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 #ifdef RE_TRACK_PATTERN_OFFSETS
155 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
157 #define RExC_emit (pRExC_state->emit)
158 #define RExC_emit_start (pRExC_state->emit_start)
159 #define RExC_naughty (pRExC_state->naughty)
160 #define RExC_sawback (pRExC_state->sawback)
161 #define RExC_seen (pRExC_state->seen)
162 #define RExC_size (pRExC_state->size)
163 #define RExC_npar (pRExC_state->npar)
164 #define RExC_nestroot (pRExC_state->nestroot)
165 #define RExC_extralen (pRExC_state->extralen)
166 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
167 #define RExC_seen_evals (pRExC_state->seen_evals)
168 #define RExC_utf8 (pRExC_state->utf8)
169 #define RExC_charnames (pRExC_state->charnames)
170 #define RExC_open_parens (pRExC_state->open_parens)
171 #define RExC_close_parens (pRExC_state->close_parens)
172 #define RExC_opend (pRExC_state->opend)
173 #define RExC_paren_names (pRExC_state->paren_names)
174 #define RExC_recurse (pRExC_state->recurse)
175 #define RExC_recurse_count (pRExC_state->recurse_count)
178 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
179 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
180 ((*s) == '{' && regcurly(s)))
183 #undef SPSTART /* dratted cpp namespace... */
186 * Flags to be passed up and down.
188 #define WORST 0 /* Worst case. */
189 #define HASWIDTH 0x1 /* Known to match non-null strings. */
190 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
191 #define SPSTART 0x4 /* Starts with * or +. */
192 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
194 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
196 /* whether trie related optimizations are enabled */
197 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
198 #define TRIE_STUDY_OPT
199 #define FULL_TRIE_STUDY
205 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
206 #define PBITVAL(paren) (1 << ((paren) & 7))
207 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
208 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
209 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
212 /* About scan_data_t.
214 During optimisation we recurse through the regexp program performing
215 various inplace (keyhole style) optimisations. In addition study_chunk
216 and scan_commit populate this data structure with information about
217 what strings MUST appear in the pattern. We look for the longest
218 string that must appear for at a fixed location, and we look for the
219 longest string that may appear at a floating location. So for instance
224 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
225 strings (because they follow a .* construct). study_chunk will identify
226 both FOO and BAR as being the longest fixed and floating strings respectively.
228 The strings can be composites, for instance
232 will result in a composite fixed substring 'foo'.
234 For each string some basic information is maintained:
236 - offset or min_offset
237 This is the position the string must appear at, or not before.
238 It also implicitly (when combined with minlenp) tells us how many
239 character must match before the string we are searching.
240 Likewise when combined with minlenp and the length of the string
241 tells us how many characters must appear after the string we have
245 Only used for floating strings. This is the rightmost point that
246 the string can appear at. Ifset to I32 max it indicates that the
247 string can occur infinitely far to the right.
250 A pointer to the minimum length of the pattern that the string
251 was found inside. This is important as in the case of positive
252 lookahead or positive lookbehind we can have multiple patterns
257 The minimum length of the pattern overall is 3, the minimum length
258 of the lookahead part is 3, but the minimum length of the part that
259 will actually match is 1. So 'FOO's minimum length is 3, but the
260 minimum length for the F is 1. This is important as the minimum length
261 is used to determine offsets in front of and behind the string being
262 looked for. Since strings can be composites this is the length of the
263 pattern at the time it was commited with a scan_commit. Note that
264 the length is calculated by study_chunk, so that the minimum lengths
265 are not known until the full pattern has been compiled, thus the
266 pointer to the value.
270 In the case of lookbehind the string being searched for can be
271 offset past the start point of the final matching string.
272 If this value was just blithely removed from the min_offset it would
273 invalidate some of the calculations for how many chars must match
274 before or after (as they are derived from min_offset and minlen and
275 the length of the string being searched for).
276 When the final pattern is compiled and the data is moved from the
277 scan_data_t structure into the regexp structure the information
278 about lookbehind is factored in, with the information that would
279 have been lost precalculated in the end_shift field for the
282 The fields pos_min and pos_delta are used to store the minimum offset
283 and the delta to the maximum offset at the current point in the pattern.
287 typedef struct scan_data_t {
288 /*I32 len_min; unused */
289 /*I32 len_delta; unused */
293 I32 last_end; /* min value, <0 unless valid. */
296 SV **longest; /* Either &l_fixed, or &l_float. */
297 SV *longest_fixed; /* longest fixed string found in pattern */
298 I32 offset_fixed; /* offset where it starts */
299 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
300 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
301 SV *longest_float; /* longest floating string found in pattern */
302 I32 offset_float_min; /* earliest point in string it can appear */
303 I32 offset_float_max; /* latest point in string it can appear */
304 I32 *minlen_float; /* pointer to the minlen relevent to the string */
305 I32 lookbehind_float; /* is the position of the string modified by LB */
309 struct regnode_charclass_class *start_class;
313 * Forward declarations for pregcomp()'s friends.
316 static const scan_data_t zero_scan_data =
317 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
319 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
320 #define SF_BEFORE_SEOL 0x0001
321 #define SF_BEFORE_MEOL 0x0002
322 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
323 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
326 # define SF_FIX_SHIFT_EOL (0+2)
327 # define SF_FL_SHIFT_EOL (0+4)
329 # define SF_FIX_SHIFT_EOL (+2)
330 # define SF_FL_SHIFT_EOL (+4)
333 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
334 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
336 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
337 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
338 #define SF_IS_INF 0x0040
339 #define SF_HAS_PAR 0x0080
340 #define SF_IN_PAR 0x0100
341 #define SF_HAS_EVAL 0x0200
342 #define SCF_DO_SUBSTR 0x0400
343 #define SCF_DO_STCLASS_AND 0x0800
344 #define SCF_DO_STCLASS_OR 0x1000
345 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
346 #define SCF_WHILEM_VISITED_POS 0x2000
348 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
349 #define SCF_SEEN_ACCEPT 0x8000
351 #define UTF (RExC_utf8 != 0)
352 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
353 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
355 #define OOB_UNICODE 12345678
356 #define OOB_NAMEDCLASS -1
358 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
359 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
362 /* length of regex to show in messages that don't mark a position within */
363 #define RegexLengthToShowInErrorMessages 127
366 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
367 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
368 * op/pragma/warn/regcomp.
370 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
371 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
373 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
376 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
377 * arg. Show regex, up to a maximum length. If it's too long, chop and add
380 #define _FAIL(code) STMT_START { \
381 const char *ellipses = ""; \
382 IV len = RExC_end - RExC_precomp; \
385 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
386 if (len > RegexLengthToShowInErrorMessages) { \
387 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
388 len = RegexLengthToShowInErrorMessages - 10; \
394 #define FAIL(msg) _FAIL( \
395 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
396 msg, (int)len, RExC_precomp, ellipses))
398 #define FAIL2(msg,arg) _FAIL( \
399 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
400 arg, (int)len, RExC_precomp, ellipses))
403 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
405 #define Simple_vFAIL(m) STMT_START { \
406 const IV offset = RExC_parse - RExC_precomp; \
407 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
408 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
412 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
414 #define vFAIL(m) STMT_START { \
416 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
421 * Like Simple_vFAIL(), but accepts two arguments.
423 #define Simple_vFAIL2(m,a1) STMT_START { \
424 const IV offset = RExC_parse - RExC_precomp; \
425 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
426 (int)offset, RExC_precomp, RExC_precomp + offset); \
430 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
432 #define vFAIL2(m,a1) STMT_START { \
434 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
435 Simple_vFAIL2(m, a1); \
440 * Like Simple_vFAIL(), but accepts three arguments.
442 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
443 const IV offset = RExC_parse - RExC_precomp; \
444 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
445 (int)offset, RExC_precomp, RExC_precomp + offset); \
449 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
451 #define vFAIL3(m,a1,a2) STMT_START { \
453 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
454 Simple_vFAIL3(m, a1, a2); \
458 * Like Simple_vFAIL(), but accepts four arguments.
460 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
461 const IV offset = RExC_parse - RExC_precomp; \
462 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
463 (int)offset, RExC_precomp, RExC_precomp + offset); \
466 #define vWARN(loc,m) STMT_START { \
467 const IV offset = loc - RExC_precomp; \
468 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
469 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
472 #define vWARNdep(loc,m) STMT_START { \
473 const IV offset = loc - RExC_precomp; \
474 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
475 "%s" REPORT_LOCATION, \
476 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
480 #define vWARN2(loc, m, a1) STMT_START { \
481 const IV offset = loc - RExC_precomp; \
482 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
483 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
486 #define vWARN3(loc, m, a1, a2) STMT_START { \
487 const IV offset = loc - RExC_precomp; \
488 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
489 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
492 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
493 const IV offset = loc - RExC_precomp; \
494 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
495 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
498 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
499 const IV offset = loc - RExC_precomp; \
500 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
501 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
505 /* Allow for side effects in s */
506 #define REGC(c,s) STMT_START { \
507 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
510 /* Macros for recording node offsets. 20001227 mjd@plover.com
511 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
512 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
513 * Element 0 holds the number n.
514 * Position is 1 indexed.
516 #ifndef RE_TRACK_PATTERN_OFFSETS
517 #define Set_Node_Offset_To_R(node,byte)
518 #define Set_Node_Offset(node,byte)
519 #define Set_Cur_Node_Offset
520 #define Set_Node_Length_To_R(node,len)
521 #define Set_Node_Length(node,len)
522 #define Set_Node_Cur_Length(node)
523 #define Node_Offset(n)
524 #define Node_Length(n)
525 #define Set_Node_Offset_Length(node,offset,len)
526 #define ProgLen(ri) ri->u.proglen
527 #define SetProgLen(ri,x) ri->u.proglen = x
529 #define ProgLen(ri) ri->u.offsets[0]
530 #define SetProgLen(ri,x) ri->u.offsets[0] = x
531 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
533 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
534 __LINE__, (int)(node), (int)(byte))); \
536 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
538 RExC_offsets[2*(node)-1] = (byte); \
543 #define Set_Node_Offset(node,byte) \
544 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
545 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
547 #define Set_Node_Length_To_R(node,len) STMT_START { \
549 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
550 __LINE__, (int)(node), (int)(len))); \
552 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
554 RExC_offsets[2*(node)] = (len); \
559 #define Set_Node_Length(node,len) \
560 Set_Node_Length_To_R((node)-RExC_emit_start, len)
561 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
562 #define Set_Node_Cur_Length(node) \
563 Set_Node_Length(node, RExC_parse - parse_start)
565 /* Get offsets and lengths */
566 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
567 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
569 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
570 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
571 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
575 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
576 #define EXPERIMENTAL_INPLACESCAN
577 #endif /*RE_TRACK_PATTERN_OFFSETS*/
579 #define DEBUG_STUDYDATA(str,data,depth) \
580 DEBUG_OPTIMISE_MORE_r(if(data){ \
581 PerlIO_printf(Perl_debug_log, \
582 "%*s" str "Pos:%"IVdf"/%"IVdf \
583 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
584 (int)(depth)*2, "", \
585 (IV)((data)->pos_min), \
586 (IV)((data)->pos_delta), \
587 (UV)((data)->flags), \
588 (IV)((data)->whilem_c), \
589 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
590 is_inf ? "INF " : "" \
592 if ((data)->last_found) \
593 PerlIO_printf(Perl_debug_log, \
594 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
595 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
596 SvPVX_const((data)->last_found), \
597 (IV)((data)->last_end), \
598 (IV)((data)->last_start_min), \
599 (IV)((data)->last_start_max), \
600 ((data)->longest && \
601 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
602 SvPVX_const((data)->longest_fixed), \
603 (IV)((data)->offset_fixed), \
604 ((data)->longest && \
605 (data)->longest==&((data)->longest_float)) ? "*" : "", \
606 SvPVX_const((data)->longest_float), \
607 (IV)((data)->offset_float_min), \
608 (IV)((data)->offset_float_max) \
610 PerlIO_printf(Perl_debug_log,"\n"); \
613 static void clear_re(pTHX_ void *r);
615 /* Mark that we cannot extend a found fixed substring at this point.
616 Update the longest found anchored substring and the longest found
617 floating substrings if needed. */
620 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
622 const STRLEN l = CHR_SVLEN(data->last_found);
623 const STRLEN old_l = CHR_SVLEN(*data->longest);
624 GET_RE_DEBUG_FLAGS_DECL;
626 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
627 SvSetMagicSV(*data->longest, data->last_found);
628 if (*data->longest == data->longest_fixed) {
629 data->offset_fixed = l ? data->last_start_min : data->pos_min;
630 if (data->flags & SF_BEFORE_EOL)
632 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
634 data->flags &= ~SF_FIX_BEFORE_EOL;
635 data->minlen_fixed=minlenp;
636 data->lookbehind_fixed=0;
638 else { /* *data->longest == data->longest_float */
639 data->offset_float_min = l ? data->last_start_min : data->pos_min;
640 data->offset_float_max = (l
641 ? data->last_start_max
642 : data->pos_min + data->pos_delta);
643 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
644 data->offset_float_max = I32_MAX;
645 if (data->flags & SF_BEFORE_EOL)
647 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
649 data->flags &= ~SF_FL_BEFORE_EOL;
650 data->minlen_float=minlenp;
651 data->lookbehind_float=0;
654 SvCUR_set(data->last_found, 0);
656 SV * const sv = data->last_found;
657 if (SvUTF8(sv) && SvMAGICAL(sv)) {
658 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
664 data->flags &= ~SF_BEFORE_EOL;
665 DEBUG_STUDYDATA("commit: ",data,0);
668 /* Can match anything (initialization) */
670 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
672 ANYOF_CLASS_ZERO(cl);
673 ANYOF_BITMAP_SETALL(cl);
674 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
676 cl->flags |= ANYOF_LOCALE;
679 /* Can match anything (initialization) */
681 S_cl_is_anything(const struct regnode_charclass_class *cl)
685 for (value = 0; value <= ANYOF_MAX; value += 2)
686 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
688 if (!(cl->flags & ANYOF_UNICODE_ALL))
690 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
695 /* Can match anything (initialization) */
697 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
699 Zero(cl, 1, struct regnode_charclass_class);
701 cl_anything(pRExC_state, cl);
705 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
707 Zero(cl, 1, struct regnode_charclass_class);
709 cl_anything(pRExC_state, cl);
711 cl->flags |= ANYOF_LOCALE;
714 /* 'And' a given class with another one. Can create false positives */
715 /* We assume that cl is not inverted */
717 S_cl_and(struct regnode_charclass_class *cl,
718 const struct regnode_charclass_class *and_with)
721 assert(and_with->type == ANYOF);
722 if (!(and_with->flags & ANYOF_CLASS)
723 && !(cl->flags & ANYOF_CLASS)
724 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
725 && !(and_with->flags & ANYOF_FOLD)
726 && !(cl->flags & ANYOF_FOLD)) {
729 if (and_with->flags & ANYOF_INVERT)
730 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
731 cl->bitmap[i] &= ~and_with->bitmap[i];
733 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
734 cl->bitmap[i] &= and_with->bitmap[i];
735 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
736 if (!(and_with->flags & ANYOF_EOS))
737 cl->flags &= ~ANYOF_EOS;
739 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
740 !(and_with->flags & ANYOF_INVERT)) {
741 cl->flags &= ~ANYOF_UNICODE_ALL;
742 cl->flags |= ANYOF_UNICODE;
743 ARG_SET(cl, ARG(and_with));
745 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
746 !(and_with->flags & ANYOF_INVERT))
747 cl->flags &= ~ANYOF_UNICODE_ALL;
748 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
749 !(and_with->flags & ANYOF_INVERT))
750 cl->flags &= ~ANYOF_UNICODE;
753 /* 'OR' a given class with another one. Can create false positives */
754 /* We assume that cl is not inverted */
756 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
758 if (or_with->flags & ANYOF_INVERT) {
760 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
761 * <= (B1 | !B2) | (CL1 | !CL2)
762 * which is wasteful if CL2 is small, but we ignore CL2:
763 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
764 * XXXX Can we handle case-fold? Unclear:
765 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
766 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
768 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
769 && !(or_with->flags & ANYOF_FOLD)
770 && !(cl->flags & ANYOF_FOLD) ) {
773 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
774 cl->bitmap[i] |= ~or_with->bitmap[i];
775 } /* XXXX: logic is complicated otherwise */
777 cl_anything(pRExC_state, cl);
780 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
781 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
782 && (!(or_with->flags & ANYOF_FOLD)
783 || (cl->flags & ANYOF_FOLD)) ) {
786 /* OR char bitmap and class bitmap separately */
787 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
788 cl->bitmap[i] |= or_with->bitmap[i];
789 if (or_with->flags & ANYOF_CLASS) {
790 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
791 cl->classflags[i] |= or_with->classflags[i];
792 cl->flags |= ANYOF_CLASS;
795 else { /* XXXX: logic is complicated, leave it along for a moment. */
796 cl_anything(pRExC_state, cl);
799 if (or_with->flags & ANYOF_EOS)
800 cl->flags |= ANYOF_EOS;
802 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
803 ARG(cl) != ARG(or_with)) {
804 cl->flags |= ANYOF_UNICODE_ALL;
805 cl->flags &= ~ANYOF_UNICODE;
807 if (or_with->flags & ANYOF_UNICODE_ALL) {
808 cl->flags |= ANYOF_UNICODE_ALL;
809 cl->flags &= ~ANYOF_UNICODE;
813 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
814 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
815 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
816 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
821 dump_trie(trie,widecharmap,revcharmap)
822 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
823 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
825 These routines dump out a trie in a somewhat readable format.
826 The _interim_ variants are used for debugging the interim
827 tables that are used to generate the final compressed
828 representation which is what dump_trie expects.
830 Part of the reason for their existance is to provide a form
831 of documentation as to how the different representations function.
836 Dumps the final compressed table form of the trie to Perl_debug_log.
837 Used for debugging make_trie().
841 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
842 AV *revcharmap, U32 depth)
845 SV *sv=sv_newmortal();
846 int colwidth= widecharmap ? 6 : 4;
847 GET_RE_DEBUG_FLAGS_DECL;
850 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
851 (int)depth * 2 + 2,"",
852 "Match","Base","Ofs" );
854 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
855 SV ** const tmp = av_fetch( revcharmap, state, 0);
857 PerlIO_printf( Perl_debug_log, "%*s",
859 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
860 PL_colors[0], PL_colors[1],
861 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
862 PERL_PV_ESCAPE_FIRSTCHAR
867 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
868 (int)depth * 2 + 2,"");
870 for( state = 0 ; state < trie->uniquecharcount ; state++ )
871 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
872 PerlIO_printf( Perl_debug_log, "\n");
874 for( state = 1 ; state < trie->statecount ; state++ ) {
875 const U32 base = trie->states[ state ].trans.base;
877 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
879 if ( trie->states[ state ].wordnum ) {
880 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
882 PerlIO_printf( Perl_debug_log, "%6s", "" );
885 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
890 while( ( base + ofs < trie->uniquecharcount ) ||
891 ( base + ofs - trie->uniquecharcount < trie->lasttrans
892 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
895 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
897 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
898 if ( ( base + ofs >= trie->uniquecharcount ) &&
899 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
900 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
902 PerlIO_printf( Perl_debug_log, "%*"UVXf,
904 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
906 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
910 PerlIO_printf( Perl_debug_log, "]");
913 PerlIO_printf( Perl_debug_log, "\n" );
917 Dumps a fully constructed but uncompressed trie in list form.
918 List tries normally only are used for construction when the number of
919 possible chars (trie->uniquecharcount) is very high.
920 Used for debugging make_trie().
923 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
924 HV *widecharmap, AV *revcharmap, U32 next_alloc,
928 SV *sv=sv_newmortal();
929 int colwidth= widecharmap ? 6 : 4;
930 GET_RE_DEBUG_FLAGS_DECL;
931 /* print out the table precompression. */
932 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
933 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
934 "------:-----+-----------------\n" );
936 for( state=1 ; state < next_alloc ; state ++ ) {
939 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
940 (int)depth * 2 + 2,"", (UV)state );
941 if ( ! trie->states[ state ].wordnum ) {
942 PerlIO_printf( Perl_debug_log, "%5s| ","");
944 PerlIO_printf( Perl_debug_log, "W%4x| ",
945 trie->states[ state ].wordnum
948 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
949 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
951 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
953 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
954 PL_colors[0], PL_colors[1],
955 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
956 PERL_PV_ESCAPE_FIRSTCHAR
958 TRIE_LIST_ITEM(state,charid).forid,
959 (UV)TRIE_LIST_ITEM(state,charid).newstate
962 PerlIO_printf(Perl_debug_log, "\n%*s| ",
963 (int)((depth * 2) + 14), "");
966 PerlIO_printf( Perl_debug_log, "\n");
971 Dumps a fully constructed but uncompressed trie in table form.
972 This is the normal DFA style state transition table, with a few
973 twists to facilitate compression later.
974 Used for debugging make_trie().
977 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
978 HV *widecharmap, AV *revcharmap, U32 next_alloc,
983 SV *sv=sv_newmortal();
984 int colwidth= widecharmap ? 6 : 4;
985 GET_RE_DEBUG_FLAGS_DECL;
988 print out the table precompression so that we can do a visual check
989 that they are identical.
992 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
994 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
995 SV ** const tmp = av_fetch( revcharmap, charid, 0);
997 PerlIO_printf( Perl_debug_log, "%*s",
999 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1000 PL_colors[0], PL_colors[1],
1001 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1002 PERL_PV_ESCAPE_FIRSTCHAR
1008 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1010 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1011 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1014 PerlIO_printf( Perl_debug_log, "\n" );
1016 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1018 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1019 (int)depth * 2 + 2,"",
1020 (UV)TRIE_NODENUM( state ) );
1022 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1023 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1025 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1027 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1029 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1030 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1032 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1033 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1040 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1041 startbranch: the first branch in the whole branch sequence
1042 first : start branch of sequence of branch-exact nodes.
1043 May be the same as startbranch
1044 last : Thing following the last branch.
1045 May be the same as tail.
1046 tail : item following the branch sequence
1047 count : words in the sequence
1048 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1049 depth : indent depth
1051 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1053 A trie is an N'ary tree where the branches are determined by digital
1054 decomposition of the key. IE, at the root node you look up the 1st character and
1055 follow that branch repeat until you find the end of the branches. Nodes can be
1056 marked as "accepting" meaning they represent a complete word. Eg:
1060 would convert into the following structure. Numbers represent states, letters
1061 following numbers represent valid transitions on the letter from that state, if
1062 the number is in square brackets it represents an accepting state, otherwise it
1063 will be in parenthesis.
1065 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1069 (1) +-i->(6)-+-s->[7]
1071 +-s->(3)-+-h->(4)-+-e->[5]
1073 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1075 This shows that when matching against the string 'hers' we will begin at state 1
1076 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1077 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1078 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1079 single traverse. We store a mapping from accepting to state to which word was
1080 matched, and then when we have multiple possibilities we try to complete the
1081 rest of the regex in the order in which they occured in the alternation.
1083 The only prior NFA like behaviour that would be changed by the TRIE support is
1084 the silent ignoring of duplicate alternations which are of the form:
1086 / (DUPE|DUPE) X? (?{ ... }) Y /x
1088 Thus EVAL blocks follwing a trie may be called a different number of times with
1089 and without the optimisation. With the optimisations dupes will be silently
1090 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1091 the following demonstrates:
1093 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1095 which prints out 'word' three times, but
1097 'words'=~/(word|word|word)(?{ print $1 })S/
1099 which doesnt print it out at all. This is due to other optimisations kicking in.
1101 Example of what happens on a structural level:
1103 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1105 1: CURLYM[1] {1,32767}(18)
1116 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1117 and should turn into:
1119 1: CURLYM[1] {1,32767}(18)
1121 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1129 Cases where tail != last would be like /(?foo|bar)baz/:
1139 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1140 and would end up looking like:
1143 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1150 d = uvuni_to_utf8_flags(d, uv, 0);
1152 is the recommended Unicode-aware way of saying
1157 #define TRIE_STORE_REVCHAR \
1159 SV *tmp = newSVpvs(""); \
1160 if (UTF) SvUTF8_on(tmp); \
1161 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1162 av_push( revcharmap, tmp ); \
1165 #define TRIE_READ_CHAR STMT_START { \
1169 if ( foldlen > 0 ) { \
1170 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1175 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1176 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1177 foldlen -= UNISKIP( uvc ); \
1178 scan = foldbuf + UNISKIP( uvc ); \
1181 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1191 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1192 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1193 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1194 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1196 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1197 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1198 TRIE_LIST_CUR( state )++; \
1201 #define TRIE_LIST_NEW(state) STMT_START { \
1202 Newxz( trie->states[ state ].trans.list, \
1203 4, reg_trie_trans_le ); \
1204 TRIE_LIST_CUR( state ) = 1; \
1205 TRIE_LIST_LEN( state ) = 4; \
1208 #define TRIE_HANDLE_WORD(state) STMT_START { \
1209 U16 dupe= trie->states[ state ].wordnum; \
1210 regnode * const noper_next = regnext( noper ); \
1212 if (trie->wordlen) \
1213 trie->wordlen[ curword ] = wordlen; \
1215 /* store the word for dumping */ \
1217 if (OP(noper) != NOTHING) \
1218 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1220 tmp = newSVpvn( "", 0 ); \
1221 if ( UTF ) SvUTF8_on( tmp ); \
1222 av_push( trie_words, tmp ); \
1227 if ( noper_next < tail ) { \
1229 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1230 trie->jump[curword] = (U16)(noper_next - convert); \
1232 jumper = noper_next; \
1234 nextbranch= regnext(cur); \
1238 /* So it's a dupe. This means we need to maintain a */\
1239 /* linked-list from the first to the next. */\
1240 /* we only allocate the nextword buffer when there */\
1241 /* a dupe, so first time we have to do the allocation */\
1242 if (!trie->nextword) \
1243 trie->nextword = (U16 *) \
1244 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1245 while ( trie->nextword[dupe] ) \
1246 dupe= trie->nextword[dupe]; \
1247 trie->nextword[dupe]= curword; \
1249 /* we haven't inserted this word yet. */ \
1250 trie->states[ state ].wordnum = curword; \
1255 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1256 ( ( base + charid >= ucharcount \
1257 && base + charid < ubound \
1258 && state == trie->trans[ base - ucharcount + charid ].check \
1259 && trie->trans[ base - ucharcount + charid ].next ) \
1260 ? trie->trans[ base - ucharcount + charid ].next \
1261 : ( state==1 ? special : 0 ) \
1265 #define MADE_JUMP_TRIE 2
1266 #define MADE_EXACT_TRIE 4
1269 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1272 /* first pass, loop through and scan words */
1273 reg_trie_data *trie;
1274 HV *widecharmap = NULL;
1275 AV *revcharmap = newAV();
1277 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1282 regnode *jumper = NULL;
1283 regnode *nextbranch = NULL;
1284 regnode *convert = NULL;
1285 /* we just use folder as a flag in utf8 */
1286 const U8 * const folder = ( flags == EXACTF
1288 : ( flags == EXACTFL
1295 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1296 AV *trie_words = NULL;
1297 /* along with revcharmap, this only used during construction but both are
1298 * useful during debugging so we store them in the struct when debugging.
1301 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1302 STRLEN trie_charcount=0;
1304 SV *re_trie_maxbuff;
1305 GET_RE_DEBUG_FLAGS_DECL;
1307 PERL_UNUSED_ARG(depth);
1310 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1312 trie->startstate = 1;
1313 trie->wordcount = word_count;
1314 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1315 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1316 if (!(UTF && folder))
1317 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1319 trie_words = newAV();
1322 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1323 if (!SvIOK(re_trie_maxbuff)) {
1324 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1327 PerlIO_printf( Perl_debug_log,
1328 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1329 (int)depth * 2 + 2, "",
1330 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1331 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1335 /* Find the node we are going to overwrite */
1336 if ( first == startbranch && OP( last ) != BRANCH ) {
1337 /* whole branch chain */
1340 /* branch sub-chain */
1341 convert = NEXTOPER( first );
1344 /* -- First loop and Setup --
1346 We first traverse the branches and scan each word to determine if it
1347 contains widechars, and how many unique chars there are, this is
1348 important as we have to build a table with at least as many columns as we
1351 We use an array of integers to represent the character codes 0..255
1352 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1353 native representation of the character value as the key and IV's for the
1356 *TODO* If we keep track of how many times each character is used we can
1357 remap the columns so that the table compression later on is more
1358 efficient in terms of memory by ensuring most common value is in the
1359 middle and the least common are on the outside. IMO this would be better
1360 than a most to least common mapping as theres a decent chance the most
1361 common letter will share a node with the least common, meaning the node
1362 will not be compressable. With a middle is most common approach the worst
1363 case is when we have the least common nodes twice.
1367 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1368 regnode * const noper = NEXTOPER( cur );
1369 const U8 *uc = (U8*)STRING( noper );
1370 const U8 * const e = uc + STR_LEN( noper );
1372 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1373 const U8 *scan = (U8*)NULL;
1374 U32 wordlen = 0; /* required init */
1377 if (OP(noper) == NOTHING) {
1382 TRIE_BITMAP_SET(trie,*uc);
1383 if ( folder ) TRIE_BITMAP_SET(trie,folder[ *uc ]);
1385 for ( ; uc < e ; uc += len ) {
1386 TRIE_CHARCOUNT(trie)++;
1390 if ( !trie->charmap[ uvc ] ) {
1391 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1393 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1399 widecharmap = newHV();
1401 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1404 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1406 if ( !SvTRUE( *svpp ) ) {
1407 sv_setiv( *svpp, ++trie->uniquecharcount );
1412 if( cur == first ) {
1415 } else if (chars < trie->minlen) {
1417 } else if (chars > trie->maxlen) {
1421 } /* end first pass */
1422 DEBUG_TRIE_COMPILE_r(
1423 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1424 (int)depth * 2 + 2,"",
1425 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1426 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1427 (int)trie->minlen, (int)trie->maxlen )
1429 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1432 We now know what we are dealing with in terms of unique chars and
1433 string sizes so we can calculate how much memory a naive
1434 representation using a flat table will take. If it's over a reasonable
1435 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1436 conservative but potentially much slower representation using an array
1439 At the end we convert both representations into the same compressed
1440 form that will be used in regexec.c for matching with. The latter
1441 is a form that cannot be used to construct with but has memory
1442 properties similar to the list form and access properties similar
1443 to the table form making it both suitable for fast searches and
1444 small enough that its feasable to store for the duration of a program.
1446 See the comment in the code where the compressed table is produced
1447 inplace from the flat tabe representation for an explanation of how
1448 the compression works.
1453 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1455 Second Pass -- Array Of Lists Representation
1457 Each state will be represented by a list of charid:state records
1458 (reg_trie_trans_le) the first such element holds the CUR and LEN
1459 points of the allocated array. (See defines above).
1461 We build the initial structure using the lists, and then convert
1462 it into the compressed table form which allows faster lookups
1463 (but cant be modified once converted).
1466 STRLEN transcount = 1;
1468 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1469 "%*sCompiling trie using list compiler\n",
1470 (int)depth * 2 + 2, ""));
1472 trie->states = (reg_trie_state *)
1473 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1474 sizeof(reg_trie_state) );
1478 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1480 regnode * const noper = NEXTOPER( cur );
1481 U8 *uc = (U8*)STRING( noper );
1482 const U8 * const e = uc + STR_LEN( noper );
1483 U32 state = 1; /* required init */
1484 U16 charid = 0; /* sanity init */
1485 U8 *scan = (U8*)NULL; /* sanity init */
1486 STRLEN foldlen = 0; /* required init */
1487 U32 wordlen = 0; /* required init */
1488 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1490 if (OP(noper) != NOTHING) {
1491 for ( ; uc < e ; uc += len ) {
1496 charid = trie->charmap[ uvc ];
1498 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1502 charid=(U16)SvIV( *svpp );
1505 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1512 if ( !trie->states[ state ].trans.list ) {
1513 TRIE_LIST_NEW( state );
1515 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1516 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1517 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1522 newstate = next_alloc++;
1523 TRIE_LIST_PUSH( state, charid, newstate );
1528 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1532 TRIE_HANDLE_WORD(state);
1534 } /* end second pass */
1536 /* next alloc is the NEXT state to be allocated */
1537 trie->statecount = next_alloc;
1538 trie->states = (reg_trie_state *)
1539 PerlMemShared_realloc( trie->states,
1541 * sizeof(reg_trie_state) );
1543 /* and now dump it out before we compress it */
1544 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1545 revcharmap, next_alloc,
1549 trie->trans = (reg_trie_trans *)
1550 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1557 for( state=1 ; state < next_alloc ; state ++ ) {
1561 DEBUG_TRIE_COMPILE_MORE_r(
1562 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1566 if (trie->states[state].trans.list) {
1567 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1571 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1572 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1573 if ( forid < minid ) {
1575 } else if ( forid > maxid ) {
1579 if ( transcount < tp + maxid - minid + 1) {
1581 trie->trans = (reg_trie_trans *)
1582 PerlMemShared_realloc( trie->trans,
1584 * sizeof(reg_trie_trans) );
1585 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1587 base = trie->uniquecharcount + tp - minid;
1588 if ( maxid == minid ) {
1590 for ( ; zp < tp ; zp++ ) {
1591 if ( ! trie->trans[ zp ].next ) {
1592 base = trie->uniquecharcount + zp - minid;
1593 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1594 trie->trans[ zp ].check = state;
1600 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1601 trie->trans[ tp ].check = state;
1606 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1607 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1608 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1609 trie->trans[ tid ].check = state;
1611 tp += ( maxid - minid + 1 );
1613 Safefree(trie->states[ state ].trans.list);
1616 DEBUG_TRIE_COMPILE_MORE_r(
1617 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1620 trie->states[ state ].trans.base=base;
1622 trie->lasttrans = tp + 1;
1626 Second Pass -- Flat Table Representation.
1628 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1629 We know that we will need Charcount+1 trans at most to store the data
1630 (one row per char at worst case) So we preallocate both structures
1631 assuming worst case.
1633 We then construct the trie using only the .next slots of the entry
1636 We use the .check field of the first entry of the node temporarily to
1637 make compression both faster and easier by keeping track of how many non
1638 zero fields are in the node.
1640 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1643 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1644 number representing the first entry of the node, and state as a
1645 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1646 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1647 are 2 entrys per node. eg:
1655 The table is internally in the right hand, idx form. However as we also
1656 have to deal with the states array which is indexed by nodenum we have to
1657 use TRIE_NODENUM() to convert.
1660 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1661 "%*sCompiling trie using table compiler\n",
1662 (int)depth * 2 + 2, ""));
1664 trie->trans = (reg_trie_trans *)
1665 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1666 * trie->uniquecharcount + 1,
1667 sizeof(reg_trie_trans) );
1668 trie->states = (reg_trie_state *)
1669 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1670 sizeof(reg_trie_state) );
1671 next_alloc = trie->uniquecharcount + 1;
1674 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1676 regnode * const noper = NEXTOPER( cur );
1677 const U8 *uc = (U8*)STRING( noper );
1678 const U8 * const e = uc + STR_LEN( noper );
1680 U32 state = 1; /* required init */
1682 U16 charid = 0; /* sanity init */
1683 U32 accept_state = 0; /* sanity init */
1684 U8 *scan = (U8*)NULL; /* sanity init */
1686 STRLEN foldlen = 0; /* required init */
1687 U32 wordlen = 0; /* required init */
1688 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1690 if ( OP(noper) != NOTHING ) {
1691 for ( ; uc < e ; uc += len ) {
1696 charid = trie->charmap[ uvc ];
1698 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1699 charid = svpp ? (U16)SvIV(*svpp) : 0;
1703 if ( !trie->trans[ state + charid ].next ) {
1704 trie->trans[ state + charid ].next = next_alloc;
1705 trie->trans[ state ].check++;
1706 next_alloc += trie->uniquecharcount;
1708 state = trie->trans[ state + charid ].next;
1710 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1712 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1715 accept_state = TRIE_NODENUM( state );
1716 TRIE_HANDLE_WORD(accept_state);
1718 } /* end second pass */
1720 /* and now dump it out before we compress it */
1721 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1723 next_alloc, depth+1));
1727 * Inplace compress the table.*
1729 For sparse data sets the table constructed by the trie algorithm will
1730 be mostly 0/FAIL transitions or to put it another way mostly empty.
1731 (Note that leaf nodes will not contain any transitions.)
1733 This algorithm compresses the tables by eliminating most such
1734 transitions, at the cost of a modest bit of extra work during lookup:
1736 - Each states[] entry contains a .base field which indicates the
1737 index in the state[] array wheres its transition data is stored.
1739 - If .base is 0 there are no valid transitions from that node.
1741 - If .base is nonzero then charid is added to it to find an entry in
1744 -If trans[states[state].base+charid].check!=state then the
1745 transition is taken to be a 0/Fail transition. Thus if there are fail
1746 transitions at the front of the node then the .base offset will point
1747 somewhere inside the previous nodes data (or maybe even into a node
1748 even earlier), but the .check field determines if the transition is
1752 The following process inplace converts the table to the compressed
1753 table: We first do not compress the root node 1,and mark its all its
1754 .check pointers as 1 and set its .base pointer as 1 as well. This
1755 allows to do a DFA construction from the compressed table later, and
1756 ensures that any .base pointers we calculate later are greater than
1759 - We set 'pos' to indicate the first entry of the second node.
1761 - We then iterate over the columns of the node, finding the first and
1762 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1763 and set the .check pointers accordingly, and advance pos
1764 appropriately and repreat for the next node. Note that when we copy
1765 the next pointers we have to convert them from the original
1766 NODEIDX form to NODENUM form as the former is not valid post
1769 - If a node has no transitions used we mark its base as 0 and do not
1770 advance the pos pointer.
1772 - If a node only has one transition we use a second pointer into the
1773 structure to fill in allocated fail transitions from other states.
1774 This pointer is independent of the main pointer and scans forward
1775 looking for null transitions that are allocated to a state. When it
1776 finds one it writes the single transition into the "hole". If the
1777 pointer doesnt find one the single transition is appended as normal.
1779 - Once compressed we can Renew/realloc the structures to release the
1782 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1783 specifically Fig 3.47 and the associated pseudocode.
1787 const U32 laststate = TRIE_NODENUM( next_alloc );
1790 trie->statecount = laststate;
1792 for ( state = 1 ; state < laststate ; state++ ) {
1794 const U32 stateidx = TRIE_NODEIDX( state );
1795 const U32 o_used = trie->trans[ stateidx ].check;
1796 U32 used = trie->trans[ stateidx ].check;
1797 trie->trans[ stateidx ].check = 0;
1799 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1800 if ( flag || trie->trans[ stateidx + charid ].next ) {
1801 if ( trie->trans[ stateidx + charid ].next ) {
1803 for ( ; zp < pos ; zp++ ) {
1804 if ( ! trie->trans[ zp ].next ) {
1808 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1809 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1810 trie->trans[ zp ].check = state;
1811 if ( ++zp > pos ) pos = zp;
1818 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1820 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1821 trie->trans[ pos ].check = state;
1826 trie->lasttrans = pos + 1;
1827 trie->states = (reg_trie_state *)
1828 PerlMemShared_realloc( trie->states, laststate
1829 * sizeof(reg_trie_state) );
1830 DEBUG_TRIE_COMPILE_MORE_r(
1831 PerlIO_printf( Perl_debug_log,
1832 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1833 (int)depth * 2 + 2,"",
1834 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1837 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1840 } /* end table compress */
1842 DEBUG_TRIE_COMPILE_MORE_r(
1843 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1844 (int)depth * 2 + 2, "",
1845 (UV)trie->statecount,
1846 (UV)trie->lasttrans)
1848 /* resize the trans array to remove unused space */
1849 trie->trans = (reg_trie_trans *)
1850 PerlMemShared_realloc( trie->trans, trie->lasttrans
1851 * sizeof(reg_trie_trans) );
1853 /* and now dump out the compressed format */
1854 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1856 { /* Modify the program and insert the new TRIE node*/
1857 U8 nodetype =(U8)(flags & 0xFF);
1861 regnode *optimize = NULL;
1862 #ifdef RE_TRACK_PATTERN_OFFSETS
1865 U32 mjd_nodelen = 0;
1866 #endif /* RE_TRACK_PATTERN_OFFSETS */
1867 #endif /* DEBUGGING */
1869 This means we convert either the first branch or the first Exact,
1870 depending on whether the thing following (in 'last') is a branch
1871 or not and whther first is the startbranch (ie is it a sub part of
1872 the alternation or is it the whole thing.)
1873 Assuming its a sub part we conver the EXACT otherwise we convert
1874 the whole branch sequence, including the first.
1876 /* Find the node we are going to overwrite */
1877 if ( first != startbranch || OP( last ) == BRANCH ) {
1878 /* branch sub-chain */
1879 NEXT_OFF( first ) = (U16)(last - first);
1880 #ifdef RE_TRACK_PATTERN_OFFSETS
1882 mjd_offset= Node_Offset((convert));
1883 mjd_nodelen= Node_Length((convert));
1886 /* whole branch chain */
1888 #ifdef RE_TRACK_PATTERN_OFFSETS
1891 const regnode *nop = NEXTOPER( convert );
1892 mjd_offset= Node_Offset((nop));
1893 mjd_nodelen= Node_Length((nop));
1897 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1898 (int)depth * 2 + 2, "",
1899 (UV)mjd_offset, (UV)mjd_nodelen)
1902 /* But first we check to see if there is a common prefix we can
1903 split out as an EXACT and put in front of the TRIE node. */
1904 trie->startstate= 1;
1905 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1907 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1911 const U32 base = trie->states[ state ].trans.base;
1913 if ( trie->states[state].wordnum )
1916 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1917 if ( ( base + ofs >= trie->uniquecharcount ) &&
1918 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1919 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1921 if ( ++count > 1 ) {
1922 SV **tmp = av_fetch( revcharmap, ofs, 0);
1923 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1924 if ( state == 1 ) break;
1926 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1928 PerlIO_printf(Perl_debug_log,
1929 "%*sNew Start State=%"UVuf" Class: [",
1930 (int)depth * 2 + 2, "",
1933 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1934 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1936 TRIE_BITMAP_SET(trie,*ch);
1938 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1940 PerlIO_printf(Perl_debug_log, (char*)ch)
1944 TRIE_BITMAP_SET(trie,*ch);
1946 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1947 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1953 SV **tmp = av_fetch( revcharmap, idx, 0);
1954 char *ch = SvPV_nolen( *tmp );
1956 SV *sv=sv_newmortal();
1957 PerlIO_printf( Perl_debug_log,
1958 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1959 (int)depth * 2 + 2, "",
1961 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1962 PL_colors[0], PL_colors[1],
1963 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1964 PERL_PV_ESCAPE_FIRSTCHAR
1969 OP( convert ) = nodetype;
1970 str=STRING(convert);
1981 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
1987 regnode *n = convert+NODE_SZ_STR(convert);
1988 NEXT_OFF(convert) = NODE_SZ_STR(convert);
1989 trie->startstate = state;
1990 trie->minlen -= (state - 1);
1991 trie->maxlen -= (state - 1);
1993 regnode *fix = convert;
1994 U32 word = trie->wordcount;
1996 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
1997 while( ++fix < n ) {
1998 Set_Node_Offset_Length(fix, 0, 0);
2001 SV ** const tmp = av_fetch( trie_words, word, 0 );
2003 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2004 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2006 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2013 NEXT_OFF(convert) = (U16)(tail - convert);
2014 DEBUG_r(optimize= n);
2020 if ( trie->maxlen ) {
2021 NEXT_OFF( convert ) = (U16)(tail - convert);
2022 ARG_SET( convert, data_slot );
2023 /* Store the offset to the first unabsorbed branch in
2024 jump[0], which is otherwise unused by the jump logic.
2025 We use this when dumping a trie and during optimisation. */
2027 trie->jump[0] = (U16)(nextbranch - convert);
2030 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2031 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2033 OP( convert ) = TRIEC;
2034 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2035 PerlMemShared_free(trie->bitmap);
2038 OP( convert ) = TRIE;
2040 /* store the type in the flags */
2041 convert->flags = nodetype;
2045 + regarglen[ OP( convert ) ];
2047 /* XXX We really should free up the resource in trie now,
2048 as we won't use them - (which resources?) dmq */
2050 /* needed for dumping*/
2051 DEBUG_r(if (optimize) {
2052 regnode *opt = convert;
2054 while ( ++opt < optimize) {
2055 Set_Node_Offset_Length(opt,0,0);
2058 Try to clean up some of the debris left after the
2061 while( optimize < jumper ) {
2062 mjd_nodelen += Node_Length((optimize));
2063 OP( optimize ) = OPTIMIZED;
2064 Set_Node_Offset_Length(optimize,0,0);
2067 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2069 } /* end node insert */
2070 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2072 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2073 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2075 SvREFCNT_dec(revcharmap);
2079 : trie->startstate>1
2085 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2087 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2089 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2090 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2093 We find the fail state for each state in the trie, this state is the longest proper
2094 suffix of the current states 'word' that is also a proper prefix of another word in our
2095 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2096 the DFA not to have to restart after its tried and failed a word at a given point, it
2097 simply continues as though it had been matching the other word in the first place.
2099 'abcdgu'=~/abcdefg|cdgu/
2100 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2101 fail, which would bring use to the state representing 'd' in the second word where we would
2102 try 'g' and succeed, prodceding to match 'cdgu'.
2104 /* add a fail transition */
2105 const U32 trie_offset = ARG(source);
2106 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2108 const U32 ucharcount = trie->uniquecharcount;
2109 const U32 numstates = trie->statecount;
2110 const U32 ubound = trie->lasttrans + ucharcount;
2114 U32 base = trie->states[ 1 ].trans.base;
2117 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2118 GET_RE_DEBUG_FLAGS_DECL;
2120 PERL_UNUSED_ARG(depth);
2124 ARG_SET( stclass, data_slot );
2125 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2126 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2127 aho->trie=trie_offset;
2128 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2129 Copy( trie->states, aho->states, numstates, reg_trie_state );
2130 Newxz( q, numstates, U32);
2131 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2134 /* initialize fail[0..1] to be 1 so that we always have
2135 a valid final fail state */
2136 fail[ 0 ] = fail[ 1 ] = 1;
2138 for ( charid = 0; charid < ucharcount ; charid++ ) {
2139 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2141 q[ q_write ] = newstate;
2142 /* set to point at the root */
2143 fail[ q[ q_write++ ] ]=1;
2146 while ( q_read < q_write) {
2147 const U32 cur = q[ q_read++ % numstates ];
2148 base = trie->states[ cur ].trans.base;
2150 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2151 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2153 U32 fail_state = cur;
2156 fail_state = fail[ fail_state ];
2157 fail_base = aho->states[ fail_state ].trans.base;
2158 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2160 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2161 fail[ ch_state ] = fail_state;
2162 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2164 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2166 q[ q_write++ % numstates] = ch_state;
2170 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2171 when we fail in state 1, this allows us to use the
2172 charclass scan to find a valid start char. This is based on the principle
2173 that theres a good chance the string being searched contains lots of stuff
2174 that cant be a start char.
2176 fail[ 0 ] = fail[ 1 ] = 0;
2177 DEBUG_TRIE_COMPILE_r({
2178 PerlIO_printf(Perl_debug_log,
2179 "%*sStclass Failtable (%"UVuf" states): 0",
2180 (int)(depth * 2), "", (UV)numstates
2182 for( q_read=1; q_read<numstates; q_read++ ) {
2183 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2185 PerlIO_printf(Perl_debug_log, "\n");
2188 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2193 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2194 * These need to be revisited when a newer toolchain becomes available.
2196 #if defined(__sparc64__) && defined(__GNUC__)
2197 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2198 # undef SPARC64_GCC_WORKAROUND
2199 # define SPARC64_GCC_WORKAROUND 1
2203 #define DEBUG_PEEP(str,scan,depth) \
2204 DEBUG_OPTIMISE_r({if (scan){ \
2205 SV * const mysv=sv_newmortal(); \
2206 regnode *Next = regnext(scan); \
2207 regprop(RExC_rx, mysv, scan); \
2208 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2209 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2210 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2217 #define JOIN_EXACT(scan,min,flags) \
2218 if (PL_regkind[OP(scan)] == EXACT) \
2219 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2222 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2223 /* Merge several consecutive EXACTish nodes into one. */
2224 regnode *n = regnext(scan);
2226 regnode *next = scan + NODE_SZ_STR(scan);
2230 regnode *stop = scan;
2231 GET_RE_DEBUG_FLAGS_DECL;
2233 PERL_UNUSED_ARG(depth);
2235 #ifndef EXPERIMENTAL_INPLACESCAN
2236 PERL_UNUSED_ARG(flags);
2237 PERL_UNUSED_ARG(val);
2239 DEBUG_PEEP("join",scan,depth);
2241 /* Skip NOTHING, merge EXACT*. */
2243 ( PL_regkind[OP(n)] == NOTHING ||
2244 (stringok && (OP(n) == OP(scan))))
2246 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2248 if (OP(n) == TAIL || n > next)
2250 if (PL_regkind[OP(n)] == NOTHING) {
2251 DEBUG_PEEP("skip:",n,depth);
2252 NEXT_OFF(scan) += NEXT_OFF(n);
2253 next = n + NODE_STEP_REGNODE;
2260 else if (stringok) {
2261 const unsigned int oldl = STR_LEN(scan);
2262 regnode * const nnext = regnext(n);
2264 DEBUG_PEEP("merg",n,depth);
2267 if (oldl + STR_LEN(n) > U8_MAX)
2269 NEXT_OFF(scan) += NEXT_OFF(n);
2270 STR_LEN(scan) += STR_LEN(n);
2271 next = n + NODE_SZ_STR(n);
2272 /* Now we can overwrite *n : */
2273 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2281 #ifdef EXPERIMENTAL_INPLACESCAN
2282 if (flags && !NEXT_OFF(n)) {
2283 DEBUG_PEEP("atch", val, depth);
2284 if (reg_off_by_arg[OP(n)]) {
2285 ARG_SET(n, val - n);
2288 NEXT_OFF(n) = val - n;
2295 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2297 Two problematic code points in Unicode casefolding of EXACT nodes:
2299 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2300 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2306 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2307 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2309 This means that in case-insensitive matching (or "loose matching",
2310 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2311 length of the above casefolded versions) can match a target string
2312 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2313 This would rather mess up the minimum length computation.
2315 What we'll do is to look for the tail four bytes, and then peek
2316 at the preceding two bytes to see whether we need to decrease
2317 the minimum length by four (six minus two).
2319 Thanks to the design of UTF-8, there cannot be false matches:
2320 A sequence of valid UTF-8 bytes cannot be a subsequence of
2321 another valid sequence of UTF-8 bytes.
2324 char * const s0 = STRING(scan), *s, *t;
2325 char * const s1 = s0 + STR_LEN(scan) - 1;
2326 char * const s2 = s1 - 4;
2327 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2328 const char t0[] = "\xaf\x49\xaf\x42";
2330 const char t0[] = "\xcc\x88\xcc\x81";
2332 const char * const t1 = t0 + 3;
2335 s < s2 && (t = ninstr(s, s1, t0, t1));
2338 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2339 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2341 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2342 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2350 n = scan + NODE_SZ_STR(scan);
2352 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2359 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2363 /* REx optimizer. Converts nodes into quickier variants "in place".
2364 Finds fixed substrings. */
2366 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2367 to the position after last scanned or to NULL. */
2369 #define INIT_AND_WITHP \
2370 assert(!and_withp); \
2371 Newx(and_withp,1,struct regnode_charclass_class); \
2372 SAVEFREEPV(and_withp)
2374 /* this is a chain of data about sub patterns we are processing that
2375 need to be handled seperately/specially in study_chunk. Its so
2376 we can simulate recursion without losing state. */
2378 typedef struct scan_frame {
2379 regnode *last; /* last node to process in this frame */
2380 regnode *next; /* next node to process when last is reached */
2381 struct scan_frame *prev; /*previous frame*/
2382 I32 stop; /* what stopparen do we use */
2386 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2389 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2390 I32 *minlenp, I32 *deltap,
2395 struct regnode_charclass_class *and_withp,
2396 U32 flags, U32 depth)
2397 /* scanp: Start here (read-write). */
2398 /* deltap: Write maxlen-minlen here. */
2399 /* last: Stop before this one. */
2400 /* data: string data about the pattern */
2401 /* stopparen: treat close N as END */
2402 /* recursed: which subroutines have we recursed into */
2403 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2406 I32 min = 0, pars = 0, code;
2407 regnode *scan = *scanp, *next;
2409 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2410 int is_inf_internal = 0; /* The studied chunk is infinite */
2411 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2412 scan_data_t data_fake;
2413 SV *re_trie_maxbuff = NULL;
2414 regnode *first_non_open = scan;
2415 I32 stopmin = I32_MAX;
2416 scan_frame *frame = NULL;
2418 GET_RE_DEBUG_FLAGS_DECL;
2421 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2425 while (first_non_open && OP(first_non_open) == OPEN)
2426 first_non_open=regnext(first_non_open);
2431 while ( scan && OP(scan) != END && scan < last ){
2432 /* Peephole optimizer: */
2433 DEBUG_STUDYDATA("Peep:", data,depth);
2434 DEBUG_PEEP("Peep",scan,depth);
2435 JOIN_EXACT(scan,&min,0);
2437 /* Follow the next-chain of the current node and optimize
2438 away all the NOTHINGs from it. */
2439 if (OP(scan) != CURLYX) {
2440 const int max = (reg_off_by_arg[OP(scan)]
2442 /* I32 may be smaller than U16 on CRAYs! */
2443 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2444 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2448 /* Skip NOTHING and LONGJMP. */
2449 while ((n = regnext(n))
2450 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2451 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2452 && off + noff < max)
2454 if (reg_off_by_arg[OP(scan)])
2457 NEXT_OFF(scan) = off;
2462 /* The principal pseudo-switch. Cannot be a switch, since we
2463 look into several different things. */
2464 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2465 || OP(scan) == IFTHEN) {
2466 next = regnext(scan);
2468 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2470 if (OP(next) == code || code == IFTHEN) {
2471 /* NOTE - There is similar code to this block below for handling
2472 TRIE nodes on a re-study. If you change stuff here check there
2474 I32 max1 = 0, min1 = I32_MAX, num = 0;
2475 struct regnode_charclass_class accum;
2476 regnode * const startbranch=scan;
2478 if (flags & SCF_DO_SUBSTR)
2479 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2480 if (flags & SCF_DO_STCLASS)
2481 cl_init_zero(pRExC_state, &accum);
2483 while (OP(scan) == code) {
2484 I32 deltanext, minnext, f = 0, fake;
2485 struct regnode_charclass_class this_class;
2488 data_fake.flags = 0;
2490 data_fake.whilem_c = data->whilem_c;
2491 data_fake.last_closep = data->last_closep;
2494 data_fake.last_closep = &fake;
2496 data_fake.pos_delta = delta;
2497 next = regnext(scan);
2498 scan = NEXTOPER(scan);
2500 scan = NEXTOPER(scan);
2501 if (flags & SCF_DO_STCLASS) {
2502 cl_init(pRExC_state, &this_class);
2503 data_fake.start_class = &this_class;
2504 f = SCF_DO_STCLASS_AND;
2506 if (flags & SCF_WHILEM_VISITED_POS)
2507 f |= SCF_WHILEM_VISITED_POS;
2509 /* we suppose the run is continuous, last=next...*/
2510 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2512 stopparen, recursed, NULL, f,depth+1);
2515 if (max1 < minnext + deltanext)
2516 max1 = minnext + deltanext;
2517 if (deltanext == I32_MAX)
2518 is_inf = is_inf_internal = 1;
2520 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2522 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2523 if ( stopmin > minnext)
2524 stopmin = min + min1;
2525 flags &= ~SCF_DO_SUBSTR;
2527 data->flags |= SCF_SEEN_ACCEPT;
2530 if (data_fake.flags & SF_HAS_EVAL)
2531 data->flags |= SF_HAS_EVAL;
2532 data->whilem_c = data_fake.whilem_c;
2534 if (flags & SCF_DO_STCLASS)
2535 cl_or(pRExC_state, &accum, &this_class);
2537 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2539 if (flags & SCF_DO_SUBSTR) {
2540 data->pos_min += min1;
2541 data->pos_delta += max1 - min1;
2542 if (max1 != min1 || is_inf)
2543 data->longest = &(data->longest_float);
2546 delta += max1 - min1;
2547 if (flags & SCF_DO_STCLASS_OR) {
2548 cl_or(pRExC_state, data->start_class, &accum);
2550 cl_and(data->start_class, and_withp);
2551 flags &= ~SCF_DO_STCLASS;
2554 else if (flags & SCF_DO_STCLASS_AND) {
2556 cl_and(data->start_class, &accum);
2557 flags &= ~SCF_DO_STCLASS;
2560 /* Switch to OR mode: cache the old value of
2561 * data->start_class */
2563 StructCopy(data->start_class, and_withp,
2564 struct regnode_charclass_class);
2565 flags &= ~SCF_DO_STCLASS_AND;
2566 StructCopy(&accum, data->start_class,
2567 struct regnode_charclass_class);
2568 flags |= SCF_DO_STCLASS_OR;
2569 data->start_class->flags |= ANYOF_EOS;
2573 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2576 Assuming this was/is a branch we are dealing with: 'scan' now
2577 points at the item that follows the branch sequence, whatever
2578 it is. We now start at the beginning of the sequence and look
2585 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2587 If we can find such a subseqence we need to turn the first
2588 element into a trie and then add the subsequent branch exact
2589 strings to the trie.
2593 1. patterns where the whole set of branch can be converted.
2595 2. patterns where only a subset can be converted.
2597 In case 1 we can replace the whole set with a single regop
2598 for the trie. In case 2 we need to keep the start and end
2601 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2602 becomes BRANCH TRIE; BRANCH X;
2604 There is an additional case, that being where there is a
2605 common prefix, which gets split out into an EXACT like node
2606 preceding the TRIE node.
2608 If x(1..n)==tail then we can do a simple trie, if not we make
2609 a "jump" trie, such that when we match the appropriate word
2610 we "jump" to the appopriate tail node. Essentailly we turn
2611 a nested if into a case structure of sorts.
2616 if (!re_trie_maxbuff) {
2617 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2618 if (!SvIOK(re_trie_maxbuff))
2619 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2621 if ( SvIV(re_trie_maxbuff)>=0 ) {
2623 regnode *first = (regnode *)NULL;
2624 regnode *last = (regnode *)NULL;
2625 regnode *tail = scan;
2630 SV * const mysv = sv_newmortal(); /* for dumping */
2632 /* var tail is used because there may be a TAIL
2633 regop in the way. Ie, the exacts will point to the
2634 thing following the TAIL, but the last branch will
2635 point at the TAIL. So we advance tail. If we
2636 have nested (?:) we may have to move through several
2640 while ( OP( tail ) == TAIL ) {
2641 /* this is the TAIL generated by (?:) */
2642 tail = regnext( tail );
2647 regprop(RExC_rx, mysv, tail );
2648 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2649 (int)depth * 2 + 2, "",
2650 "Looking for TRIE'able sequences. Tail node is: ",
2651 SvPV_nolen_const( mysv )
2657 step through the branches, cur represents each
2658 branch, noper is the first thing to be matched
2659 as part of that branch and noper_next is the
2660 regnext() of that node. if noper is an EXACT
2661 and noper_next is the same as scan (our current
2662 position in the regex) then the EXACT branch is
2663 a possible optimization target. Once we have
2664 two or more consequetive such branches we can
2665 create a trie of the EXACT's contents and stich
2666 it in place. If the sequence represents all of
2667 the branches we eliminate the whole thing and
2668 replace it with a single TRIE. If it is a
2669 subsequence then we need to stitch it in. This
2670 means the first branch has to remain, and needs
2671 to be repointed at the item on the branch chain
2672 following the last branch optimized. This could
2673 be either a BRANCH, in which case the
2674 subsequence is internal, or it could be the
2675 item following the branch sequence in which
2676 case the subsequence is at the end.
2680 /* dont use tail as the end marker for this traverse */
2681 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2682 regnode * const noper = NEXTOPER( cur );
2683 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2684 regnode * const noper_next = regnext( noper );
2688 regprop(RExC_rx, mysv, cur);
2689 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2690 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2692 regprop(RExC_rx, mysv, noper);
2693 PerlIO_printf( Perl_debug_log, " -> %s",
2694 SvPV_nolen_const(mysv));
2697 regprop(RExC_rx, mysv, noper_next );
2698 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2699 SvPV_nolen_const(mysv));
2701 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2702 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2704 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2705 : PL_regkind[ OP( noper ) ] == EXACT )
2706 || OP(noper) == NOTHING )
2708 && noper_next == tail
2713 if ( !first || optype == NOTHING ) {
2714 if (!first) first = cur;
2715 optype = OP( noper );
2721 make_trie( pRExC_state,
2722 startbranch, first, cur, tail, count,
2725 if ( PL_regkind[ OP( noper ) ] == EXACT
2727 && noper_next == tail
2732 optype = OP( noper );
2742 regprop(RExC_rx, mysv, cur);
2743 PerlIO_printf( Perl_debug_log,
2744 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2745 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2749 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2750 #ifdef TRIE_STUDY_OPT
2751 if ( ((made == MADE_EXACT_TRIE &&
2752 startbranch == first)
2753 || ( first_non_open == first )) &&
2755 flags |= SCF_TRIE_RESTUDY;
2756 if ( startbranch == first
2759 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2769 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2770 scan = NEXTOPER(NEXTOPER(scan));
2771 } else /* single branch is optimized. */
2772 scan = NEXTOPER(scan);
2774 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2775 scan_frame *newframe = NULL;
2780 if (OP(scan) != SUSPEND) {
2781 /* set the pointer */
2782 if (OP(scan) == GOSUB) {
2784 RExC_recurse[ARG2L(scan)] = scan;
2785 start = RExC_open_parens[paren-1];
2786 end = RExC_close_parens[paren-1];
2789 start = RExC_rxi->program + 1;
2793 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2794 SAVEFREEPV(recursed);
2796 if (!PAREN_TEST(recursed,paren+1)) {
2797 PAREN_SET(recursed,paren+1);
2798 Newx(newframe,1,scan_frame);
2800 if (flags & SCF_DO_SUBSTR) {
2801 SCAN_COMMIT(pRExC_state,data,minlenp);
2802 data->longest = &(data->longest_float);
2804 is_inf = is_inf_internal = 1;
2805 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2806 cl_anything(pRExC_state, data->start_class);
2807 flags &= ~SCF_DO_STCLASS;
2810 Newx(newframe,1,scan_frame);
2813 end = regnext(scan);
2818 SAVEFREEPV(newframe);
2819 newframe->next = regnext(scan);
2820 newframe->last = last;
2821 newframe->stop = stopparen;
2822 newframe->prev = frame;
2832 else if (OP(scan) == EXACT) {
2833 I32 l = STR_LEN(scan);
2836 const U8 * const s = (U8*)STRING(scan);
2837 l = utf8_length(s, s + l);
2838 uc = utf8_to_uvchr(s, NULL);
2840 uc = *((U8*)STRING(scan));
2843 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2844 /* The code below prefers earlier match for fixed
2845 offset, later match for variable offset. */
2846 if (data->last_end == -1) { /* Update the start info. */
2847 data->last_start_min = data->pos_min;
2848 data->last_start_max = is_inf
2849 ? I32_MAX : data->pos_min + data->pos_delta;
2851 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2853 SvUTF8_on(data->last_found);
2855 SV * const sv = data->last_found;
2856 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2857 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2858 if (mg && mg->mg_len >= 0)
2859 mg->mg_len += utf8_length((U8*)STRING(scan),
2860 (U8*)STRING(scan)+STR_LEN(scan));
2862 data->last_end = data->pos_min + l;
2863 data->pos_min += l; /* As in the first entry. */
2864 data->flags &= ~SF_BEFORE_EOL;
2866 if (flags & SCF_DO_STCLASS_AND) {
2867 /* Check whether it is compatible with what we know already! */
2871 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2872 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2873 && (!(data->start_class->flags & ANYOF_FOLD)
2874 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2877 ANYOF_CLASS_ZERO(data->start_class);
2878 ANYOF_BITMAP_ZERO(data->start_class);
2880 ANYOF_BITMAP_SET(data->start_class, uc);
2881 data->start_class->flags &= ~ANYOF_EOS;
2883 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2885 else if (flags & SCF_DO_STCLASS_OR) {
2886 /* false positive possible if the class is case-folded */
2888 ANYOF_BITMAP_SET(data->start_class, uc);
2890 data->start_class->flags |= ANYOF_UNICODE_ALL;
2891 data->start_class->flags &= ~ANYOF_EOS;
2892 cl_and(data->start_class, and_withp);
2894 flags &= ~SCF_DO_STCLASS;
2896 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2897 I32 l = STR_LEN(scan);
2898 UV uc = *((U8*)STRING(scan));
2900 /* Search for fixed substrings supports EXACT only. */
2901 if (flags & SCF_DO_SUBSTR) {
2903 SCAN_COMMIT(pRExC_state, data, minlenp);
2906 const U8 * const s = (U8 *)STRING(scan);
2907 l = utf8_length(s, s + l);
2908 uc = utf8_to_uvchr(s, NULL);
2911 if (flags & SCF_DO_SUBSTR)
2913 if (flags & SCF_DO_STCLASS_AND) {
2914 /* Check whether it is compatible with what we know already! */
2918 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2919 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2920 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2922 ANYOF_CLASS_ZERO(data->start_class);
2923 ANYOF_BITMAP_ZERO(data->start_class);
2925 ANYOF_BITMAP_SET(data->start_class, uc);
2926 data->start_class->flags &= ~ANYOF_EOS;
2927 data->start_class->flags |= ANYOF_FOLD;
2928 if (OP(scan) == EXACTFL)
2929 data->start_class->flags |= ANYOF_LOCALE;
2932 else if (flags & SCF_DO_STCLASS_OR) {
2933 if (data->start_class->flags & ANYOF_FOLD) {
2934 /* false positive possible if the class is case-folded.
2935 Assume that the locale settings are the same... */
2937 ANYOF_BITMAP_SET(data->start_class, uc);
2938 data->start_class->flags &= ~ANYOF_EOS;
2940 cl_and(data->start_class, and_withp);
2942 flags &= ~SCF_DO_STCLASS;
2944 else if (strchr((const char*)PL_varies,OP(scan))) {
2945 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2946 I32 f = flags, pos_before = 0;
2947 regnode * const oscan = scan;
2948 struct regnode_charclass_class this_class;
2949 struct regnode_charclass_class *oclass = NULL;
2950 I32 next_is_eval = 0;
2952 switch (PL_regkind[OP(scan)]) {
2953 case WHILEM: /* End of (?:...)* . */
2954 scan = NEXTOPER(scan);
2957 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2958 next = NEXTOPER(scan);
2959 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2961 maxcount = REG_INFTY;
2962 next = regnext(scan);
2963 scan = NEXTOPER(scan);
2967 if (flags & SCF_DO_SUBSTR)
2972 if (flags & SCF_DO_STCLASS) {
2974 maxcount = REG_INFTY;
2975 next = regnext(scan);
2976 scan = NEXTOPER(scan);
2979 is_inf = is_inf_internal = 1;
2980 scan = regnext(scan);
2981 if (flags & SCF_DO_SUBSTR) {
2982 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2983 data->longest = &(data->longest_float);
2985 goto optimize_curly_tail;
2987 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
2988 && (scan->flags == stopparen))
2993 mincount = ARG1(scan);
2994 maxcount = ARG2(scan);
2996 next = regnext(scan);
2997 if (OP(scan) == CURLYX) {
2998 I32 lp = (data ? *(data->last_closep) : 0);
2999 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3001 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3002 next_is_eval = (OP(scan) == EVAL);
3004 if (flags & SCF_DO_SUBSTR) {
3005 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3006 pos_before = data->pos_min;
3010 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3012 data->flags |= SF_IS_INF;
3014 if (flags & SCF_DO_STCLASS) {
3015 cl_init(pRExC_state, &this_class);
3016 oclass = data->start_class;
3017 data->start_class = &this_class;
3018 f |= SCF_DO_STCLASS_AND;
3019 f &= ~SCF_DO_STCLASS_OR;
3021 /* These are the cases when once a subexpression
3022 fails at a particular position, it cannot succeed
3023 even after backtracking at the enclosing scope.
3025 XXXX what if minimal match and we are at the
3026 initial run of {n,m}? */
3027 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3028 f &= ~SCF_WHILEM_VISITED_POS;
3030 /* This will finish on WHILEM, setting scan, or on NULL: */
3031 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3032 last, data, stopparen, recursed, NULL,
3034 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3036 if (flags & SCF_DO_STCLASS)
3037 data->start_class = oclass;
3038 if (mincount == 0 || minnext == 0) {
3039 if (flags & SCF_DO_STCLASS_OR) {
3040 cl_or(pRExC_state, data->start_class, &this_class);
3042 else if (flags & SCF_DO_STCLASS_AND) {
3043 /* Switch to OR mode: cache the old value of
3044 * data->start_class */
3046 StructCopy(data->start_class, and_withp,
3047 struct regnode_charclass_class);
3048 flags &= ~SCF_DO_STCLASS_AND;
3049 StructCopy(&this_class, data->start_class,
3050 struct regnode_charclass_class);
3051 flags |= SCF_DO_STCLASS_OR;
3052 data->start_class->flags |= ANYOF_EOS;
3054 } else { /* Non-zero len */
3055 if (flags & SCF_DO_STCLASS_OR) {
3056 cl_or(pRExC_state, data->start_class, &this_class);
3057 cl_and(data->start_class, and_withp);
3059 else if (flags & SCF_DO_STCLASS_AND)
3060 cl_and(data->start_class, &this_class);
3061 flags &= ~SCF_DO_STCLASS;
3063 if (!scan) /* It was not CURLYX, but CURLY. */
3065 if ( /* ? quantifier ok, except for (?{ ... }) */
3066 (next_is_eval || !(mincount == 0 && maxcount == 1))
3067 && (minnext == 0) && (deltanext == 0)
3068 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3069 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3070 && ckWARN(WARN_REGEXP))
3073 "Quantifier unexpected on zero-length expression");
3076 min += minnext * mincount;
3077 is_inf_internal |= ((maxcount == REG_INFTY
3078 && (minnext + deltanext) > 0)
3079 || deltanext == I32_MAX);
3080 is_inf |= is_inf_internal;
3081 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3083 /* Try powerful optimization CURLYX => CURLYN. */
3084 if ( OP(oscan) == CURLYX && data
3085 && data->flags & SF_IN_PAR
3086 && !(data->flags & SF_HAS_EVAL)
3087 && !deltanext && minnext == 1 ) {
3088 /* Try to optimize to CURLYN. */
3089 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3090 regnode * const nxt1 = nxt;
3097 if (!strchr((const char*)PL_simple,OP(nxt))
3098 && !(PL_regkind[OP(nxt)] == EXACT
3099 && STR_LEN(nxt) == 1))
3105 if (OP(nxt) != CLOSE)
3107 if (RExC_open_parens) {
3108 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3109 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3111 /* Now we know that nxt2 is the only contents: */
3112 oscan->flags = (U8)ARG(nxt);
3114 OP(nxt1) = NOTHING; /* was OPEN. */
3117 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3118 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3119 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3120 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3121 OP(nxt + 1) = OPTIMIZED; /* was count. */
3122 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3127 /* Try optimization CURLYX => CURLYM. */
3128 if ( OP(oscan) == CURLYX && data
3129 && !(data->flags & SF_HAS_PAR)
3130 && !(data->flags & SF_HAS_EVAL)
3131 && !deltanext /* atom is fixed width */
3132 && minnext != 0 /* CURLYM can't handle zero width */
3134 /* XXXX How to optimize if data == 0? */
3135 /* Optimize to a simpler form. */
3136 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3140 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3141 && (OP(nxt2) != WHILEM))
3143 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3144 /* Need to optimize away parenths. */
3145 if (data->flags & SF_IN_PAR) {
3146 /* Set the parenth number. */
3147 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3149 if (OP(nxt) != CLOSE)
3150 FAIL("Panic opt close");
3151 oscan->flags = (U8)ARG(nxt);
3152 if (RExC_open_parens) {
3153 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3154 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3156 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3157 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3160 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3161 OP(nxt + 1) = OPTIMIZED; /* was count. */
3162 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3163 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3166 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3167 regnode *nnxt = regnext(nxt1);
3170 if (reg_off_by_arg[OP(nxt1)])
3171 ARG_SET(nxt1, nxt2 - nxt1);
3172 else if (nxt2 - nxt1 < U16_MAX)
3173 NEXT_OFF(nxt1) = nxt2 - nxt1;
3175 OP(nxt) = NOTHING; /* Cannot beautify */
3180 /* Optimize again: */
3181 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3182 NULL, stopparen, recursed, NULL, 0,depth+1);
3187 else if ((OP(oscan) == CURLYX)
3188 && (flags & SCF_WHILEM_VISITED_POS)
3189 /* See the comment on a similar expression above.
3190 However, this time it not a subexpression
3191 we care about, but the expression itself. */
3192 && (maxcount == REG_INFTY)
3193 && data && ++data->whilem_c < 16) {
3194 /* This stays as CURLYX, we can put the count/of pair. */
3195 /* Find WHILEM (as in regexec.c) */
3196 regnode *nxt = oscan + NEXT_OFF(oscan);
3198 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3200 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3201 | (RExC_whilem_seen << 4)); /* On WHILEM */
3203 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3205 if (flags & SCF_DO_SUBSTR) {
3206 SV *last_str = NULL;
3207 int counted = mincount != 0;
3209 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3210 #if defined(SPARC64_GCC_WORKAROUND)
3213 const char *s = NULL;
3216 if (pos_before >= data->last_start_min)
3219 b = data->last_start_min;
3222 s = SvPV_const(data->last_found, l);
3223 old = b - data->last_start_min;
3226 I32 b = pos_before >= data->last_start_min
3227 ? pos_before : data->last_start_min;
3229 const char * const s = SvPV_const(data->last_found, l);
3230 I32 old = b - data->last_start_min;
3234 old = utf8_hop((U8*)s, old) - (U8*)s;
3237 /* Get the added string: */
3238 last_str = newSVpvn(s + old, l);
3240 SvUTF8_on(last_str);
3241 if (deltanext == 0 && pos_before == b) {
3242 /* What was added is a constant string */
3244 SvGROW(last_str, (mincount * l) + 1);
3245 repeatcpy(SvPVX(last_str) + l,
3246 SvPVX_const(last_str), l, mincount - 1);
3247 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3248 /* Add additional parts. */
3249 SvCUR_set(data->last_found,
3250 SvCUR(data->last_found) - l);
3251 sv_catsv(data->last_found, last_str);
3253 SV * sv = data->last_found;
3255 SvUTF8(sv) && SvMAGICAL(sv) ?
3256 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3257 if (mg && mg->mg_len >= 0)
3258 mg->mg_len += CHR_SVLEN(last_str);
3260 data->last_end += l * (mincount - 1);
3263 /* start offset must point into the last copy */
3264 data->last_start_min += minnext * (mincount - 1);
3265 data->last_start_max += is_inf ? I32_MAX
3266 : (maxcount - 1) * (minnext + data->pos_delta);
3269 /* It is counted once already... */
3270 data->pos_min += minnext * (mincount - counted);
3271 data->pos_delta += - counted * deltanext +
3272 (minnext + deltanext) * maxcount - minnext * mincount;
3273 if (mincount != maxcount) {
3274 /* Cannot extend fixed substrings found inside
3276 SCAN_COMMIT(pRExC_state,data,minlenp);
3277 if (mincount && last_str) {
3278 SV * const sv = data->last_found;
3279 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3280 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3284 sv_setsv(sv, last_str);
3285 data->last_end = data->pos_min;
3286 data->last_start_min =
3287 data->pos_min - CHR_SVLEN(last_str);
3288 data->last_start_max = is_inf
3290 : data->pos_min + data->pos_delta
3291 - CHR_SVLEN(last_str);
3293 data->longest = &(data->longest_float);
3295 SvREFCNT_dec(last_str);
3297 if (data && (fl & SF_HAS_EVAL))
3298 data->flags |= SF_HAS_EVAL;
3299 optimize_curly_tail:
3300 if (OP(oscan) != CURLYX) {
3301 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3303 NEXT_OFF(oscan) += NEXT_OFF(next);
3306 default: /* REF and CLUMP only? */
3307 if (flags & SCF_DO_SUBSTR) {
3308 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3309 data->longest = &(data->longest_float);
3311 is_inf = is_inf_internal = 1;
3312 if (flags & SCF_DO_STCLASS_OR)
3313 cl_anything(pRExC_state, data->start_class);
3314 flags &= ~SCF_DO_STCLASS;
3318 else if (strchr((const char*)PL_simple,OP(scan))) {
3321 if (flags & SCF_DO_SUBSTR) {
3322 SCAN_COMMIT(pRExC_state,data,minlenp);
3326 if (flags & SCF_DO_STCLASS) {
3327 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3329 /* Some of the logic below assumes that switching
3330 locale on will only add false positives. */
3331 switch (PL_regkind[OP(scan)]) {
3335 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3336 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3337 cl_anything(pRExC_state, data->start_class);
3340 if (OP(scan) == SANY)
3342 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3343 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3344 || (data->start_class->flags & ANYOF_CLASS));
3345 cl_anything(pRExC_state, data->start_class);
3347 if (flags & SCF_DO_STCLASS_AND || !value)
3348 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3351 if (flags & SCF_DO_STCLASS_AND)
3352 cl_and(data->start_class,
3353 (struct regnode_charclass_class*)scan);
3355 cl_or(pRExC_state, data->start_class,
3356 (struct regnode_charclass_class*)scan);
3359 if (flags & SCF_DO_STCLASS_AND) {
3360 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3361 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3362 for (value = 0; value < 256; value++)
3363 if (!isALNUM(value))
3364 ANYOF_BITMAP_CLEAR(data->start_class, value);
3368 if (data->start_class->flags & ANYOF_LOCALE)
3369 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3371 for (value = 0; value < 256; value++)
3373 ANYOF_BITMAP_SET(data->start_class, value);
3378 if (flags & SCF_DO_STCLASS_AND) {
3379 if (data->start_class->flags & ANYOF_LOCALE)
3380 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3383 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3384 data->start_class->flags |= ANYOF_LOCALE;
3388 if (flags & SCF_DO_STCLASS_AND) {
3389 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3390 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3391 for (value = 0; value < 256; value++)
3393 ANYOF_BITMAP_CLEAR(data->start_class, value);
3397 if (data->start_class->flags & ANYOF_LOCALE)
3398 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3400 for (value = 0; value < 256; value++)
3401 if (!isALNUM(value))
3402 ANYOF_BITMAP_SET(data->start_class, value);
3407 if (flags & SCF_DO_STCLASS_AND) {
3408 if (data->start_class->flags & ANYOF_LOCALE)
3409 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3412 data->start_class->flags |= ANYOF_LOCALE;
3413 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3417 if (flags & SCF_DO_STCLASS_AND) {
3418 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3419 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3420 for (value = 0; value < 256; value++)
3421 if (!isSPACE(value))
3422 ANYOF_BITMAP_CLEAR(data->start_class, value);
3426 if (data->start_class->flags & ANYOF_LOCALE)
3427 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3429 for (value = 0; value < 256; value++)
3431 ANYOF_BITMAP_SET(data->start_class, value);
3436 if (flags & SCF_DO_STCLASS_AND) {
3437 if (data->start_class->flags & ANYOF_LOCALE)
3438 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3441 data->start_class->flags |= ANYOF_LOCALE;
3442 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3446 if (flags & SCF_DO_STCLASS_AND) {
3447 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3448 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3449 for (value = 0; value < 256; value++)
3451 ANYOF_BITMAP_CLEAR(data->start_class, value);
3455 if (data->start_class->flags & ANYOF_LOCALE)
3456 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3458 for (value = 0; value < 256; value++)
3459 if (!isSPACE(value))
3460 ANYOF_BITMAP_SET(data->start_class, value);
3465 if (flags & SCF_DO_STCLASS_AND) {
3466 if (data->start_class->flags & ANYOF_LOCALE) {
3467 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3468 for (value = 0; value < 256; value++)
3469 if (!isSPACE(value))
3470 ANYOF_BITMAP_CLEAR(data->start_class, value);
3474 data->start_class->flags |= ANYOF_LOCALE;
3475 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3479 if (flags & SCF_DO_STCLASS_AND) {
3480 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3481 for (value = 0; value < 256; value++)
3482 if (!isDIGIT(value))
3483 ANYOF_BITMAP_CLEAR(data->start_class, value);
3486 if (data->start_class->flags & ANYOF_LOCALE)
3487 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3489 for (value = 0; value < 256; value++)
3491 ANYOF_BITMAP_SET(data->start_class, value);
3496 if (flags & SCF_DO_STCLASS_AND) {
3497 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3498 for (value = 0; value < 256; value++)
3500 ANYOF_BITMAP_CLEAR(data->start_class, value);
3503 if (data->start_class->flags & ANYOF_LOCALE)
3504 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3506 for (value = 0; value < 256; value++)
3507 if (!isDIGIT(value))
3508 ANYOF_BITMAP_SET(data->start_class, value);
3513 if (flags & SCF_DO_STCLASS_OR)
3514 cl_and(data->start_class, and_withp);
3515 flags &= ~SCF_DO_STCLASS;
3518 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3519 data->flags |= (OP(scan) == MEOL
3523 else if ( PL_regkind[OP(scan)] == BRANCHJ
3524 /* Lookbehind, or need to calculate parens/evals/stclass: */
3525 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3526 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3527 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3528 || OP(scan) == UNLESSM )
3530 /* Negative Lookahead/lookbehind
3531 In this case we can't do fixed string optimisation.
3534 I32 deltanext, minnext, fake = 0;
3536 struct regnode_charclass_class intrnl;
3539 data_fake.flags = 0;
3541 data_fake.whilem_c = data->whilem_c;
3542 data_fake.last_closep = data->last_closep;
3545 data_fake.last_closep = &fake;
3546 data_fake.pos_delta = delta;
3547 if ( flags & SCF_DO_STCLASS && !scan->flags
3548 && OP(scan) == IFMATCH ) { /* Lookahead */
3549 cl_init(pRExC_state, &intrnl);
3550 data_fake.start_class = &intrnl;
3551 f |= SCF_DO_STCLASS_AND;
3553 if (flags & SCF_WHILEM_VISITED_POS)
3554 f |= SCF_WHILEM_VISITED_POS;
3555 next = regnext(scan);
3556 nscan = NEXTOPER(NEXTOPER(scan));
3557 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3558 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3561 FAIL("Variable length lookbehind not implemented");
3563 else if (minnext > (I32)U8_MAX) {
3564 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3566 scan->flags = (U8)minnext;
3569 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3571 if (data_fake.flags & SF_HAS_EVAL)
3572 data->flags |= SF_HAS_EVAL;
3573 data->whilem_c = data_fake.whilem_c;
3575 if (f & SCF_DO_STCLASS_AND) {
3576 const int was = (data->start_class->flags & ANYOF_EOS);
3578 cl_and(data->start_class, &intrnl);
3580 data->start_class->flags |= ANYOF_EOS;
3583 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3585 /* Positive Lookahead/lookbehind
3586 In this case we can do fixed string optimisation,
3587 but we must be careful about it. Note in the case of
3588 lookbehind the positions will be offset by the minimum
3589 length of the pattern, something we won't know about
3590 until after the recurse.
3592 I32 deltanext, fake = 0;
3594 struct regnode_charclass_class intrnl;
3596 /* We use SAVEFREEPV so that when the full compile
3597 is finished perl will clean up the allocated
3598 minlens when its all done. This was we don't
3599 have to worry about freeing them when we know
3600 they wont be used, which would be a pain.
3603 Newx( minnextp, 1, I32 );
3604 SAVEFREEPV(minnextp);
3607 StructCopy(data, &data_fake, scan_data_t);
3608 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3611 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3612 data_fake.last_found=newSVsv(data->last_found);
3616 data_fake.last_closep = &fake;
3617 data_fake.flags = 0;
3618 data_fake.pos_delta = delta;
3620 data_fake.flags |= SF_IS_INF;
3621 if ( flags & SCF_DO_STCLASS && !scan->flags
3622 && OP(scan) == IFMATCH ) { /* Lookahead */
3623 cl_init(pRExC_state, &intrnl);
3624 data_fake.start_class = &intrnl;
3625 f |= SCF_DO_STCLASS_AND;
3627 if (flags & SCF_WHILEM_VISITED_POS)
3628 f |= SCF_WHILEM_VISITED_POS;
3629 next = regnext(scan);
3630 nscan = NEXTOPER(NEXTOPER(scan));
3632 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3633 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3636 FAIL("Variable length lookbehind not implemented");
3638 else if (*minnextp > (I32)U8_MAX) {
3639 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3641 scan->flags = (U8)*minnextp;
3646 if (f & SCF_DO_STCLASS_AND) {
3647 const int was = (data->start_class->flags & ANYOF_EOS);
3649 cl_and(data->start_class, &intrnl);
3651 data->start_class->flags |= ANYOF_EOS;
3654 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3656 if (data_fake.flags & SF_HAS_EVAL)
3657 data->flags |= SF_HAS_EVAL;
3658 data->whilem_c = data_fake.whilem_c;
3659 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3660 if (RExC_rx->minlen<*minnextp)
3661 RExC_rx->minlen=*minnextp;
3662 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3663 SvREFCNT_dec(data_fake.last_found);
3665 if ( data_fake.minlen_fixed != minlenp )
3667 data->offset_fixed= data_fake.offset_fixed;
3668 data->minlen_fixed= data_fake.minlen_fixed;
3669 data->lookbehind_fixed+= scan->flags;
3671 if ( data_fake.minlen_float != minlenp )
3673 data->minlen_float= data_fake.minlen_float;
3674 data->offset_float_min=data_fake.offset_float_min;
3675 data->offset_float_max=data_fake.offset_float_max;
3676 data->lookbehind_float+= scan->flags;
3685 else if (OP(scan) == OPEN) {
3686 if (stopparen != (I32)ARG(scan))
3689 else if (OP(scan) == CLOSE) {
3690 if (stopparen == (I32)ARG(scan)) {
3693 if ((I32)ARG(scan) == is_par) {
3694 next = regnext(scan);
3696 if ( next && (OP(next) != WHILEM) && next < last)
3697 is_par = 0; /* Disable optimization */
3700 *(data->last_closep) = ARG(scan);
3702 else if (OP(scan) == EVAL) {
3704 data->flags |= SF_HAS_EVAL;
3706 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3707 if (flags & SCF_DO_SUBSTR) {
3708 SCAN_COMMIT(pRExC_state,data,minlenp);
3709 flags &= ~SCF_DO_SUBSTR;
3711 if (data && OP(scan)==ACCEPT) {
3712 data->flags |= SCF_SEEN_ACCEPT;
3717 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3719 if (flags & SCF_DO_SUBSTR) {
3720 SCAN_COMMIT(pRExC_state,data,minlenp);
3721 data->longest = &(data->longest_float);
3723 is_inf = is_inf_internal = 1;
3724 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3725 cl_anything(pRExC_state, data->start_class);
3726 flags &= ~SCF_DO_STCLASS;
3728 else if (OP(scan) == GPOS) {
3729 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3730 !(delta || is_inf || (data && data->pos_delta)))
3732 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3733 RExC_rx->extflags |= RXf_ANCH_GPOS;
3734 if (RExC_rx->gofs < (U32)min)
3735 RExC_rx->gofs = min;
3737 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3741 #ifdef TRIE_STUDY_OPT
3742 #ifdef FULL_TRIE_STUDY
3743 else if (PL_regkind[OP(scan)] == TRIE) {
3744 /* NOTE - There is similar code to this block above for handling
3745 BRANCH nodes on the initial study. If you change stuff here
3747 regnode *trie_node= scan;
3748 regnode *tail= regnext(scan);
3749 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3750 I32 max1 = 0, min1 = I32_MAX;
3751 struct regnode_charclass_class accum;
3753 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3754 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3755 if (flags & SCF_DO_STCLASS)
3756 cl_init_zero(pRExC_state, &accum);
3762 const regnode *nextbranch= NULL;
3765 for ( word=1 ; word <= trie->wordcount ; word++)
3767 I32 deltanext=0, minnext=0, f = 0, fake;
3768 struct regnode_charclass_class this_class;
3770 data_fake.flags = 0;
3772 data_fake.whilem_c = data->whilem_c;
3773 data_fake.last_closep = data->last_closep;
3776 data_fake.last_closep = &fake;
3777 data_fake.pos_delta = delta;
3778 if (flags & SCF_DO_STCLASS) {
3779 cl_init(pRExC_state, &this_class);
3780 data_fake.start_class = &this_class;
3781 f = SCF_DO_STCLASS_AND;
3783 if (flags & SCF_WHILEM_VISITED_POS)
3784 f |= SCF_WHILEM_VISITED_POS;
3786 if (trie->jump[word]) {
3788 nextbranch = trie_node + trie->jump[0];
3789 scan= trie_node + trie->jump[word];
3790 /* We go from the jump point to the branch that follows
3791 it. Note this means we need the vestigal unused branches
3792 even though they arent otherwise used.
3794 minnext = study_chunk(pRExC_state, &scan, minlenp,
3795 &deltanext, (regnode *)nextbranch, &data_fake,
3796 stopparen, recursed, NULL, f,depth+1);
3798 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3799 nextbranch= regnext((regnode*)nextbranch);
3801 if (min1 > (I32)(minnext + trie->minlen))
3802 min1 = minnext + trie->minlen;
3803 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3804 max1 = minnext + deltanext + trie->maxlen;
3805 if (deltanext == I32_MAX)
3806 is_inf = is_inf_internal = 1;
3808 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3810 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3811 if ( stopmin > min + min1)
3812 stopmin = min + min1;
3813 flags &= ~SCF_DO_SUBSTR;
3815 data->flags |= SCF_SEEN_ACCEPT;
3818 if (data_fake.flags & SF_HAS_EVAL)
3819 data->flags |= SF_HAS_EVAL;
3820 data->whilem_c = data_fake.whilem_c;
3822 if (flags & SCF_DO_STCLASS)
3823 cl_or(pRExC_state, &accum, &this_class);
3826 if (flags & SCF_DO_SUBSTR) {
3827 data->pos_min += min1;
3828 data->pos_delta += max1 - min1;
3829 if (max1 != min1 || is_inf)
3830 data->longest = &(data->longest_float);
3833 delta += max1 - min1;
3834 if (flags & SCF_DO_STCLASS_OR) {
3835 cl_or(pRExC_state, data->start_class, &accum);
3837 cl_and(data->start_class, and_withp);
3838 flags &= ~SCF_DO_STCLASS;
3841 else if (flags & SCF_DO_STCLASS_AND) {
3843 cl_and(data->start_class, &accum);
3844 flags &= ~SCF_DO_STCLASS;
3847 /* Switch to OR mode: cache the old value of
3848 * data->start_class */
3850 StructCopy(data->start_class, and_withp,
3851 struct regnode_charclass_class);
3852 flags &= ~SCF_DO_STCLASS_AND;
3853 StructCopy(&accum, data->start_class,
3854 struct regnode_charclass_class);
3855 flags |= SCF_DO_STCLASS_OR;
3856 data->start_class->flags |= ANYOF_EOS;
3863 else if (PL_regkind[OP(scan)] == TRIE) {
3864 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3867 min += trie->minlen;
3868 delta += (trie->maxlen - trie->minlen);
3869 flags &= ~SCF_DO_STCLASS; /* xxx */
3870 if (flags & SCF_DO_SUBSTR) {
3871 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3872 data->pos_min += trie->minlen;
3873 data->pos_delta += (trie->maxlen - trie->minlen);
3874 if (trie->maxlen != trie->minlen)
3875 data->longest = &(data->longest_float);
3877 if (trie->jump) /* no more substrings -- for now /grr*/
3878 flags &= ~SCF_DO_SUBSTR;
3880 #endif /* old or new */
3881 #endif /* TRIE_STUDY_OPT */
3882 /* Else: zero-length, ignore. */
3883 scan = regnext(scan);
3888 stopparen = frame->stop;
3889 frame = frame->prev;
3890 goto fake_study_recurse;
3895 DEBUG_STUDYDATA("pre-fin:",data,depth);
3898 *deltap = is_inf_internal ? I32_MAX : delta;
3899 if (flags & SCF_DO_SUBSTR && is_inf)
3900 data->pos_delta = I32_MAX - data->pos_min;
3901 if (is_par > (I32)U8_MAX)
3903 if (is_par && pars==1 && data) {
3904 data->flags |= SF_IN_PAR;
3905 data->flags &= ~SF_HAS_PAR;
3907 else if (pars && data) {
3908 data->flags |= SF_HAS_PAR;
3909 data->flags &= ~SF_IN_PAR;
3911 if (flags & SCF_DO_STCLASS_OR)
3912 cl_and(data->start_class, and_withp);
3913 if (flags & SCF_TRIE_RESTUDY)
3914 data->flags |= SCF_TRIE_RESTUDY;
3916 DEBUG_STUDYDATA("post-fin:",data,depth);
3918 return min < stopmin ? min : stopmin;
3922 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3924 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3926 Renewc(RExC_rxi->data,
3927 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3928 char, struct reg_data);
3930 Renew(RExC_rxi->data->what, count + n, U8);
3932 Newx(RExC_rxi->data->what, n, U8);
3933 RExC_rxi->data->count = count + n;
3934 Copy(s, RExC_rxi->data->what + count, n, U8);
3938 /*XXX: todo make this not included in a non debugging perl */
3939 #ifndef PERL_IN_XSUB_RE
3941 Perl_reginitcolors(pTHX)
3944 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3946 char *t = savepv(s);
3950 t = strchr(t, '\t');
3956 PL_colors[i] = t = (char *)"";
3961 PL_colors[i++] = (char *)"";
3968 #ifdef TRIE_STUDY_OPT
3969 #define CHECK_RESTUDY_GOTO \
3971 (data.flags & SCF_TRIE_RESTUDY) \
3975 #define CHECK_RESTUDY_GOTO
3979 - pregcomp - compile a regular expression into internal code
3981 * We can't allocate space until we know how big the compiled form will be,
3982 * but we can't compile it (and thus know how big it is) until we've got a
3983 * place to put the code. So we cheat: we compile it twice, once with code
3984 * generation turned off and size counting turned on, and once "for real".
3985 * This also means that we don't allocate space until we are sure that the
3986 * thing really will compile successfully, and we never have to move the
3987 * code and thus invalidate pointers into it. (Note that it has to be in
3988 * one piece because free() must be able to free it all.) [NB: not true in perl]
3990 * Beware that the optimization-preparation code in here knows about some
3991 * of the structure of the compiled regexp. [I'll say.]
3996 #ifndef PERL_IN_XSUB_RE
3997 #define RE_ENGINE_PTR &PL_core_reg_engine
3999 extern const struct regexp_engine my_reg_engine;
4000 #define RE_ENGINE_PTR &my_reg_engine
4003 #ifndef PERL_IN_XSUB_RE
4005 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
4008 HV * const table = GvHV(PL_hintgv);
4009 /* Dispatch a request to compile a regexp to correct
4012 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4013 GET_RE_DEBUG_FLAGS_DECL;
4014 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4015 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4017 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4020 return CALLREGCOMP_ENG(eng, exp, xend, pm);
4023 return Perl_re_compile(aTHX_ exp, xend, pm);
4028 Perl_re_compile(pTHX_ char *exp, char *xend, PMOP *pm)
4032 register regexp_internal *ri;
4040 RExC_state_t RExC_state;
4041 RExC_state_t * const pRExC_state = &RExC_state;
4042 #ifdef TRIE_STUDY_OPT
4044 RExC_state_t copyRExC_state;
4046 GET_RE_DEBUG_FLAGS_DECL;
4047 DEBUG_r(if (!PL_colorset) reginitcolors());
4050 FAIL("NULL regexp argument");
4052 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4056 SV *dsv= sv_newmortal();
4057 RE_PV_QUOTED_DECL(s, RExC_utf8,
4058 dsv, RExC_precomp, (xend - exp), 60);
4059 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4060 PL_colors[4],PL_colors[5],s);
4062 RExC_flags = pm->op_pmflags;
4066 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4067 RExC_seen_evals = 0;
4070 /* First pass: determine size, legality. */
4078 RExC_emit = &PL_regdummy;
4079 RExC_whilem_seen = 0;
4080 RExC_charnames = NULL;
4081 RExC_open_parens = NULL;
4082 RExC_close_parens = NULL;
4084 RExC_paren_names = NULL;
4086 RExC_paren_name_list = NULL;
4088 RExC_recurse = NULL;
4089 RExC_recurse_count = 0;
4091 #if 0 /* REGC() is (currently) a NOP at the first pass.
4092 * Clever compilers notice this and complain. --jhi */
4093 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4095 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4096 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4097 RExC_precomp = NULL;
4101 PerlIO_printf(Perl_debug_log,
4102 "Required size %"IVdf" nodes\n"
4103 "Starting second pass (creation)\n",
4106 RExC_lastparse=NULL;
4108 /* Small enough for pointer-storage convention?
4109 If extralen==0, this means that we will not need long jumps. */
4110 if (RExC_size >= 0x10000L && RExC_extralen)
4111 RExC_size += RExC_extralen;
4114 if (RExC_whilem_seen > 15)
4115 RExC_whilem_seen = 15;
4118 /* Make room for a sentinel value at the end of the program */
4122 /* Allocate space and zero-initialize. Note, the two step process
4123 of zeroing when in debug mode, thus anything assigned has to
4124 happen after that */
4125 Newxz(r, 1, regexp);
4126 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4127 char, regexp_internal);
4128 if ( r == NULL || ri == NULL )
4129 FAIL("Regexp out of space");
4131 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4132 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4134 /* bulk initialize base fields with 0. */
4135 Zero(ri, sizeof(regexp_internal), char);
4138 /* non-zero initialization begins here */
4140 r->engine= RE_ENGINE_PTR;
4142 r->prelen = xend - exp;
4143 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4145 bool has_k = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4146 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4147 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4148 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) >> 12);
4149 const char *fptr = STD_PAT_MODS; /*"msix"*/
4151 r->wraplen = r->prelen + has_minus + has_k + has_runon
4152 + (sizeof(STD_PAT_MODS) - 1)
4153 + (sizeof("(?:)") - 1);
4155 Newx(r->wrapped, r->wraplen, char );
4159 *p++ = KEEPCOPY_PAT_MOD; /*'k'*/
4161 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4162 char *colon = r + 1;
4165 while((ch = *fptr++)) {
4179 Copy(RExC_precomp, p, r->prelen, char);
4188 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4190 if (RExC_seen & REG_SEEN_RECURSE) {
4191 Newxz(RExC_open_parens, RExC_npar,regnode *);
4192 SAVEFREEPV(RExC_open_parens);
4193 Newxz(RExC_close_parens,RExC_npar,regnode *);
4194 SAVEFREEPV(RExC_close_parens);
4197 /* Useful during FAIL. */
4198 #ifdef RE_TRACK_PATTERN_OFFSETS
4199 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4200 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4201 "%s %"UVuf" bytes for offset annotations.\n",
4202 ri->u.offsets ? "Got" : "Couldn't get",
4203 (UV)((2*RExC_size+1) * sizeof(U32))));
4205 SetProgLen(ri,RExC_size);
4209 /* Second pass: emit code. */
4210 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4215 RExC_emit_start = ri->program;
4216 RExC_emit = ri->program;
4218 /* put a sentinal on the end of the program so we can check for
4220 ri->program[RExC_size].type = 255;
4222 /* Store the count of eval-groups for security checks: */
4223 RExC_rx->seen_evals = RExC_seen_evals;
4224 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4225 if (reg(pRExC_state, 0, &flags,1) == NULL)
4228 /* XXXX To minimize changes to RE engine we always allocate
4229 3-units-long substrs field. */
4230 Newx(r->substrs, 1, struct reg_substr_data);
4231 if (RExC_recurse_count) {
4232 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4233 SAVEFREEPV(RExC_recurse);
4237 r->minlen = minlen = sawplus = sawopen = 0;
4238 Zero(r->substrs, 1, struct reg_substr_data);
4240 #ifdef TRIE_STUDY_OPT
4243 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4245 RExC_state = copyRExC_state;
4246 if (seen & REG_TOP_LEVEL_BRANCHES)
4247 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4249 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4250 if (data.last_found) {
4251 SvREFCNT_dec(data.longest_fixed);
4252 SvREFCNT_dec(data.longest_float);
4253 SvREFCNT_dec(data.last_found);
4255 StructCopy(&zero_scan_data, &data, scan_data_t);
4257 StructCopy(&zero_scan_data, &data, scan_data_t);
4258 copyRExC_state = RExC_state;
4261 StructCopy(&zero_scan_data, &data, scan_data_t);
4264 /* Dig out information for optimizations. */
4265 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4266 pm->op_pmflags = RExC_flags;
4268 r->extflags |= RXf_UTF8; /* Unicode in it? */
4269 ri->regstclass = NULL;
4270 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4271 r->intflags |= PREGf_NAUGHTY;
4272 scan = ri->program + 1; /* First BRANCH. */
4274 /* testing for BRANCH here tells us whether there is "must appear"
4275 data in the pattern. If there is then we can use it for optimisations */
4276 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4278 STRLEN longest_float_length, longest_fixed_length;
4279 struct regnode_charclass_class ch_class; /* pointed to by data */
4281 I32 last_close = 0; /* pointed to by data */
4284 /* Skip introductions and multiplicators >= 1. */
4285 while ((OP(first) == OPEN && (sawopen = 1)) ||
4286 /* An OR of *one* alternative - should not happen now. */
4287 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4288 /* for now we can't handle lookbehind IFMATCH*/
4289 (OP(first) == IFMATCH && !first->flags) ||
4290 (OP(first) == PLUS) ||
4291 (OP(first) == MINMOD) ||
4292 /* An {n,m} with n>0 */
4293 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4296 if (OP(first) == PLUS)
4299 first += regarglen[OP(first)];
4300 if (OP(first) == IFMATCH) {
4301 first = NEXTOPER(first);
4302 first += EXTRA_STEP_2ARGS;
4303 } else /* XXX possible optimisation for /(?=)/ */
4304 first = NEXTOPER(first);
4307 /* Starting-point info. */
4309 DEBUG_PEEP("first:",first,0);
4310 /* Ignore EXACT as we deal with it later. */
4311 if (PL_regkind[OP(first)] == EXACT) {
4312 if (OP(first) == EXACT)
4313 NOOP; /* Empty, get anchored substr later. */
4314 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4315 ri->regstclass = first;
4318 else if (PL_regkind[OP(first)] == TRIE &&
4319 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4322 /* this can happen only on restudy */
4323 if ( OP(first) == TRIE ) {
4324 struct regnode_1 *trieop = (struct regnode_1 *)
4325 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4326 StructCopy(first,trieop,struct regnode_1);
4327 trie_op=(regnode *)trieop;
4329 struct regnode_charclass *trieop = (struct regnode_charclass *)
4330 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4331 StructCopy(first,trieop,struct regnode_charclass);
4332 trie_op=(regnode *)trieop;
4335 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4336 ri->regstclass = trie_op;
4339 else if (strchr((const char*)PL_simple,OP(first)))
4340 ri->regstclass = first;
4341 else if (PL_regkind[OP(first)] == BOUND ||
4342 PL_regkind[OP(first)] == NBOUND)
4343 ri->regstclass = first;
4344 else if (PL_regkind[OP(first)] == BOL) {
4345 r->extflags |= (OP(first) == MBOL
4347 : (OP(first) == SBOL
4350 first = NEXTOPER(first);
4353 else if (OP(first) == GPOS) {
4354 r->extflags |= RXf_ANCH_GPOS;
4355 first = NEXTOPER(first);
4358 else if ((!sawopen || !RExC_sawback) &&
4359 (OP(first) == STAR &&
4360 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4361 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4363 /* turn .* into ^.* with an implied $*=1 */
4365 (OP(NEXTOPER(first)) == REG_ANY)
4368 r->extflags |= type;
4369 r->intflags |= PREGf_IMPLICIT;
4370 first = NEXTOPER(first);
4373 if (sawplus && (!sawopen || !RExC_sawback)
4374 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4375 /* x+ must match at the 1st pos of run of x's */
4376 r->intflags |= PREGf_SKIP;
4378 /* Scan is after the zeroth branch, first is atomic matcher. */
4379 #ifdef TRIE_STUDY_OPT
4382 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4383 (IV)(first - scan + 1))
4387 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4388 (IV)(first - scan + 1))
4394 * If there's something expensive in the r.e., find the
4395 * longest literal string that must appear and make it the
4396 * regmust. Resolve ties in favor of later strings, since
4397 * the regstart check works with the beginning of the r.e.
4398 * and avoiding duplication strengthens checking. Not a
4399 * strong reason, but sufficient in the absence of others.
4400 * [Now we resolve ties in favor of the earlier string if
4401 * it happens that c_offset_min has been invalidated, since the
4402 * earlier string may buy us something the later one won't.]
4405 data.longest_fixed = newSVpvs("");
4406 data.longest_float = newSVpvs("");
4407 data.last_found = newSVpvs("");
4408 data.longest = &(data.longest_fixed);
4410 if (!ri->regstclass) {
4411 cl_init(pRExC_state, &ch_class);
4412 data.start_class = &ch_class;
4413 stclass_flag = SCF_DO_STCLASS_AND;
4414 } else /* XXXX Check for BOUND? */
4416 data.last_closep = &last_close;
4418 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4419 &data, -1, NULL, NULL,
4420 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4426 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4427 && data.last_start_min == 0 && data.last_end > 0
4428 && !RExC_seen_zerolen
4429 && !(RExC_seen & REG_SEEN_VERBARG)
4430 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4431 r->extflags |= RXf_CHECK_ALL;
4432 scan_commit(pRExC_state, &data,&minlen,0);
4433 SvREFCNT_dec(data.last_found);
4435 /* Note that code very similar to this but for anchored string
4436 follows immediately below, changes may need to be made to both.
4439 longest_float_length = CHR_SVLEN(data.longest_float);
4440 if (longest_float_length
4441 || (data.flags & SF_FL_BEFORE_EOL
4442 && (!(data.flags & SF_FL_BEFORE_MEOL)
4443 || (RExC_flags & RXf_PMf_MULTILINE))))
4447 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4448 && data.offset_fixed == data.offset_float_min
4449 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4450 goto remove_float; /* As in (a)+. */
4452 /* copy the information about the longest float from the reg_scan_data
4453 over to the program. */
4454 if (SvUTF8(data.longest_float)) {
4455 r->float_utf8 = data.longest_float;
4456 r->float_substr = NULL;
4458 r->float_substr = data.longest_float;
4459 r->float_utf8 = NULL;
4461 /* float_end_shift is how many chars that must be matched that
4462 follow this item. We calculate it ahead of time as once the
4463 lookbehind offset is added in we lose the ability to correctly
4465 ml = data.minlen_float ? *(data.minlen_float)
4466 : (I32)longest_float_length;
4467 r->float_end_shift = ml - data.offset_float_min
4468 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4469 + data.lookbehind_float;
4470 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4471 r->float_max_offset = data.offset_float_max;
4472 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4473 r->float_max_offset -= data.lookbehind_float;
4475 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4476 && (!(data.flags & SF_FL_BEFORE_MEOL)
4477 || (RExC_flags & RXf_PMf_MULTILINE)));
4478 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4482 r->float_substr = r->float_utf8 = NULL;
4483 SvREFCNT_dec(data.longest_float);
4484 longest_float_length = 0;
4487 /* Note that code very similar to this but for floating string
4488 is immediately above, changes may need to be made to both.
4491 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4492 if (longest_fixed_length
4493 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4494 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4495 || (RExC_flags & RXf_PMf_MULTILINE))))
4499 /* copy the information about the longest fixed
4500 from the reg_scan_data over to the program. */
4501 if (SvUTF8(data.longest_fixed)) {
4502 r->anchored_utf8 = data.longest_fixed;
4503 r->anchored_substr = NULL;
4505 r->anchored_substr = data.longest_fixed;
4506 r->anchored_utf8 = NULL;
4508 /* fixed_end_shift is how many chars that must be matched that
4509 follow this item. We calculate it ahead of time as once the
4510 lookbehind offset is added in we lose the ability to correctly
4512 ml = data.minlen_fixed ? *(data.minlen_fixed)
4513 : (I32)longest_fixed_length;
4514 r->anchored_end_shift = ml - data.offset_fixed
4515 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4516 + data.lookbehind_fixed;
4517 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4519 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4520 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4521 || (RExC_flags & RXf_PMf_MULTILINE)));
4522 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4525 r->anchored_substr = r->anchored_utf8 = NULL;
4526 SvREFCNT_dec(data.longest_fixed);
4527 longest_fixed_length = 0;
4530 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4531 ri->regstclass = NULL;
4532 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4534 && !(data.start_class->flags & ANYOF_EOS)
4535 && !cl_is_anything(data.start_class))
4537 const U32 n = add_data(pRExC_state, 1, "f");
4539 Newx(RExC_rxi->data->data[n], 1,
4540 struct regnode_charclass_class);
4541 StructCopy(data.start_class,
4542 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4543 struct regnode_charclass_class);
4544 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4545 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4546 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4547 regprop(r, sv, (regnode*)data.start_class);
4548 PerlIO_printf(Perl_debug_log,
4549 "synthetic stclass \"%s\".\n",
4550 SvPVX_const(sv));});
4553 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4554 if (longest_fixed_length > longest_float_length) {
4555 r->check_end_shift = r->anchored_end_shift;
4556 r->check_substr = r->anchored_substr;
4557 r->check_utf8 = r->anchored_utf8;
4558 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4559 if (r->extflags & RXf_ANCH_SINGLE)
4560 r->extflags |= RXf_NOSCAN;
4563 r->check_end_shift = r->float_end_shift;
4564 r->check_substr = r->float_substr;
4565 r->check_utf8 = r->float_utf8;
4566 r->check_offset_min = r->float_min_offset;
4567 r->check_offset_max = r->float_max_offset;
4569 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4570 This should be changed ASAP! */
4571 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4572 r->extflags |= RXf_USE_INTUIT;
4573 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4574 r->extflags |= RXf_INTUIT_TAIL;
4576 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4577 if ( (STRLEN)minlen < longest_float_length )
4578 minlen= longest_float_length;
4579 if ( (STRLEN)minlen < longest_fixed_length )
4580 minlen= longest_fixed_length;
4584 /* Several toplevels. Best we can is to set minlen. */
4586 struct regnode_charclass_class ch_class;
4589 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4591 scan = ri->program + 1;
4592 cl_init(pRExC_state, &ch_class);
4593 data.start_class = &ch_class;
4594 data.last_closep = &last_close;
4597 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4598 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4602 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4603 = r->float_substr = r->float_utf8 = NULL;
4604 if (!(data.start_class->flags & ANYOF_EOS)
4605 && !cl_is_anything(data.start_class))
4607 const U32 n = add_data(pRExC_state, 1, "f");
4609 Newx(RExC_rxi->data->data[n], 1,
4610 struct regnode_charclass_class);
4611 StructCopy(data.start_class,
4612 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4613 struct regnode_charclass_class);
4614 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4615 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4616 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4617 regprop(r, sv, (regnode*)data.start_class);
4618 PerlIO_printf(Perl_debug_log,
4619 "synthetic stclass \"%s\".\n",
4620 SvPVX_const(sv));});
4624 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4625 the "real" pattern. */
4627 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4628 (IV)minlen, (IV)r->minlen);
4630 r->minlenret = minlen;
4631 if (r->minlen < minlen)
4634 if (RExC_seen & REG_SEEN_GPOS)
4635 r->extflags |= RXf_GPOS_SEEN;
4636 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4637 r->extflags |= RXf_LOOKBEHIND_SEEN;
4638 if (RExC_seen & REG_SEEN_EVAL)
4639 r->extflags |= RXf_EVAL_SEEN;
4640 if (RExC_seen & REG_SEEN_CANY)
4641 r->extflags |= RXf_CANY_SEEN;
4642 if (RExC_seen & REG_SEEN_VERBARG)
4643 r->intflags |= PREGf_VERBARG_SEEN;
4644 if (RExC_seen & REG_SEEN_CUTGROUP)
4645 r->intflags |= PREGf_CUTGROUP_SEEN;
4646 if (RExC_paren_names)
4647 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4649 r->paren_names = NULL;
4650 if (r->prelen == 3 && strEQ("\\s+", r->precomp))
4651 r->extflags |= RXf_WHITE;
4652 else if (r->prelen == 1 && r->precomp[0] == '^')
4653 r->extflags |= RXf_START_ONLY;
4656 if (RExC_paren_names) {
4657 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4658 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4661 ri->name_list_idx = 0;
4663 if (RExC_recurse_count) {
4664 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4665 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4666 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4669 Newxz(r->startp, RExC_npar, I32);
4670 Newxz(r->endp, RExC_npar, I32);
4671 /* assume we don't need to swap parens around before we match */
4674 PerlIO_printf(Perl_debug_log,"Final program:\n");
4677 #ifdef RE_TRACK_PATTERN_OFFSETS
4678 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4679 const U32 len = ri->u.offsets[0];
4681 GET_RE_DEBUG_FLAGS_DECL;
4682 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4683 for (i = 1; i <= len; i++) {
4684 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4685 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4686 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4688 PerlIO_printf(Perl_debug_log, "\n");
4694 #undef CORE_ONLY_BLOCK
4695 #undef RE_ENGINE_PTR
4697 #ifndef PERL_IN_XSUB_RE
4699 Perl_reg_named_buff_get(pTHX_ SV* namesv, const REGEXP * const from_re, U32 flags)
4701 AV *retarray = NULL;
4706 if (from_re || PL_curpm) {
4707 const REGEXP * const rx = from_re ? from_re : PM_GETRE(PL_curpm);
4708 if (rx && rx->paren_names) {
4709 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4712 SV* sv_dat=HeVAL(he_str);
4713 I32 *nums=(I32*)SvPVX(sv_dat);
4714 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4715 if ((I32)(rx->nparens) >= nums[i]
4716 && rx->startp[nums[i]] != -1
4717 && rx->endp[nums[i]] != -1)
4719 ret = reg_numbered_buff_get(nums[i],rx,NULL,0);
4723 ret = newSVsv(&PL_sv_undef);
4727 av_push(retarray, ret);
4731 return (SV*)retarray;
4739 Perl_reg_numbered_buff_get(pTHX_ I32 paren, const REGEXP * const rx, SV* usesv, U32 flags)
4744 SV *sv = usesv ? usesv : newSVpvs("");
4745 PERL_UNUSED_ARG(flags);
4748 sv_setsv(sv,&PL_sv_undef);
4752 if (paren == -2 && rx->startp[0] != -1) {
4758 if (paren == -1 && rx->endp[0] != -1) {
4760 s = rx->subbeg + rx->endp[0];
4761 i = rx->sublen - rx->endp[0];
4764 if ( 0 <= paren && paren <= (I32)rx->nparens &&
4765 (s1 = rx->startp[paren]) != -1 &&
4766 (t1 = rx->endp[paren]) != -1)
4770 s = rx->subbeg + s1;
4772 sv_setsv(sv,&PL_sv_undef);
4775 assert(rx->sublen >= (s - rx->subbeg) + i );
4777 const int oldtainted = PL_tainted;
4779 sv_setpvn(sv, s, i);
4780 PL_tainted = oldtainted;
4781 if ( (rx->extflags & RXf_CANY_SEEN)
4782 ? (RX_MATCH_UTF8(rx)
4783 && (!i || is_utf8_string((U8*)s, i)))
4784 : (RX_MATCH_UTF8(rx)) )
4791 if (RX_MATCH_TAINTED(rx)) {
4792 if (SvTYPE(sv) >= SVt_PVMG) {
4793 MAGIC* const mg = SvMAGIC(sv);
4796 SvMAGIC_set(sv, mg->mg_moremagic);
4798 if ((mgt = SvMAGIC(sv))) {
4799 mg->mg_moremagic = mgt;
4800 SvMAGIC_set(sv, mg);
4810 sv_setsv(sv,&PL_sv_undef);
4816 /* Scans the name of a named buffer from the pattern.
4817 * If flags is REG_RSN_RETURN_NULL returns null.
4818 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4819 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4820 * to the parsed name as looked up in the RExC_paren_names hash.
4821 * If there is an error throws a vFAIL().. type exception.
4824 #define REG_RSN_RETURN_NULL 0
4825 #define REG_RSN_RETURN_NAME 1
4826 #define REG_RSN_RETURN_DATA 2
4829 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4830 char *name_start = RExC_parse;
4832 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
4833 /* skip IDFIRST by using do...while */
4836 RExC_parse += UTF8SKIP(RExC_parse);
4837 } while (isALNUM_utf8((U8*)RExC_parse));
4841 } while (isALNUM(*RExC_parse));
4845 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4846 (int)(RExC_parse - name_start)));
4849 if ( flags == REG_RSN_RETURN_NAME)
4851 else if (flags==REG_RSN_RETURN_DATA) {
4854 if ( ! sv_name ) /* should not happen*/
4855 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4856 if (RExC_paren_names)
4857 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4859 sv_dat = HeVAL(he_str);
4861 vFAIL("Reference to nonexistent named group");
4865 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4872 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4873 int rem=(int)(RExC_end - RExC_parse); \
4882 if (RExC_lastparse!=RExC_parse) \
4883 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4886 iscut ? "..." : "<" \
4889 PerlIO_printf(Perl_debug_log,"%16s",""); \
4894 num=REG_NODE_NUM(RExC_emit); \
4895 if (RExC_lastnum!=num) \
4896 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4898 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4899 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4900 (int)((depth*2)), "", \
4904 RExC_lastparse=RExC_parse; \
4909 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4910 DEBUG_PARSE_MSG((funcname)); \
4911 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4913 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4914 DEBUG_PARSE_MSG((funcname)); \
4915 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4918 - reg - regular expression, i.e. main body or parenthesized thing
4920 * Caller must absorb opening parenthesis.
4922 * Combining parenthesis handling with the base level of regular expression
4923 * is a trifle forced, but the need to tie the tails of the branches to what
4924 * follows makes it hard to avoid.
4926 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4928 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4930 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4933 /* this idea is borrowed from STR_WITH_LEN in handy.h */
4934 #define CHECK_WORD(s,v,l) \
4935 (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
4938 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4939 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4942 register regnode *ret; /* Will be the head of the group. */
4943 register regnode *br;
4944 register regnode *lastbr;
4945 register regnode *ender = NULL;
4946 register I32 parno = 0;
4948 const I32 oregflags = RExC_flags;
4949 bool have_branch = 0;
4951 I32 freeze_paren = 0;
4952 I32 after_freeze = 0;
4954 /* for (?g), (?gc), and (?o) warnings; warning
4955 about (?c) will warn about (?g) -- japhy */
4957 #define WASTED_O 0x01
4958 #define WASTED_G 0x02
4959 #define WASTED_C 0x04
4960 #define WASTED_GC (0x02|0x04)
4961 I32 wastedflags = 0x00;
4963 char * parse_start = RExC_parse; /* MJD */
4964 char * const oregcomp_parse = RExC_parse;
4966 GET_RE_DEBUG_FLAGS_DECL;
4967 DEBUG_PARSE("reg ");
4970 *flagp = 0; /* Tentatively. */
4973 /* Make an OPEN node, if parenthesized. */
4975 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4976 char *start_verb = RExC_parse;
4977 STRLEN verb_len = 0;
4978 char *start_arg = NULL;
4979 unsigned char op = 0;
4981 int internal_argval = 0; /* internal_argval is only useful if !argok */
4982 while ( *RExC_parse && *RExC_parse != ')' ) {
4983 if ( *RExC_parse == ':' ) {
4984 start_arg = RExC_parse + 1;
4990 verb_len = RExC_parse - start_verb;
4993 while ( *RExC_parse && *RExC_parse != ')' )
4995 if ( *RExC_parse != ')' )
4996 vFAIL("Unterminated verb pattern argument");
4997 if ( RExC_parse == start_arg )
5000 if ( *RExC_parse != ')' )
5001 vFAIL("Unterminated verb pattern");
5004 switch ( *start_verb ) {
5005 case 'A': /* (*ACCEPT) */
5006 if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
5008 internal_argval = RExC_nestroot;
5011 case 'C': /* (*COMMIT) */
5012 if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
5015 case 'F': /* (*FAIL) */
5016 if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
5021 case ':': /* (*:NAME) */
5022 case 'M': /* (*MARK:NAME) */
5023 if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
5028 case 'P': /* (*PRUNE) */
5029 if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
5032 case 'S': /* (*SKIP) */
5033 if ( CHECK_WORD("SKIP",start_verb,verb_len) )
5036 case 'T': /* (*THEN) */
5037 /* [19:06] <TimToady> :: is then */
5038 if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
5040 RExC_seen |= REG_SEEN_CUTGROUP;
5046 vFAIL3("Unknown verb pattern '%.*s'",
5047 verb_len, start_verb);
5050 if ( start_arg && internal_argval ) {
5051 vFAIL3("Verb pattern '%.*s' may not have an argument",
5052 verb_len, start_verb);
5053 } else if ( argok < 0 && !start_arg ) {
5054 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5055 verb_len, start_verb);
5057 ret = reganode(pRExC_state, op, internal_argval);
5058 if ( ! internal_argval && ! SIZE_ONLY ) {
5060 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5061 ARG(ret) = add_data( pRExC_state, 1, "S" );
5062 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5069 if (!internal_argval)
5070 RExC_seen |= REG_SEEN_VERBARG;
5071 } else if ( start_arg ) {
5072 vFAIL3("Verb pattern '%.*s' may not have an argument",
5073 verb_len, start_verb);
5075 ret = reg_node(pRExC_state, op);
5077 nextchar(pRExC_state);
5080 if (*RExC_parse == '?') { /* (?...) */
5081 bool is_logical = 0;
5082 const char * const seqstart = RExC_parse;
5085 paren = *RExC_parse++;
5086 ret = NULL; /* For look-ahead/behind. */
5089 case 'P': /* (?P...) variants for those used to PCRE/Python */
5090 paren = *RExC_parse++;
5091 if ( paren == '<') /* (?P<...>) named capture */
5093 else if (paren == '>') { /* (?P>name) named recursion */
5094 goto named_recursion;
5096 else if (paren == '=') { /* (?P=...) named backref */
5097 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5098 you change this make sure you change that */
5099 char* name_start = RExC_parse;
5101 SV *sv_dat = reg_scan_name(pRExC_state,
5102 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5103 if (RExC_parse == name_start || *RExC_parse != ')')
5104 vFAIL2("Sequence %.3s... not terminated",parse_start);
5107 num = add_data( pRExC_state, 1, "S" );
5108 RExC_rxi->data->data[num]=(void*)sv_dat;
5109 SvREFCNT_inc(sv_dat);
5112 ret = reganode(pRExC_state,
5113 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5117 Set_Node_Offset(ret, parse_start+1);
5118 Set_Node_Cur_Length(ret); /* MJD */
5120 nextchar(pRExC_state);
5124 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5126 case '<': /* (?<...) */
5127 if (*RExC_parse == '!')
5129 else if (*RExC_parse != '=')
5135 case '\'': /* (?'...') */
5136 name_start= RExC_parse;
5137 svname = reg_scan_name(pRExC_state,
5138 SIZE_ONLY ? /* reverse test from the others */
5139 REG_RSN_RETURN_NAME :
5140 REG_RSN_RETURN_NULL);
5141 if (RExC_parse == name_start) {
5143 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5146 if (*RExC_parse != paren)
5147 vFAIL2("Sequence (?%c... not terminated",
5148 paren=='>' ? '<' : paren);
5152 if (!svname) /* shouldnt happen */
5154 "panic: reg_scan_name returned NULL");
5155 if (!RExC_paren_names) {
5156 RExC_paren_names= newHV();
5157 sv_2mortal((SV*)RExC_paren_names);
5159 RExC_paren_name_list= newAV();
5160 sv_2mortal((SV*)RExC_paren_name_list);
5163 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5165 sv_dat = HeVAL(he_str);
5167 /* croak baby croak */
5169 "panic: paren_name hash element allocation failed");
5170 } else if ( SvPOK(sv_dat) ) {
5171 IV count=SvIV(sv_dat);
5172 I32 *pv=(I32*)SvGROW(sv_dat,SvCUR(sv_dat)+sizeof(I32)+1);
5173 SvCUR_set(sv_dat,SvCUR(sv_dat)+sizeof(I32));
5174 pv[count]=RExC_npar;
5177 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5178 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5183 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5184 SvREFCNT_dec(svname);
5187 /*sv_dump(sv_dat);*/
5189 nextchar(pRExC_state);
5191 goto capturing_parens;
5193 RExC_seen |= REG_SEEN_LOOKBEHIND;
5195 case '=': /* (?=...) */
5196 case '!': /* (?!...) */
5197 RExC_seen_zerolen++;
5198 if (*RExC_parse == ')') {
5199 ret=reg_node(pRExC_state, OPFAIL);
5200 nextchar(pRExC_state);
5204 case '|': /* (?|...) */
5205 /* branch reset, behave like a (?:...) except that
5206 buffers in alternations share the same numbers */
5208 after_freeze = freeze_paren = RExC_npar;
5210 case ':': /* (?:...) */
5211 case '>': /* (?>...) */
5213 case '$': /* (?$...) */
5214 case '@': /* (?@...) */
5215 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5217 case '#': /* (?#...) */
5218 while (*RExC_parse && *RExC_parse != ')')
5220 if (*RExC_parse != ')')
5221 FAIL("Sequence (?#... not terminated");
5222 nextchar(pRExC_state);
5225 case '0' : /* (?0) */
5226 case 'R' : /* (?R) */
5227 if (*RExC_parse != ')')
5228 FAIL("Sequence (?R) not terminated");
5229 ret = reg_node(pRExC_state, GOSTART);
5230 nextchar(pRExC_state);
5233 { /* named and numeric backreferences */
5235 case '&': /* (?&NAME) */
5236 parse_start = RExC_parse - 1;
5239 SV *sv_dat = reg_scan_name(pRExC_state,
5240 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5241 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5243 goto gen_recurse_regop;
5246 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5248 vFAIL("Illegal pattern");
5250 goto parse_recursion;
5252 case '-': /* (?-1) */
5253 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5254 RExC_parse--; /* rewind to let it be handled later */
5258 case '1': case '2': case '3': case '4': /* (?1) */
5259 case '5': case '6': case '7': case '8': case '9':
5262 num = atoi(RExC_parse);
5263 parse_start = RExC_parse - 1; /* MJD */
5264 if (*RExC_parse == '-')
5266 while (isDIGIT(*RExC_parse))
5268 if (*RExC_parse!=')')
5269 vFAIL("Expecting close bracket");
5272 if ( paren == '-' ) {
5274 Diagram of capture buffer numbering.
5275 Top line is the normal capture buffer numbers
5276 Botton line is the negative indexing as from
5280 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5284 num = RExC_npar + num;
5287 vFAIL("Reference to nonexistent group");
5289 } else if ( paren == '+' ) {
5290 num = RExC_npar + num - 1;
5293 ret = reganode(pRExC_state, GOSUB, num);
5295 if (num > (I32)RExC_rx->nparens) {
5297 vFAIL("Reference to nonexistent group");
5299 ARG2L_SET( ret, RExC_recurse_count++);
5301 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5302 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5306 RExC_seen |= REG_SEEN_RECURSE;
5307 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5308 Set_Node_Offset(ret, parse_start); /* MJD */
5310 nextchar(pRExC_state);
5312 } /* named and numeric backreferences */
5315 case 'p': /* (?p...) */
5316 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
5317 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
5319 case '?': /* (??...) */
5321 if (*RExC_parse != '{') {
5323 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5326 paren = *RExC_parse++;
5328 case '{': /* (?{...}) */
5333 char *s = RExC_parse;
5335 RExC_seen_zerolen++;
5336 RExC_seen |= REG_SEEN_EVAL;
5337 while (count && (c = *RExC_parse)) {
5348 if (*RExC_parse != ')') {
5350 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5354 OP_4tree *sop, *rop;
5355 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5358 Perl_save_re_context(aTHX);
5359 rop = sv_compile_2op(sv, &sop, "re", &pad);
5360 sop->op_private |= OPpREFCOUNTED;
5361 /* re_dup will OpREFCNT_inc */
5362 OpREFCNT_set(sop, 1);
5365 n = add_data(pRExC_state, 3, "nop");
5366 RExC_rxi->data->data[n] = (void*)rop;
5367 RExC_rxi->data->data[n+1] = (void*)sop;
5368 RExC_rxi->data->data[n+2] = (void*)pad;
5371 else { /* First pass */
5372 if (PL_reginterp_cnt < ++RExC_seen_evals
5374 /* No compiled RE interpolated, has runtime
5375 components ===> unsafe. */
5376 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5377 if (PL_tainting && PL_tainted)
5378 FAIL("Eval-group in insecure regular expression");
5379 #if PERL_VERSION > 8
5380 if (IN_PERL_COMPILETIME)
5385 nextchar(pRExC_state);
5387 ret = reg_node(pRExC_state, LOGICAL);
5390 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5391 /* deal with the length of this later - MJD */
5394 ret = reganode(pRExC_state, EVAL, n);
5395 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5396 Set_Node_Offset(ret, parse_start);
5399 case '(': /* (?(?{...})...) and (?(?=...)...) */
5402 if (RExC_parse[0] == '?') { /* (?(?...)) */
5403 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5404 || RExC_parse[1] == '<'
5405 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5408 ret = reg_node(pRExC_state, LOGICAL);
5411 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5415 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5416 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5418 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5419 char *name_start= RExC_parse++;
5421 SV *sv_dat=reg_scan_name(pRExC_state,
5422 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5423 if (RExC_parse == name_start || *RExC_parse != ch)
5424 vFAIL2("Sequence (?(%c... not terminated",
5425 (ch == '>' ? '<' : ch));
5428 num = add_data( pRExC_state, 1, "S" );
5429 RExC_rxi->data->data[num]=(void*)sv_dat;
5430 SvREFCNT_inc(sv_dat);
5432 ret = reganode(pRExC_state,NGROUPP,num);
5433 goto insert_if_check_paren;
5435 else if (RExC_parse[0] == 'D' &&
5436 RExC_parse[1] == 'E' &&
5437 RExC_parse[2] == 'F' &&
5438 RExC_parse[3] == 'I' &&
5439 RExC_parse[4] == 'N' &&
5440 RExC_parse[5] == 'E')
5442 ret = reganode(pRExC_state,DEFINEP,0);
5445 goto insert_if_check_paren;
5447 else if (RExC_parse[0] == 'R') {
5450 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5451 parno = atoi(RExC_parse++);
5452 while (isDIGIT(*RExC_parse))
5454 } else if (RExC_parse[0] == '&') {
5457 sv_dat = reg_scan_name(pRExC_state,
5458 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5459 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5461 ret = reganode(pRExC_state,INSUBP,parno);
5462 goto insert_if_check_paren;
5464 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5467 parno = atoi(RExC_parse++);
5469 while (isDIGIT(*RExC_parse))
5471 ret = reganode(pRExC_state, GROUPP, parno);
5473 insert_if_check_paren:
5474 if ((c = *nextchar(pRExC_state)) != ')')
5475 vFAIL("Switch condition not recognized");
5477 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5478 br = regbranch(pRExC_state, &flags, 1,depth+1);
5480 br = reganode(pRExC_state, LONGJMP, 0);
5482 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5483 c = *nextchar(pRExC_state);
5488 vFAIL("(?(DEFINE)....) does not allow branches");
5489 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5490 regbranch(pRExC_state, &flags, 1,depth+1);
5491 REGTAIL(pRExC_state, ret, lastbr);
5494 c = *nextchar(pRExC_state);
5499 vFAIL("Switch (?(condition)... contains too many branches");
5500 ender = reg_node(pRExC_state, TAIL);
5501 REGTAIL(pRExC_state, br, ender);
5503 REGTAIL(pRExC_state, lastbr, ender);
5504 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5507 REGTAIL(pRExC_state, ret, ender);
5511 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5515 RExC_parse--; /* for vFAIL to print correctly */
5516 vFAIL("Sequence (? incomplete");
5520 parse_flags: /* (?i) */
5522 U32 posflags = 0, negflags = 0;
5523 U32 *flagsp = &posflags;
5525 while (*RExC_parse) {
5526 /* && strchr("iogcmsx", *RExC_parse) */
5527 /* (?g), (?gc) and (?o) are useless here
5528 and must be globally applied -- japhy */
5529 switch (*RExC_parse) {
5530 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5533 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5534 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5535 if (! (wastedflags & wflagbit) ) {
5536 wastedflags |= wflagbit;
5539 "Useless (%s%c) - %suse /%c modifier",
5540 flagsp == &negflags ? "?-" : "?",
5542 flagsp == &negflags ? "don't " : "",
5550 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5551 if (! (wastedflags & WASTED_C) ) {
5552 wastedflags |= WASTED_GC;
5555 "Useless (%sc) - %suse /gc modifier",
5556 flagsp == &negflags ? "?-" : "?",
5557 flagsp == &negflags ? "don't " : ""
5563 if (flagsp == &negflags) {
5564 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5565 vWARN(RExC_parse + 1,"Useless use of (?-k)");
5567 *flagsp |= RXf_PMf_KEEPCOPY;
5571 if (flagsp == &negflags) {
5573 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5577 wastedflags = 0; /* reset so (?g-c) warns twice */
5583 RExC_flags |= posflags;
5584 RExC_flags &= ~negflags;
5585 nextchar(pRExC_state);
5596 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5601 }} /* one for the default block, one for the switch */
5608 ret = reganode(pRExC_state, OPEN, parno);
5611 RExC_nestroot = parno;
5612 if (RExC_seen & REG_SEEN_RECURSE) {
5613 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5614 "Setting open paren #%"IVdf" to %d\n",
5615 (IV)parno, REG_NODE_NUM(ret)));
5616 RExC_open_parens[parno-1]= ret;
5619 Set_Node_Length(ret, 1); /* MJD */
5620 Set_Node_Offset(ret, RExC_parse); /* MJD */
5628 /* Pick up the branches, linking them together. */
5629 parse_start = RExC_parse; /* MJD */
5630 br = regbranch(pRExC_state, &flags, 1,depth+1);
5631 /* branch_len = (paren != 0); */
5635 if (*RExC_parse == '|') {
5636 if (!SIZE_ONLY && RExC_extralen) {
5637 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5640 reginsert(pRExC_state, BRANCH, br, depth+1);
5641 Set_Node_Length(br, paren != 0);
5642 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5646 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5648 else if (paren == ':') {
5649 *flagp |= flags&SIMPLE;
5651 if (is_open) { /* Starts with OPEN. */
5652 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5654 else if (paren != '?') /* Not Conditional */
5656 *flagp |= flags & (SPSTART | HASWIDTH);
5658 while (*RExC_parse == '|') {
5659 if (!SIZE_ONLY && RExC_extralen) {
5660 ender = reganode(pRExC_state, LONGJMP,0);
5661 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5664 RExC_extralen += 2; /* Account for LONGJMP. */
5665 nextchar(pRExC_state);
5667 if (RExC_npar > after_freeze)
5668 after_freeze = RExC_npar;
5669 RExC_npar = freeze_paren;
5671 br = regbranch(pRExC_state, &flags, 0, depth+1);
5675 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5679 *flagp |= flags&SPSTART;
5682 if (have_branch || paren != ':') {
5683 /* Make a closing node, and hook it on the end. */
5686 ender = reg_node(pRExC_state, TAIL);
5689 ender = reganode(pRExC_state, CLOSE, parno);
5690 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5691 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5692 "Setting close paren #%"IVdf" to %d\n",
5693 (IV)parno, REG_NODE_NUM(ender)));
5694 RExC_close_parens[parno-1]= ender;
5695 if (RExC_nestroot == parno)
5698 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5699 Set_Node_Length(ender,1); /* MJD */
5705 *flagp &= ~HASWIDTH;
5708 ender = reg_node(pRExC_state, SUCCEED);
5711 ender = reg_node(pRExC_state, END);
5713 assert(!RExC_opend); /* there can only be one! */
5718 REGTAIL(pRExC_state, lastbr, ender);
5720 if (have_branch && !SIZE_ONLY) {
5722 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5724 /* Hook the tails of the branches to the closing node. */
5725 for (br = ret; br; br = regnext(br)) {
5726 const U8 op = PL_regkind[OP(br)];
5728 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5730 else if (op == BRANCHJ) {
5731 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5739 static const char parens[] = "=!<,>";
5741 if (paren && (p = strchr(parens, paren))) {
5742 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5743 int flag = (p - parens) > 1;
5746 node = SUSPEND, flag = 0;
5747 reginsert(pRExC_state, node,ret, depth+1);
5748 Set_Node_Cur_Length(ret);
5749 Set_Node_Offset(ret, parse_start + 1);
5751 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5755 /* Check for proper termination. */
5757 RExC_flags = oregflags;
5758 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5759 RExC_parse = oregcomp_parse;
5760 vFAIL("Unmatched (");
5763 else if (!paren && RExC_parse < RExC_end) {
5764 if (*RExC_parse == ')') {
5766 vFAIL("Unmatched )");
5769 FAIL("Junk on end of regexp"); /* "Can't happen". */
5773 RExC_npar = after_freeze;
5778 - regbranch - one alternative of an | operator
5780 * Implements the concatenation operator.
5783 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5786 register regnode *ret;
5787 register regnode *chain = NULL;
5788 register regnode *latest;
5789 I32 flags = 0, c = 0;
5790 GET_RE_DEBUG_FLAGS_DECL;
5791 DEBUG_PARSE("brnc");
5795 if (!SIZE_ONLY && RExC_extralen)
5796 ret = reganode(pRExC_state, BRANCHJ,0);
5798 ret = reg_node(pRExC_state, BRANCH);
5799 Set_Node_Length(ret, 1);
5803 if (!first && SIZE_ONLY)
5804 RExC_extralen += 1; /* BRANCHJ */
5806 *flagp = WORST; /* Tentatively. */
5809 nextchar(pRExC_state);
5810 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5812 latest = regpiece(pRExC_state, &flags,depth+1);
5813 if (latest == NULL) {
5814 if (flags & TRYAGAIN)
5818 else if (ret == NULL)
5820 *flagp |= flags&HASWIDTH;
5821 if (chain == NULL) /* First piece. */
5822 *flagp |= flags&SPSTART;
5825 REGTAIL(pRExC_state, chain, latest);
5830 if (chain == NULL) { /* Loop ran zero times. */
5831 chain = reg_node(pRExC_state, NOTHING);
5836 *flagp |= flags&SIMPLE;
5843 - regpiece - something followed by possible [*+?]
5845 * Note that the branching code sequences used for ? and the general cases
5846 * of * and + are somewhat optimized: they use the same NOTHING node as
5847 * both the endmarker for their branch list and the body of the last branch.
5848 * It might seem that this node could be dispensed with entirely, but the
5849 * endmarker role is not redundant.
5852 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5855 register regnode *ret;
5857 register char *next;
5859 const char * const origparse = RExC_parse;
5861 I32 max = REG_INFTY;
5863 const char *maxpos = NULL;
5864 GET_RE_DEBUG_FLAGS_DECL;
5865 DEBUG_PARSE("piec");
5867 ret = regatom(pRExC_state, &flags,depth+1);
5869 if (flags & TRYAGAIN)
5876 if (op == '{' && regcurly(RExC_parse)) {
5878 parse_start = RExC_parse; /* MJD */
5879 next = RExC_parse + 1;
5880 while (isDIGIT(*next) || *next == ',') {
5889 if (*next == '}') { /* got one */
5893 min = atoi(RExC_parse);
5897 maxpos = RExC_parse;
5899 if (!max && *maxpos != '0')
5900 max = REG_INFTY; /* meaning "infinity" */
5901 else if (max >= REG_INFTY)
5902 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5904 nextchar(pRExC_state);
5907 if ((flags&SIMPLE)) {
5908 RExC_naughty += 2 + RExC_naughty / 2;
5909 reginsert(pRExC_state, CURLY, ret, depth+1);
5910 Set_Node_Offset(ret, parse_start+1); /* MJD */
5911 Set_Node_Cur_Length(ret);
5914 regnode * const w = reg_node(pRExC_state, WHILEM);
5917 REGTAIL(pRExC_state, ret, w);
5918 if (!SIZE_ONLY && RExC_extralen) {
5919 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5920 reginsert(pRExC_state, NOTHING,ret, depth+1);
5921 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5923 reginsert(pRExC_state, CURLYX,ret, depth+1);
5925 Set_Node_Offset(ret, parse_start+1);
5926 Set_Node_Length(ret,
5927 op == '{' ? (RExC_parse - parse_start) : 1);
5929 if (!SIZE_ONLY && RExC_extralen)
5930 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5931 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5933 RExC_whilem_seen++, RExC_extralen += 3;
5934 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5942 if (max && max < min)
5943 vFAIL("Can't do {n,m} with n > m");
5945 ARG1_SET(ret, (U16)min);
5946 ARG2_SET(ret, (U16)max);
5958 #if 0 /* Now runtime fix should be reliable. */
5960 /* if this is reinstated, don't forget to put this back into perldiag:
5962 =item Regexp *+ operand could be empty at {#} in regex m/%s/
5964 (F) The part of the regexp subject to either the * or + quantifier
5965 could match an empty string. The {#} shows in the regular
5966 expression about where the problem was discovered.
5970 if (!(flags&HASWIDTH) && op != '?')
5971 vFAIL("Regexp *+ operand could be empty");
5974 parse_start = RExC_parse;
5975 nextchar(pRExC_state);
5977 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
5979 if (op == '*' && (flags&SIMPLE)) {
5980 reginsert(pRExC_state, STAR, ret, depth+1);
5984 else if (op == '*') {
5988 else if (op == '+' && (flags&SIMPLE)) {
5989 reginsert(pRExC_state, PLUS, ret, depth+1);
5993 else if (op == '+') {
5997 else if (op == '?') {
6002 if (!SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6004 "%.*s matches null string many times",
6005 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6009 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6010 nextchar(pRExC_state);
6011 reginsert(pRExC_state, MINMOD, ret, depth+1);
6012 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6014 #ifndef REG_ALLOW_MINMOD_SUSPEND
6017 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6019 nextchar(pRExC_state);
6020 ender = reg_node(pRExC_state, SUCCEED);
6021 REGTAIL(pRExC_state, ret, ender);
6022 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6024 ender = reg_node(pRExC_state, TAIL);
6025 REGTAIL(pRExC_state, ret, ender);
6029 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6031 vFAIL("Nested quantifiers");
6038 /* reg_namedseq(pRExC_state,UVp)
6040 This is expected to be called by a parser routine that has
6041 recognized'\N' and needs to handle the rest. RExC_parse is
6042 expected to point at the first char following the N at the time
6045 If valuep is non-null then it is assumed that we are parsing inside
6046 of a charclass definition and the first codepoint in the resolved
6047 string is returned via *valuep and the routine will return NULL.
6048 In this mode if a multichar string is returned from the charnames
6049 handler a warning will be issued, and only the first char in the
6050 sequence will be examined. If the string returned is zero length
6051 then the value of *valuep is undefined and NON-NULL will
6052 be returned to indicate failure. (This will NOT be a valid pointer
6055 If value is null then it is assumed that we are parsing normal text
6056 and inserts a new EXACT node into the program containing the resolved
6057 string and returns a pointer to the new node. If the string is
6058 zerolength a NOTHING node is emitted.
6060 On success RExC_parse is set to the char following the endbrace.
6061 Parsing failures will generate a fatal errorvia vFAIL(...)
6063 NOTE: We cache all results from the charnames handler locally in
6064 the RExC_charnames hash (created on first use) to prevent a charnames
6065 handler from playing silly-buggers and returning a short string and
6066 then a long string for a given pattern. Since the regexp program
6067 size is calculated during an initial parse this would result
6068 in a buffer overrun so we cache to prevent the charname result from
6069 changing during the course of the parse.
6073 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6075 char * name; /* start of the content of the name */
6076 char * endbrace; /* endbrace following the name */
6079 STRLEN len; /* this has various purposes throughout the code */
6080 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6081 regnode *ret = NULL;
6083 if (*RExC_parse != '{') {
6084 vFAIL("Missing braces on \\N{}");
6086 name = RExC_parse+1;
6087 endbrace = strchr(RExC_parse, '}');
6090 vFAIL("Missing right brace on \\N{}");
6092 RExC_parse = endbrace + 1;
6095 /* RExC_parse points at the beginning brace,
6096 endbrace points at the last */
6097 if ( name[0]=='U' && name[1]=='+' ) {
6098 /* its a "unicode hex" notation {U+89AB} */
6099 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6100 | PERL_SCAN_DISALLOW_PREFIX
6101 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6103 len = (STRLEN)(endbrace - name - 2);
6104 cp = grok_hex(name + 2, &len, &fl, NULL);
6105 if ( len != (STRLEN)(endbrace - name - 2) ) {
6114 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6116 /* fetch the charnames handler for this scope */
6117 HV * const table = GvHV(PL_hintgv);
6119 hv_fetchs(table, "charnames", FALSE) :
6121 SV *cv= cvp ? *cvp : NULL;
6124 /* create an SV with the name as argument */
6125 sv_name = newSVpvn(name, endbrace - name);
6127 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6128 vFAIL2("Constant(\\N{%s}) unknown: "
6129 "(possibly a missing \"use charnames ...\")",
6132 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6133 vFAIL2("Constant(\\N{%s}): "
6134 "$^H{charnames} is not defined",SvPVX(sv_name));
6139 if (!RExC_charnames) {
6140 /* make sure our cache is allocated */
6141 RExC_charnames = newHV();
6142 sv_2mortal((SV*)RExC_charnames);
6144 /* see if we have looked this one up before */
6145 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6147 sv_str = HeVAL(he_str);
6160 count= call_sv(cv, G_SCALAR);
6162 if (count == 1) { /* XXXX is this right? dmq */
6164 SvREFCNT_inc_simple_void(sv_str);
6172 if ( !sv_str || !SvOK(sv_str) ) {
6173 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6174 "did not return a defined value",SvPVX(sv_name));
6176 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6181 char *p = SvPV(sv_str, len);
6184 if ( SvUTF8(sv_str) ) {
6185 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6189 We have to turn on utf8 for high bit chars otherwise
6190 we get failures with
6192 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6193 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6195 This is different from what \x{} would do with the same
6196 codepoint, where the condition is > 0xFF.
6203 /* warn if we havent used the whole string? */
6205 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6207 "Ignoring excess chars from \\N{%s} in character class",
6211 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6213 "Ignoring zero length \\N{%s} in character class",
6218 SvREFCNT_dec(sv_name);
6220 SvREFCNT_dec(sv_str);
6221 return len ? NULL : (regnode *)&len;
6222 } else if(SvCUR(sv_str)) {
6228 char * parse_start = name-3; /* needed for the offsets */
6230 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6232 ret = reg_node(pRExC_state,
6233 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6236 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6237 sv_utf8_upgrade(sv_str);
6238 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6242 p = SvPV(sv_str, len);
6244 /* len is the length written, charlen is the size the char read */
6245 for ( len = 0; p < pend; p += charlen ) {
6247 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6249 STRLEN foldlen,numlen;
6250 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6251 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6252 /* Emit all the Unicode characters. */
6254 for (foldbuf = tmpbuf;
6258 uvc = utf8_to_uvchr(foldbuf, &numlen);
6260 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6263 /* In EBCDIC the numlen
6264 * and unilen can differ. */
6266 if (numlen >= foldlen)
6270 break; /* "Can't happen." */
6273 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6285 RExC_size += STR_SZ(len);
6288 RExC_emit += STR_SZ(len);
6290 Set_Node_Cur_Length(ret); /* MJD */
6292 nextchar(pRExC_state);
6294 ret = reg_node(pRExC_state,NOTHING);
6297 SvREFCNT_dec(sv_str);
6300 SvREFCNT_dec(sv_name);
6310 * It returns the code point in utf8 for the value in *encp.
6311 * value: a code value in the source encoding
6312 * encp: a pointer to an Encode object
6314 * If the result from Encode is not a single character,
6315 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6318 S_reg_recode(pTHX_ const char value, SV **encp)
6321 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6322 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6324 const STRLEN newlen = SvCUR(sv);
6325 UV uv = UNICODE_REPLACEMENT;
6329 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6332 if (!newlen || numlen != newlen) {
6333 uv = UNICODE_REPLACEMENT;
6342 - regatom - the lowest level
6344 Try to identify anything special at the start of the pattern. If there
6345 is, then handle it as required. This may involve generating a single regop,
6346 such as for an assertion; or it may involve recursing, such as to
6347 handle a () structure.
6349 If the string doesn't start with something special then we gobble up
6350 as much literal text as we can.
6352 Once we have been able to handle whatever type of thing started the
6353 sequence, we return.
6355 Note: we have to be careful with escapes, as they can be both literal
6356 and special, and in the case of \10 and friends can either, depending
6357 on context. Specifically there are two seperate switches for handling
6358 escape sequences, with the one for handling literal escapes requiring
6359 a dummy entry for all of the special escapes that are actually handled
6364 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6367 register regnode *ret = NULL;
6369 char *parse_start = RExC_parse;
6370 GET_RE_DEBUG_FLAGS_DECL;
6371 DEBUG_PARSE("atom");
6372 *flagp = WORST; /* Tentatively. */
6376 switch (*RExC_parse) {
6378 RExC_seen_zerolen++;
6379 nextchar(pRExC_state);
6380 if (RExC_flags & RXf_PMf_MULTILINE)
6381 ret = reg_node(pRExC_state, MBOL);
6382 else if (RExC_flags & RXf_PMf_SINGLELINE)
6383 ret = reg_node(pRExC_state, SBOL);
6385 ret = reg_node(pRExC_state, BOL);
6386 Set_Node_Length(ret, 1); /* MJD */
6389 nextchar(pRExC_state);
6391 RExC_seen_zerolen++;
6392 if (RExC_flags & RXf_PMf_MULTILINE)
6393 ret = reg_node(pRExC_state, MEOL);
6394 else if (RExC_flags & RXf_PMf_SINGLELINE)
6395 ret = reg_node(pRExC_state, SEOL);
6397 ret = reg_node(pRExC_state, EOL);
6398 Set_Node_Length(ret, 1); /* MJD */
6401 nextchar(pRExC_state);
6402 if (RExC_flags & RXf_PMf_SINGLELINE)
6403 ret = reg_node(pRExC_state, SANY);
6405 ret = reg_node(pRExC_state, REG_ANY);
6406 *flagp |= HASWIDTH|SIMPLE;
6408 Set_Node_Length(ret, 1); /* MJD */
6412 char * const oregcomp_parse = ++RExC_parse;
6413 ret = regclass(pRExC_state,depth+1);
6414 if (*RExC_parse != ']') {
6415 RExC_parse = oregcomp_parse;
6416 vFAIL("Unmatched [");
6418 nextchar(pRExC_state);
6419 *flagp |= HASWIDTH|SIMPLE;
6420 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6424 nextchar(pRExC_state);
6425 ret = reg(pRExC_state, 1, &flags,depth+1);
6427 if (flags & TRYAGAIN) {
6428 if (RExC_parse == RExC_end) {
6429 /* Make parent create an empty node if needed. */
6437 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
6441 if (flags & TRYAGAIN) {
6445 vFAIL("Internal urp");
6446 /* Supposed to be caught earlier. */
6449 if (!regcurly(RExC_parse)) {
6458 vFAIL("Quantifier follows nothing");
6463 This switch handles escape sequences that resolve to some kind
6464 of special regop and not to literal text. Escape sequnces that
6465 resolve to literal text are handled below in the switch marked
6468 Every entry in this switch *must* have a corresponding entry
6469 in the literal escape switch. However, the opposite is not
6470 required, as the default for this switch is to jump to the
6471 literal text handling code.
6473 switch (*++RExC_parse) {
6474 /* Special Escapes */
6476 RExC_seen_zerolen++;
6477 ret = reg_node(pRExC_state, SBOL);
6479 goto finish_meta_pat;
6481 ret = reg_node(pRExC_state, GPOS);
6482 RExC_seen |= REG_SEEN_GPOS;
6484 goto finish_meta_pat;
6486 RExC_seen_zerolen++;
6487 ret = reg_node(pRExC_state, KEEPS);
6489 goto finish_meta_pat;
6491 ret = reg_node(pRExC_state, SEOL);
6493 RExC_seen_zerolen++; /* Do not optimize RE away */
6494 goto finish_meta_pat;
6496 ret = reg_node(pRExC_state, EOS);
6498 RExC_seen_zerolen++; /* Do not optimize RE away */
6499 goto finish_meta_pat;
6501 ret = reg_node(pRExC_state, CANY);
6502 RExC_seen |= REG_SEEN_CANY;
6503 *flagp |= HASWIDTH|SIMPLE;
6504 goto finish_meta_pat;
6506 ret = reg_node(pRExC_state, CLUMP);
6508 goto finish_meta_pat;
6510 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6511 *flagp |= HASWIDTH|SIMPLE;
6512 goto finish_meta_pat;
6514 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6515 *flagp |= HASWIDTH|SIMPLE;
6516 goto finish_meta_pat;
6518 RExC_seen_zerolen++;
6519 RExC_seen |= REG_SEEN_LOOKBEHIND;
6520 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6522 goto finish_meta_pat;
6524 RExC_seen_zerolen++;
6525 RExC_seen |= REG_SEEN_LOOKBEHIND;
6526 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6528 goto finish_meta_pat;
6530 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6531 *flagp |= HASWIDTH|SIMPLE;
6532 goto finish_meta_pat;
6534 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6535 *flagp |= HASWIDTH|SIMPLE;
6536 goto finish_meta_pat;
6538 ret = reg_node(pRExC_state, DIGIT);
6539 *flagp |= HASWIDTH|SIMPLE;
6540 goto finish_meta_pat;
6542 ret = reg_node(pRExC_state, NDIGIT);
6543 *flagp |= HASWIDTH|SIMPLE;
6544 goto finish_meta_pat;
6546 ret = reganode(pRExC_state, PRUNE, 0);
6549 goto finish_meta_pat;
6551 ret = reganode(pRExC_state, SKIP, 0);
6555 nextchar(pRExC_state);
6556 Set_Node_Length(ret, 2); /* MJD */
6561 char* const oldregxend = RExC_end;
6563 char* parse_start = RExC_parse - 2;
6566 if (RExC_parse[1] == '{') {
6567 /* a lovely hack--pretend we saw [\pX] instead */
6568 RExC_end = strchr(RExC_parse, '}');
6570 const U8 c = (U8)*RExC_parse;
6572 RExC_end = oldregxend;
6573 vFAIL2("Missing right brace on \\%c{}", c);
6578 RExC_end = RExC_parse + 2;
6579 if (RExC_end > oldregxend)
6580 RExC_end = oldregxend;
6584 ret = regclass(pRExC_state,depth+1);
6586 RExC_end = oldregxend;
6589 Set_Node_Offset(ret, parse_start + 2);
6590 Set_Node_Cur_Length(ret);
6591 nextchar(pRExC_state);
6592 *flagp |= HASWIDTH|SIMPLE;
6596 /* Handle \N{NAME} here and not below because it can be
6597 multicharacter. join_exact() will join them up later on.
6598 Also this makes sure that things like /\N{BLAH}+/ and
6599 \N{BLAH} being multi char Just Happen. dmq*/
6601 ret= reg_namedseq(pRExC_state, NULL);
6603 case 'k': /* Handle \k<NAME> and \k'NAME' */
6606 char ch= RExC_parse[1];
6607 if (ch != '<' && ch != '\'' && ch != '{') {
6609 vFAIL2("Sequence %.2s... not terminated",parse_start);
6611 /* this pretty much dupes the code for (?P=...) in reg(), if
6612 you change this make sure you change that */
6613 char* name_start = (RExC_parse += 2);
6615 SV *sv_dat = reg_scan_name(pRExC_state,
6616 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6617 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
6618 if (RExC_parse == name_start || *RExC_parse != ch)
6619 vFAIL2("Sequence %.3s... not terminated",parse_start);
6622 num = add_data( pRExC_state, 1, "S" );
6623 RExC_rxi->data->data[num]=(void*)sv_dat;
6624 SvREFCNT_inc(sv_dat);
6628 ret = reganode(pRExC_state,
6629 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6633 /* override incorrect value set in reganode MJD */
6634 Set_Node_Offset(ret, parse_start+1);
6635 Set_Node_Cur_Length(ret); /* MJD */
6636 nextchar(pRExC_state);
6642 case '1': case '2': case '3': case '4':
6643 case '5': case '6': case '7': case '8': case '9':
6646 bool isg = *RExC_parse == 'g';
6651 if (*RExC_parse == '{') {
6655 if (*RExC_parse == '-') {
6659 if (hasbrace && !isDIGIT(*RExC_parse)) {
6660 if (isrel) RExC_parse--;
6662 goto parse_named_seq;
6664 num = atoi(RExC_parse);
6666 num = RExC_npar - num;
6668 vFAIL("Reference to nonexistent or unclosed group");
6670 if (!isg && num > 9 && num >= RExC_npar)
6673 char * const parse_start = RExC_parse - 1; /* MJD */
6674 while (isDIGIT(*RExC_parse))
6676 if (parse_start == RExC_parse - 1)
6677 vFAIL("Unterminated \\g... pattern");
6679 if (*RExC_parse != '}')
6680 vFAIL("Unterminated \\g{...} pattern");
6684 if (num > (I32)RExC_rx->nparens)
6685 vFAIL("Reference to nonexistent group");
6688 ret = reganode(pRExC_state,
6689 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6693 /* override incorrect value set in reganode MJD */
6694 Set_Node_Offset(ret, parse_start+1);
6695 Set_Node_Cur_Length(ret); /* MJD */
6697 nextchar(pRExC_state);
6702 if (RExC_parse >= RExC_end)
6703 FAIL("Trailing \\");
6706 /* Do not generate "unrecognized" warnings here, we fall
6707 back into the quick-grab loop below */
6714 if (RExC_flags & RXf_PMf_EXTENDED) {
6715 if ( reg_skipcomment( pRExC_state ) )
6721 register STRLEN len;
6726 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6728 parse_start = RExC_parse - 1;
6734 ret = reg_node(pRExC_state,
6735 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6737 for (len = 0, p = RExC_parse - 1;
6738 len < 127 && p < RExC_end;
6741 char * const oldp = p;
6743 if (RExC_flags & RXf_PMf_EXTENDED)
6744 p = regwhite( pRExC_state, p );
6755 /* Literal Escapes Switch
6757 This switch is meant to handle escape sequences that
6758 resolve to a literal character.
6760 Every escape sequence that represents something
6761 else, like an assertion or a char class, is handled
6762 in the switch marked 'Special Escapes' above in this
6763 routine, but also has an entry here as anything that
6764 isn't explicitly mentioned here will be treated as
6765 an unescaped equivalent literal.
6769 /* These are all the special escapes. */
6770 case 'A': /* Start assertion */
6771 case 'b': case 'B': /* Word-boundary assertion*/
6772 case 'C': /* Single char !DANGEROUS! */
6773 case 'd': case 'D': /* digit class */
6774 case 'g': case 'G': /* generic-backref, pos assertion */
6775 case 'k': case 'K': /* named backref, keep marker */
6776 case 'N': /* named char sequence */
6777 case 'p': case 'P': /* unicode property */
6778 case 's': case 'S': /* space class */
6779 case 'v': case 'V': /* (*PRUNE) and (*SKIP) */
6780 case 'w': case 'W': /* word class */
6781 case 'X': /* eXtended Unicode "combining character sequence" */
6782 case 'z': case 'Z': /* End of line/string assertion */
6786 /* Anything after here is an escape that resolves to a
6787 literal. (Except digits, which may or may not)
6806 ender = ASCII_TO_NATIVE('\033');
6810 ender = ASCII_TO_NATIVE('\007');
6815 char* const e = strchr(p, '}');
6819 vFAIL("Missing right brace on \\x{}");
6822 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6823 | PERL_SCAN_DISALLOW_PREFIX;
6824 STRLEN numlen = e - p - 1;
6825 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6832 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6834 ender = grok_hex(p, &numlen, &flags, NULL);
6837 if (PL_encoding && ender < 0x100)
6838 goto recode_encoding;
6842 ender = UCHARAT(p++);
6843 ender = toCTRL(ender);
6845 case '0': case '1': case '2': case '3':case '4':
6846 case '5': case '6': case '7': case '8':case '9':
6848 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6851 ender = grok_oct(p, &numlen, &flags, NULL);
6858 if (PL_encoding && ender < 0x100)
6859 goto recode_encoding;
6863 SV* enc = PL_encoding;
6864 ender = reg_recode((const char)(U8)ender, &enc);
6865 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6866 vWARN(p, "Invalid escape in the specified encoding");
6872 FAIL("Trailing \\");
6875 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6876 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6877 goto normal_default;
6882 if (UTF8_IS_START(*p) && UTF) {
6884 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6885 &numlen, UTF8_ALLOW_DEFAULT);
6892 if ( RExC_flags & RXf_PMf_EXTENDED)
6893 p = regwhite( pRExC_state, p );
6895 /* Prime the casefolded buffer. */
6896 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6898 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
6903 /* Emit all the Unicode characters. */
6905 for (foldbuf = tmpbuf;
6907 foldlen -= numlen) {
6908 ender = utf8_to_uvchr(foldbuf, &numlen);
6910 const STRLEN unilen = reguni(pRExC_state, ender, s);
6913 /* In EBCDIC the numlen
6914 * and unilen can differ. */
6916 if (numlen >= foldlen)
6920 break; /* "Can't happen." */
6924 const STRLEN unilen = reguni(pRExC_state, ender, s);
6933 REGC((char)ender, s++);
6939 /* Emit all the Unicode characters. */
6941 for (foldbuf = tmpbuf;
6943 foldlen -= numlen) {
6944 ender = utf8_to_uvchr(foldbuf, &numlen);
6946 const STRLEN unilen = reguni(pRExC_state, ender, s);
6949 /* In EBCDIC the numlen
6950 * and unilen can differ. */
6952 if (numlen >= foldlen)
6960 const STRLEN unilen = reguni(pRExC_state, ender, s);
6969 REGC((char)ender, s++);
6973 Set_Node_Cur_Length(ret); /* MJD */
6974 nextchar(pRExC_state);
6976 /* len is STRLEN which is unsigned, need to copy to signed */
6979 vFAIL("Internal disaster");
6983 if (len == 1 && UNI_IS_INVARIANT(ender))
6987 RExC_size += STR_SZ(len);
6990 RExC_emit += STR_SZ(len);
7000 S_regwhite( RExC_state_t *pRExC_state, char *p )
7002 const char *e = RExC_end;
7006 else if (*p == '#') {
7015 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7023 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7024 Character classes ([:foo:]) can also be negated ([:^foo:]).
7025 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7026 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7027 but trigger failures because they are currently unimplemented. */
7029 #define POSIXCC_DONE(c) ((c) == ':')
7030 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7031 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7034 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7037 I32 namedclass = OOB_NAMEDCLASS;
7039 if (value == '[' && RExC_parse + 1 < RExC_end &&
7040 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7041 POSIXCC(UCHARAT(RExC_parse))) {
7042 const char c = UCHARAT(RExC_parse);
7043 char* const s = RExC_parse++;
7045 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7047 if (RExC_parse == RExC_end)
7048 /* Grandfather lone [:, [=, [. */
7051 const char* const t = RExC_parse++; /* skip over the c */
7054 if (UCHARAT(RExC_parse) == ']') {
7055 const char *posixcc = s + 1;
7056 RExC_parse++; /* skip over the ending ] */
7059 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7060 const I32 skip = t - posixcc;
7062 /* Initially switch on the length of the name. */
7065 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7066 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7069 /* Names all of length 5. */
7070 /* alnum alpha ascii blank cntrl digit graph lower
7071 print punct space upper */
7072 /* Offset 4 gives the best switch position. */
7073 switch (posixcc[4]) {
7075 if (memEQ(posixcc, "alph", 4)) /* alpha */
7076 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7079 if (memEQ(posixcc, "spac", 4)) /* space */
7080 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7083 if (memEQ(posixcc, "grap", 4)) /* graph */
7084 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7087 if (memEQ(posixcc, "asci", 4)) /* ascii */
7088 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7091 if (memEQ(posixcc, "blan", 4)) /* blank */
7092 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7095 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7096 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7099 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7100 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7103 if (memEQ(posixcc, "lowe", 4)) /* lower */
7104 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7105 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7106 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7109 if (memEQ(posixcc, "digi", 4)) /* digit */
7110 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7111 else if (memEQ(posixcc, "prin", 4)) /* print */
7112 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7113 else if (memEQ(posixcc, "punc", 4)) /* punct */
7114 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7119 if (memEQ(posixcc, "xdigit", 6))
7120 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7124 if (namedclass == OOB_NAMEDCLASS)
7125 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7127 assert (posixcc[skip] == ':');
7128 assert (posixcc[skip+1] == ']');
7129 } else if (!SIZE_ONLY) {
7130 /* [[=foo=]] and [[.foo.]] are still future. */
7132 /* adjust RExC_parse so the warning shows after
7134 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7136 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7139 /* Maternal grandfather:
7140 * "[:" ending in ":" but not in ":]" */
7150 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7153 if (POSIXCC(UCHARAT(RExC_parse))) {
7154 const char *s = RExC_parse;
7155 const char c = *s++;
7159 if (*s && c == *s && s[1] == ']') {
7160 if (ckWARN(WARN_REGEXP))
7162 "POSIX syntax [%c %c] belongs inside character classes",
7165 /* [[=foo=]] and [[.foo.]] are still future. */
7166 if (POSIXCC_NOTYET(c)) {
7167 /* adjust RExC_parse so the error shows after
7169 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7171 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7178 #define _C_C_T_(NAME,TEST,WORD) \
7181 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7183 for (value = 0; value < 256; value++) \
7185 ANYOF_BITMAP_SET(ret, value); \
7190 case ANYOF_N##NAME: \
7192 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7194 for (value = 0; value < 256; value++) \
7196 ANYOF_BITMAP_SET(ret, value); \
7204 parse a class specification and produce either an ANYOF node that
7205 matches the pattern or if the pattern matches a single char only and
7206 that char is < 256 and we are case insensitive then we produce an
7211 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7214 register UV value = 0;
7215 register UV nextvalue;
7216 register IV prevvalue = OOB_UNICODE;
7217 register IV range = 0;
7218 register regnode *ret;
7221 char *rangebegin = NULL;
7222 bool need_class = 0;
7225 bool optimize_invert = TRUE;
7226 AV* unicode_alternate = NULL;
7228 UV literal_endpoint = 0;
7230 UV stored = 0; /* number of chars stored in the class */
7232 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7233 case we need to change the emitted regop to an EXACT. */
7234 const char * orig_parse = RExC_parse;
7235 GET_RE_DEBUG_FLAGS_DECL;
7237 PERL_UNUSED_ARG(depth);
7240 DEBUG_PARSE("clas");
7242 /* Assume we are going to generate an ANYOF node. */
7243 ret = reganode(pRExC_state, ANYOF, 0);
7246 ANYOF_FLAGS(ret) = 0;
7248 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7252 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7256 RExC_size += ANYOF_SKIP;
7257 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7260 RExC_emit += ANYOF_SKIP;
7262 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7264 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7265 ANYOF_BITMAP_ZERO(ret);
7266 listsv = newSVpvs("# comment\n");
7269 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7271 if (!SIZE_ONLY && POSIXCC(nextvalue))
7272 checkposixcc(pRExC_state);
7274 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7275 if (UCHARAT(RExC_parse) == ']')
7279 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7283 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7286 rangebegin = RExC_parse;
7288 value = utf8n_to_uvchr((U8*)RExC_parse,
7289 RExC_end - RExC_parse,
7290 &numlen, UTF8_ALLOW_DEFAULT);
7291 RExC_parse += numlen;
7294 value = UCHARAT(RExC_parse++);
7296 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7297 if (value == '[' && POSIXCC(nextvalue))
7298 namedclass = regpposixcc(pRExC_state, value);
7299 else if (value == '\\') {
7301 value = utf8n_to_uvchr((U8*)RExC_parse,
7302 RExC_end - RExC_parse,
7303 &numlen, UTF8_ALLOW_DEFAULT);
7304 RExC_parse += numlen;
7307 value = UCHARAT(RExC_parse++);
7308 /* Some compilers cannot handle switching on 64-bit integer
7309 * values, therefore value cannot be an UV. Yes, this will
7310 * be a problem later if we want switch on Unicode.
7311 * A similar issue a little bit later when switching on
7312 * namedclass. --jhi */
7313 switch ((I32)value) {
7314 case 'w': namedclass = ANYOF_ALNUM; break;
7315 case 'W': namedclass = ANYOF_NALNUM; break;
7316 case 's': namedclass = ANYOF_SPACE; break;
7317 case 'S': namedclass = ANYOF_NSPACE; break;
7318 case 'd': namedclass = ANYOF_DIGIT; break;
7319 case 'D': namedclass = ANYOF_NDIGIT; break;
7320 case 'N': /* Handle \N{NAME} in class */
7322 /* We only pay attention to the first char of
7323 multichar strings being returned. I kinda wonder
7324 if this makes sense as it does change the behaviour
7325 from earlier versions, OTOH that behaviour was broken
7327 UV v; /* value is register so we cant & it /grrr */
7328 if (reg_namedseq(pRExC_state, &v)) {
7338 if (RExC_parse >= RExC_end)
7339 vFAIL2("Empty \\%c{}", (U8)value);
7340 if (*RExC_parse == '{') {
7341 const U8 c = (U8)value;
7342 e = strchr(RExC_parse++, '}');
7344 vFAIL2("Missing right brace on \\%c{}", c);
7345 while (isSPACE(UCHARAT(RExC_parse)))
7347 if (e == RExC_parse)
7348 vFAIL2("Empty \\%c{}", c);
7350 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7358 if (UCHARAT(RExC_parse) == '^') {
7361 value = value == 'p' ? 'P' : 'p'; /* toggle */
7362 while (isSPACE(UCHARAT(RExC_parse))) {
7367 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7368 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7371 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7372 namedclass = ANYOF_MAX; /* no official name, but it's named */
7375 case 'n': value = '\n'; break;
7376 case 'r': value = '\r'; break;
7377 case 't': value = '\t'; break;
7378 case 'f': value = '\f'; break;
7379 case 'b': value = '\b'; break;
7380 case 'e': value = ASCII_TO_NATIVE('\033');break;
7381 case 'a': value = ASCII_TO_NATIVE('\007');break;
7383 if (*RExC_parse == '{') {
7384 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7385 | PERL_SCAN_DISALLOW_PREFIX;
7386 char * const e = strchr(RExC_parse++, '}');
7388 vFAIL("Missing right brace on \\x{}");
7390 numlen = e - RExC_parse;
7391 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7395 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7397 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7398 RExC_parse += numlen;
7400 if (PL_encoding && value < 0x100)
7401 goto recode_encoding;
7404 value = UCHARAT(RExC_parse++);
7405 value = toCTRL(value);
7407 case '0': case '1': case '2': case '3': case '4':
7408 case '5': case '6': case '7': case '8': case '9':
7412 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7413 RExC_parse += numlen;
7414 if (PL_encoding && value < 0x100)
7415 goto recode_encoding;
7420 SV* enc = PL_encoding;
7421 value = reg_recode((const char)(U8)value, &enc);
7422 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7424 "Invalid escape in the specified encoding");
7428 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7430 "Unrecognized escape \\%c in character class passed through",
7434 } /* end of \blah */
7440 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7442 if (!SIZE_ONLY && !need_class)
7443 ANYOF_CLASS_ZERO(ret);
7447 /* a bad range like a-\d, a-[:digit:] ? */
7450 if (ckWARN(WARN_REGEXP)) {
7452 RExC_parse >= rangebegin ?
7453 RExC_parse - rangebegin : 0;
7455 "False [] range \"%*.*s\"",
7458 if (prevvalue < 256) {
7459 ANYOF_BITMAP_SET(ret, prevvalue);
7460 ANYOF_BITMAP_SET(ret, '-');
7463 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7464 Perl_sv_catpvf(aTHX_ listsv,
7465 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7469 range = 0; /* this was not a true range */
7475 const char *what = NULL;
7478 if (namedclass > OOB_NAMEDCLASS)
7479 optimize_invert = FALSE;
7480 /* Possible truncation here but in some 64-bit environments
7481 * the compiler gets heartburn about switch on 64-bit values.
7482 * A similar issue a little earlier when switching on value.
7484 switch ((I32)namedclass) {
7485 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7486 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7487 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7488 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7489 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7490 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7491 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7492 case _C_C_T_(PRINT, isPRINT(value), "Print");
7493 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7494 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7495 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7496 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7497 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7500 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7503 for (value = 0; value < 128; value++)
7504 ANYOF_BITMAP_SET(ret, value);
7506 for (value = 0; value < 256; value++) {
7508 ANYOF_BITMAP_SET(ret, value);
7517 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7520 for (value = 128; value < 256; value++)
7521 ANYOF_BITMAP_SET(ret, value);
7523 for (value = 0; value < 256; value++) {
7524 if (!isASCII(value))
7525 ANYOF_BITMAP_SET(ret, value);
7534 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7536 /* consecutive digits assumed */
7537 for (value = '0'; value <= '9'; value++)
7538 ANYOF_BITMAP_SET(ret, value);
7545 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7547 /* consecutive digits assumed */
7548 for (value = 0; value < '0'; value++)
7549 ANYOF_BITMAP_SET(ret, value);
7550 for (value = '9' + 1; value < 256; value++)
7551 ANYOF_BITMAP_SET(ret, value);
7557 /* this is to handle \p and \P */
7560 vFAIL("Invalid [::] class");
7564 /* Strings such as "+utf8::isWord\n" */
7565 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7568 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7571 } /* end of namedclass \blah */
7574 if (prevvalue > (IV)value) /* b-a */ {
7575 const int w = RExC_parse - rangebegin;
7576 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7577 range = 0; /* not a valid range */
7581 prevvalue = value; /* save the beginning of the range */
7582 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7583 RExC_parse[1] != ']') {
7586 /* a bad range like \w-, [:word:]- ? */
7587 if (namedclass > OOB_NAMEDCLASS) {
7588 if (ckWARN(WARN_REGEXP)) {
7590 RExC_parse >= rangebegin ?
7591 RExC_parse - rangebegin : 0;
7593 "False [] range \"%*.*s\"",
7597 ANYOF_BITMAP_SET(ret, '-');
7599 range = 1; /* yeah, it's a range! */
7600 continue; /* but do it the next time */
7604 /* now is the next time */
7605 /*stored += (value - prevvalue + 1);*/
7607 if (prevvalue < 256) {
7608 const IV ceilvalue = value < 256 ? value : 255;
7611 /* In EBCDIC [\x89-\x91] should include
7612 * the \x8e but [i-j] should not. */
7613 if (literal_endpoint == 2 &&
7614 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7615 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7617 if (isLOWER(prevvalue)) {
7618 for (i = prevvalue; i <= ceilvalue; i++)
7620 ANYOF_BITMAP_SET(ret, i);
7622 for (i = prevvalue; i <= ceilvalue; i++)
7624 ANYOF_BITMAP_SET(ret, i);
7629 for (i = prevvalue; i <= ceilvalue; i++) {
7630 if (!ANYOF_BITMAP_TEST(ret,i)) {
7632 ANYOF_BITMAP_SET(ret, i);
7636 if (value > 255 || UTF) {
7637 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7638 const UV natvalue = NATIVE_TO_UNI(value);
7639 stored+=2; /* can't optimize this class */
7640 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7641 if (prevnatvalue < natvalue) { /* what about > ? */
7642 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7643 prevnatvalue, natvalue);
7645 else if (prevnatvalue == natvalue) {
7646 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7648 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7650 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7652 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7653 if (RExC_precomp[0] == ':' &&
7654 RExC_precomp[1] == '[' &&
7655 (f == 0xDF || f == 0x92)) {
7656 f = NATIVE_TO_UNI(f);
7659 /* If folding and foldable and a single
7660 * character, insert also the folded version
7661 * to the charclass. */
7663 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7664 if ((RExC_precomp[0] == ':' &&
7665 RExC_precomp[1] == '[' &&
7667 (value == 0xFB05 || value == 0xFB06))) ?
7668 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7669 foldlen == (STRLEN)UNISKIP(f) )
7671 if (foldlen == (STRLEN)UNISKIP(f))
7673 Perl_sv_catpvf(aTHX_ listsv,
7676 /* Any multicharacter foldings
7677 * require the following transform:
7678 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7679 * where E folds into "pq" and F folds
7680 * into "rst", all other characters
7681 * fold to single characters. We save
7682 * away these multicharacter foldings,
7683 * to be later saved as part of the
7684 * additional "s" data. */
7687 if (!unicode_alternate)
7688 unicode_alternate = newAV();
7689 sv = newSVpvn((char*)foldbuf, foldlen);
7691 av_push(unicode_alternate, sv);
7695 /* If folding and the value is one of the Greek
7696 * sigmas insert a few more sigmas to make the
7697 * folding rules of the sigmas to work right.
7698 * Note that not all the possible combinations
7699 * are handled here: some of them are handled
7700 * by the standard folding rules, and some of
7701 * them (literal or EXACTF cases) are handled
7702 * during runtime in regexec.c:S_find_byclass(). */
7703 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7704 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7705 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7706 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7707 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7709 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7710 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7711 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7716 literal_endpoint = 0;
7720 range = 0; /* this range (if it was one) is done now */
7724 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7726 RExC_size += ANYOF_CLASS_ADD_SKIP;
7728 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7734 /****** !SIZE_ONLY AFTER HERE *********/
7736 if( stored == 1 && value < 256
7737 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7739 /* optimize single char class to an EXACT node
7740 but *only* when its not a UTF/high char */
7741 const char * cur_parse= RExC_parse;
7742 RExC_emit = (regnode *)orig_emit;
7743 RExC_parse = (char *)orig_parse;
7744 ret = reg_node(pRExC_state,
7745 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7746 RExC_parse = (char *)cur_parse;
7747 *STRING(ret)= (char)value;
7749 RExC_emit += STR_SZ(1);
7752 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7753 if ( /* If the only flag is folding (plus possibly inversion). */
7754 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7756 for (value = 0; value < 256; ++value) {
7757 if (ANYOF_BITMAP_TEST(ret, value)) {
7758 UV fold = PL_fold[value];
7761 ANYOF_BITMAP_SET(ret, fold);
7764 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7767 /* optimize inverted simple patterns (e.g. [^a-z]) */
7768 if (optimize_invert &&
7769 /* If the only flag is inversion. */
7770 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7771 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7772 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7773 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7776 AV * const av = newAV();
7778 /* The 0th element stores the character class description
7779 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7780 * to initialize the appropriate swash (which gets stored in
7781 * the 1st element), and also useful for dumping the regnode.
7782 * The 2nd element stores the multicharacter foldings,
7783 * used later (regexec.c:S_reginclass()). */
7784 av_store(av, 0, listsv);
7785 av_store(av, 1, NULL);
7786 av_store(av, 2, (SV*)unicode_alternate);
7787 rv = newRV_noinc((SV*)av);
7788 n = add_data(pRExC_state, 1, "s");
7789 RExC_rxi->data->data[n] = (void*)rv;
7797 /* reg_skipcomment()
7799 Absorbs an /x style # comments from the input stream.
7800 Returns true if there is more text remaining in the stream.
7801 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
7802 terminates the pattern without including a newline.
7804 Note its the callers responsibility to ensure that we are
7810 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
7813 while (RExC_parse < RExC_end)
7814 if (*RExC_parse++ == '\n') {
7819 /* we ran off the end of the pattern without ending
7820 the comment, so we have to add an \n when wrapping */
7821 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7829 Advance that parse position, and optionall absorbs
7830 "whitespace" from the inputstream.
7832 Without /x "whitespace" means (?#...) style comments only,
7833 with /x this means (?#...) and # comments and whitespace proper.
7835 Returns the RExC_parse point from BEFORE the scan occurs.
7837 This is the /x friendly way of saying RExC_parse++.
7841 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7843 char* const retval = RExC_parse++;
7846 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7847 RExC_parse[2] == '#') {
7848 while (*RExC_parse != ')') {
7849 if (RExC_parse == RExC_end)
7850 FAIL("Sequence (?#... not terminated");
7856 if (RExC_flags & RXf_PMf_EXTENDED) {
7857 if (isSPACE(*RExC_parse)) {
7861 else if (*RExC_parse == '#') {
7862 if ( reg_skipcomment( pRExC_state ) )
7871 - reg_node - emit a node
7873 STATIC regnode * /* Location. */
7874 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7877 register regnode *ptr;
7878 regnode * const ret = RExC_emit;
7879 GET_RE_DEBUG_FLAGS_DECL;
7882 SIZE_ALIGN(RExC_size);
7887 if (OP(RExC_emit) == 255)
7888 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %s: %d ",
7889 reg_name[op], OP(RExC_emit));
7891 NODE_ALIGN_FILL(ret);
7893 FILL_ADVANCE_NODE(ptr, op);
7894 #ifdef RE_TRACK_PATTERN_OFFSETS
7895 if (RExC_offsets) { /* MJD */
7896 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7897 "reg_node", __LINE__,
7899 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7900 ? "Overwriting end of array!\n" : "OK",
7901 (UV)(RExC_emit - RExC_emit_start),
7902 (UV)(RExC_parse - RExC_start),
7903 (UV)RExC_offsets[0]));
7904 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7912 - reganode - emit a node with an argument
7914 STATIC regnode * /* Location. */
7915 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7918 register regnode *ptr;
7919 regnode * const ret = RExC_emit;
7920 GET_RE_DEBUG_FLAGS_DECL;
7923 SIZE_ALIGN(RExC_size);
7928 assert(2==regarglen[op]+1);
7930 Anything larger than this has to allocate the extra amount.
7931 If we changed this to be:
7933 RExC_size += (1 + regarglen[op]);
7935 then it wouldn't matter. Its not clear what side effect
7936 might come from that so its not done so far.
7942 if (OP(RExC_emit) == 255)
7943 Perl_croak(aTHX_ "panic: reganode overwriting end of allocated program space");
7945 NODE_ALIGN_FILL(ret);
7947 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7948 #ifdef RE_TRACK_PATTERN_OFFSETS
7949 if (RExC_offsets) { /* MJD */
7950 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7954 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7955 "Overwriting end of array!\n" : "OK",
7956 (UV)(RExC_emit - RExC_emit_start),
7957 (UV)(RExC_parse - RExC_start),
7958 (UV)RExC_offsets[0]));
7959 Set_Cur_Node_Offset;
7967 - reguni - emit (if appropriate) a Unicode character
7970 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
7973 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
7977 - reginsert - insert an operator in front of already-emitted operand
7979 * Means relocating the operand.
7982 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
7985 register regnode *src;
7986 register regnode *dst;
7987 register regnode *place;
7988 const int offset = regarglen[(U8)op];
7989 const int size = NODE_STEP_REGNODE + offset;
7990 GET_RE_DEBUG_FLAGS_DECL;
7991 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
7992 DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
8001 if (RExC_open_parens) {
8003 DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);
8004 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8005 if ( RExC_open_parens[paren] >= opnd ) {
8006 DEBUG_PARSE_FMT("open"," - %d",size);
8007 RExC_open_parens[paren] += size;
8009 DEBUG_PARSE_FMT("open"," - %s","ok");
8011 if ( RExC_close_parens[paren] >= opnd ) {
8012 DEBUG_PARSE_FMT("close"," - %d",size);
8013 RExC_close_parens[paren] += size;
8015 DEBUG_PARSE_FMT("close"," - %s","ok");
8020 while (src > opnd) {
8021 StructCopy(--src, --dst, regnode);
8022 #ifdef RE_TRACK_PATTERN_OFFSETS
8023 if (RExC_offsets) { /* MJD 20010112 */
8024 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8028 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8029 ? "Overwriting end of array!\n" : "OK",
8030 (UV)(src - RExC_emit_start),
8031 (UV)(dst - RExC_emit_start),
8032 (UV)RExC_offsets[0]));
8033 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8034 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8040 place = opnd; /* Op node, where operand used to be. */
8041 #ifdef RE_TRACK_PATTERN_OFFSETS
8042 if (RExC_offsets) { /* MJD */
8043 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8047 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8048 ? "Overwriting end of array!\n" : "OK",
8049 (UV)(place - RExC_emit_start),
8050 (UV)(RExC_parse - RExC_start),
8051 (UV)RExC_offsets[0]));
8052 Set_Node_Offset(place, RExC_parse);
8053 Set_Node_Length(place, 1);
8056 src = NEXTOPER(place);
8057 FILL_ADVANCE_NODE(place, op);
8058 Zero(src, offset, regnode);
8062 - regtail - set the next-pointer at the end of a node chain of p to val.
8063 - SEE ALSO: regtail_study
8065 /* TODO: All three parms should be const */
8067 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8070 register regnode *scan;
8071 GET_RE_DEBUG_FLAGS_DECL;
8073 PERL_UNUSED_ARG(depth);
8079 /* Find last node. */
8082 regnode * const temp = regnext(scan);
8084 SV * const mysv=sv_newmortal();
8085 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8086 regprop(RExC_rx, mysv, scan);
8087 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8088 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8089 (temp == NULL ? "->" : ""),
8090 (temp == NULL ? reg_name[OP(val)] : "")
8098 if (reg_off_by_arg[OP(scan)]) {
8099 ARG_SET(scan, val - scan);
8102 NEXT_OFF(scan) = val - scan;
8108 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8109 - Look for optimizable sequences at the same time.
8110 - currently only looks for EXACT chains.
8112 This is expermental code. The idea is to use this routine to perform
8113 in place optimizations on branches and groups as they are constructed,
8114 with the long term intention of removing optimization from study_chunk so
8115 that it is purely analytical.
8117 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8118 to control which is which.
8121 /* TODO: All four parms should be const */
8124 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8127 register regnode *scan;
8129 #ifdef EXPERIMENTAL_INPLACESCAN
8133 GET_RE_DEBUG_FLAGS_DECL;
8139 /* Find last node. */
8143 regnode * const temp = regnext(scan);
8144 #ifdef EXPERIMENTAL_INPLACESCAN
8145 if (PL_regkind[OP(scan)] == EXACT)
8146 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8154 if( exact == PSEUDO )
8156 else if ( exact != OP(scan) )
8165 SV * const mysv=sv_newmortal();
8166 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8167 regprop(RExC_rx, mysv, scan);
8168 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8169 SvPV_nolen_const(mysv),
8178 SV * const mysv_val=sv_newmortal();
8179 DEBUG_PARSE_MSG("");
8180 regprop(RExC_rx, mysv_val, val);
8181 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8182 SvPV_nolen_const(mysv_val),
8183 (IV)REG_NODE_NUM(val),
8187 if (reg_off_by_arg[OP(scan)]) {
8188 ARG_SET(scan, val - scan);
8191 NEXT_OFF(scan) = val - scan;
8199 - regcurly - a little FSA that accepts {\d+,?\d*}
8202 S_regcurly(register const char *s)
8221 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8224 Perl_regdump(pTHX_ const regexp *r)
8228 SV * const sv = sv_newmortal();
8229 SV *dsv= sv_newmortal();
8232 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8234 /* Header fields of interest. */
8235 if (r->anchored_substr) {
8236 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8237 RE_SV_DUMPLEN(r->anchored_substr), 30);
8238 PerlIO_printf(Perl_debug_log,
8239 "anchored %s%s at %"IVdf" ",
8240 s, RE_SV_TAIL(r->anchored_substr),
8241 (IV)r->anchored_offset);
8242 } else if (r->anchored_utf8) {
8243 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8244 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8245 PerlIO_printf(Perl_debug_log,
8246 "anchored utf8 %s%s at %"IVdf" ",
8247 s, RE_SV_TAIL(r->anchored_utf8),
8248 (IV)r->anchored_offset);
8250 if (r->float_substr) {
8251 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8252 RE_SV_DUMPLEN(r->float_substr), 30);
8253 PerlIO_printf(Perl_debug_log,
8254 "floating %s%s at %"IVdf"..%"UVuf" ",
8255 s, RE_SV_TAIL(r->float_substr),
8256 (IV)r->float_min_offset, (UV)r->float_max_offset);
8257 } else if (r->float_utf8) {
8258 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8259 RE_SV_DUMPLEN(r->float_utf8), 30);
8260 PerlIO_printf(Perl_debug_log,
8261 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8262 s, RE_SV_TAIL(r->float_utf8),
8263 (IV)r->float_min_offset, (UV)r->float_max_offset);
8265 if (r->check_substr || r->check_utf8)
8266 PerlIO_printf(Perl_debug_log,
8268 (r->check_substr == r->float_substr
8269 && r->check_utf8 == r->float_utf8
8270 ? "(checking floating" : "(checking anchored"));
8271 if (r->extflags & RXf_NOSCAN)
8272 PerlIO_printf(Perl_debug_log, " noscan");
8273 if (r->extflags & RXf_CHECK_ALL)
8274 PerlIO_printf(Perl_debug_log, " isall");
8275 if (r->check_substr || r->check_utf8)
8276 PerlIO_printf(Perl_debug_log, ") ");
8278 if (ri->regstclass) {
8279 regprop(r, sv, ri->regstclass);
8280 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8282 if (r->extflags & RXf_ANCH) {
8283 PerlIO_printf(Perl_debug_log, "anchored");
8284 if (r->extflags & RXf_ANCH_BOL)
8285 PerlIO_printf(Perl_debug_log, "(BOL)");
8286 if (r->extflags & RXf_ANCH_MBOL)
8287 PerlIO_printf(Perl_debug_log, "(MBOL)");
8288 if (r->extflags & RXf_ANCH_SBOL)
8289 PerlIO_printf(Perl_debug_log, "(SBOL)");
8290 if (r->extflags & RXf_ANCH_GPOS)
8291 PerlIO_printf(Perl_debug_log, "(GPOS)");
8292 PerlIO_putc(Perl_debug_log, ' ');
8294 if (r->extflags & RXf_GPOS_SEEN)
8295 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8296 if (r->intflags & PREGf_SKIP)
8297 PerlIO_printf(Perl_debug_log, "plus ");
8298 if (r->intflags & PREGf_IMPLICIT)
8299 PerlIO_printf(Perl_debug_log, "implicit ");
8300 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8301 if (r->extflags & RXf_EVAL_SEEN)
8302 PerlIO_printf(Perl_debug_log, "with eval ");
8303 PerlIO_printf(Perl_debug_log, "\n");
8305 PERL_UNUSED_CONTEXT;
8307 #endif /* DEBUGGING */
8311 - regprop - printable representation of opcode
8314 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8319 RXi_GET_DECL(prog,progi);
8320 GET_RE_DEBUG_FLAGS_DECL;
8323 sv_setpvn(sv, "", 0);
8325 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8326 /* It would be nice to FAIL() here, but this may be called from
8327 regexec.c, and it would be hard to supply pRExC_state. */
8328 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8329 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
8331 k = PL_regkind[OP(o)];
8334 SV * const dsv = sv_2mortal(newSVpvs(""));
8335 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8336 * is a crude hack but it may be the best for now since
8337 * we have no flag "this EXACTish node was UTF-8"
8339 const char * const s =
8340 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8341 PL_colors[0], PL_colors[1],
8342 PERL_PV_ESCAPE_UNI_DETECT |
8343 PERL_PV_PRETTY_ELIPSES |
8346 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8347 } else if (k == TRIE) {
8348 /* print the details of the trie in dumpuntil instead, as
8349 * progi->data isn't available here */
8350 const char op = OP(o);
8351 const U32 n = ARG(o);
8352 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8353 (reg_ac_data *)progi->data->data[n] :
8355 const reg_trie_data * const trie
8356 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8358 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
8359 DEBUG_TRIE_COMPILE_r(
8360 Perl_sv_catpvf(aTHX_ sv,
8361 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8362 (UV)trie->startstate,
8363 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8364 (UV)trie->wordcount,
8367 (UV)TRIE_CHARCOUNT(trie),
8368 (UV)trie->uniquecharcount
8371 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8373 int rangestart = -1;
8374 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8375 Perl_sv_catpvf(aTHX_ sv, "[");
8376 for (i = 0; i <= 256; i++) {
8377 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8378 if (rangestart == -1)
8380 } else if (rangestart != -1) {
8381 if (i <= rangestart + 3)
8382 for (; rangestart < i; rangestart++)
8383 put_byte(sv, rangestart);
8385 put_byte(sv, rangestart);
8387 put_byte(sv, i - 1);
8392 Perl_sv_catpvf(aTHX_ sv, "]");
8395 } else if (k == CURLY) {
8396 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8397 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8398 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8400 else if (k == WHILEM && o->flags) /* Ordinal/of */
8401 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8402 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8403 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8404 if ( prog->paren_names ) {
8405 if ( k != REF || OP(o) < NREF) {
8406 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8407 SV **name= av_fetch(list, ARG(o), 0 );
8409 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8412 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8413 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8414 I32 *nums=(I32*)SvPVX(sv_dat);
8415 SV **name= av_fetch(list, nums[0], 0 );
8418 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8419 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8420 (n ? "," : ""), (IV)nums[n]);
8422 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8426 } else if (k == GOSUB)
8427 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8428 else if (k == VERB) {
8430 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8431 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8432 } else if (k == LOGICAL)
8433 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8434 else if (k == ANYOF) {
8435 int i, rangestart = -1;
8436 const U8 flags = ANYOF_FLAGS(o);
8438 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8439 static const char * const anyofs[] = {
8472 if (flags & ANYOF_LOCALE)
8473 sv_catpvs(sv, "{loc}");
8474 if (flags & ANYOF_FOLD)
8475 sv_catpvs(sv, "{i}");
8476 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8477 if (flags & ANYOF_INVERT)
8479 for (i = 0; i <= 256; i++) {
8480 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8481 if (rangestart == -1)
8483 } else if (rangestart != -1) {
8484 if (i <= rangestart + 3)
8485 for (; rangestart < i; rangestart++)
8486 put_byte(sv, rangestart);
8488 put_byte(sv, rangestart);
8490 put_byte(sv, i - 1);
8496 if (o->flags & ANYOF_CLASS)
8497 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8498 if (ANYOF_CLASS_TEST(o,i))
8499 sv_catpv(sv, anyofs[i]);
8501 if (flags & ANYOF_UNICODE)
8502 sv_catpvs(sv, "{unicode}");
8503 else if (flags & ANYOF_UNICODE_ALL)
8504 sv_catpvs(sv, "{unicode_all}");
8508 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8512 U8 s[UTF8_MAXBYTES_CASE+1];
8514 for (i = 0; i <= 256; i++) { /* just the first 256 */
8515 uvchr_to_utf8(s, i);
8517 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8518 if (rangestart == -1)
8520 } else if (rangestart != -1) {
8521 if (i <= rangestart + 3)
8522 for (; rangestart < i; rangestart++) {
8523 const U8 * const e = uvchr_to_utf8(s,rangestart);
8525 for(p = s; p < e; p++)
8529 const U8 *e = uvchr_to_utf8(s,rangestart);
8531 for (p = s; p < e; p++)
8534 e = uvchr_to_utf8(s, i-1);
8535 for (p = s; p < e; p++)
8542 sv_catpvs(sv, "..."); /* et cetera */
8546 char *s = savesvpv(lv);
8547 char * const origs = s;
8549 while (*s && *s != '\n')
8553 const char * const t = ++s;
8571 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8573 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8574 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8576 PERL_UNUSED_CONTEXT;
8577 PERL_UNUSED_ARG(sv);
8579 PERL_UNUSED_ARG(prog);
8580 #endif /* DEBUGGING */
8584 Perl_re_intuit_string(pTHX_ regexp *prog)
8585 { /* Assume that RE_INTUIT is set */
8587 GET_RE_DEBUG_FLAGS_DECL;
8588 PERL_UNUSED_CONTEXT;
8592 const char * const s = SvPV_nolen_const(prog->check_substr
8593 ? prog->check_substr : prog->check_utf8);
8595 if (!PL_colorset) reginitcolors();
8596 PerlIO_printf(Perl_debug_log,
8597 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8599 prog->check_substr ? "" : "utf8 ",
8600 PL_colors[5],PL_colors[0],
8603 (strlen(s) > 60 ? "..." : ""));
8606 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8612 handles refcounting and freeing the perl core regexp structure. When
8613 it is necessary to actually free the structure the first thing it
8614 does is call the 'free' method of the regexp_engine associated to to
8615 the regexp, allowing the handling of the void *pprivate; member
8616 first. (This routine is not overridable by extensions, which is why
8617 the extensions free is called first.)
8619 See regdupe and regdupe_internal if you change anything here.
8621 #ifndef PERL_IN_XSUB_RE
8623 Perl_pregfree(pTHX_ struct regexp *r)
8626 GET_RE_DEBUG_FLAGS_DECL;
8628 if (!r || (--r->refcnt > 0))
8631 CALLREGFREE_PVT(r); /* free the private data */
8632 RX_MATCH_COPY_FREE(r);
8633 #ifdef PERL_OLD_COPY_ON_WRITE
8635 SvREFCNT_dec(r->saved_copy);
8638 if (r->anchored_substr)
8639 SvREFCNT_dec(r->anchored_substr);
8640 if (r->anchored_utf8)
8641 SvREFCNT_dec(r->anchored_utf8);
8642 if (r->float_substr)
8643 SvREFCNT_dec(r->float_substr);
8645 SvREFCNT_dec(r->float_utf8);
8646 Safefree(r->substrs);
8649 SvREFCNT_dec(r->paren_names);
8650 Safefree(r->wrapped);
8651 Safefree(r->startp);
8657 /* regfree_internal()
8659 Free the private data in a regexp. This is overloadable by
8660 extensions. Perl takes care of the regexp structure in pregfree(),
8661 this covers the *pprivate pointer which technically perldoesnt
8662 know about, however of course we have to handle the
8663 regexp_internal structure when no extension is in use.
8665 Note this is called before freeing anything in the regexp
8670 Perl_regfree_internal(pTHX_ struct regexp *r)
8674 GET_RE_DEBUG_FLAGS_DECL;
8680 SV *dsv= sv_newmortal();
8681 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8682 dsv, r->precomp, r->prelen, 60);
8683 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8684 PL_colors[4],PL_colors[5],s);
8687 #ifdef RE_TRACK_PATTERN_OFFSETS
8689 Safefree(ri->u.offsets); /* 20010421 MJD */
8692 int n = ri->data->count;
8693 PAD* new_comppad = NULL;
8698 /* If you add a ->what type here, update the comment in regcomp.h */
8699 switch (ri->data->what[n]) {
8703 SvREFCNT_dec((SV*)ri->data->data[n]);
8706 Safefree(ri->data->data[n]);
8709 new_comppad = (AV*)ri->data->data[n];
8712 if (new_comppad == NULL)
8713 Perl_croak(aTHX_ "panic: pregfree comppad");
8714 PAD_SAVE_LOCAL(old_comppad,
8715 /* Watch out for global destruction's random ordering. */
8716 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8719 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8722 op_free((OP_4tree*)ri->data->data[n]);
8724 PAD_RESTORE_LOCAL(old_comppad);
8725 SvREFCNT_dec((SV*)new_comppad);
8731 { /* Aho Corasick add-on structure for a trie node.
8732 Used in stclass optimization only */
8734 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8736 refcount = --aho->refcount;
8739 PerlMemShared_free(aho->states);
8740 PerlMemShared_free(aho->fail);
8741 /* do this last!!!! */
8742 PerlMemShared_free(ri->data->data[n]);
8743 PerlMemShared_free(ri->regstclass);
8749 /* trie structure. */
8751 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8753 refcount = --trie->refcount;
8756 PerlMemShared_free(trie->charmap);
8757 PerlMemShared_free(trie->states);
8758 PerlMemShared_free(trie->trans);
8760 PerlMemShared_free(trie->bitmap);
8762 PerlMemShared_free(trie->wordlen);
8764 PerlMemShared_free(trie->jump);
8766 PerlMemShared_free(trie->nextword);
8767 /* do this last!!!! */
8768 PerlMemShared_free(ri->data->data[n]);
8773 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8776 Safefree(ri->data->what);
8780 Safefree(ri->swap->startp);
8781 Safefree(ri->swap->endp);
8787 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8788 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8789 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8790 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8793 regdupe - duplicate a regexp.
8795 This routine is called by sv.c's re_dup and is expected to clone a
8796 given regexp structure. It is a no-op when not under USE_ITHREADS.
8797 (Originally this *was* re_dup() for change history see sv.c)
8799 After all of the core data stored in struct regexp is duplicated
8800 the regexp_engine.dupe method is used to copy any private data
8801 stored in the *pprivate pointer. This allows extensions to handle
8802 any duplication it needs to do.
8804 See pregfree() and regfree_internal() if you change anything here.
8806 #if defined(USE_ITHREADS)
8807 #ifndef PERL_IN_XSUB_RE
8809 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
8814 struct reg_substr_datum *s;
8817 return (REGEXP *)NULL;
8819 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8823 npar = r->nparens+1;
8824 Newxz(ret, 1, regexp);
8825 Newx(ret->startp, npar, I32);
8826 Copy(r->startp, ret->startp, npar, I32);
8827 Newx(ret->endp, npar, I32);
8828 Copy(r->endp, ret->endp, npar, I32);
8831 Newx(ret->substrs, 1, struct reg_substr_data);
8832 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8833 s->min_offset = r->substrs->data[i].min_offset;
8834 s->max_offset = r->substrs->data[i].max_offset;
8835 s->end_shift = r->substrs->data[i].end_shift;
8836 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8837 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8840 ret->substrs = NULL;
8842 ret->wrapped = SAVEPVN(r->wrapped, r->wraplen);
8843 ret->precomp = ret->wrapped + (r->precomp - r->wrapped);
8844 ret->prelen = r->prelen;
8845 ret->wraplen = r->wraplen;
8847 ret->refcnt = r->refcnt;
8848 ret->minlen = r->minlen;
8849 ret->minlenret = r->minlenret;
8850 ret->nparens = r->nparens;
8851 ret->lastparen = r->lastparen;
8852 ret->lastcloseparen = r->lastcloseparen;
8853 ret->intflags = r->intflags;
8854 ret->extflags = r->extflags;
8856 ret->sublen = r->sublen;
8858 ret->engine = r->engine;
8860 ret->paren_names = hv_dup_inc(r->paren_names, param);
8862 if (RX_MATCH_COPIED(ret))
8863 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8866 #ifdef PERL_OLD_COPY_ON_WRITE
8867 ret->saved_copy = NULL;
8870 ret->pprivate = r->pprivate;
8872 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
8874 ptr_table_store(PL_ptr_table, r, ret);
8877 #endif /* PERL_IN_XSUB_RE */
8882 This is the internal complement to regdupe() which is used to copy
8883 the structure pointed to by the *pprivate pointer in the regexp.
8884 This is the core version of the extension overridable cloning hook.
8885 The regexp structure being duplicated will be copied by perl prior
8886 to this and will be provided as the regexp *r argument, however
8887 with the /old/ structures pprivate pointer value. Thus this routine
8888 may override any copying normally done by perl.
8890 It returns a pointer to the new regexp_internal structure.
8894 Perl_regdupe_internal(pTHX_ const regexp *r, CLONE_PARAMS *param)
8897 regexp_internal *reti;
8901 npar = r->nparens+1;
8904 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8905 Copy(ri->program, reti->program, len+1, regnode);
8908 Newx(reti->swap, 1, regexp_paren_ofs);
8909 /* no need to copy these */
8910 Newx(reti->swap->startp, npar, I32);
8911 Newx(reti->swap->endp, npar, I32);
8916 reti->regstclass = NULL;
8920 const int count = ri->data->count;
8923 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8924 char, struct reg_data);
8925 Newx(d->what, count, U8);
8928 for (i = 0; i < count; i++) {
8929 d->what[i] = ri->data->what[i];
8930 switch (d->what[i]) {
8931 /* legal options are one of: sSfpontTu
8932 see also regcomp.h and pregfree() */
8935 case 'p': /* actually an AV, but the dup function is identical. */
8936 case 'u': /* actually an HV, but the dup function is identical. */
8937 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
8940 /* This is cheating. */
8941 Newx(d->data[i], 1, struct regnode_charclass_class);
8942 StructCopy(ri->data->data[i], d->data[i],
8943 struct regnode_charclass_class);
8944 reti->regstclass = (regnode*)d->data[i];
8947 /* Compiled op trees are readonly and in shared memory,
8948 and can thus be shared without duplication. */
8950 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
8954 /* Trie stclasses are readonly and can thus be shared
8955 * without duplication. We free the stclass in pregfree
8956 * when the corresponding reg_ac_data struct is freed.
8958 reti->regstclass= ri->regstclass;
8962 ((reg_trie_data*)ri->data->data[i])->refcount++;
8966 d->data[i] = ri->data->data[i];
8969 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
8978 reti->name_list_idx = ri->name_list_idx;
8980 #ifdef RE_TRACK_PATTERN_OFFSETS
8981 if (ri->u.offsets) {
8982 Newx(reti->u.offsets, 2*len+1, U32);
8983 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
8986 SetProgLen(reti,len);
8992 #endif /* USE_ITHREADS */
8997 converts a regexp embedded in a MAGIC struct to its stringified form,
8998 caching the converted form in the struct and returns the cached
9001 If lp is nonnull then it is used to return the length of the
9004 If flags is nonnull and the returned string contains UTF8 then
9005 (*flags & 1) will be true.
9007 If haseval is nonnull then it is used to return whether the pattern
9010 Normally called via macro:
9012 CALLREG_STRINGIFY(mg,&len,&utf8);
9016 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9018 See sv_2pv_flags() in sv.c for an example of internal usage.
9021 #ifndef PERL_IN_XSUB_RE
9024 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9026 const regexp * const re = (regexp *)mg->mg_obj;
9028 *haseval = re->seen_evals;
9030 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
9037 - regnext - dig the "next" pointer out of a node
9040 Perl_regnext(pTHX_ register regnode *p)
9043 register I32 offset;
9048 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9057 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9060 STRLEN l1 = strlen(pat1);
9061 STRLEN l2 = strlen(pat2);
9064 const char *message;
9070 Copy(pat1, buf, l1 , char);
9071 Copy(pat2, buf + l1, l2 , char);
9072 buf[l1 + l2] = '\n';
9073 buf[l1 + l2 + 1] = '\0';
9075 /* ANSI variant takes additional second argument */
9076 va_start(args, pat2);
9080 msv = vmess(buf, &args);
9082 message = SvPV_const(msv,l1);
9085 Copy(message, buf, l1 , char);
9086 buf[l1-1] = '\0'; /* Overwrite \n */
9087 Perl_croak(aTHX_ "%s", buf);
9090 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9092 #ifndef PERL_IN_XSUB_RE
9094 Perl_save_re_context(pTHX)
9098 struct re_save_state *state;
9100 SAVEVPTR(PL_curcop);
9101 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9103 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9104 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9105 SSPUSHINT(SAVEt_RE_STATE);
9107 Copy(&PL_reg_state, state, 1, struct re_save_state);
9109 PL_reg_start_tmp = 0;
9110 PL_reg_start_tmpl = 0;
9111 PL_reg_oldsaved = NULL;
9112 PL_reg_oldsavedlen = 0;
9114 PL_reg_leftiter = 0;
9115 PL_reg_poscache = NULL;
9116 PL_reg_poscache_size = 0;
9117 #ifdef PERL_OLD_COPY_ON_WRITE
9121 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9123 const REGEXP * const rx = PM_GETRE(PL_curpm);
9126 for (i = 1; i <= rx->nparens; i++) {
9127 char digits[TYPE_CHARS(long)];
9128 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9129 GV *const *const gvp
9130 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9133 GV * const gv = *gvp;
9134 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9144 clear_re(pTHX_ void *r)
9147 ReREFCNT_dec((regexp *)r);
9153 S_put_byte(pTHX_ SV *sv, int c)
9155 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9156 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9157 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9158 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9160 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9164 #define CLEAR_OPTSTART \
9165 if (optstart) STMT_START { \
9166 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9170 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9172 STATIC const regnode *
9173 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9174 const regnode *last, const regnode *plast,
9175 SV* sv, I32 indent, U32 depth)
9178 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9179 register const regnode *next;
9180 const regnode *optstart= NULL;
9183 GET_RE_DEBUG_FLAGS_DECL;
9185 #ifdef DEBUG_DUMPUNTIL
9186 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9187 last ? last-start : 0,plast ? plast-start : 0);
9190 if (plast && plast < last)
9193 while (PL_regkind[op] != END && (!last || node < last)) {
9194 /* While that wasn't END last time... */
9197 if (op == CLOSE || op == WHILEM)
9199 next = regnext((regnode *)node);
9202 if (OP(node) == OPTIMIZED) {
9203 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9210 regprop(r, sv, node);
9211 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9212 (int)(2*indent + 1), "", SvPVX_const(sv));
9214 if (OP(node) != OPTIMIZED) {
9215 if (next == NULL) /* Next ptr. */
9216 PerlIO_printf(Perl_debug_log, " (0)");
9217 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9218 PerlIO_printf(Perl_debug_log, " (FAIL)");
9220 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9221 (void)PerlIO_putc(Perl_debug_log, '\n');
9225 if (PL_regkind[(U8)op] == BRANCHJ) {
9228 register const regnode *nnode = (OP(next) == LONGJMP
9229 ? regnext((regnode *)next)
9231 if (last && nnode > last)
9233 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9236 else if (PL_regkind[(U8)op] == BRANCH) {
9238 DUMPUNTIL(NEXTOPER(node), next);
9240 else if ( PL_regkind[(U8)op] == TRIE ) {
9241 const regnode *this_trie = node;
9242 const char op = OP(node);
9243 const U32 n = ARG(node);
9244 const reg_ac_data * const ac = op>=AHOCORASICK ?
9245 (reg_ac_data *)ri->data->data[n] :
9247 const reg_trie_data * const trie =
9248 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9250 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9252 const regnode *nextbranch= NULL;
9254 sv_setpvn(sv, "", 0);
9255 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9256 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9258 PerlIO_printf(Perl_debug_log, "%*s%s ",
9259 (int)(2*(indent+3)), "",
9260 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9261 PL_colors[0], PL_colors[1],
9262 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9263 PERL_PV_PRETTY_ELIPSES |
9269 U16 dist= trie->jump[word_idx+1];
9270 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9271 (UV)((dist ? this_trie + dist : next) - start));
9274 nextbranch= this_trie + trie->jump[0];
9275 DUMPUNTIL(this_trie + dist, nextbranch);
9277 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9278 nextbranch= regnext((regnode *)nextbranch);
9280 PerlIO_printf(Perl_debug_log, "\n");
9283 if (last && next > last)
9288 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9289 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9290 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9292 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9294 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9296 else if ( op == PLUS || op == STAR) {
9297 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9299 else if (op == ANYOF) {
9300 /* arglen 1 + class block */
9301 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9302 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9303 node = NEXTOPER(node);
9305 else if (PL_regkind[(U8)op] == EXACT) {
9306 /* Literal string, where present. */
9307 node += NODE_SZ_STR(node) - 1;
9308 node = NEXTOPER(node);
9311 node = NEXTOPER(node);
9312 node += regarglen[(U8)op];
9314 if (op == CURLYX || op == OPEN)
9318 #ifdef DEBUG_DUMPUNTIL
9319 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9324 #endif /* DEBUGGING */
9328 * c-indentation-style: bsd
9330 * indent-tabs-mode: t
9333 * ex: set ts=8 sts=4 sw=4 noet: