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4043 RExC_seen_evals = 0;
4046 /* First pass: determine size, legality. */
4054 RExC_emit = &PL_regdummy;
4055 RExC_whilem_seen = 0;
4056 RExC_charnames = NULL;
4057 RExC_open_parens = NULL;
4058 RExC_close_parens = NULL;
4060 RExC_paren_names = NULL;
4062 RExC_paren_name_list = NULL;
4064 RExC_recurse = NULL;
4065 RExC_recurse_count = 0;
4067 #if 0 /* REGC() is (currently) a NOP at the first pass.
4068 * Clever compilers notice this and complain. --jhi */
4069 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4071 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4072 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4073 RExC_precomp = NULL;
4077 PerlIO_printf(Perl_debug_log,
4078 "Required size %"IVdf" nodes\n"
4079 "Starting second pass (creation)\n",
4082 RExC_lastparse=NULL;
4084 /* Small enough for pointer-storage convention?
4085 If extralen==0, this means that we will not need long jumps. */
4086 if (RExC_size >= 0x10000L && RExC_extralen)
4087 RExC_size += RExC_extralen;
4090 if (RExC_whilem_seen > 15)
4091 RExC_whilem_seen = 15;
4094 /* Make room for a sentinel value at the end of the program */
4098 /* Allocate space and zero-initialize. Note, the two step process
4099 of zeroing when in debug mode, thus anything assigned has to
4100 happen after that */
4101 Newxz(r, 1, regexp);
4102 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4103 char, regexp_internal);
4104 if ( r == NULL || ri == NULL )
4105 FAIL("Regexp out of space");
4107 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4108 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4110 /* bulk initialize base fields with 0. */
4111 Zero(ri, sizeof(regexp_internal), char);
4114 /* non-zero initialization begins here */
4116 r->engine= RE_ENGINE_PTR;
4118 r->prelen = xend - exp;
4119 r->precomp = savepvn(RExC_precomp, r->prelen);
4120 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4122 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4124 if (RExC_seen & REG_SEEN_RECURSE) {
4125 Newxz(RExC_open_parens, RExC_npar,regnode *);
4126 SAVEFREEPV(RExC_open_parens);
4127 Newxz(RExC_close_parens,RExC_npar,regnode *);
4128 SAVEFREEPV(RExC_close_parens);
4131 /* Useful during FAIL. */
4132 Newxz(ri->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4134 ri->offsets[0] = RExC_size;
4136 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4137 "%s %"UVuf" bytes for offset annotations.\n",
4138 ri->offsets ? "Got" : "Couldn't get",
4139 (UV)((2*RExC_size+1) * sizeof(U32))));
4144 /* Second pass: emit code. */
4145 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4150 RExC_emit_start = ri->program;
4151 RExC_emit = ri->program;
4153 /* put a sentinal on the end of the program so we can check for
4155 ri->program[RExC_size].type = 255;
4157 /* Store the count of eval-groups for security checks: */
4158 RExC_rx->seen_evals = RExC_seen_evals;
4159 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4160 if (reg(pRExC_state, 0, &flags,1) == NULL)
4163 /* XXXX To minimize changes to RE engine we always allocate
4164 3-units-long substrs field. */
4165 Newx(r->substrs, 1, struct reg_substr_data);
4166 if (RExC_recurse_count) {
4167 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4168 SAVEFREEPV(RExC_recurse);
4172 r->minlen = minlen = sawplus = sawopen = 0;
4173 Zero(r->substrs, 1, struct reg_substr_data);
4175 #ifdef TRIE_STUDY_OPT
4178 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4180 RExC_state = copyRExC_state;
4181 if (seen & REG_TOP_LEVEL_BRANCHES)
4182 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4184 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4185 if (data.last_found) {
4186 SvREFCNT_dec(data.longest_fixed);
4187 SvREFCNT_dec(data.longest_float);
4188 SvREFCNT_dec(data.last_found);
4190 StructCopy(&zero_scan_data, &data, scan_data_t);
4192 StructCopy(&zero_scan_data, &data, scan_data_t);
4193 copyRExC_state = RExC_state;
4196 StructCopy(&zero_scan_data, &data, scan_data_t);
4199 /* Dig out information for optimizations. */
4200 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4201 pm->op_pmflags = RExC_flags;
4203 r->extflags |= RXf_UTF8; /* Unicode in it? */
4204 ri->regstclass = NULL;
4205 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4206 r->intflags |= PREGf_NAUGHTY;
4207 scan = ri->program + 1; /* First BRANCH. */
4209 /* testing for BRANCH here tells us whether there is "must appear"
4210 data in the pattern. If there is then we can use it for optimisations */
4211 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4213 STRLEN longest_float_length, longest_fixed_length;
4214 struct regnode_charclass_class ch_class; /* pointed to by data */
4216 I32 last_close = 0; /* pointed to by data */
4219 /* Skip introductions and multiplicators >= 1. */
4220 while ((OP(first) == OPEN && (sawopen = 1)) ||
4221 /* An OR of *one* alternative - should not happen now. */
4222 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4223 /* for now we can't handle lookbehind IFMATCH*/
4224 (OP(first) == IFMATCH && !first->flags) ||
4225 (OP(first) == PLUS) ||
4226 (OP(first) == MINMOD) ||
4227 /* An {n,m} with n>0 */
4228 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4231 if (OP(first) == PLUS)
4234 first += regarglen[OP(first)];
4235 if (OP(first) == IFMATCH) {
4236 first = NEXTOPER(first);
4237 first += EXTRA_STEP_2ARGS;
4238 } else /* XXX possible optimisation for /(?=)/ */
4239 first = NEXTOPER(first);
4242 /* Starting-point info. */
4244 DEBUG_PEEP("first:",first,0);
4245 /* Ignore EXACT as we deal with it later. */
4246 if (PL_regkind[OP(first)] == EXACT) {
4247 if (OP(first) == EXACT)
4248 NOOP; /* Empty, get anchored substr later. */
4249 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4250 ri->regstclass = first;
4253 else if (PL_regkind[OP(first)] == TRIE &&
4254 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4257 /* this can happen only on restudy */
4258 if ( OP(first) == TRIE ) {
4259 struct regnode_1 *trieop = (struct regnode_1 *)
4260 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4261 StructCopy(first,trieop,struct regnode_1);
4262 trie_op=(regnode *)trieop;
4264 struct regnode_charclass *trieop = (struct regnode_charclass *)
4265 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4266 StructCopy(first,trieop,struct regnode_charclass);
4267 trie_op=(regnode *)trieop;
4270 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4271 ri->regstclass = trie_op;
4274 else if (strchr((const char*)PL_simple,OP(first)))
4275 ri->regstclass = first;
4276 else if (PL_regkind[OP(first)] == BOUND ||
4277 PL_regkind[OP(first)] == NBOUND)
4278 ri->regstclass = first;
4279 else if (PL_regkind[OP(first)] == BOL) {
4280 r->extflags |= (OP(first) == MBOL
4282 : (OP(first) == SBOL
4285 first = NEXTOPER(first);
4288 else if (OP(first) == GPOS) {
4289 r->extflags |= RXf_ANCH_GPOS;
4290 first = NEXTOPER(first);
4293 else if ((!sawopen || !RExC_sawback) &&
4294 (OP(first) == STAR &&
4295 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4296 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4298 /* turn .* into ^.* with an implied $*=1 */
4300 (OP(NEXTOPER(first)) == REG_ANY)
4303 r->extflags |= type;
4304 r->intflags |= PREGf_IMPLICIT;
4305 first = NEXTOPER(first);
4308 if (sawplus && (!sawopen || !RExC_sawback)
4309 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4310 /* x+ must match at the 1st pos of run of x's */
4311 r->intflags |= PREGf_SKIP;
4313 /* Scan is after the zeroth branch, first is atomic matcher. */
4314 #ifdef TRIE_STUDY_OPT
4317 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4318 (IV)(first - scan + 1))
4322 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4323 (IV)(first - scan + 1))
4329 * If there's something expensive in the r.e., find the
4330 * longest literal string that must appear and make it the
4331 * regmust. Resolve ties in favor of later strings, since
4332 * the regstart check works with the beginning of the r.e.
4333 * and avoiding duplication strengthens checking. Not a
4334 * strong reason, but sufficient in the absence of others.
4335 * [Now we resolve ties in favor of the earlier string if
4336 * it happens that c_offset_min has been invalidated, since the
4337 * earlier string may buy us something the later one won't.]
4340 data.longest_fixed = newSVpvs("");
4341 data.longest_float = newSVpvs("");
4342 data.last_found = newSVpvs("");
4343 data.longest = &(data.longest_fixed);
4345 if (!ri->regstclass) {
4346 cl_init(pRExC_state, &ch_class);
4347 data.start_class = &ch_class;
4348 stclass_flag = SCF_DO_STCLASS_AND;
4349 } else /* XXXX Check for BOUND? */
4351 data.last_closep = &last_close;
4353 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4354 &data, -1, NULL, NULL,
4355 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4361 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4362 && data.last_start_min == 0 && data.last_end > 0
4363 && !RExC_seen_zerolen
4364 && !(RExC_seen & REG_SEEN_VERBARG)
4365 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4366 r->extflags |= RXf_CHECK_ALL;
4367 scan_commit(pRExC_state, &data,&minlen,0);
4368 SvREFCNT_dec(data.last_found);
4370 /* Note that code very similar to this but for anchored string
4371 follows immediately below, changes may need to be made to both.
4374 longest_float_length = CHR_SVLEN(data.longest_float);
4375 if (longest_float_length
4376 || (data.flags & SF_FL_BEFORE_EOL
4377 && (!(data.flags & SF_FL_BEFORE_MEOL)
4378 || (RExC_flags & RXf_PMf_MULTILINE))))
4382 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4383 && data.offset_fixed == data.offset_float_min
4384 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4385 goto remove_float; /* As in (a)+. */
4387 /* copy the information about the longest float from the reg_scan_data
4388 over to the program. */
4389 if (SvUTF8(data.longest_float)) {
4390 r->float_utf8 = data.longest_float;
4391 r->float_substr = NULL;
4393 r->float_substr = data.longest_float;
4394 r->float_utf8 = NULL;
4396 /* float_end_shift is how many chars that must be matched that
4397 follow this item. We calculate it ahead of time as once the
4398 lookbehind offset is added in we lose the ability to correctly
4400 ml = data.minlen_float ? *(data.minlen_float)
4401 : (I32)longest_float_length;
4402 r->float_end_shift = ml - data.offset_float_min
4403 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4404 + data.lookbehind_float;
4405 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4406 r->float_max_offset = data.offset_float_max;
4407 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4408 r->float_max_offset -= data.lookbehind_float;
4410 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4411 && (!(data.flags & SF_FL_BEFORE_MEOL)
4412 || (RExC_flags & RXf_PMf_MULTILINE)));
4413 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4417 r->float_substr = r->float_utf8 = NULL;
4418 SvREFCNT_dec(data.longest_float);
4419 longest_float_length = 0;
4422 /* Note that code very similar to this but for floating string
4423 is immediately above, changes may need to be made to both.
4426 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4427 if (longest_fixed_length
4428 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4429 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4430 || (RExC_flags & RXf_PMf_MULTILINE))))
4434 /* copy the information about the longest fixed
4435 from the reg_scan_data over to the program. */
4436 if (SvUTF8(data.longest_fixed)) {
4437 r->anchored_utf8 = data.longest_fixed;
4438 r->anchored_substr = NULL;
4440 r->anchored_substr = data.longest_fixed;
4441 r->anchored_utf8 = NULL;
4443 /* fixed_end_shift is how many chars that must be matched that
4444 follow this item. We calculate it ahead of time as once the
4445 lookbehind offset is added in we lose the ability to correctly
4447 ml = data.minlen_fixed ? *(data.minlen_fixed)
4448 : (I32)longest_fixed_length;
4449 r->anchored_end_shift = ml - data.offset_fixed
4450 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4451 + data.lookbehind_fixed;
4452 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4454 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4455 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4456 || (RExC_flags & RXf_PMf_MULTILINE)));
4457 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4460 r->anchored_substr = r->anchored_utf8 = NULL;
4461 SvREFCNT_dec(data.longest_fixed);
4462 longest_fixed_length = 0;
4465 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4466 ri->regstclass = NULL;
4467 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4469 && !(data.start_class->flags & ANYOF_EOS)
4470 && !cl_is_anything(data.start_class))
4472 const U32 n = add_data(pRExC_state, 1, "f");
4474 Newx(RExC_rxi->data->data[n], 1,
4475 struct regnode_charclass_class);
4476 StructCopy(data.start_class,
4477 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4478 struct regnode_charclass_class);
4479 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4480 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4481 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4482 regprop(r, sv, (regnode*)data.start_class);
4483 PerlIO_printf(Perl_debug_log,
4484 "synthetic stclass \"%s\".\n",
4485 SvPVX_const(sv));});
4488 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4489 if (longest_fixed_length > longest_float_length) {
4490 r->check_end_shift = r->anchored_end_shift;
4491 r->check_substr = r->anchored_substr;
4492 r->check_utf8 = r->anchored_utf8;
4493 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4494 if (r->extflags & RXf_ANCH_SINGLE)
4495 r->extflags |= RXf_NOSCAN;
4498 r->check_end_shift = r->float_end_shift;
4499 r->check_substr = r->float_substr;
4500 r->check_utf8 = r->float_utf8;
4501 r->check_offset_min = r->float_min_offset;
4502 r->check_offset_max = r->float_max_offset;
4504 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4505 This should be changed ASAP! */
4506 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4507 r->extflags |= RXf_USE_INTUIT;
4508 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4509 r->extflags |= RXf_INTUIT_TAIL;
4511 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4512 if ( (STRLEN)minlen < longest_float_length )
4513 minlen= longest_float_length;
4514 if ( (STRLEN)minlen < longest_fixed_length )
4515 minlen= longest_fixed_length;
4519 /* Several toplevels. Best we can is to set minlen. */
4521 struct regnode_charclass_class ch_class;
4524 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4526 scan = ri->program + 1;
4527 cl_init(pRExC_state, &ch_class);
4528 data.start_class = &ch_class;
4529 data.last_closep = &last_close;
4532 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4533 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4537 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4538 = r->float_substr = r->float_utf8 = NULL;
4539 if (!(data.start_class->flags & ANYOF_EOS)
4540 && !cl_is_anything(data.start_class))
4542 const U32 n = add_data(pRExC_state, 1, "f");
4544 Newx(RExC_rxi->data->data[n], 1,
4545 struct regnode_charclass_class);
4546 StructCopy(data.start_class,
4547 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4548 struct regnode_charclass_class);
4549 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4550 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4551 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4552 regprop(r, sv, (regnode*)data.start_class);
4553 PerlIO_printf(Perl_debug_log,
4554 "synthetic stclass \"%s\".\n",
4555 SvPVX_const(sv));});
4559 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4560 the "real" pattern. */
4562 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4563 (IV)minlen, (IV)r->minlen);
4565 r->minlenret = minlen;
4566 if (r->minlen < minlen)
4569 if (RExC_seen & REG_SEEN_GPOS)
4570 r->extflags |= RXf_GPOS_SEEN;
4571 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4572 r->extflags |= RXf_LOOKBEHIND_SEEN;
4573 if (RExC_seen & REG_SEEN_EVAL)
4574 r->extflags |= RXf_EVAL_SEEN;
4575 if (RExC_seen & REG_SEEN_CANY)
4576 r->extflags |= RXf_CANY_SEEN;
4577 if (RExC_seen & REG_SEEN_VERBARG)
4578 r->intflags |= PREGf_VERBARG_SEEN;
4579 if (RExC_seen & REG_SEEN_CUTGROUP)
4580 r->intflags |= PREGf_CUTGROUP_SEEN;
4581 if (RExC_paren_names)
4582 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4584 r->paren_names = NULL;
4585 if (r->prelen == 3 && strEQ("\\s+", r->precomp))
4586 r->extflags |= RXf_WHITE;
4587 else if (r->prelen == 1 && r->precomp[0] == '^')
4588 r->extflags |= RXf_START_ONLY;
4591 if (RExC_paren_names) {
4592 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4593 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4595 ri->name_list_idx = 0;
4598 if (RExC_recurse_count) {
4599 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4600 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4601 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4604 Newxz(r->startp, RExC_npar, I32);
4605 Newxz(r->endp, RExC_npar, I32);
4606 /* assume we don't need to swap parens around before we match */
4609 PerlIO_printf(Perl_debug_log,"Final program:\n");
4612 DEBUG_OFFSETS_r(if (ri->offsets) {
4613 const U32 len = ri->offsets[0];
4615 GET_RE_DEBUG_FLAGS_DECL;
4616 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->offsets[0]);
4617 for (i = 1; i <= len; i++) {
4618 if (ri->offsets[i*2-1] || ri->offsets[i*2])
4619 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4620 (UV)i, (UV)ri->offsets[i*2-1], (UV)ri->offsets[i*2]);
4622 PerlIO_printf(Perl_debug_log, "\n");
4627 #undef CORE_ONLY_BLOCK
4628 #undef RE_ENGINE_PTR
4630 #ifndef PERL_IN_XSUB_RE
4632 Perl_reg_named_buff_get(pTHX_ SV* namesv, const REGEXP * const from_re, U32 flags)
4634 AV *retarray = NULL;
4639 if (from_re || PL_curpm) {
4640 const REGEXP * const rx = from_re ? from_re : PM_GETRE(PL_curpm);
4641 if (rx && rx->paren_names) {
4642 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4645 SV* sv_dat=HeVAL(he_str);
4646 I32 *nums=(I32*)SvPVX(sv_dat);
4647 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4648 if ((I32)(rx->lastparen) >= nums[i] &&
4649 rx->endp[nums[i]] != -1)
4651 ret = reg_numbered_buff_get(nums[i],rx,NULL,0);
4655 ret = newSVsv(&PL_sv_undef);
4659 av_push(retarray, ret);
4663 return (SV*)retarray;
4671 Perl_reg_numbered_buff_get(pTHX_ I32 paren, const REGEXP * const rx, SV* usesv, U32 flags)
4676 SV *sv = usesv ? usesv : newSVpvs("");
4678 if (paren == -2 && (s = rx->subbeg) && rx->startp[0] != -1) {
4683 if (paren == -1 && rx->subbeg && rx->endp[0] != -1) {
4685 s = rx->subbeg + rx->endp[0];
4686 i = rx->sublen - rx->endp[0];
4689 if ( 0 <= paren && paren <= (I32)rx->nparens &&
4690 (s1 = rx->startp[paren]) != -1 &&
4691 (t1 = rx->endp[paren]) != -1)
4695 s = rx->subbeg + s1;
4700 assert(rx->sublen >= (s - rx->subbeg) + i );
4703 const int oldtainted = PL_tainted;
4705 sv_setpvn(sv, s, i);
4706 PL_tainted = oldtainted;
4707 if ( (rx->extflags & RXf_CANY_SEEN)
4708 ? (RX_MATCH_UTF8(rx)
4709 && (!i || is_utf8_string((U8*)s, i)))
4710 : (RX_MATCH_UTF8(rx)) )
4717 if (RX_MATCH_TAINTED(rx)) {
4718 if (SvTYPE(sv) >= SVt_PVMG) {
4719 MAGIC* const mg = SvMAGIC(sv);
4722 SvMAGIC_set(sv, mg->mg_moremagic);
4724 if ((mgt = SvMAGIC(sv))) {
4725 mg->mg_moremagic = mgt;
4726 SvMAGIC_set(sv, mg);
4736 sv_setsv(sv,&PL_sv_undef);
4739 sv_setsv(sv,&PL_sv_undef);
4745 /* Scans the name of a named buffer from the pattern.
4746 * If flags is REG_RSN_RETURN_NULL returns null.
4747 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4748 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4749 * to the parsed name as looked up in the RExC_paren_names hash.
4750 * If there is an error throws a vFAIL().. type exception.
4753 #define REG_RSN_RETURN_NULL 0
4754 #define REG_RSN_RETURN_NAME 1
4755 #define REG_RSN_RETURN_DATA 2
4758 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4759 char *name_start = RExC_parse;
4761 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
4762 /* skip IDFIRST by using do...while */
4765 RExC_parse += UTF8SKIP(RExC_parse);
4766 } while (isALNUM_utf8((U8*)RExC_parse));
4770 } while (isALNUM(*RExC_parse));
4774 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4775 (int)(RExC_parse - name_start)));
4778 if ( flags == REG_RSN_RETURN_NAME)
4780 else if (flags==REG_RSN_RETURN_DATA) {
4783 if ( ! sv_name ) /* should not happen*/
4784 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4785 if (RExC_paren_names)
4786 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4788 sv_dat = HeVAL(he_str);
4790 vFAIL("Reference to nonexistent named group");
4794 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4801 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4802 int rem=(int)(RExC_end - RExC_parse); \
4811 if (RExC_lastparse!=RExC_parse) \
4812 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4815 iscut ? "..." : "<" \
4818 PerlIO_printf(Perl_debug_log,"%16s",""); \
4823 num=REG_NODE_NUM(RExC_emit); \
4824 if (RExC_lastnum!=num) \
4825 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4827 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4828 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4829 (int)((depth*2)), "", \
4833 RExC_lastparse=RExC_parse; \
4838 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4839 DEBUG_PARSE_MSG((funcname)); \
4840 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4842 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4843 DEBUG_PARSE_MSG((funcname)); \
4844 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4847 - reg - regular expression, i.e. main body or parenthesized thing
4849 * Caller must absorb opening parenthesis.
4851 * Combining parenthesis handling with the base level of regular expression
4852 * is a trifle forced, but the need to tie the tails of the branches to what
4853 * follows makes it hard to avoid.
4855 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4857 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4859 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4862 /* this idea is borrowed from STR_WITH_LEN in handy.h */
4863 #define CHECK_WORD(s,v,l) \
4864 (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
4867 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4868 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4871 register regnode *ret; /* Will be the head of the group. */
4872 register regnode *br;
4873 register regnode *lastbr;
4874 register regnode *ender = NULL;
4875 register I32 parno = 0;
4877 const I32 oregflags = RExC_flags;
4878 bool have_branch = 0;
4881 /* for (?g), (?gc), and (?o) warnings; warning
4882 about (?c) will warn about (?g) -- japhy */
4884 #define WASTED_O 0x01
4885 #define WASTED_G 0x02
4886 #define WASTED_C 0x04
4887 #define WASTED_GC (0x02|0x04)
4888 I32 wastedflags = 0x00;
4890 char * parse_start = RExC_parse; /* MJD */
4891 char * const oregcomp_parse = RExC_parse;
4893 GET_RE_DEBUG_FLAGS_DECL;
4894 DEBUG_PARSE("reg ");
4897 *flagp = 0; /* Tentatively. */
4900 /* Make an OPEN node, if parenthesized. */
4902 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4903 char *start_verb = RExC_parse;
4904 STRLEN verb_len = 0;
4905 char *start_arg = NULL;
4906 unsigned char op = 0;
4908 int internal_argval = 0; /* internal_argval is only useful if !argok */
4909 while ( *RExC_parse && *RExC_parse != ')' ) {
4910 if ( *RExC_parse == ':' ) {
4911 start_arg = RExC_parse + 1;
4917 verb_len = RExC_parse - start_verb;
4920 while ( *RExC_parse && *RExC_parse != ')' )
4922 if ( *RExC_parse != ')' )
4923 vFAIL("Unterminated verb pattern argument");
4924 if ( RExC_parse == start_arg )
4927 if ( *RExC_parse != ')' )
4928 vFAIL("Unterminated verb pattern");
4931 switch ( *start_verb ) {
4932 case 'A': /* (*ACCEPT) */
4933 if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
4935 internal_argval = RExC_nestroot;
4938 case 'C': /* (*COMMIT) */
4939 if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
4942 case 'F': /* (*FAIL) */
4943 if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
4948 case ':': /* (*:NAME) */
4949 case 'M': /* (*MARK:NAME) */
4950 if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
4955 case 'P': /* (*PRUNE) */
4956 if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
4959 case 'S': /* (*SKIP) */
4960 if ( CHECK_WORD("SKIP",start_verb,verb_len) )
4963 case 'T': /* (*THEN) */
4964 /* [19:06] <TimToady> :: is then */
4965 if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
4967 RExC_seen |= REG_SEEN_CUTGROUP;
4973 vFAIL3("Unknown verb pattern '%.*s'",
4974 verb_len, start_verb);
4977 if ( start_arg && internal_argval ) {
4978 vFAIL3("Verb pattern '%.*s' may not have an argument",
4979 verb_len, start_verb);
4980 } else if ( argok < 0 && !start_arg ) {
4981 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
4982 verb_len, start_verb);
4984 ret = reganode(pRExC_state, op, internal_argval);
4985 if ( ! internal_argval && ! SIZE_ONLY ) {
4987 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
4988 ARG(ret) = add_data( pRExC_state, 1, "S" );
4989 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
4996 if (!internal_argval)
4997 RExC_seen |= REG_SEEN_VERBARG;
4998 } else if ( start_arg ) {
4999 vFAIL3("Verb pattern '%.*s' may not have an argument",
5000 verb_len, start_verb);
5002 ret = reg_node(pRExC_state, op);
5004 nextchar(pRExC_state);
5007 if (*RExC_parse == '?') { /* (?...) */
5008 U32 posflags = 0, negflags = 0;
5009 U32 *flagsp = &posflags;
5010 bool is_logical = 0;
5011 const char * const seqstart = RExC_parse;
5014 paren = *RExC_parse++;
5015 ret = NULL; /* For look-ahead/behind. */
5018 case 'P': /* (?P...) variants for those used to PCRE/Python */
5019 paren = *RExC_parse++;
5020 if ( paren == '<') /* (?P<...>) named capture */
5022 else if (paren == '>') { /* (?P>name) named recursion */
5023 goto named_recursion;
5025 else if (paren == '=') { /* (?P=...) named backref */
5026 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5027 you change this make sure you change that */
5028 char* name_start = RExC_parse;
5030 SV *sv_dat = reg_scan_name(pRExC_state,
5031 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5032 if (RExC_parse == name_start || *RExC_parse != ')')
5033 vFAIL2("Sequence %.3s... not terminated",parse_start);
5036 num = add_data( pRExC_state, 1, "S" );
5037 RExC_rxi->data->data[num]=(void*)sv_dat;
5038 SvREFCNT_inc(sv_dat);
5041 ret = reganode(pRExC_state,
5042 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5046 Set_Node_Offset(ret, parse_start+1);
5047 Set_Node_Cur_Length(ret); /* MJD */
5049 nextchar(pRExC_state);
5053 case '<': /* (?<...) */
5054 if (*RExC_parse == '!')
5056 else if (*RExC_parse != '=')
5062 case '\'': /* (?'...') */
5063 name_start= RExC_parse;
5064 svname = reg_scan_name(pRExC_state,
5065 SIZE_ONLY ? /* reverse test from the others */
5066 REG_RSN_RETURN_NAME :
5067 REG_RSN_RETURN_NULL);
5068 if (RExC_parse == name_start)
5070 if (*RExC_parse != paren)
5071 vFAIL2("Sequence (?%c... not terminated",
5072 paren=='>' ? '<' : paren);
5076 if (!svname) /* shouldnt happen */
5078 "panic: reg_scan_name returned NULL");
5079 if (!RExC_paren_names) {
5080 RExC_paren_names= newHV();
5081 sv_2mortal((SV*)RExC_paren_names);
5083 RExC_paren_name_list= newAV();
5084 sv_2mortal((SV*)RExC_paren_name_list);
5087 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5089 sv_dat = HeVAL(he_str);
5091 /* croak baby croak */
5093 "panic: paren_name hash element allocation failed");
5094 } else if ( SvPOK(sv_dat) ) {
5095 IV count=SvIV(sv_dat);
5096 I32 *pv=(I32*)SvGROW(sv_dat,SvCUR(sv_dat)+sizeof(I32)+1);
5097 SvCUR_set(sv_dat,SvCUR(sv_dat)+sizeof(I32));
5098 pv[count]=RExC_npar;
5101 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5102 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5107 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5108 SvREFCNT_dec(svname);
5111 /*sv_dump(sv_dat);*/
5113 nextchar(pRExC_state);
5115 goto capturing_parens;
5117 RExC_seen |= REG_SEEN_LOOKBEHIND;
5119 case '=': /* (?=...) */
5120 case '!': /* (?!...) */
5121 RExC_seen_zerolen++;
5122 if (*RExC_parse == ')') {
5123 ret=reg_node(pRExC_state, OPFAIL);
5124 nextchar(pRExC_state);
5127 case ':': /* (?:...) */
5128 case '>': /* (?>...) */
5130 case '$': /* (?$...) */
5131 case '@': /* (?@...) */
5132 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5134 case '#': /* (?#...) */
5135 while (*RExC_parse && *RExC_parse != ')')
5137 if (*RExC_parse != ')')
5138 FAIL("Sequence (?#... not terminated");
5139 nextchar(pRExC_state);
5142 case '0' : /* (?0) */
5143 case 'R' : /* (?R) */
5144 if (*RExC_parse != ')')
5145 FAIL("Sequence (?R) not terminated");
5146 ret = reg_node(pRExC_state, GOSTART);
5147 nextchar(pRExC_state);
5150 { /* named and numeric backreferences */
5152 case '&': /* (?&NAME) */
5153 parse_start = RExC_parse - 1;
5156 SV *sv_dat = reg_scan_name(pRExC_state,
5157 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5158 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5160 goto gen_recurse_regop;
5163 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5165 vFAIL("Illegal pattern");
5167 goto parse_recursion;
5169 case '-': /* (?-1) */
5170 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5171 RExC_parse--; /* rewind to let it be handled later */
5175 case '1': case '2': case '3': case '4': /* (?1) */
5176 case '5': case '6': case '7': case '8': case '9':
5179 num = atoi(RExC_parse);
5180 parse_start = RExC_parse - 1; /* MJD */
5181 if (*RExC_parse == '-')
5183 while (isDIGIT(*RExC_parse))
5185 if (*RExC_parse!=')')
5186 vFAIL("Expecting close bracket");
5189 if ( paren == '-' ) {
5191 Diagram of capture buffer numbering.
5192 Top line is the normal capture buffer numbers
5193 Botton line is the negative indexing as from
5197 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5201 num = RExC_npar + num;
5204 vFAIL("Reference to nonexistent group");
5206 } else if ( paren == '+' ) {
5207 num = RExC_npar + num - 1;
5210 ret = reganode(pRExC_state, GOSUB, num);
5212 if (num > (I32)RExC_rx->nparens) {
5214 vFAIL("Reference to nonexistent group");
5216 ARG2L_SET( ret, RExC_recurse_count++);
5218 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5219 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5223 RExC_seen |= REG_SEEN_RECURSE;
5224 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5225 Set_Node_Offset(ret, parse_start); /* MJD */
5227 nextchar(pRExC_state);
5229 } /* named and numeric backreferences */
5232 case 'p': /* (?p...) */
5233 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
5234 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
5236 case '?': /* (??...) */
5238 if (*RExC_parse != '{')
5240 paren = *RExC_parse++;
5242 case '{': /* (?{...}) */
5247 char *s = RExC_parse;
5249 RExC_seen_zerolen++;
5250 RExC_seen |= REG_SEEN_EVAL;
5251 while (count && (c = *RExC_parse)) {
5262 if (*RExC_parse != ')') {
5264 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5268 OP_4tree *sop, *rop;
5269 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5272 Perl_save_re_context(aTHX);
5273 rop = sv_compile_2op(sv, &sop, "re", &pad);
5274 sop->op_private |= OPpREFCOUNTED;
5275 /* re_dup will OpREFCNT_inc */
5276 OpREFCNT_set(sop, 1);
5279 n = add_data(pRExC_state, 3, "nop");
5280 RExC_rxi->data->data[n] = (void*)rop;
5281 RExC_rxi->data->data[n+1] = (void*)sop;
5282 RExC_rxi->data->data[n+2] = (void*)pad;
5285 else { /* First pass */
5286 if (PL_reginterp_cnt < ++RExC_seen_evals
5288 /* No compiled RE interpolated, has runtime
5289 components ===> unsafe. */
5290 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5291 if (PL_tainting && PL_tainted)
5292 FAIL("Eval-group in insecure regular expression");
5293 #if PERL_VERSION > 8
5294 if (IN_PERL_COMPILETIME)
5299 nextchar(pRExC_state);
5301 ret = reg_node(pRExC_state, LOGICAL);
5304 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5305 /* deal with the length of this later - MJD */
5308 ret = reganode(pRExC_state, EVAL, n);
5309 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5310 Set_Node_Offset(ret, parse_start);
5313 case '(': /* (?(?{...})...) and (?(?=...)...) */
5316 if (RExC_parse[0] == '?') { /* (?(?...)) */
5317 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5318 || RExC_parse[1] == '<'
5319 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5322 ret = reg_node(pRExC_state, LOGICAL);
5325 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5329 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5330 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5332 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5333 char *name_start= RExC_parse++;
5335 SV *sv_dat=reg_scan_name(pRExC_state,
5336 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5337 if (RExC_parse == name_start || *RExC_parse != ch)
5338 vFAIL2("Sequence (?(%c... not terminated",
5339 (ch == '>' ? '<' : ch));
5342 num = add_data( pRExC_state, 1, "S" );
5343 RExC_rxi->data->data[num]=(void*)sv_dat;
5344 SvREFCNT_inc(sv_dat);
5346 ret = reganode(pRExC_state,NGROUPP,num);
5347 goto insert_if_check_paren;
5349 else if (RExC_parse[0] == 'D' &&
5350 RExC_parse[1] == 'E' &&
5351 RExC_parse[2] == 'F' &&
5352 RExC_parse[3] == 'I' &&
5353 RExC_parse[4] == 'N' &&
5354 RExC_parse[5] == 'E')
5356 ret = reganode(pRExC_state,DEFINEP,0);
5359 goto insert_if_check_paren;
5361 else if (RExC_parse[0] == 'R') {
5364 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5365 parno = atoi(RExC_parse++);
5366 while (isDIGIT(*RExC_parse))
5368 } else if (RExC_parse[0] == '&') {
5371 sv_dat = reg_scan_name(pRExC_state,
5372 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5373 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5375 ret = reganode(pRExC_state,INSUBP,parno);
5376 goto insert_if_check_paren;
5378 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5381 parno = atoi(RExC_parse++);
5383 while (isDIGIT(*RExC_parse))
5385 ret = reganode(pRExC_state, GROUPP, parno);
5387 insert_if_check_paren:
5388 if ((c = *nextchar(pRExC_state)) != ')')
5389 vFAIL("Switch condition not recognized");
5391 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5392 br = regbranch(pRExC_state, &flags, 1,depth+1);
5394 br = reganode(pRExC_state, LONGJMP, 0);
5396 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5397 c = *nextchar(pRExC_state);
5402 vFAIL("(?(DEFINE)....) does not allow branches");
5403 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5404 regbranch(pRExC_state, &flags, 1,depth+1);
5405 REGTAIL(pRExC_state, ret, lastbr);
5408 c = *nextchar(pRExC_state);
5413 vFAIL("Switch (?(condition)... contains too many branches");
5414 ender = reg_node(pRExC_state, TAIL);
5415 REGTAIL(pRExC_state, br, ender);
5417 REGTAIL(pRExC_state, lastbr, ender);
5418 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5421 REGTAIL(pRExC_state, ret, ender);
5425 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5429 RExC_parse--; /* for vFAIL to print correctly */
5430 vFAIL("Sequence (? incomplete");
5434 parse_flags: /* (?i) */
5435 while (*RExC_parse && strchr("iogcmsx", *RExC_parse)) {
5436 /* (?g), (?gc) and (?o) are useless here
5437 and must be globally applied -- japhy */
5439 if (*RExC_parse == 'o' || *RExC_parse == 'g') {
5440 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5441 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5442 if (! (wastedflags & wflagbit) ) {
5443 wastedflags |= wflagbit;
5446 "Useless (%s%c) - %suse /%c modifier",
5447 flagsp == &negflags ? "?-" : "?",
5449 flagsp == &negflags ? "don't " : "",
5455 else if (*RExC_parse == 'c') {
5456 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5457 if (! (wastedflags & WASTED_C) ) {
5458 wastedflags |= WASTED_GC;
5461 "Useless (%sc) - %suse /gc modifier",
5462 flagsp == &negflags ? "?-" : "?",
5463 flagsp == &negflags ? "don't " : ""
5468 else { pmflag(flagsp, *RExC_parse); }
5472 if (*RExC_parse == '-') {
5474 wastedflags = 0; /* reset so (?g-c) warns twice */
5478 RExC_flags |= posflags;
5479 RExC_flags &= ~negflags;
5480 if (*RExC_parse == ':') {
5486 if (*RExC_parse != ')') {
5488 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5490 nextchar(pRExC_state);
5500 ret = reganode(pRExC_state, OPEN, parno);
5503 RExC_nestroot = parno;
5504 if (RExC_seen & REG_SEEN_RECURSE) {
5505 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5506 "Setting open paren #%"IVdf" to %d\n",
5507 (IV)parno, REG_NODE_NUM(ret)));
5508 RExC_open_parens[parno-1]= ret;
5511 Set_Node_Length(ret, 1); /* MJD */
5512 Set_Node_Offset(ret, RExC_parse); /* MJD */
5519 /* Pick up the branches, linking them together. */
5520 parse_start = RExC_parse; /* MJD */
5521 br = regbranch(pRExC_state, &flags, 1,depth+1);
5522 /* branch_len = (paren != 0); */
5526 if (*RExC_parse == '|') {
5527 if (!SIZE_ONLY && RExC_extralen) {
5528 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5531 reginsert(pRExC_state, BRANCH, br, depth+1);
5532 Set_Node_Length(br, paren != 0);
5533 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5537 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5539 else if (paren == ':') {
5540 *flagp |= flags&SIMPLE;
5542 if (is_open) { /* Starts with OPEN. */
5543 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5545 else if (paren != '?') /* Not Conditional */
5547 *flagp |= flags & (SPSTART | HASWIDTH);
5549 while (*RExC_parse == '|') {
5550 if (!SIZE_ONLY && RExC_extralen) {
5551 ender = reganode(pRExC_state, LONGJMP,0);
5552 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5555 RExC_extralen += 2; /* Account for LONGJMP. */
5556 nextchar(pRExC_state);
5557 br = regbranch(pRExC_state, &flags, 0, depth+1);
5561 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5565 *flagp |= flags&SPSTART;
5568 if (have_branch || paren != ':') {
5569 /* Make a closing node, and hook it on the end. */
5572 ender = reg_node(pRExC_state, TAIL);
5575 ender = reganode(pRExC_state, CLOSE, parno);
5576 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5577 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5578 "Setting close paren #%"IVdf" to %d\n",
5579 (IV)parno, REG_NODE_NUM(ender)));
5580 RExC_close_parens[parno-1]= ender;
5581 if (RExC_nestroot == parno)
5584 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5585 Set_Node_Length(ender,1); /* MJD */
5591 *flagp &= ~HASWIDTH;
5594 ender = reg_node(pRExC_state, SUCCEED);
5597 ender = reg_node(pRExC_state, END);
5599 assert(!RExC_opend); /* there can only be one! */
5604 REGTAIL(pRExC_state, lastbr, ender);
5606 if (have_branch && !SIZE_ONLY) {
5608 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5610 /* Hook the tails of the branches to the closing node. */
5611 for (br = ret; br; br = regnext(br)) {
5612 const U8 op = PL_regkind[OP(br)];
5614 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5616 else if (op == BRANCHJ) {
5617 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5625 static const char parens[] = "=!<,>";
5627 if (paren && (p = strchr(parens, paren))) {
5628 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5629 int flag = (p - parens) > 1;
5632 node = SUSPEND, flag = 0;
5633 reginsert(pRExC_state, node,ret, depth+1);
5634 Set_Node_Cur_Length(ret);
5635 Set_Node_Offset(ret, parse_start + 1);
5637 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5641 /* Check for proper termination. */
5643 RExC_flags = oregflags;
5644 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5645 RExC_parse = oregcomp_parse;
5646 vFAIL("Unmatched (");
5649 else if (!paren && RExC_parse < RExC_end) {
5650 if (*RExC_parse == ')') {
5652 vFAIL("Unmatched )");
5655 FAIL("Junk on end of regexp"); /* "Can't happen". */
5663 - regbranch - one alternative of an | operator
5665 * Implements the concatenation operator.
5668 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5671 register regnode *ret;
5672 register regnode *chain = NULL;
5673 register regnode *latest;
5674 I32 flags = 0, c = 0;
5675 GET_RE_DEBUG_FLAGS_DECL;
5676 DEBUG_PARSE("brnc");
5680 if (!SIZE_ONLY && RExC_extralen)
5681 ret = reganode(pRExC_state, BRANCHJ,0);
5683 ret = reg_node(pRExC_state, BRANCH);
5684 Set_Node_Length(ret, 1);
5688 if (!first && SIZE_ONLY)
5689 RExC_extralen += 1; /* BRANCHJ */
5691 *flagp = WORST; /* Tentatively. */
5694 nextchar(pRExC_state);
5695 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5697 latest = regpiece(pRExC_state, &flags,depth+1);
5698 if (latest == NULL) {
5699 if (flags & TRYAGAIN)
5703 else if (ret == NULL)
5705 *flagp |= flags&HASWIDTH;
5706 if (chain == NULL) /* First piece. */
5707 *flagp |= flags&SPSTART;
5710 REGTAIL(pRExC_state, chain, latest);
5715 if (chain == NULL) { /* Loop ran zero times. */
5716 chain = reg_node(pRExC_state, NOTHING);
5721 *flagp |= flags&SIMPLE;
5728 - regpiece - something followed by possible [*+?]
5730 * Note that the branching code sequences used for ? and the general cases
5731 * of * and + are somewhat optimized: they use the same NOTHING node as
5732 * both the endmarker for their branch list and the body of the last branch.
5733 * It might seem that this node could be dispensed with entirely, but the
5734 * endmarker role is not redundant.
5737 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5740 register regnode *ret;
5742 register char *next;
5744 const char * const origparse = RExC_parse;
5746 I32 max = REG_INFTY;
5748 const char *maxpos = NULL;
5749 GET_RE_DEBUG_FLAGS_DECL;
5750 DEBUG_PARSE("piec");
5752 ret = regatom(pRExC_state, &flags,depth+1);
5754 if (flags & TRYAGAIN)
5761 if (op == '{' && regcurly(RExC_parse)) {
5763 parse_start = RExC_parse; /* MJD */
5764 next = RExC_parse + 1;
5765 while (isDIGIT(*next) || *next == ',') {
5774 if (*next == '}') { /* got one */
5778 min = atoi(RExC_parse);
5782 maxpos = RExC_parse;
5784 if (!max && *maxpos != '0')
5785 max = REG_INFTY; /* meaning "infinity" */
5786 else if (max >= REG_INFTY)
5787 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5789 nextchar(pRExC_state);
5792 if ((flags&SIMPLE)) {
5793 RExC_naughty += 2 + RExC_naughty / 2;
5794 reginsert(pRExC_state, CURLY, ret, depth+1);
5795 Set_Node_Offset(ret, parse_start+1); /* MJD */
5796 Set_Node_Cur_Length(ret);
5799 regnode * const w = reg_node(pRExC_state, WHILEM);
5802 REGTAIL(pRExC_state, ret, w);
5803 if (!SIZE_ONLY && RExC_extralen) {
5804 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5805 reginsert(pRExC_state, NOTHING,ret, depth+1);
5806 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5808 reginsert(pRExC_state, CURLYX,ret, depth+1);
5810 Set_Node_Offset(ret, parse_start+1);
5811 Set_Node_Length(ret,
5812 op == '{' ? (RExC_parse - parse_start) : 1);
5814 if (!SIZE_ONLY && RExC_extralen)
5815 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5816 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5818 RExC_whilem_seen++, RExC_extralen += 3;
5819 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5827 if (max && max < min)
5828 vFAIL("Can't do {n,m} with n > m");
5830 ARG1_SET(ret, (U16)min);
5831 ARG2_SET(ret, (U16)max);
5843 #if 0 /* Now runtime fix should be reliable. */
5845 /* if this is reinstated, don't forget to put this back into perldiag:
5847 =item Regexp *+ operand could be empty at {#} in regex m/%s/
5849 (F) The part of the regexp subject to either the * or + quantifier
5850 could match an empty string. The {#} shows in the regular
5851 expression about where the problem was discovered.
5855 if (!(flags&HASWIDTH) && op != '?')
5856 vFAIL("Regexp *+ operand could be empty");
5859 parse_start = RExC_parse;
5860 nextchar(pRExC_state);
5862 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
5864 if (op == '*' && (flags&SIMPLE)) {
5865 reginsert(pRExC_state, STAR, ret, depth+1);
5869 else if (op == '*') {
5873 else if (op == '+' && (flags&SIMPLE)) {
5874 reginsert(pRExC_state, PLUS, ret, depth+1);
5878 else if (op == '+') {
5882 else if (op == '?') {
5887 if (!SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
5889 "%.*s matches null string many times",
5890 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
5894 if (RExC_parse < RExC_end && *RExC_parse == '?') {
5895 nextchar(pRExC_state);
5896 reginsert(pRExC_state, MINMOD, ret, depth+1);
5897 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
5899 #ifndef REG_ALLOW_MINMOD_SUSPEND
5902 if (RExC_parse < RExC_end && *RExC_parse == '+') {
5904 nextchar(pRExC_state);
5905 ender = reg_node(pRExC_state, SUCCEED);
5906 REGTAIL(pRExC_state, ret, ender);
5907 reginsert(pRExC_state, SUSPEND, ret, depth+1);
5909 ender = reg_node(pRExC_state, TAIL);
5910 REGTAIL(pRExC_state, ret, ender);
5914 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
5916 vFAIL("Nested quantifiers");
5923 /* reg_namedseq(pRExC_state,UVp)
5925 This is expected to be called by a parser routine that has
5926 recognized'\N' and needs to handle the rest. RExC_parse is
5927 expected to point at the first char following the N at the time
5930 If valuep is non-null then it is assumed that we are parsing inside
5931 of a charclass definition and the first codepoint in the resolved
5932 string is returned via *valuep and the routine will return NULL.
5933 In this mode if a multichar string is returned from the charnames
5934 handler a warning will be issued, and only the first char in the
5935 sequence will be examined. If the string returned is zero length
5936 then the value of *valuep is undefined and NON-NULL will
5937 be returned to indicate failure. (This will NOT be a valid pointer
5940 If value is null then it is assumed that we are parsing normal text
5941 and inserts a new EXACT node into the program containing the resolved
5942 string and returns a pointer to the new node. If the string is
5943 zerolength a NOTHING node is emitted.
5945 On success RExC_parse is set to the char following the endbrace.
5946 Parsing failures will generate a fatal errorvia vFAIL(...)
5948 NOTE: We cache all results from the charnames handler locally in
5949 the RExC_charnames hash (created on first use) to prevent a charnames
5950 handler from playing silly-buggers and returning a short string and
5951 then a long string for a given pattern. Since the regexp program
5952 size is calculated during an initial parse this would result
5953 in a buffer overrun so we cache to prevent the charname result from
5954 changing during the course of the parse.
5958 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
5960 char * name; /* start of the content of the name */
5961 char * endbrace; /* endbrace following the name */
5964 STRLEN len; /* this has various purposes throughout the code */
5965 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
5966 regnode *ret = NULL;
5968 if (*RExC_parse != '{') {
5969 vFAIL("Missing braces on \\N{}");
5971 name = RExC_parse+1;
5972 endbrace = strchr(RExC_parse, '}');
5975 vFAIL("Missing right brace on \\N{}");
5977 RExC_parse = endbrace + 1;
5980 /* RExC_parse points at the beginning brace,
5981 endbrace points at the last */
5982 if ( name[0]=='U' && name[1]=='+' ) {
5983 /* its a "unicode hex" notation {U+89AB} */
5984 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
5985 | PERL_SCAN_DISALLOW_PREFIX
5986 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
5988 len = (STRLEN)(endbrace - name - 2);
5989 cp = grok_hex(name + 2, &len, &fl, NULL);
5990 if ( len != (STRLEN)(endbrace - name - 2) ) {
5999 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6001 /* fetch the charnames handler for this scope */
6002 HV * const table = GvHV(PL_hintgv);
6004 hv_fetchs(table, "charnames", FALSE) :
6006 SV *cv= cvp ? *cvp : NULL;
6009 /* create an SV with the name as argument */
6010 sv_name = newSVpvn(name, endbrace - name);
6012 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6013 vFAIL2("Constant(\\N{%s}) unknown: "
6014 "(possibly a missing \"use charnames ...\")",
6017 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6018 vFAIL2("Constant(\\N{%s}): "
6019 "$^H{charnames} is not defined",SvPVX(sv_name));
6024 if (!RExC_charnames) {
6025 /* make sure our cache is allocated */
6026 RExC_charnames = newHV();
6027 sv_2mortal((SV*)RExC_charnames);
6029 /* see if we have looked this one up before */
6030 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6032 sv_str = HeVAL(he_str);
6045 count= call_sv(cv, G_SCALAR);
6047 if (count == 1) { /* XXXX is this right? dmq */
6049 SvREFCNT_inc_simple_void(sv_str);
6057 if ( !sv_str || !SvOK(sv_str) ) {
6058 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6059 "did not return a defined value",SvPVX(sv_name));
6061 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6066 char *p = SvPV(sv_str, len);
6069 if ( SvUTF8(sv_str) ) {
6070 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6074 We have to turn on utf8 for high bit chars otherwise
6075 we get failures with
6077 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6078 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6080 This is different from what \x{} would do with the same
6081 codepoint, where the condition is > 0xFF.
6088 /* warn if we havent used the whole string? */
6090 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6092 "Ignoring excess chars from \\N{%s} in character class",
6096 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6098 "Ignoring zero length \\N{%s} in character class",
6103 SvREFCNT_dec(sv_name);
6105 SvREFCNT_dec(sv_str);
6106 return len ? NULL : (regnode *)&len;
6107 } else if(SvCUR(sv_str)) {
6112 char * parse_start = name-3; /* needed for the offsets */
6113 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6115 ret = reg_node(pRExC_state,
6116 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6119 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6120 sv_utf8_upgrade(sv_str);
6121 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6125 p = SvPV(sv_str, len);
6127 /* len is the length written, charlen is the size the char read */
6128 for ( len = 0; p < pend; p += charlen ) {
6130 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6132 STRLEN foldlen,numlen;
6133 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6134 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6135 /* Emit all the Unicode characters. */
6137 for (foldbuf = tmpbuf;
6141 uvc = utf8_to_uvchr(foldbuf, &numlen);
6143 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6146 /* In EBCDIC the numlen
6147 * and unilen can differ. */
6149 if (numlen >= foldlen)
6153 break; /* "Can't happen." */
6156 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6168 RExC_size += STR_SZ(len);
6171 RExC_emit += STR_SZ(len);
6173 Set_Node_Cur_Length(ret); /* MJD */
6175 nextchar(pRExC_state);
6177 ret = reg_node(pRExC_state,NOTHING);
6180 SvREFCNT_dec(sv_str);
6183 SvREFCNT_dec(sv_name);
6193 * It returns the code point in utf8 for the value in *encp.
6194 * value: a code value in the source encoding
6195 * encp: a pointer to an Encode object
6197 * If the result from Encode is not a single character,
6198 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6201 S_reg_recode(pTHX_ const char value, SV **encp)
6204 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6205 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6207 const STRLEN newlen = SvCUR(sv);
6208 UV uv = UNICODE_REPLACEMENT;
6212 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6215 if (!newlen || numlen != newlen) {
6216 uv = UNICODE_REPLACEMENT;
6225 - regatom - the lowest level
6227 * Optimization: gobbles an entire sequence of ordinary characters so that
6228 * it can turn them into a single node, which is smaller to store and
6229 * faster to run. Backslashed characters are exceptions, each becoming a
6230 * separate node; the code is simpler that way and it's not worth fixing.
6232 * [Yes, it is worth fixing, some scripts can run twice the speed.]
6233 * [It looks like its ok, as in S_study_chunk we merge adjacent EXACT nodes]
6236 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6239 register regnode *ret = NULL;
6241 char *parse_start = RExC_parse;
6242 GET_RE_DEBUG_FLAGS_DECL;
6243 DEBUG_PARSE("atom");
6244 *flagp = WORST; /* Tentatively. */
6247 switch (*RExC_parse) {
6249 RExC_seen_zerolen++;
6250 nextchar(pRExC_state);
6251 if (RExC_flags & RXf_PMf_MULTILINE)
6252 ret = reg_node(pRExC_state, MBOL);
6253 else if (RExC_flags & RXf_PMf_SINGLELINE)
6254 ret = reg_node(pRExC_state, SBOL);
6256 ret = reg_node(pRExC_state, BOL);
6257 Set_Node_Length(ret, 1); /* MJD */
6260 nextchar(pRExC_state);
6262 RExC_seen_zerolen++;
6263 if (RExC_flags & RXf_PMf_MULTILINE)
6264 ret = reg_node(pRExC_state, MEOL);
6265 else if (RExC_flags & RXf_PMf_SINGLELINE)
6266 ret = reg_node(pRExC_state, SEOL);
6268 ret = reg_node(pRExC_state, EOL);
6269 Set_Node_Length(ret, 1); /* MJD */
6272 nextchar(pRExC_state);
6273 if (RExC_flags & RXf_PMf_SINGLELINE)
6274 ret = reg_node(pRExC_state, SANY);
6276 ret = reg_node(pRExC_state, REG_ANY);
6277 *flagp |= HASWIDTH|SIMPLE;
6279 Set_Node_Length(ret, 1); /* MJD */
6283 char * const oregcomp_parse = ++RExC_parse;
6284 ret = regclass(pRExC_state,depth+1);
6285 if (*RExC_parse != ']') {
6286 RExC_parse = oregcomp_parse;
6287 vFAIL("Unmatched [");
6289 nextchar(pRExC_state);
6290 *flagp |= HASWIDTH|SIMPLE;
6291 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6295 nextchar(pRExC_state);
6296 ret = reg(pRExC_state, 1, &flags,depth+1);
6298 if (flags & TRYAGAIN) {
6299 if (RExC_parse == RExC_end) {
6300 /* Make parent create an empty node if needed. */
6308 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
6312 if (flags & TRYAGAIN) {
6316 vFAIL("Internal urp");
6317 /* Supposed to be caught earlier. */
6320 if (!regcurly(RExC_parse)) {
6329 vFAIL("Quantifier follows nothing");
6332 switch (*++RExC_parse) {
6334 RExC_seen_zerolen++;
6335 ret = reg_node(pRExC_state, SBOL);
6337 nextchar(pRExC_state);
6338 Set_Node_Length(ret, 2); /* MJD */
6341 ret = reg_node(pRExC_state, GPOS);
6342 RExC_seen |= REG_SEEN_GPOS;
6344 nextchar(pRExC_state);
6345 Set_Node_Length(ret, 2); /* MJD */
6348 ret = reg_node(pRExC_state, SEOL);
6350 RExC_seen_zerolen++; /* Do not optimize RE away */
6351 nextchar(pRExC_state);
6354 ret = reg_node(pRExC_state, EOS);
6356 RExC_seen_zerolen++; /* Do not optimize RE away */
6357 nextchar(pRExC_state);
6358 Set_Node_Length(ret, 2); /* MJD */
6361 ret = reg_node(pRExC_state, CANY);
6362 RExC_seen |= REG_SEEN_CANY;
6363 *flagp |= HASWIDTH|SIMPLE;
6364 nextchar(pRExC_state);
6365 Set_Node_Length(ret, 2); /* MJD */
6368 ret = reg_node(pRExC_state, CLUMP);
6370 nextchar(pRExC_state);
6371 Set_Node_Length(ret, 2); /* MJD */
6374 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6375 *flagp |= HASWIDTH|SIMPLE;
6376 nextchar(pRExC_state);
6377 Set_Node_Length(ret, 2); /* MJD */
6380 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6381 *flagp |= HASWIDTH|SIMPLE;
6382 nextchar(pRExC_state);
6383 Set_Node_Length(ret, 2); /* MJD */
6386 RExC_seen_zerolen++;
6387 RExC_seen |= REG_SEEN_LOOKBEHIND;
6388 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6390 nextchar(pRExC_state);
6391 Set_Node_Length(ret, 2); /* MJD */
6394 RExC_seen_zerolen++;
6395 RExC_seen |= REG_SEEN_LOOKBEHIND;
6396 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6398 nextchar(pRExC_state);
6399 Set_Node_Length(ret, 2); /* MJD */
6402 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6403 *flagp |= HASWIDTH|SIMPLE;
6404 nextchar(pRExC_state);
6405 Set_Node_Length(ret, 2); /* MJD */
6408 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6409 *flagp |= HASWIDTH|SIMPLE;
6410 nextchar(pRExC_state);
6411 Set_Node_Length(ret, 2); /* MJD */
6414 ret = reg_node(pRExC_state, DIGIT);
6415 *flagp |= HASWIDTH|SIMPLE;
6416 nextchar(pRExC_state);
6417 Set_Node_Length(ret, 2); /* MJD */
6420 ret = reg_node(pRExC_state, NDIGIT);
6421 *flagp |= HASWIDTH|SIMPLE;
6422 nextchar(pRExC_state);
6423 Set_Node_Length(ret, 2); /* MJD */
6428 char* const oldregxend = RExC_end;
6429 char* parse_start = RExC_parse - 2;
6431 if (RExC_parse[1] == '{') {
6432 /* a lovely hack--pretend we saw [\pX] instead */
6433 RExC_end = strchr(RExC_parse, '}');
6435 const U8 c = (U8)*RExC_parse;
6437 RExC_end = oldregxend;
6438 vFAIL2("Missing right brace on \\%c{}", c);
6443 RExC_end = RExC_parse + 2;
6444 if (RExC_end > oldregxend)
6445 RExC_end = oldregxend;
6449 ret = regclass(pRExC_state,depth+1);
6451 RExC_end = oldregxend;
6454 Set_Node_Offset(ret, parse_start + 2);
6455 Set_Node_Cur_Length(ret);
6456 nextchar(pRExC_state);
6457 *flagp |= HASWIDTH|SIMPLE;
6461 /* Handle \N{NAME} here and not below because it can be
6462 multicharacter. join_exact() will join them up later on.
6463 Also this makes sure that things like /\N{BLAH}+/ and
6464 \N{BLAH} being multi char Just Happen. dmq*/
6466 ret= reg_namedseq(pRExC_state, NULL);
6468 case 'k': /* Handle \k<NAME> and \k'NAME' */
6471 char ch= RExC_parse[1];
6472 if (ch != '<' && ch != '\'' && ch != '{') {
6474 vFAIL2("Sequence %.2s... not terminated",parse_start);
6476 /* this pretty much dupes the code for (?P=...) in reg(), if
6477 you change this make sure you change that */
6478 char* name_start = (RExC_parse += 2);
6480 SV *sv_dat = reg_scan_name(pRExC_state,
6481 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6482 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
6483 if (RExC_parse == name_start || *RExC_parse != ch)
6484 vFAIL2("Sequence %.3s... not terminated",parse_start);
6487 num = add_data( pRExC_state, 1, "S" );
6488 RExC_rxi->data->data[num]=(void*)sv_dat;
6489 SvREFCNT_inc(sv_dat);
6493 ret = reganode(pRExC_state,
6494 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6498 /* override incorrect value set in reganode MJD */
6499 Set_Node_Offset(ret, parse_start+1);
6500 Set_Node_Cur_Length(ret); /* MJD */
6501 nextchar(pRExC_state);
6517 case '1': case '2': case '3': case '4':
6518 case '5': case '6': case '7': case '8': case '9':
6521 bool isg = *RExC_parse == 'g';
6526 if (*RExC_parse == '{') {
6530 if (*RExC_parse == '-') {
6534 if (hasbrace && !isDIGIT(*RExC_parse)) {
6535 if (isrel) RExC_parse--;
6537 goto parse_named_seq;
6539 num = atoi(RExC_parse);
6541 num = RExC_npar - num;
6543 vFAIL("Reference to nonexistent or unclosed group");
6545 if (!isg && num > 9 && num >= RExC_npar)
6548 char * const parse_start = RExC_parse - 1; /* MJD */
6549 while (isDIGIT(*RExC_parse))
6551 if (parse_start == RExC_parse - 1)
6552 vFAIL("Unterminated \\g... pattern");
6554 if (*RExC_parse != '}')
6555 vFAIL("Unterminated \\g{...} pattern");
6559 if (num > (I32)RExC_rx->nparens)
6560 vFAIL("Reference to nonexistent group");
6563 ret = reganode(pRExC_state,
6564 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6568 /* override incorrect value set in reganode MJD */
6569 Set_Node_Offset(ret, parse_start+1);
6570 Set_Node_Cur_Length(ret); /* MJD */
6572 nextchar(pRExC_state);
6577 if (RExC_parse >= RExC_end)
6578 FAIL("Trailing \\");
6581 /* Do not generate "unrecognized" warnings here, we fall
6582 back into the quick-grab loop below */
6589 if (RExC_flags & RXf_PMf_EXTENDED) {
6590 while (RExC_parse < RExC_end && *RExC_parse != '\n')
6592 if (RExC_parse < RExC_end)
6598 register STRLEN len;
6603 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6605 parse_start = RExC_parse - 1;
6611 ret = reg_node(pRExC_state,
6612 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6614 for (len = 0, p = RExC_parse - 1;
6615 len < 127 && p < RExC_end;
6618 char * const oldp = p;
6620 if (RExC_flags & RXf_PMf_EXTENDED)
6621 p = regwhite(p, RExC_end);
6672 ender = ASCII_TO_NATIVE('\033');
6676 ender = ASCII_TO_NATIVE('\007');
6681 char* const e = strchr(p, '}');
6685 vFAIL("Missing right brace on \\x{}");
6688 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6689 | PERL_SCAN_DISALLOW_PREFIX;
6690 STRLEN numlen = e - p - 1;
6691 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6698 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6700 ender = grok_hex(p, &numlen, &flags, NULL);
6703 if (PL_encoding && ender < 0x100)
6704 goto recode_encoding;
6708 ender = UCHARAT(p++);
6709 ender = toCTRL(ender);
6711 case '0': case '1': case '2': case '3':case '4':
6712 case '5': case '6': case '7': case '8':case '9':
6714 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6717 ender = grok_oct(p, &numlen, &flags, NULL);
6724 if (PL_encoding && ender < 0x100)
6725 goto recode_encoding;
6729 SV* enc = PL_encoding;
6730 ender = reg_recode((const char)(U8)ender, &enc);
6731 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6732 vWARN(p, "Invalid escape in the specified encoding");
6738 FAIL("Trailing \\");
6741 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6742 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6743 goto normal_default;
6748 if (UTF8_IS_START(*p) && UTF) {
6750 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6751 &numlen, UTF8_ALLOW_DEFAULT);
6758 if (RExC_flags & RXf_PMf_EXTENDED)
6759 p = regwhite(p, RExC_end);
6761 /* Prime the casefolded buffer. */
6762 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6764 if (ISMULT2(p)) { /* Back off on ?+*. */
6769 /* Emit all the Unicode characters. */
6771 for (foldbuf = tmpbuf;
6773 foldlen -= numlen) {
6774 ender = utf8_to_uvchr(foldbuf, &numlen);
6776 const STRLEN unilen = reguni(pRExC_state, ender, s);
6779 /* In EBCDIC the numlen
6780 * and unilen can differ. */
6782 if (numlen >= foldlen)
6786 break; /* "Can't happen." */
6790 const STRLEN unilen = reguni(pRExC_state, ender, s);
6799 REGC((char)ender, s++);
6805 /* Emit all the Unicode characters. */
6807 for (foldbuf = tmpbuf;
6809 foldlen -= numlen) {
6810 ender = utf8_to_uvchr(foldbuf, &numlen);
6812 const STRLEN unilen = reguni(pRExC_state, ender, s);
6815 /* In EBCDIC the numlen
6816 * and unilen can differ. */
6818 if (numlen >= foldlen)
6826 const STRLEN unilen = reguni(pRExC_state, ender, s);
6835 REGC((char)ender, s++);
6839 Set_Node_Cur_Length(ret); /* MJD */
6840 nextchar(pRExC_state);
6842 /* len is STRLEN which is unsigned, need to copy to signed */
6845 vFAIL("Internal disaster");
6849 if (len == 1 && UNI_IS_INVARIANT(ender))
6853 RExC_size += STR_SZ(len);
6856 RExC_emit += STR_SZ(len);
6866 S_regwhite(char *p, const char *e)
6871 else if (*p == '#') {
6874 } while (p < e && *p != '\n');
6882 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
6883 Character classes ([:foo:]) can also be negated ([:^foo:]).
6884 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
6885 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
6886 but trigger failures because they are currently unimplemented. */
6888 #define POSIXCC_DONE(c) ((c) == ':')
6889 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
6890 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
6893 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
6896 I32 namedclass = OOB_NAMEDCLASS;
6898 if (value == '[' && RExC_parse + 1 < RExC_end &&
6899 /* I smell either [: or [= or [. -- POSIX has been here, right? */
6900 POSIXCC(UCHARAT(RExC_parse))) {
6901 const char c = UCHARAT(RExC_parse);
6902 char* const s = RExC_parse++;
6904 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
6906 if (RExC_parse == RExC_end)
6907 /* Grandfather lone [:, [=, [. */
6910 const char* const t = RExC_parse++; /* skip over the c */
6913 if (UCHARAT(RExC_parse) == ']') {
6914 const char *posixcc = s + 1;
6915 RExC_parse++; /* skip over the ending ] */
6918 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
6919 const I32 skip = t - posixcc;
6921 /* Initially switch on the length of the name. */
6924 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
6925 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
6928 /* Names all of length 5. */
6929 /* alnum alpha ascii blank cntrl digit graph lower
6930 print punct space upper */
6931 /* Offset 4 gives the best switch position. */
6932 switch (posixcc[4]) {
6934 if (memEQ(posixcc, "alph", 4)) /* alpha */
6935 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
6938 if (memEQ(posixcc, "spac", 4)) /* space */
6939 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
6942 if (memEQ(posixcc, "grap", 4)) /* graph */
6943 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
6946 if (memEQ(posixcc, "asci", 4)) /* ascii */
6947 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
6950 if (memEQ(posixcc, "blan", 4)) /* blank */
6951 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
6954 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
6955 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
6958 if (memEQ(posixcc, "alnu", 4)) /* alnum */
6959 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
6962 if (memEQ(posixcc, "lowe", 4)) /* lower */
6963 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
6964 else if (memEQ(posixcc, "uppe", 4)) /* upper */
6965 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
6968 if (memEQ(posixcc, "digi", 4)) /* digit */
6969 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
6970 else if (memEQ(posixcc, "prin", 4)) /* print */
6971 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
6972 else if (memEQ(posixcc, "punc", 4)) /* punct */
6973 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
6978 if (memEQ(posixcc, "xdigit", 6))
6979 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
6983 if (namedclass == OOB_NAMEDCLASS)
6984 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
6986 assert (posixcc[skip] == ':');
6987 assert (posixcc[skip+1] == ']');
6988 } else if (!SIZE_ONLY) {
6989 /* [[=foo=]] and [[.foo.]] are still future. */
6991 /* adjust RExC_parse so the warning shows after
6993 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
6995 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6998 /* Maternal grandfather:
6999 * "[:" ending in ":" but not in ":]" */
7009 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7012 if (POSIXCC(UCHARAT(RExC_parse))) {
7013 const char *s = RExC_parse;
7014 const char c = *s++;
7018 if (*s && c == *s && s[1] == ']') {
7019 if (ckWARN(WARN_REGEXP))
7021 "POSIX syntax [%c %c] belongs inside character classes",
7024 /* [[=foo=]] and [[.foo.]] are still future. */
7025 if (POSIXCC_NOTYET(c)) {
7026 /* adjust RExC_parse so the error shows after
7028 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7030 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7037 #define _C_C_T_(NAME,TEST,WORD) \
7040 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7042 for (value = 0; value < 256; value++) \
7044 ANYOF_BITMAP_SET(ret, value); \
7049 case ANYOF_N##NAME: \
7051 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7053 for (value = 0; value < 256; value++) \
7055 ANYOF_BITMAP_SET(ret, value); \
7063 parse a class specification and produce either an ANYOF node that
7064 matches the pattern or if the pattern matches a single char only and
7065 that char is < 256 and we are case insensitive then we produce an
7070 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7073 register UV value = 0;
7074 register UV nextvalue;
7075 register IV prevvalue = OOB_UNICODE;
7076 register IV range = 0;
7077 register regnode *ret;
7080 char *rangebegin = NULL;
7081 bool need_class = 0;
7084 bool optimize_invert = TRUE;
7085 AV* unicode_alternate = NULL;
7087 UV literal_endpoint = 0;
7089 UV stored = 0; /* number of chars stored in the class */
7091 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7092 case we need to change the emitted regop to an EXACT. */
7093 const char * orig_parse = RExC_parse;
7094 GET_RE_DEBUG_FLAGS_DECL;
7096 PERL_UNUSED_ARG(depth);
7099 DEBUG_PARSE("clas");
7101 /* Assume we are going to generate an ANYOF node. */
7102 ret = reganode(pRExC_state, ANYOF, 0);
7105 ANYOF_FLAGS(ret) = 0;
7107 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7111 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7115 RExC_size += ANYOF_SKIP;
7116 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7119 RExC_emit += ANYOF_SKIP;
7121 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7123 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7124 ANYOF_BITMAP_ZERO(ret);
7125 listsv = newSVpvs("# comment\n");
7128 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7130 if (!SIZE_ONLY && POSIXCC(nextvalue))
7131 checkposixcc(pRExC_state);
7133 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7134 if (UCHARAT(RExC_parse) == ']')
7138 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7142 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7145 rangebegin = RExC_parse;
7147 value = utf8n_to_uvchr((U8*)RExC_parse,
7148 RExC_end - RExC_parse,
7149 &numlen, UTF8_ALLOW_DEFAULT);
7150 RExC_parse += numlen;
7153 value = UCHARAT(RExC_parse++);
7155 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7156 if (value == '[' && POSIXCC(nextvalue))
7157 namedclass = regpposixcc(pRExC_state, value);
7158 else if (value == '\\') {
7160 value = utf8n_to_uvchr((U8*)RExC_parse,
7161 RExC_end - RExC_parse,
7162 &numlen, UTF8_ALLOW_DEFAULT);
7163 RExC_parse += numlen;
7166 value = UCHARAT(RExC_parse++);
7167 /* Some compilers cannot handle switching on 64-bit integer
7168 * values, therefore value cannot be an UV. Yes, this will
7169 * be a problem later if we want switch on Unicode.
7170 * A similar issue a little bit later when switching on
7171 * namedclass. --jhi */
7172 switch ((I32)value) {
7173 case 'w': namedclass = ANYOF_ALNUM; break;
7174 case 'W': namedclass = ANYOF_NALNUM; break;
7175 case 's': namedclass = ANYOF_SPACE; break;
7176 case 'S': namedclass = ANYOF_NSPACE; break;
7177 case 'd': namedclass = ANYOF_DIGIT; break;
7178 case 'D': namedclass = ANYOF_NDIGIT; break;
7179 case 'N': /* Handle \N{NAME} in class */
7181 /* We only pay attention to the first char of
7182 multichar strings being returned. I kinda wonder
7183 if this makes sense as it does change the behaviour
7184 from earlier versions, OTOH that behaviour was broken
7186 UV v; /* value is register so we cant & it /grrr */
7187 if (reg_namedseq(pRExC_state, &v)) {
7197 if (RExC_parse >= RExC_end)
7198 vFAIL2("Empty \\%c{}", (U8)value);
7199 if (*RExC_parse == '{') {
7200 const U8 c = (U8)value;
7201 e = strchr(RExC_parse++, '}');
7203 vFAIL2("Missing right brace on \\%c{}", c);
7204 while (isSPACE(UCHARAT(RExC_parse)))
7206 if (e == RExC_parse)
7207 vFAIL2("Empty \\%c{}", c);
7209 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7217 if (UCHARAT(RExC_parse) == '^') {
7220 value = value == 'p' ? 'P' : 'p'; /* toggle */
7221 while (isSPACE(UCHARAT(RExC_parse))) {
7226 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7227 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7230 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7231 namedclass = ANYOF_MAX; /* no official name, but it's named */
7234 case 'n': value = '\n'; break;
7235 case 'r': value = '\r'; break;
7236 case 't': value = '\t'; break;
7237 case 'f': value = '\f'; break;
7238 case 'b': value = '\b'; break;
7239 case 'e': value = ASCII_TO_NATIVE('\033');break;
7240 case 'a': value = ASCII_TO_NATIVE('\007');break;
7242 if (*RExC_parse == '{') {
7243 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7244 | PERL_SCAN_DISALLOW_PREFIX;
7245 char * const e = strchr(RExC_parse++, '}');
7247 vFAIL("Missing right brace on \\x{}");
7249 numlen = e - RExC_parse;
7250 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7254 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7256 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7257 RExC_parse += numlen;
7259 if (PL_encoding && value < 0x100)
7260 goto recode_encoding;
7263 value = UCHARAT(RExC_parse++);
7264 value = toCTRL(value);
7266 case '0': case '1': case '2': case '3': case '4':
7267 case '5': case '6': case '7': case '8': case '9':
7271 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7272 RExC_parse += numlen;
7273 if (PL_encoding && value < 0x100)
7274 goto recode_encoding;
7279 SV* enc = PL_encoding;
7280 value = reg_recode((const char)(U8)value, &enc);
7281 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7283 "Invalid escape in the specified encoding");
7287 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7289 "Unrecognized escape \\%c in character class passed through",
7293 } /* end of \blah */
7299 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7301 if (!SIZE_ONLY && !need_class)
7302 ANYOF_CLASS_ZERO(ret);
7306 /* a bad range like a-\d, a-[:digit:] ? */
7309 if (ckWARN(WARN_REGEXP)) {
7311 RExC_parse >= rangebegin ?
7312 RExC_parse - rangebegin : 0;
7314 "False [] range \"%*.*s\"",
7317 if (prevvalue < 256) {
7318 ANYOF_BITMAP_SET(ret, prevvalue);
7319 ANYOF_BITMAP_SET(ret, '-');
7322 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7323 Perl_sv_catpvf(aTHX_ listsv,
7324 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7328 range = 0; /* this was not a true range */
7334 const char *what = NULL;
7337 if (namedclass > OOB_NAMEDCLASS)
7338 optimize_invert = FALSE;
7339 /* Possible truncation here but in some 64-bit environments
7340 * the compiler gets heartburn about switch on 64-bit values.
7341 * A similar issue a little earlier when switching on value.
7343 switch ((I32)namedclass) {
7344 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7345 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7346 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7347 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7348 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7349 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7350 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7351 case _C_C_T_(PRINT, isPRINT(value), "Print");
7352 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7353 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7354 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7355 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7356 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7359 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7362 for (value = 0; value < 128; value++)
7363 ANYOF_BITMAP_SET(ret, value);
7365 for (value = 0; value < 256; value++) {
7367 ANYOF_BITMAP_SET(ret, value);
7376 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7379 for (value = 128; value < 256; value++)
7380 ANYOF_BITMAP_SET(ret, value);
7382 for (value = 0; value < 256; value++) {
7383 if (!isASCII(value))
7384 ANYOF_BITMAP_SET(ret, value);
7393 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7395 /* consecutive digits assumed */
7396 for (value = '0'; value <= '9'; value++)
7397 ANYOF_BITMAP_SET(ret, value);
7404 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7406 /* consecutive digits assumed */
7407 for (value = 0; value < '0'; value++)
7408 ANYOF_BITMAP_SET(ret, value);
7409 for (value = '9' + 1; value < 256; value++)
7410 ANYOF_BITMAP_SET(ret, value);
7416 /* this is to handle \p and \P */
7419 vFAIL("Invalid [::] class");
7423 /* Strings such as "+utf8::isWord\n" */
7424 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7427 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7430 } /* end of namedclass \blah */
7433 if (prevvalue > (IV)value) /* b-a */ {
7434 const int w = RExC_parse - rangebegin;
7435 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7436 range = 0; /* not a valid range */
7440 prevvalue = value; /* save the beginning of the range */
7441 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7442 RExC_parse[1] != ']') {
7445 /* a bad range like \w-, [:word:]- ? */
7446 if (namedclass > OOB_NAMEDCLASS) {
7447 if (ckWARN(WARN_REGEXP)) {
7449 RExC_parse >= rangebegin ?
7450 RExC_parse - rangebegin : 0;
7452 "False [] range \"%*.*s\"",
7456 ANYOF_BITMAP_SET(ret, '-');
7458 range = 1; /* yeah, it's a range! */
7459 continue; /* but do it the next time */
7463 /* now is the next time */
7464 /*stored += (value - prevvalue + 1);*/
7466 if (prevvalue < 256) {
7467 const IV ceilvalue = value < 256 ? value : 255;
7470 /* In EBCDIC [\x89-\x91] should include
7471 * the \x8e but [i-j] should not. */
7472 if (literal_endpoint == 2 &&
7473 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7474 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7476 if (isLOWER(prevvalue)) {
7477 for (i = prevvalue; i <= ceilvalue; i++)
7479 ANYOF_BITMAP_SET(ret, i);
7481 for (i = prevvalue; i <= ceilvalue; i++)
7483 ANYOF_BITMAP_SET(ret, i);
7488 for (i = prevvalue; i <= ceilvalue; i++) {
7489 if (!ANYOF_BITMAP_TEST(ret,i)) {
7491 ANYOF_BITMAP_SET(ret, i);
7495 if (value > 255 || UTF) {
7496 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7497 const UV natvalue = NATIVE_TO_UNI(value);
7498 stored+=2; /* can't optimize this class */
7499 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7500 if (prevnatvalue < natvalue) { /* what about > ? */
7501 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7502 prevnatvalue, natvalue);
7504 else if (prevnatvalue == natvalue) {
7505 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7507 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7509 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7511 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7512 if (RExC_precomp[0] == ':' &&
7513 RExC_precomp[1] == '[' &&
7514 (f == 0xDF || f == 0x92)) {
7515 f = NATIVE_TO_UNI(f);
7518 /* If folding and foldable and a single
7519 * character, insert also the folded version
7520 * to the charclass. */
7522 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7523 if ((RExC_precomp[0] == ':' &&
7524 RExC_precomp[1] == '[' &&
7526 (value == 0xFB05 || value == 0xFB06))) ?
7527 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7528 foldlen == (STRLEN)UNISKIP(f) )
7530 if (foldlen == (STRLEN)UNISKIP(f))
7532 Perl_sv_catpvf(aTHX_ listsv,
7535 /* Any multicharacter foldings
7536 * require the following transform:
7537 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7538 * where E folds into "pq" and F folds
7539 * into "rst", all other characters
7540 * fold to single characters. We save
7541 * away these multicharacter foldings,
7542 * to be later saved as part of the
7543 * additional "s" data. */
7546 if (!unicode_alternate)
7547 unicode_alternate = newAV();
7548 sv = newSVpvn((char*)foldbuf, foldlen);
7550 av_push(unicode_alternate, sv);
7554 /* If folding and the value is one of the Greek
7555 * sigmas insert a few more sigmas to make the
7556 * folding rules of the sigmas to work right.
7557 * Note that not all the possible combinations
7558 * are handled here: some of them are handled
7559 * by the standard folding rules, and some of
7560 * them (literal or EXACTF cases) are handled
7561 * during runtime in regexec.c:S_find_byclass(). */
7562 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7563 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7564 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7565 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7566 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7568 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7569 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7570 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7575 literal_endpoint = 0;
7579 range = 0; /* this range (if it was one) is done now */
7583 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7585 RExC_size += ANYOF_CLASS_ADD_SKIP;
7587 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7593 /****** !SIZE_ONLY AFTER HERE *********/
7595 if( stored == 1 && value < 256
7596 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7598 /* optimize single char class to an EXACT node
7599 but *only* when its not a UTF/high char */
7600 const char * cur_parse= RExC_parse;
7601 RExC_emit = (regnode *)orig_emit;
7602 RExC_parse = (char *)orig_parse;
7603 ret = reg_node(pRExC_state,
7604 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7605 RExC_parse = (char *)cur_parse;
7606 *STRING(ret)= (char)value;
7608 RExC_emit += STR_SZ(1);
7611 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7612 if ( /* If the only flag is folding (plus possibly inversion). */
7613 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7615 for (value = 0; value < 256; ++value) {
7616 if (ANYOF_BITMAP_TEST(ret, value)) {
7617 UV fold = PL_fold[value];
7620 ANYOF_BITMAP_SET(ret, fold);
7623 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7626 /* optimize inverted simple patterns (e.g. [^a-z]) */
7627 if (optimize_invert &&
7628 /* If the only flag is inversion. */
7629 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7630 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7631 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7632 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7635 AV * const av = newAV();
7637 /* The 0th element stores the character class description
7638 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7639 * to initialize the appropriate swash (which gets stored in
7640 * the 1st element), and also useful for dumping the regnode.
7641 * The 2nd element stores the multicharacter foldings,
7642 * used later (regexec.c:S_reginclass()). */
7643 av_store(av, 0, listsv);
7644 av_store(av, 1, NULL);
7645 av_store(av, 2, (SV*)unicode_alternate);
7646 rv = newRV_noinc((SV*)av);
7647 n = add_data(pRExC_state, 1, "s");
7648 RExC_rxi->data->data[n] = (void*)rv;
7657 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7659 char* const retval = RExC_parse++;
7662 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7663 RExC_parse[2] == '#') {
7664 while (*RExC_parse != ')') {
7665 if (RExC_parse == RExC_end)
7666 FAIL("Sequence (?#... not terminated");
7672 if (RExC_flags & RXf_PMf_EXTENDED) {
7673 if (isSPACE(*RExC_parse)) {
7677 else if (*RExC_parse == '#') {
7678 while (RExC_parse < RExC_end)
7679 if (*RExC_parse++ == '\n') break;
7688 - reg_node - emit a node
7690 STATIC regnode * /* Location. */
7691 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7694 register regnode *ptr;
7695 regnode * const ret = RExC_emit;
7696 GET_RE_DEBUG_FLAGS_DECL;
7699 SIZE_ALIGN(RExC_size);
7704 if (OP(RExC_emit) == 255)
7705 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %s: %d ",
7706 reg_name[op], OP(RExC_emit));
7708 NODE_ALIGN_FILL(ret);
7710 FILL_ADVANCE_NODE(ptr, op);
7711 if (RExC_offsets) { /* MJD */
7712 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7713 "reg_node", __LINE__,
7715 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7716 ? "Overwriting end of array!\n" : "OK",
7717 (UV)(RExC_emit - RExC_emit_start),
7718 (UV)(RExC_parse - RExC_start),
7719 (UV)RExC_offsets[0]));
7720 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7728 - reganode - emit a node with an argument
7730 STATIC regnode * /* Location. */
7731 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7734 register regnode *ptr;
7735 regnode * const ret = RExC_emit;
7736 GET_RE_DEBUG_FLAGS_DECL;
7739 SIZE_ALIGN(RExC_size);
7744 assert(2==regarglen[op]+1);
7746 Anything larger than this has to allocate the extra amount.
7747 If we changed this to be:
7749 RExC_size += (1 + regarglen[op]);
7751 then it wouldn't matter. Its not clear what side effect
7752 might come from that so its not done so far.
7758 if (OP(RExC_emit) == 255)
7759 Perl_croak(aTHX_ "panic: reganode overwriting end of allocated program space");
7761 NODE_ALIGN_FILL(ret);
7763 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7764 if (RExC_offsets) { /* MJD */
7765 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7769 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7770 "Overwriting end of array!\n" : "OK",
7771 (UV)(RExC_emit - RExC_emit_start),
7772 (UV)(RExC_parse - RExC_start),
7773 (UV)RExC_offsets[0]));
7774 Set_Cur_Node_Offset;
7782 - reguni - emit (if appropriate) a Unicode character
7785 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
7788 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
7792 - reginsert - insert an operator in front of already-emitted operand
7794 * Means relocating the operand.
7797 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
7800 register regnode *src;
7801 register regnode *dst;
7802 register regnode *place;
7803 const int offset = regarglen[(U8)op];
7804 const int size = NODE_STEP_REGNODE + offset;
7805 GET_RE_DEBUG_FLAGS_DECL;
7806 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
7807 DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
7816 if (RExC_open_parens) {
7818 DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);
7819 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
7820 if ( RExC_open_parens[paren] >= opnd ) {
7821 DEBUG_PARSE_FMT("open"," - %d",size);
7822 RExC_open_parens[paren] += size;
7824 DEBUG_PARSE_FMT("open"," - %s","ok");
7826 if ( RExC_close_parens[paren] >= opnd ) {
7827 DEBUG_PARSE_FMT("close"," - %d",size);
7828 RExC_close_parens[paren] += size;
7830 DEBUG_PARSE_FMT("close"," - %s","ok");
7835 while (src > opnd) {
7836 StructCopy(--src, --dst, regnode);
7837 if (RExC_offsets) { /* MJD 20010112 */
7838 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
7842 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
7843 ? "Overwriting end of array!\n" : "OK",
7844 (UV)(src - RExC_emit_start),
7845 (UV)(dst - RExC_emit_start),
7846 (UV)RExC_offsets[0]));
7847 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
7848 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
7853 place = opnd; /* Op node, where operand used to be. */
7854 if (RExC_offsets) { /* MJD */
7855 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7859 (UV)(place - RExC_emit_start) > RExC_offsets[0]
7860 ? "Overwriting end of array!\n" : "OK",
7861 (UV)(place - RExC_emit_start),
7862 (UV)(RExC_parse - RExC_start),
7863 (UV)RExC_offsets[0]));
7864 Set_Node_Offset(place, RExC_parse);
7865 Set_Node_Length(place, 1);
7867 src = NEXTOPER(place);
7868 FILL_ADVANCE_NODE(place, op);
7869 Zero(src, offset, regnode);
7873 - regtail - set the next-pointer at the end of a node chain of p to val.
7874 - SEE ALSO: regtail_study
7876 /* TODO: All three parms should be const */
7878 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7881 register regnode *scan;
7882 GET_RE_DEBUG_FLAGS_DECL;
7884 PERL_UNUSED_ARG(depth);
7890 /* Find last node. */
7893 regnode * const temp = regnext(scan);
7895 SV * const mysv=sv_newmortal();
7896 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
7897 regprop(RExC_rx, mysv, scan);
7898 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
7899 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
7900 (temp == NULL ? "->" : ""),
7901 (temp == NULL ? reg_name[OP(val)] : "")
7909 if (reg_off_by_arg[OP(scan)]) {
7910 ARG_SET(scan, val - scan);
7913 NEXT_OFF(scan) = val - scan;
7919 - regtail_study - set the next-pointer at the end of a node chain of p to val.
7920 - Look for optimizable sequences at the same time.
7921 - currently only looks for EXACT chains.
7923 This is expermental code. The idea is to use this routine to perform
7924 in place optimizations on branches and groups as they are constructed,
7925 with the long term intention of removing optimization from study_chunk so
7926 that it is purely analytical.
7928 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
7929 to control which is which.
7932 /* TODO: All four parms should be const */
7935 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7938 register regnode *scan;
7940 #ifdef EXPERIMENTAL_INPLACESCAN
7944 GET_RE_DEBUG_FLAGS_DECL;
7950 /* Find last node. */
7954 regnode * const temp = regnext(scan);
7955 #ifdef EXPERIMENTAL_INPLACESCAN
7956 if (PL_regkind[OP(scan)] == EXACT)
7957 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
7965 if( exact == PSEUDO )
7967 else if ( exact != OP(scan) )
7976 SV * const mysv=sv_newmortal();
7977 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
7978 regprop(RExC_rx, mysv, scan);
7979 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
7980 SvPV_nolen_const(mysv),
7989 SV * const mysv_val=sv_newmortal();
7990 DEBUG_PARSE_MSG("");
7991 regprop(RExC_rx, mysv_val, val);
7992 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
7993 SvPV_nolen_const(mysv_val),
7994 (IV)REG_NODE_NUM(val),
7998 if (reg_off_by_arg[OP(scan)]) {
7999 ARG_SET(scan, val - scan);
8002 NEXT_OFF(scan) = val - scan;
8010 - regcurly - a little FSA that accepts {\d+,?\d*}
8013 S_regcurly(register const char *s)
8032 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8035 Perl_regdump(pTHX_ const regexp *r)
8039 SV * const sv = sv_newmortal();
8040 SV *dsv= sv_newmortal();
8043 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8045 /* Header fields of interest. */
8046 if (r->anchored_substr) {
8047 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8048 RE_SV_DUMPLEN(r->anchored_substr), 30);
8049 PerlIO_printf(Perl_debug_log,
8050 "anchored %s%s at %"IVdf" ",
8051 s, RE_SV_TAIL(r->anchored_substr),
8052 (IV)r->anchored_offset);
8053 } else if (r->anchored_utf8) {
8054 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8055 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8056 PerlIO_printf(Perl_debug_log,
8057 "anchored utf8 %s%s at %"IVdf" ",
8058 s, RE_SV_TAIL(r->anchored_utf8),
8059 (IV)r->anchored_offset);
8061 if (r->float_substr) {
8062 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8063 RE_SV_DUMPLEN(r->float_substr), 30);
8064 PerlIO_printf(Perl_debug_log,
8065 "floating %s%s at %"IVdf"..%"UVuf" ",
8066 s, RE_SV_TAIL(r->float_substr),
8067 (IV)r->float_min_offset, (UV)r->float_max_offset);
8068 } else if (r->float_utf8) {
8069 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8070 RE_SV_DUMPLEN(r->float_utf8), 30);
8071 PerlIO_printf(Perl_debug_log,
8072 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8073 s, RE_SV_TAIL(r->float_utf8),
8074 (IV)r->float_min_offset, (UV)r->float_max_offset);
8076 if (r->check_substr || r->check_utf8)
8077 PerlIO_printf(Perl_debug_log,
8079 (r->check_substr == r->float_substr
8080 && r->check_utf8 == r->float_utf8
8081 ? "(checking floating" : "(checking anchored"));
8082 if (r->extflags & RXf_NOSCAN)
8083 PerlIO_printf(Perl_debug_log, " noscan");
8084 if (r->extflags & RXf_CHECK_ALL)
8085 PerlIO_printf(Perl_debug_log, " isall");
8086 if (r->check_substr || r->check_utf8)
8087 PerlIO_printf(Perl_debug_log, ") ");
8089 if (ri->regstclass) {
8090 regprop(r, sv, ri->regstclass);
8091 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8093 if (r->extflags & RXf_ANCH) {
8094 PerlIO_printf(Perl_debug_log, "anchored");
8095 if (r->extflags & RXf_ANCH_BOL)
8096 PerlIO_printf(Perl_debug_log, "(BOL)");
8097 if (r->extflags & RXf_ANCH_MBOL)
8098 PerlIO_printf(Perl_debug_log, "(MBOL)");
8099 if (r->extflags & RXf_ANCH_SBOL)
8100 PerlIO_printf(Perl_debug_log, "(SBOL)");
8101 if (r->extflags & RXf_ANCH_GPOS)
8102 PerlIO_printf(Perl_debug_log, "(GPOS)");
8103 PerlIO_putc(Perl_debug_log, ' ');
8105 if (r->extflags & RXf_GPOS_SEEN)
8106 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8107 if (r->intflags & PREGf_SKIP)
8108 PerlIO_printf(Perl_debug_log, "plus ");
8109 if (r->intflags & PREGf_IMPLICIT)
8110 PerlIO_printf(Perl_debug_log, "implicit ");
8111 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8112 if (r->extflags & RXf_EVAL_SEEN)
8113 PerlIO_printf(Perl_debug_log, "with eval ");
8114 PerlIO_printf(Perl_debug_log, "\n");
8116 PERL_UNUSED_CONTEXT;
8118 #endif /* DEBUGGING */
8122 - regprop - printable representation of opcode
8125 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8130 RXi_GET_DECL(prog,progi);
8131 GET_RE_DEBUG_FLAGS_DECL;
8134 sv_setpvn(sv, "", 0);
8136 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8137 /* It would be nice to FAIL() here, but this may be called from
8138 regexec.c, and it would be hard to supply pRExC_state. */
8139 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8140 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
8142 k = PL_regkind[OP(o)];
8145 SV * const dsv = sv_2mortal(newSVpvs(""));
8146 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8147 * is a crude hack but it may be the best for now since
8148 * we have no flag "this EXACTish node was UTF-8"
8150 const char * const s =
8151 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8152 PL_colors[0], PL_colors[1],
8153 PERL_PV_ESCAPE_UNI_DETECT |
8154 PERL_PV_PRETTY_ELIPSES |
8157 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8158 } else if (k == TRIE) {
8159 /* print the details of the trie in dumpuntil instead, as
8160 * progi->data isn't available here */
8161 const char op = OP(o);
8162 const U32 n = ARG(o);
8163 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8164 (reg_ac_data *)progi->data->data[n] :
8166 const reg_trie_data * const trie
8167 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8169 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
8170 DEBUG_TRIE_COMPILE_r(
8171 Perl_sv_catpvf(aTHX_ sv,
8172 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8173 (UV)trie->startstate,
8174 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8175 (UV)trie->wordcount,
8178 (UV)TRIE_CHARCOUNT(trie),
8179 (UV)trie->uniquecharcount
8182 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8184 int rangestart = -1;
8185 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8186 Perl_sv_catpvf(aTHX_ sv, "[");
8187 for (i = 0; i <= 256; i++) {
8188 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8189 if (rangestart == -1)
8191 } else if (rangestart != -1) {
8192 if (i <= rangestart + 3)
8193 for (; rangestart < i; rangestart++)
8194 put_byte(sv, rangestart);
8196 put_byte(sv, rangestart);
8198 put_byte(sv, i - 1);
8203 Perl_sv_catpvf(aTHX_ sv, "]");
8206 } else if (k == CURLY) {
8207 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8208 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8209 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8211 else if (k == WHILEM && o->flags) /* Ordinal/of */
8212 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8213 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8214 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8215 if ( prog->paren_names ) {
8216 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8217 SV **name= av_fetch(list, ARG(o), 0 );
8219 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8221 } else if (k == NREF) {
8222 if ( prog->paren_names ) {
8223 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8224 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8225 I32 *nums=(I32*)SvPVX(sv_dat);
8226 SV **name= av_fetch(list, nums[0], 0 );
8229 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8230 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8231 (n ? "," : ""), (IV)nums[n]);
8233 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8236 } else if (k == GOSUB)
8237 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8238 else if (k == VERB) {
8240 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8241 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8242 } else if (k == LOGICAL)
8243 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8244 else if (k == ANYOF) {
8245 int i, rangestart = -1;
8246 const U8 flags = ANYOF_FLAGS(o);
8248 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8249 static const char * const anyofs[] = {
8282 if (flags & ANYOF_LOCALE)
8283 sv_catpvs(sv, "{loc}");
8284 if (flags & ANYOF_FOLD)
8285 sv_catpvs(sv, "{i}");
8286 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8287 if (flags & ANYOF_INVERT)
8289 for (i = 0; i <= 256; i++) {
8290 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8291 if (rangestart == -1)
8293 } else if (rangestart != -1) {
8294 if (i <= rangestart + 3)
8295 for (; rangestart < i; rangestart++)
8296 put_byte(sv, rangestart);
8298 put_byte(sv, rangestart);
8300 put_byte(sv, i - 1);
8306 if (o->flags & ANYOF_CLASS)
8307 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8308 if (ANYOF_CLASS_TEST(o,i))
8309 sv_catpv(sv, anyofs[i]);
8311 if (flags & ANYOF_UNICODE)
8312 sv_catpvs(sv, "{unicode}");
8313 else if (flags & ANYOF_UNICODE_ALL)
8314 sv_catpvs(sv, "{unicode_all}");
8318 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8322 U8 s[UTF8_MAXBYTES_CASE+1];
8324 for (i = 0; i <= 256; i++) { /* just the first 256 */
8325 uvchr_to_utf8(s, i);
8327 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8328 if (rangestart == -1)
8330 } else if (rangestart != -1) {
8331 if (i <= rangestart + 3)
8332 for (; rangestart < i; rangestart++) {
8333 const U8 * const e = uvchr_to_utf8(s,rangestart);
8335 for(p = s; p < e; p++)
8339 const U8 *e = uvchr_to_utf8(s,rangestart);
8341 for (p = s; p < e; p++)
8344 e = uvchr_to_utf8(s, i-1);
8345 for (p = s; p < e; p++)
8352 sv_catpvs(sv, "..."); /* et cetera */
8356 char *s = savesvpv(lv);
8357 char * const origs = s;
8359 while (*s && *s != '\n')
8363 const char * const t = ++s;
8381 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8383 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8384 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8386 PERL_UNUSED_CONTEXT;
8387 PERL_UNUSED_ARG(sv);
8389 PERL_UNUSED_ARG(prog);
8390 #endif /* DEBUGGING */
8394 Perl_re_intuit_string(pTHX_ regexp *prog)
8395 { /* Assume that RE_INTUIT is set */
8397 GET_RE_DEBUG_FLAGS_DECL;
8398 PERL_UNUSED_CONTEXT;
8402 const char * const s = SvPV_nolen_const(prog->check_substr
8403 ? prog->check_substr : prog->check_utf8);
8405 if (!PL_colorset) reginitcolors();
8406 PerlIO_printf(Perl_debug_log,
8407 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8409 prog->check_substr ? "" : "utf8 ",
8410 PL_colors[5],PL_colors[0],
8413 (strlen(s) > 60 ? "..." : ""));
8416 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8422 handles refcounting and freeing the perl core regexp structure. When
8423 it is necessary to actually free the structure the first thing it
8424 does is call the 'free' method of the regexp_engine associated to to
8425 the regexp, allowing the handling of the void *pprivate; member
8426 first. (This routine is not overridable by extensions, which is why
8427 the extensions free is called first.)
8429 See regdupe and regdupe_internal if you change anything here.
8431 #ifndef PERL_IN_XSUB_RE
8433 Perl_pregfree(pTHX_ struct regexp *r)
8436 GET_RE_DEBUG_FLAGS_DECL;
8438 if (!r || (--r->refcnt > 0))
8441 CALLREGFREE_PVT(r); /* free the private data */
8443 /* gcov results gave these as non-null 100% of the time, so there's no
8444 optimisation in checking them before calling Safefree */
8445 Safefree(r->precomp);
8446 RX_MATCH_COPY_FREE(r);
8447 #ifdef PERL_OLD_COPY_ON_WRITE
8449 SvREFCNT_dec(r->saved_copy);
8452 if (r->anchored_substr)
8453 SvREFCNT_dec(r->anchored_substr);
8454 if (r->anchored_utf8)
8455 SvREFCNT_dec(r->anchored_utf8);
8456 if (r->float_substr)
8457 SvREFCNT_dec(r->float_substr);
8459 SvREFCNT_dec(r->float_utf8);
8460 Safefree(r->substrs);
8463 SvREFCNT_dec(r->paren_names);
8465 Safefree(r->startp);
8471 /* regfree_internal()
8473 Free the private data in a regexp. This is overloadable by
8474 extensions. Perl takes care of the regexp structure in pregfree(),
8475 this covers the *pprivate pointer which technically perldoesnt
8476 know about, however of course we have to handle the
8477 regexp_internal structure when no extension is in use.
8479 Note this is called before freeing anything in the regexp
8484 Perl_regfree_internal(pTHX_ struct regexp *r)
8488 GET_RE_DEBUG_FLAGS_DECL;
8494 SV *dsv= sv_newmortal();
8495 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8496 dsv, r->precomp, r->prelen, 60);
8497 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8498 PL_colors[4],PL_colors[5],s);
8502 Safefree(ri->offsets); /* 20010421 MJD */
8504 int n = ri->data->count;
8505 PAD* new_comppad = NULL;
8510 /* If you add a ->what type here, update the comment in regcomp.h */
8511 switch (ri->data->what[n]) {
8515 SvREFCNT_dec((SV*)ri->data->data[n]);
8518 Safefree(ri->data->data[n]);
8521 new_comppad = (AV*)ri->data->data[n];
8524 if (new_comppad == NULL)
8525 Perl_croak(aTHX_ "panic: pregfree comppad");
8526 PAD_SAVE_LOCAL(old_comppad,
8527 /* Watch out for global destruction's random ordering. */
8528 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8531 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8534 op_free((OP_4tree*)ri->data->data[n]);
8536 PAD_RESTORE_LOCAL(old_comppad);
8537 SvREFCNT_dec((SV*)new_comppad);
8543 { /* Aho Corasick add-on structure for a trie node.
8544 Used in stclass optimization only */
8546 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8548 refcount = --aho->refcount;
8551 PerlMemShared_free(aho->states);
8552 PerlMemShared_free(aho->fail);
8553 /* do this last!!!! */
8554 PerlMemShared_free(ri->data->data[n]);
8555 PerlMemShared_free(ri->regstclass);
8561 /* trie structure. */
8563 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8565 refcount = --trie->refcount;
8568 PerlMemShared_free(trie->charmap);
8569 PerlMemShared_free(trie->states);
8570 PerlMemShared_free(trie->trans);
8572 PerlMemShared_free(trie->bitmap);
8574 PerlMemShared_free(trie->wordlen);
8576 PerlMemShared_free(trie->jump);
8578 PerlMemShared_free(trie->nextword);
8579 /* do this last!!!! */
8580 PerlMemShared_free(ri->data->data[n]);
8585 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8588 Safefree(ri->data->what);
8592 Safefree(ri->swap->startp);
8593 Safefree(ri->swap->endp);
8599 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8600 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8601 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8602 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8605 regdupe - duplicate a regexp.
8607 This routine is called by sv.c's re_dup and is expected to clone a
8608 given regexp structure. It is a no-op when not under USE_ITHREADS.
8609 (Originally this *was* re_dup() for change history see sv.c)
8611 After all of the core data stored in struct regexp is duplicated
8612 the regexp_engine.dupe method is used to copy any private data
8613 stored in the *pprivate pointer. This allows extensions to handle
8614 any duplication it needs to do.
8616 See pregfree() and regfree_internal() if you change anything here.
8618 #if defined(USE_ITHREADS)
8619 #ifndef PERL_IN_XSUB_RE
8621 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
8626 struct reg_substr_datum *s;
8629 return (REGEXP *)NULL;
8631 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8635 npar = r->nparens+1;
8636 Newxz(ret, 1, regexp);
8637 Newx(ret->startp, npar, I32);
8638 Copy(r->startp, ret->startp, npar, I32);
8639 Newx(ret->endp, npar, I32);
8640 Copy(r->endp, ret->endp, npar, I32);
8643 Newx(ret->substrs, 1, struct reg_substr_data);
8644 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8645 s->min_offset = r->substrs->data[i].min_offset;
8646 s->max_offset = r->substrs->data[i].max_offset;
8647 s->end_shift = r->substrs->data[i].end_shift;
8648 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8649 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8652 ret->substrs = NULL;
8654 ret->precomp = SAVEPVN(r->precomp, r->prelen);
8655 ret->refcnt = r->refcnt;
8656 ret->minlen = r->minlen;
8657 ret->minlenret = r->minlenret;
8658 ret->prelen = r->prelen;
8659 ret->nparens = r->nparens;
8660 ret->lastparen = r->lastparen;
8661 ret->lastcloseparen = r->lastcloseparen;
8662 ret->intflags = r->intflags;
8663 ret->extflags = r->extflags;
8665 ret->sublen = r->sublen;
8667 ret->engine = r->engine;
8669 ret->paren_names = hv_dup_inc(r->paren_names, param);
8671 if (RX_MATCH_COPIED(ret))
8672 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8675 #ifdef PERL_OLD_COPY_ON_WRITE
8676 ret->saved_copy = NULL;
8679 ret->pprivate = r->pprivate;
8681 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
8683 ptr_table_store(PL_ptr_table, r, ret);
8686 #endif /* PERL_IN_XSUB_RE */
8691 This is the internal complement to regdupe() which is used to copy
8692 the structure pointed to by the *pprivate pointer in the regexp.
8693 This is the core version of the extension overridable cloning hook.
8694 The regexp structure being duplicated will be copied by perl prior
8695 to this and will be provided as the regexp *r argument, however
8696 with the /old/ structures pprivate pointer value. Thus this routine
8697 may override any copying normally done by perl.
8699 It returns a pointer to the new regexp_internal structure.
8703 Perl_regdupe_internal(pTHX_ const regexp *r, CLONE_PARAMS *param)
8706 regexp_internal *reti;
8710 npar = r->nparens+1;
8711 len = ri->offsets[0];
8713 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8714 Copy(ri->program, reti->program, len+1, regnode);
8717 Newx(reti->swap, 1, regexp_paren_ofs);
8718 /* no need to copy these */
8719 Newx(reti->swap->startp, npar, I32);
8720 Newx(reti->swap->endp, npar, I32);
8726 reti->regstclass = NULL;
8729 const int count = ri->data->count;
8732 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8733 char, struct reg_data);
8734 Newx(d->what, count, U8);
8737 for (i = 0; i < count; i++) {
8738 d->what[i] = ri->data->what[i];
8739 switch (d->what[i]) {
8740 /* legal options are one of: sSfpontTu
8741 see also regcomp.h and pregfree() */
8744 case 'p': /* actually an AV, but the dup function is identical. */
8745 case 'u': /* actually an HV, but the dup function is identical. */
8746 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
8749 /* This is cheating. */
8750 Newx(d->data[i], 1, struct regnode_charclass_class);
8751 StructCopy(ri->data->data[i], d->data[i],
8752 struct regnode_charclass_class);
8753 reti->regstclass = (regnode*)d->data[i];
8756 /* Compiled op trees are readonly and in shared memory,
8757 and can thus be shared without duplication. */
8759 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
8763 /* Trie stclasses are readonly and can thus be shared
8764 * without duplication. We free the stclass in pregfree
8765 * when the corresponding reg_ac_data struct is freed.
8767 reti->regstclass= ri->regstclass;
8771 ((reg_trie_data*)ri->data->data[i])->refcount++;
8775 d->data[i] = ri->data->data[i];
8778 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
8787 Newx(reti->offsets, 2*len+1, U32);
8788 Copy(ri->offsets, reti->offsets, 2*len+1, U32);
8793 #endif /* USE_ITHREADS */
8798 converts a regexp embedded in a MAGIC struct to its stringified form,
8799 caching the converted form in the struct and returns the cached
8802 If lp is nonnull then it is used to return the length of the
8805 If flags is nonnull and the returned string contains UTF8 then
8806 (*flags & 1) will be true.
8808 If haseval is nonnull then it is used to return whether the pattern
8811 Normally called via macro:
8813 CALLREG_STRINGIFY(mg,&len,&utf8);
8817 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
8819 See sv_2pv_flags() in sv.c for an example of internal usage.
8822 #ifndef PERL_IN_XSUB_RE
8824 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
8826 const regexp * const re = (regexp *)mg->mg_obj;
8829 const char *fptr = "msix";
8834 bool need_newline = 0;
8835 U16 reganch = (U16)((re->extflags & RXf_PMf_COMPILETIME) >> 12);
8837 while((ch = *fptr++)) {
8839 reflags[left++] = ch;
8842 reflags[right--] = ch;
8847 reflags[left] = '-';
8851 mg->mg_len = re->prelen + 4 + left;
8853 * If /x was used, we have to worry about a regex ending with a
8854 * comment later being embedded within another regex. If so, we don't
8855 * want this regex's "commentization" to leak out to the right part of
8856 * the enclosing regex, we must cap it with a newline.
8858 * So, if /x was used, we scan backwards from the end of the regex. If
8859 * we find a '#' before we find a newline, we need to add a newline
8860 * ourself. If we find a '\n' first (or if we don't find '#' or '\n'),
8861 * we don't need to add anything. -jfriedl
8863 if (PMf_EXTENDED & re->extflags) {
8864 const char *endptr = re->precomp + re->prelen;
8865 while (endptr >= re->precomp) {
8866 const char c = *(endptr--);
8868 break; /* don't need another */
8870 /* we end while in a comment, so we need a newline */
8871 mg->mg_len++; /* save space for it */
8872 need_newline = 1; /* note to add it */
8878 Newx(mg->mg_ptr, mg->mg_len + 1 + left, char);
8879 mg->mg_ptr[0] = '(';
8880 mg->mg_ptr[1] = '?';
8881 Copy(reflags, mg->mg_ptr+2, left, char);
8882 *(mg->mg_ptr+left+2) = ':';
8883 Copy(re->precomp, mg->mg_ptr+3+left, re->prelen, char);
8885 mg->mg_ptr[mg->mg_len - 2] = '\n';
8886 mg->mg_ptr[mg->mg_len - 1] = ')';
8887 mg->mg_ptr[mg->mg_len] = 0;
8890 *haseval = re->seen_evals;
8892 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
8900 - regnext - dig the "next" pointer out of a node
8903 Perl_regnext(pTHX_ register regnode *p)
8906 register I32 offset;
8911 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
8920 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
8923 STRLEN l1 = strlen(pat1);
8924 STRLEN l2 = strlen(pat2);
8927 const char *message;
8933 Copy(pat1, buf, l1 , char);
8934 Copy(pat2, buf + l1, l2 , char);
8935 buf[l1 + l2] = '\n';
8936 buf[l1 + l2 + 1] = '\0';
8938 /* ANSI variant takes additional second argument */
8939 va_start(args, pat2);
8943 msv = vmess(buf, &args);
8945 message = SvPV_const(msv,l1);
8948 Copy(message, buf, l1 , char);
8949 buf[l1-1] = '\0'; /* Overwrite \n */
8950 Perl_croak(aTHX_ "%s", buf);
8953 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
8955 #ifndef PERL_IN_XSUB_RE
8957 Perl_save_re_context(pTHX)
8961 struct re_save_state *state;
8963 SAVEVPTR(PL_curcop);
8964 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
8966 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
8967 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
8968 SSPUSHINT(SAVEt_RE_STATE);
8970 Copy(&PL_reg_state, state, 1, struct re_save_state);
8972 PL_reg_start_tmp = 0;
8973 PL_reg_start_tmpl = 0;
8974 PL_reg_oldsaved = NULL;
8975 PL_reg_oldsavedlen = 0;
8977 PL_reg_leftiter = 0;
8978 PL_reg_poscache = NULL;
8979 PL_reg_poscache_size = 0;
8980 #ifdef PERL_OLD_COPY_ON_WRITE
8984 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
8986 const REGEXP * const rx = PM_GETRE(PL_curpm);
8989 for (i = 1; i <= rx->nparens; i++) {
8990 char digits[TYPE_CHARS(long)];
8991 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
8992 GV *const *const gvp
8993 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
8996 GV * const gv = *gvp;
8997 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9007 clear_re(pTHX_ void *r)
9010 ReREFCNT_dec((regexp *)r);
9016 S_put_byte(pTHX_ SV *sv, int c)
9018 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9019 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9020 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9021 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9023 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9027 #define CLEAR_OPTSTART \
9028 if (optstart) STMT_START { \
9029 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9033 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9035 STATIC const regnode *
9036 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9037 const regnode *last, const regnode *plast,
9038 SV* sv, I32 indent, U32 depth)
9041 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9042 register const regnode *next;
9043 const regnode *optstart= NULL;
9046 GET_RE_DEBUG_FLAGS_DECL;
9048 #ifdef DEBUG_DUMPUNTIL
9049 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9050 last ? last-start : 0,plast ? plast-start : 0);
9053 if (plast && plast < last)
9056 while (PL_regkind[op] != END && (!last || node < last)) {
9057 /* While that wasn't END last time... */
9060 if (op == CLOSE || op == WHILEM)
9062 next = regnext((regnode *)node);
9065 if (OP(node) == OPTIMIZED) {
9066 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9073 regprop(r, sv, node);
9074 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9075 (int)(2*indent + 1), "", SvPVX_const(sv));
9077 if (OP(node) != OPTIMIZED) {
9078 if (next == NULL) /* Next ptr. */
9079 PerlIO_printf(Perl_debug_log, " (0)");
9080 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9081 PerlIO_printf(Perl_debug_log, " (FAIL)");
9083 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9084 (void)PerlIO_putc(Perl_debug_log, '\n');
9088 if (PL_regkind[(U8)op] == BRANCHJ) {
9091 register const regnode *nnode = (OP(next) == LONGJMP
9092 ? regnext((regnode *)next)
9094 if (last && nnode > last)
9096 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9099 else if (PL_regkind[(U8)op] == BRANCH) {
9101 DUMPUNTIL(NEXTOPER(node), next);
9103 else if ( PL_regkind[(U8)op] == TRIE ) {
9104 const regnode *this_trie = node;
9105 const char op = OP(node);
9106 const U32 n = ARG(node);
9107 const reg_ac_data * const ac = op>=AHOCORASICK ?
9108 (reg_ac_data *)ri->data->data[n] :
9110 const reg_trie_data * const trie =
9111 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9113 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9115 const regnode *nextbranch= NULL;
9117 sv_setpvn(sv, "", 0);
9118 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9119 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9121 PerlIO_printf(Perl_debug_log, "%*s%s ",
9122 (int)(2*(indent+3)), "",
9123 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9124 PL_colors[0], PL_colors[1],
9125 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9126 PERL_PV_PRETTY_ELIPSES |
9132 U16 dist= trie->jump[word_idx+1];
9133 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9134 (UV)((dist ? this_trie + dist : next) - start));
9137 nextbranch= this_trie + trie->jump[0];
9138 DUMPUNTIL(this_trie + dist, nextbranch);
9140 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9141 nextbranch= regnext((regnode *)nextbranch);
9143 PerlIO_printf(Perl_debug_log, "\n");
9146 if (last && next > last)
9151 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9152 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9153 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9155 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9157 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9159 else if ( op == PLUS || op == STAR) {
9160 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9162 else if (op == ANYOF) {
9163 /* arglen 1 + class block */
9164 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9165 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9166 node = NEXTOPER(node);
9168 else if (PL_regkind[(U8)op] == EXACT) {
9169 /* Literal string, where present. */
9170 node += NODE_SZ_STR(node) - 1;
9171 node = NEXTOPER(node);
9174 node = NEXTOPER(node);
9175 node += regarglen[(U8)op];
9177 if (op == CURLYX || op == OPEN)
9181 #ifdef DEBUG_DUMPUNTIL
9182 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9187 #endif /* DEBUGGING */
9191 * c-indentation-style: bsd
9193 * indent-tabs-mode: t
9196 * ex: set ts=8 sts=4 sw=4 noet: