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("cl_anything: ",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;
2053 while ( ++opt < optimize) {
2054 Set_Node_Offset_Length(opt,0,0);
2057 Try to clean up some of the debris left after the
2060 while( optimize < jumper ) {
2061 mjd_nodelen += Node_Length((optimize));
2062 OP( optimize ) = OPTIMIZED;
2063 Set_Node_Offset_Length(optimize,0,0);
2066 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2068 } /* end node insert */
2069 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2071 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2072 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2074 SvREFCNT_dec(revcharmap);
2078 : trie->startstate>1
2084 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2086 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2088 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2089 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2092 We find the fail state for each state in the trie, this state is the longest proper
2093 suffix of the current states 'word' that is also a proper prefix of another word in our
2094 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2095 the DFA not to have to restart after its tried and failed a word at a given point, it
2096 simply continues as though it had been matching the other word in the first place.
2098 'abcdgu'=~/abcdefg|cdgu/
2099 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2100 fail, which would bring use to the state representing 'd' in the second word where we would
2101 try 'g' and succeed, prodceding to match 'cdgu'.
2103 /* add a fail transition */
2104 const U32 trie_offset = ARG(source);
2105 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2107 const U32 ucharcount = trie->uniquecharcount;
2108 const U32 numstates = trie->statecount;
2109 const U32 ubound = trie->lasttrans + ucharcount;
2113 U32 base = trie->states[ 1 ].trans.base;
2116 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2117 GET_RE_DEBUG_FLAGS_DECL;
2119 PERL_UNUSED_ARG(depth);
2123 ARG_SET( stclass, data_slot );
2124 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2125 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2126 aho->trie=trie_offset;
2127 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2128 Copy( trie->states, aho->states, numstates, reg_trie_state );
2129 Newxz( q, numstates, U32);
2130 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2133 /* initialize fail[0..1] to be 1 so that we always have
2134 a valid final fail state */
2135 fail[ 0 ] = fail[ 1 ] = 1;
2137 for ( charid = 0; charid < ucharcount ; charid++ ) {
2138 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2140 q[ q_write ] = newstate;
2141 /* set to point at the root */
2142 fail[ q[ q_write++ ] ]=1;
2145 while ( q_read < q_write) {
2146 const U32 cur = q[ q_read++ % numstates ];
2147 base = trie->states[ cur ].trans.base;
2149 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2150 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2152 U32 fail_state = cur;
2155 fail_state = fail[ fail_state ];
2156 fail_base = aho->states[ fail_state ].trans.base;
2157 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2159 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2160 fail[ ch_state ] = fail_state;
2161 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2163 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2165 q[ q_write++ % numstates] = ch_state;
2169 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2170 when we fail in state 1, this allows us to use the
2171 charclass scan to find a valid start char. This is based on the principle
2172 that theres a good chance the string being searched contains lots of stuff
2173 that cant be a start char.
2175 fail[ 0 ] = fail[ 1 ] = 0;
2176 DEBUG_TRIE_COMPILE_r({
2177 PerlIO_printf(Perl_debug_log,
2178 "%*sStclass Failtable (%"UVuf" states): 0",
2179 (int)(depth * 2), "", (UV)numstates
2181 for( q_read=1; q_read<numstates; q_read++ ) {
2182 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2184 PerlIO_printf(Perl_debug_log, "\n");
2187 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2192 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2193 * These need to be revisited when a newer toolchain becomes available.
2195 #if defined(__sparc64__) && defined(__GNUC__)
2196 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2197 # undef SPARC64_GCC_WORKAROUND
2198 # define SPARC64_GCC_WORKAROUND 1
2202 #define DEBUG_PEEP(str,scan,depth) \
2203 DEBUG_OPTIMISE_r({if (scan){ \
2204 SV * const mysv=sv_newmortal(); \
2205 regnode *Next = regnext(scan); \
2206 regprop(RExC_rx, mysv, scan); \
2207 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2208 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2209 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2216 #define JOIN_EXACT(scan,min,flags) \
2217 if (PL_regkind[OP(scan)] == EXACT) \
2218 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2221 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2222 /* Merge several consecutive EXACTish nodes into one. */
2223 regnode *n = regnext(scan);
2225 regnode *next = scan + NODE_SZ_STR(scan);
2229 regnode *stop = scan;
2230 GET_RE_DEBUG_FLAGS_DECL;
2232 PERL_UNUSED_ARG(depth);
2234 #ifndef EXPERIMENTAL_INPLACESCAN
2235 PERL_UNUSED_ARG(flags);
2236 PERL_UNUSED_ARG(val);
2238 DEBUG_PEEP("join",scan,depth);
2240 /* Skip NOTHING, merge EXACT*. */
2242 ( PL_regkind[OP(n)] == NOTHING ||
2243 (stringok && (OP(n) == OP(scan))))
2245 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2247 if (OP(n) == TAIL || n > next)
2249 if (PL_regkind[OP(n)] == NOTHING) {
2250 DEBUG_PEEP("skip:",n,depth);
2251 NEXT_OFF(scan) += NEXT_OFF(n);
2252 next = n + NODE_STEP_REGNODE;
2259 else if (stringok) {
2260 const unsigned int oldl = STR_LEN(scan);
2261 regnode * const nnext = regnext(n);
2263 DEBUG_PEEP("merg",n,depth);
2266 if (oldl + STR_LEN(n) > U8_MAX)
2268 NEXT_OFF(scan) += NEXT_OFF(n);
2269 STR_LEN(scan) += STR_LEN(n);
2270 next = n + NODE_SZ_STR(n);
2271 /* Now we can overwrite *n : */
2272 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2280 #ifdef EXPERIMENTAL_INPLACESCAN
2281 if (flags && !NEXT_OFF(n)) {
2282 DEBUG_PEEP("atch", val, depth);
2283 if (reg_off_by_arg[OP(n)]) {
2284 ARG_SET(n, val - n);
2287 NEXT_OFF(n) = val - n;
2294 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2296 Two problematic code points in Unicode casefolding of EXACT nodes:
2298 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2299 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2305 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2306 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2308 This means that in case-insensitive matching (or "loose matching",
2309 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2310 length of the above casefolded versions) can match a target string
2311 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2312 This would rather mess up the minimum length computation.
2314 What we'll do is to look for the tail four bytes, and then peek
2315 at the preceding two bytes to see whether we need to decrease
2316 the minimum length by four (six minus two).
2318 Thanks to the design of UTF-8, there cannot be false matches:
2319 A sequence of valid UTF-8 bytes cannot be a subsequence of
2320 another valid sequence of UTF-8 bytes.
2323 char * const s0 = STRING(scan), *s, *t;
2324 char * const s1 = s0 + STR_LEN(scan) - 1;
2325 char * const s2 = s1 - 4;
2326 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2327 const char t0[] = "\xaf\x49\xaf\x42";
2329 const char t0[] = "\xcc\x88\xcc\x81";
2331 const char * const t1 = t0 + 3;
2334 s < s2 && (t = ninstr(s, s1, t0, t1));
2337 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2338 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2340 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2341 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2349 n = scan + NODE_SZ_STR(scan);
2351 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2358 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2362 /* REx optimizer. Converts nodes into quickier variants "in place".
2363 Finds fixed substrings. */
2365 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2366 to the position after last scanned or to NULL. */
2368 #define INIT_AND_WITHP \
2369 assert(!and_withp); \
2370 Newx(and_withp,1,struct regnode_charclass_class); \
2371 SAVEFREEPV(and_withp)
2373 /* this is a chain of data about sub patterns we are processing that
2374 need to be handled seperately/specially in study_chunk. Its so
2375 we can simulate recursion without losing state. */
2377 typedef struct scan_frame {
2378 regnode *last; /* last node to process in this frame */
2379 regnode *next; /* next node to process when last is reached */
2380 struct scan_frame *prev; /*previous frame*/
2381 I32 stop; /* what stopparen do we use */
2385 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2388 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2389 I32 *minlenp, I32 *deltap,
2394 struct regnode_charclass_class *and_withp,
2395 U32 flags, U32 depth)
2396 /* scanp: Start here (read-write). */
2397 /* deltap: Write maxlen-minlen here. */
2398 /* last: Stop before this one. */
2399 /* data: string data about the pattern */
2400 /* stopparen: treat close N as END */
2401 /* recursed: which subroutines have we recursed into */
2402 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2405 I32 min = 0, pars = 0, code;
2406 regnode *scan = *scanp, *next;
2408 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2409 int is_inf_internal = 0; /* The studied chunk is infinite */
2410 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2411 scan_data_t data_fake;
2412 SV *re_trie_maxbuff = NULL;
2413 regnode *first_non_open = scan;
2414 I32 stopmin = I32_MAX;
2415 scan_frame *frame = NULL;
2417 GET_RE_DEBUG_FLAGS_DECL;
2420 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2424 while (first_non_open && OP(first_non_open) == OPEN)
2425 first_non_open=regnext(first_non_open);
2430 while ( scan && OP(scan) != END && scan < last ){
2431 /* Peephole optimizer: */
2432 DEBUG_STUDYDATA("Peep:", data,depth);
2433 DEBUG_PEEP("Peep",scan,depth);
2434 JOIN_EXACT(scan,&min,0);
2436 /* Follow the next-chain of the current node and optimize
2437 away all the NOTHINGs from it. */
2438 if (OP(scan) != CURLYX) {
2439 const int max = (reg_off_by_arg[OP(scan)]
2441 /* I32 may be smaller than U16 on CRAYs! */
2442 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2443 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2447 /* Skip NOTHING and LONGJMP. */
2448 while ((n = regnext(n))
2449 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2450 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2451 && off + noff < max)
2453 if (reg_off_by_arg[OP(scan)])
2456 NEXT_OFF(scan) = off;
2461 /* The principal pseudo-switch. Cannot be a switch, since we
2462 look into several different things. */
2463 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2464 || OP(scan) == IFTHEN) {
2465 next = regnext(scan);
2467 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2469 if (OP(next) == code || code == IFTHEN) {
2470 /* NOTE - There is similar code to this block below for handling
2471 TRIE nodes on a re-study. If you change stuff here check there
2473 I32 max1 = 0, min1 = I32_MAX, num = 0;
2474 struct regnode_charclass_class accum;
2475 regnode * const startbranch=scan;
2477 if (flags & SCF_DO_SUBSTR)
2478 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2479 if (flags & SCF_DO_STCLASS)
2480 cl_init_zero(pRExC_state, &accum);
2482 while (OP(scan) == code) {
2483 I32 deltanext, minnext, f = 0, fake;
2484 struct regnode_charclass_class this_class;
2487 data_fake.flags = 0;
2489 data_fake.whilem_c = data->whilem_c;
2490 data_fake.last_closep = data->last_closep;
2493 data_fake.last_closep = &fake;
2495 data_fake.pos_delta = delta;
2496 next = regnext(scan);
2497 scan = NEXTOPER(scan);
2499 scan = NEXTOPER(scan);
2500 if (flags & SCF_DO_STCLASS) {
2501 cl_init(pRExC_state, &this_class);
2502 data_fake.start_class = &this_class;
2503 f = SCF_DO_STCLASS_AND;
2505 if (flags & SCF_WHILEM_VISITED_POS)
2506 f |= SCF_WHILEM_VISITED_POS;
2508 /* we suppose the run is continuous, last=next...*/
2509 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2511 stopparen, recursed, NULL, f,depth+1);
2514 if (max1 < minnext + deltanext)
2515 max1 = minnext + deltanext;
2516 if (deltanext == I32_MAX)
2517 is_inf = is_inf_internal = 1;
2519 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2521 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2522 if ( stopmin > minnext)
2523 stopmin = min + min1;
2524 flags &= ~SCF_DO_SUBSTR;
2526 data->flags |= SCF_SEEN_ACCEPT;
2529 if (data_fake.flags & SF_HAS_EVAL)
2530 data->flags |= SF_HAS_EVAL;
2531 data->whilem_c = data_fake.whilem_c;
2533 if (flags & SCF_DO_STCLASS)
2534 cl_or(pRExC_state, &accum, &this_class);
2536 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2538 if (flags & SCF_DO_SUBSTR) {
2539 data->pos_min += min1;
2540 data->pos_delta += max1 - min1;
2541 if (max1 != min1 || is_inf)
2542 data->longest = &(data->longest_float);
2545 delta += max1 - min1;
2546 if (flags & SCF_DO_STCLASS_OR) {
2547 cl_or(pRExC_state, data->start_class, &accum);
2549 cl_and(data->start_class, and_withp);
2550 flags &= ~SCF_DO_STCLASS;
2553 else if (flags & SCF_DO_STCLASS_AND) {
2555 cl_and(data->start_class, &accum);
2556 flags &= ~SCF_DO_STCLASS;
2559 /* Switch to OR mode: cache the old value of
2560 * data->start_class */
2562 StructCopy(data->start_class, and_withp,
2563 struct regnode_charclass_class);
2564 flags &= ~SCF_DO_STCLASS_AND;
2565 StructCopy(&accum, data->start_class,
2566 struct regnode_charclass_class);
2567 flags |= SCF_DO_STCLASS_OR;
2568 data->start_class->flags |= ANYOF_EOS;
2572 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2575 Assuming this was/is a branch we are dealing with: 'scan' now
2576 points at the item that follows the branch sequence, whatever
2577 it is. We now start at the beginning of the sequence and look
2584 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2586 If we can find such a subseqence we need to turn the first
2587 element into a trie and then add the subsequent branch exact
2588 strings to the trie.
2592 1. patterns where the whole set of branch can be converted.
2594 2. patterns where only a subset can be converted.
2596 In case 1 we can replace the whole set with a single regop
2597 for the trie. In case 2 we need to keep the start and end
2600 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2601 becomes BRANCH TRIE; BRANCH X;
2603 There is an additional case, that being where there is a
2604 common prefix, which gets split out into an EXACT like node
2605 preceding the TRIE node.
2607 If x(1..n)==tail then we can do a simple trie, if not we make
2608 a "jump" trie, such that when we match the appropriate word
2609 we "jump" to the appopriate tail node. Essentailly we turn
2610 a nested if into a case structure of sorts.
2615 if (!re_trie_maxbuff) {
2616 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2617 if (!SvIOK(re_trie_maxbuff))
2618 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2620 if ( SvIV(re_trie_maxbuff)>=0 ) {
2622 regnode *first = (regnode *)NULL;
2623 regnode *last = (regnode *)NULL;
2624 regnode *tail = scan;
2629 SV * const mysv = sv_newmortal(); /* for dumping */
2631 /* var tail is used because there may be a TAIL
2632 regop in the way. Ie, the exacts will point to the
2633 thing following the TAIL, but the last branch will
2634 point at the TAIL. So we advance tail. If we
2635 have nested (?:) we may have to move through several
2639 while ( OP( tail ) == TAIL ) {
2640 /* this is the TAIL generated by (?:) */
2641 tail = regnext( tail );
2646 regprop(RExC_rx, mysv, tail );
2647 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2648 (int)depth * 2 + 2, "",
2649 "Looking for TRIE'able sequences. Tail node is: ",
2650 SvPV_nolen_const( mysv )
2656 step through the branches, cur represents each
2657 branch, noper is the first thing to be matched
2658 as part of that branch and noper_next is the
2659 regnext() of that node. if noper is an EXACT
2660 and noper_next is the same as scan (our current
2661 position in the regex) then the EXACT branch is
2662 a possible optimization target. Once we have
2663 two or more consequetive such branches we can
2664 create a trie of the EXACT's contents and stich
2665 it in place. If the sequence represents all of
2666 the branches we eliminate the whole thing and
2667 replace it with a single TRIE. If it is a
2668 subsequence then we need to stitch it in. This
2669 means the first branch has to remain, and needs
2670 to be repointed at the item on the branch chain
2671 following the last branch optimized. This could
2672 be either a BRANCH, in which case the
2673 subsequence is internal, or it could be the
2674 item following the branch sequence in which
2675 case the subsequence is at the end.
2679 /* dont use tail as the end marker for this traverse */
2680 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2681 regnode * const noper = NEXTOPER( cur );
2682 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2683 regnode * const noper_next = regnext( noper );
2687 regprop(RExC_rx, mysv, cur);
2688 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2689 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2691 regprop(RExC_rx, mysv, noper);
2692 PerlIO_printf( Perl_debug_log, " -> %s",
2693 SvPV_nolen_const(mysv));
2696 regprop(RExC_rx, mysv, noper_next );
2697 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2698 SvPV_nolen_const(mysv));
2700 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2701 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2703 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2704 : PL_regkind[ OP( noper ) ] == EXACT )
2705 || OP(noper) == NOTHING )
2707 && noper_next == tail
2712 if ( !first || optype == NOTHING ) {
2713 if (!first) first = cur;
2714 optype = OP( noper );
2720 make_trie( pRExC_state,
2721 startbranch, first, cur, tail, count,
2724 if ( PL_regkind[ OP( noper ) ] == EXACT
2726 && noper_next == tail
2731 optype = OP( noper );
2741 regprop(RExC_rx, mysv, cur);
2742 PerlIO_printf( Perl_debug_log,
2743 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2744 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2748 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2749 #ifdef TRIE_STUDY_OPT
2750 if ( ((made == MADE_EXACT_TRIE &&
2751 startbranch == first)
2752 || ( first_non_open == first )) &&
2754 flags |= SCF_TRIE_RESTUDY;
2755 if ( startbranch == first
2758 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2768 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2769 scan = NEXTOPER(NEXTOPER(scan));
2770 } else /* single branch is optimized. */
2771 scan = NEXTOPER(scan);
2773 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2774 scan_frame *newframe = NULL;
2779 if (OP(scan) != SUSPEND) {
2780 /* set the pointer */
2781 if (OP(scan) == GOSUB) {
2783 RExC_recurse[ARG2L(scan)] = scan;
2784 start = RExC_open_parens[paren-1];
2785 end = RExC_close_parens[paren-1];
2788 start = RExC_rxi->program + 1;
2792 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2793 SAVEFREEPV(recursed);
2795 if (!PAREN_TEST(recursed,paren+1)) {
2796 PAREN_SET(recursed,paren+1);
2797 Newx(newframe,1,scan_frame);
2799 if (flags & SCF_DO_SUBSTR) {
2800 SCAN_COMMIT(pRExC_state,data,minlenp);
2801 data->longest = &(data->longest_float);
2803 is_inf = is_inf_internal = 1;
2804 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2805 cl_anything(pRExC_state, data->start_class);
2806 flags &= ~SCF_DO_STCLASS;
2809 Newx(newframe,1,scan_frame);
2812 end = regnext(scan);
2817 SAVEFREEPV(newframe);
2818 newframe->next = regnext(scan);
2819 newframe->last = last;
2820 newframe->stop = stopparen;
2821 newframe->prev = frame;
2831 else if (OP(scan) == EXACT) {
2832 I32 l = STR_LEN(scan);
2835 const U8 * const s = (U8*)STRING(scan);
2836 l = utf8_length(s, s + l);
2837 uc = utf8_to_uvchr(s, NULL);
2839 uc = *((U8*)STRING(scan));
2842 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2843 /* The code below prefers earlier match for fixed
2844 offset, later match for variable offset. */
2845 if (data->last_end == -1) { /* Update the start info. */
2846 data->last_start_min = data->pos_min;
2847 data->last_start_max = is_inf
2848 ? I32_MAX : data->pos_min + data->pos_delta;
2850 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2852 SvUTF8_on(data->last_found);
2854 SV * const sv = data->last_found;
2855 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2856 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2857 if (mg && mg->mg_len >= 0)
2858 mg->mg_len += utf8_length((U8*)STRING(scan),
2859 (U8*)STRING(scan)+STR_LEN(scan));
2861 data->last_end = data->pos_min + l;
2862 data->pos_min += l; /* As in the first entry. */
2863 data->flags &= ~SF_BEFORE_EOL;
2865 if (flags & SCF_DO_STCLASS_AND) {
2866 /* Check whether it is compatible with what we know already! */
2870 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2871 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2872 && (!(data->start_class->flags & ANYOF_FOLD)
2873 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2876 ANYOF_CLASS_ZERO(data->start_class);
2877 ANYOF_BITMAP_ZERO(data->start_class);
2879 ANYOF_BITMAP_SET(data->start_class, uc);
2880 data->start_class->flags &= ~ANYOF_EOS;
2882 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2884 else if (flags & SCF_DO_STCLASS_OR) {
2885 /* false positive possible if the class is case-folded */
2887 ANYOF_BITMAP_SET(data->start_class, uc);
2889 data->start_class->flags |= ANYOF_UNICODE_ALL;
2890 data->start_class->flags &= ~ANYOF_EOS;
2891 cl_and(data->start_class, and_withp);
2893 flags &= ~SCF_DO_STCLASS;
2895 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2896 I32 l = STR_LEN(scan);
2897 UV uc = *((U8*)STRING(scan));
2899 /* Search for fixed substrings supports EXACT only. */
2900 if (flags & SCF_DO_SUBSTR) {
2902 SCAN_COMMIT(pRExC_state, data, minlenp);
2905 const U8 * const s = (U8 *)STRING(scan);
2906 l = utf8_length(s, s + l);
2907 uc = utf8_to_uvchr(s, NULL);
2910 if (flags & SCF_DO_SUBSTR)
2912 if (flags & SCF_DO_STCLASS_AND) {
2913 /* Check whether it is compatible with what we know already! */
2917 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2918 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2919 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2921 ANYOF_CLASS_ZERO(data->start_class);
2922 ANYOF_BITMAP_ZERO(data->start_class);
2924 ANYOF_BITMAP_SET(data->start_class, uc);
2925 data->start_class->flags &= ~ANYOF_EOS;
2926 data->start_class->flags |= ANYOF_FOLD;
2927 if (OP(scan) == EXACTFL)
2928 data->start_class->flags |= ANYOF_LOCALE;
2931 else if (flags & SCF_DO_STCLASS_OR) {
2932 if (data->start_class->flags & ANYOF_FOLD) {
2933 /* false positive possible if the class is case-folded.
2934 Assume that the locale settings are the same... */
2936 ANYOF_BITMAP_SET(data->start_class, uc);
2937 data->start_class->flags &= ~ANYOF_EOS;
2939 cl_and(data->start_class, and_withp);
2941 flags &= ~SCF_DO_STCLASS;
2943 else if (strchr((const char*)PL_varies,OP(scan))) {
2944 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2945 I32 f = flags, pos_before = 0;
2946 regnode * const oscan = scan;
2947 struct regnode_charclass_class this_class;
2948 struct regnode_charclass_class *oclass = NULL;
2949 I32 next_is_eval = 0;
2951 switch (PL_regkind[OP(scan)]) {
2952 case WHILEM: /* End of (?:...)* . */
2953 scan = NEXTOPER(scan);
2956 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2957 next = NEXTOPER(scan);
2958 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2960 maxcount = REG_INFTY;
2961 next = regnext(scan);
2962 scan = NEXTOPER(scan);
2966 if (flags & SCF_DO_SUBSTR)
2971 if (flags & SCF_DO_STCLASS) {
2973 maxcount = REG_INFTY;
2974 next = regnext(scan);
2975 scan = NEXTOPER(scan);
2978 is_inf = is_inf_internal = 1;
2979 scan = regnext(scan);
2980 if (flags & SCF_DO_SUBSTR) {
2981 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2982 data->longest = &(data->longest_float);
2984 goto optimize_curly_tail;
2986 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
2987 && (scan->flags == stopparen))
2992 mincount = ARG1(scan);
2993 maxcount = ARG2(scan);
2995 next = regnext(scan);
2996 if (OP(scan) == CURLYX) {
2997 I32 lp = (data ? *(data->last_closep) : 0);
2998 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3000 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3001 next_is_eval = (OP(scan) == EVAL);
3003 if (flags & SCF_DO_SUBSTR) {
3004 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3005 pos_before = data->pos_min;
3009 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3011 data->flags |= SF_IS_INF;
3013 if (flags & SCF_DO_STCLASS) {
3014 cl_init(pRExC_state, &this_class);
3015 oclass = data->start_class;
3016 data->start_class = &this_class;
3017 f |= SCF_DO_STCLASS_AND;
3018 f &= ~SCF_DO_STCLASS_OR;
3020 /* These are the cases when once a subexpression
3021 fails at a particular position, it cannot succeed
3022 even after backtracking at the enclosing scope.
3024 XXXX what if minimal match and we are at the
3025 initial run of {n,m}? */
3026 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3027 f &= ~SCF_WHILEM_VISITED_POS;
3029 /* This will finish on WHILEM, setting scan, or on NULL: */
3030 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3031 last, data, stopparen, recursed, NULL,
3033 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3035 if (flags & SCF_DO_STCLASS)
3036 data->start_class = oclass;
3037 if (mincount == 0 || minnext == 0) {
3038 if (flags & SCF_DO_STCLASS_OR) {
3039 cl_or(pRExC_state, data->start_class, &this_class);
3041 else if (flags & SCF_DO_STCLASS_AND) {
3042 /* Switch to OR mode: cache the old value of
3043 * data->start_class */
3045 StructCopy(data->start_class, and_withp,
3046 struct regnode_charclass_class);
3047 flags &= ~SCF_DO_STCLASS_AND;
3048 StructCopy(&this_class, data->start_class,
3049 struct regnode_charclass_class);
3050 flags |= SCF_DO_STCLASS_OR;
3051 data->start_class->flags |= ANYOF_EOS;
3053 } else { /* Non-zero len */
3054 if (flags & SCF_DO_STCLASS_OR) {
3055 cl_or(pRExC_state, data->start_class, &this_class);
3056 cl_and(data->start_class, and_withp);
3058 else if (flags & SCF_DO_STCLASS_AND)
3059 cl_and(data->start_class, &this_class);
3060 flags &= ~SCF_DO_STCLASS;
3062 if (!scan) /* It was not CURLYX, but CURLY. */
3064 if ( /* ? quantifier ok, except for (?{ ... }) */
3065 (next_is_eval || !(mincount == 0 && maxcount == 1))
3066 && (minnext == 0) && (deltanext == 0)
3067 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3068 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3069 && ckWARN(WARN_REGEXP))
3072 "Quantifier unexpected on zero-length expression");
3075 min += minnext * mincount;
3076 is_inf_internal |= ((maxcount == REG_INFTY
3077 && (minnext + deltanext) > 0)
3078 || deltanext == I32_MAX);
3079 is_inf |= is_inf_internal;
3080 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3082 /* Try powerful optimization CURLYX => CURLYN. */
3083 if ( OP(oscan) == CURLYX && data
3084 && data->flags & SF_IN_PAR
3085 && !(data->flags & SF_HAS_EVAL)
3086 && !deltanext && minnext == 1 ) {
3087 /* Try to optimize to CURLYN. */
3088 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3089 regnode * const nxt1 = nxt;
3096 if (!strchr((const char*)PL_simple,OP(nxt))
3097 && !(PL_regkind[OP(nxt)] == EXACT
3098 && STR_LEN(nxt) == 1))
3104 if (OP(nxt) != CLOSE)
3106 if (RExC_open_parens) {
3107 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3108 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3110 /* Now we know that nxt2 is the only contents: */
3111 oscan->flags = (U8)ARG(nxt);
3113 OP(nxt1) = NOTHING; /* was OPEN. */
3116 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3117 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3118 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3119 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3120 OP(nxt + 1) = OPTIMIZED; /* was count. */
3121 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3126 /* Try optimization CURLYX => CURLYM. */
3127 if ( OP(oscan) == CURLYX && data
3128 && !(data->flags & SF_HAS_PAR)
3129 && !(data->flags & SF_HAS_EVAL)
3130 && !deltanext /* atom is fixed width */
3131 && minnext != 0 /* CURLYM can't handle zero width */
3133 /* XXXX How to optimize if data == 0? */
3134 /* Optimize to a simpler form. */
3135 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3139 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3140 && (OP(nxt2) != WHILEM))
3142 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3143 /* Need to optimize away parenths. */
3144 if (data->flags & SF_IN_PAR) {
3145 /* Set the parenth number. */
3146 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3148 if (OP(nxt) != CLOSE)
3149 FAIL("Panic opt close");
3150 oscan->flags = (U8)ARG(nxt);
3151 if (RExC_open_parens) {
3152 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3153 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3155 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3156 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3159 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3160 OP(nxt + 1) = OPTIMIZED; /* was count. */
3161 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3162 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3165 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3166 regnode *nnxt = regnext(nxt1);
3169 if (reg_off_by_arg[OP(nxt1)])
3170 ARG_SET(nxt1, nxt2 - nxt1);
3171 else if (nxt2 - nxt1 < U16_MAX)
3172 NEXT_OFF(nxt1) = nxt2 - nxt1;
3174 OP(nxt) = NOTHING; /* Cannot beautify */
3179 /* Optimize again: */
3180 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3181 NULL, stopparen, recursed, NULL, 0,depth+1);
3186 else if ((OP(oscan) == CURLYX)
3187 && (flags & SCF_WHILEM_VISITED_POS)
3188 /* See the comment on a similar expression above.
3189 However, this time it not a subexpression
3190 we care about, but the expression itself. */
3191 && (maxcount == REG_INFTY)
3192 && data && ++data->whilem_c < 16) {
3193 /* This stays as CURLYX, we can put the count/of pair. */
3194 /* Find WHILEM (as in regexec.c) */
3195 regnode *nxt = oscan + NEXT_OFF(oscan);
3197 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3199 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3200 | (RExC_whilem_seen << 4)); /* On WHILEM */
3202 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3204 if (flags & SCF_DO_SUBSTR) {
3205 SV *last_str = NULL;
3206 int counted = mincount != 0;
3208 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3209 #if defined(SPARC64_GCC_WORKAROUND)
3212 const char *s = NULL;
3215 if (pos_before >= data->last_start_min)
3218 b = data->last_start_min;
3221 s = SvPV_const(data->last_found, l);
3222 old = b - data->last_start_min;
3225 I32 b = pos_before >= data->last_start_min
3226 ? pos_before : data->last_start_min;
3228 const char * const s = SvPV_const(data->last_found, l);
3229 I32 old = b - data->last_start_min;
3233 old = utf8_hop((U8*)s, old) - (U8*)s;
3236 /* Get the added string: */
3237 last_str = newSVpvn(s + old, l);
3239 SvUTF8_on(last_str);
3240 if (deltanext == 0 && pos_before == b) {
3241 /* What was added is a constant string */
3243 SvGROW(last_str, (mincount * l) + 1);
3244 repeatcpy(SvPVX(last_str) + l,
3245 SvPVX_const(last_str), l, mincount - 1);
3246 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3247 /* Add additional parts. */
3248 SvCUR_set(data->last_found,
3249 SvCUR(data->last_found) - l);
3250 sv_catsv(data->last_found, last_str);
3252 SV * sv = data->last_found;
3254 SvUTF8(sv) && SvMAGICAL(sv) ?
3255 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3256 if (mg && mg->mg_len >= 0)
3257 mg->mg_len += CHR_SVLEN(last_str);
3259 data->last_end += l * (mincount - 1);
3262 /* start offset must point into the last copy */
3263 data->last_start_min += minnext * (mincount - 1);
3264 data->last_start_max += is_inf ? I32_MAX
3265 : (maxcount - 1) * (minnext + data->pos_delta);
3268 /* It is counted once already... */
3269 data->pos_min += minnext * (mincount - counted);
3270 data->pos_delta += - counted * deltanext +
3271 (minnext + deltanext) * maxcount - minnext * mincount;
3272 if (mincount != maxcount) {
3273 /* Cannot extend fixed substrings found inside
3275 SCAN_COMMIT(pRExC_state,data,minlenp);
3276 if (mincount && last_str) {
3277 SV * const sv = data->last_found;
3278 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3279 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3283 sv_setsv(sv, last_str);
3284 data->last_end = data->pos_min;
3285 data->last_start_min =
3286 data->pos_min - CHR_SVLEN(last_str);
3287 data->last_start_max = is_inf
3289 : data->pos_min + data->pos_delta
3290 - CHR_SVLEN(last_str);
3292 data->longest = &(data->longest_float);
3294 SvREFCNT_dec(last_str);
3296 if (data && (fl & SF_HAS_EVAL))
3297 data->flags |= SF_HAS_EVAL;
3298 optimize_curly_tail:
3299 if (OP(oscan) != CURLYX) {
3300 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3302 NEXT_OFF(oscan) += NEXT_OFF(next);
3305 default: /* REF and CLUMP only? */
3306 if (flags & SCF_DO_SUBSTR) {
3307 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3308 data->longest = &(data->longest_float);
3310 is_inf = is_inf_internal = 1;
3311 if (flags & SCF_DO_STCLASS_OR)
3312 cl_anything(pRExC_state, data->start_class);
3313 flags &= ~SCF_DO_STCLASS;
3317 else if (strchr((const char*)PL_simple,OP(scan))) {
3320 if (flags & SCF_DO_SUBSTR) {
3321 SCAN_COMMIT(pRExC_state,data,minlenp);
3325 if (flags & SCF_DO_STCLASS) {
3326 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3328 /* Some of the logic below assumes that switching
3329 locale on will only add false positives. */
3330 switch (PL_regkind[OP(scan)]) {
3334 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3335 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3336 cl_anything(pRExC_state, data->start_class);
3339 if (OP(scan) == SANY)
3341 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3342 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3343 || (data->start_class->flags & ANYOF_CLASS));
3344 cl_anything(pRExC_state, data->start_class);
3346 if (flags & SCF_DO_STCLASS_AND || !value)
3347 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3350 if (flags & SCF_DO_STCLASS_AND)
3351 cl_and(data->start_class,
3352 (struct regnode_charclass_class*)scan);
3354 cl_or(pRExC_state, data->start_class,
3355 (struct regnode_charclass_class*)scan);
3358 if (flags & SCF_DO_STCLASS_AND) {
3359 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3360 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3361 for (value = 0; value < 256; value++)
3362 if (!isALNUM(value))
3363 ANYOF_BITMAP_CLEAR(data->start_class, value);
3367 if (data->start_class->flags & ANYOF_LOCALE)
3368 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3370 for (value = 0; value < 256; value++)
3372 ANYOF_BITMAP_SET(data->start_class, value);
3377 if (flags & SCF_DO_STCLASS_AND) {
3378 if (data->start_class->flags & ANYOF_LOCALE)
3379 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3382 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3383 data->start_class->flags |= ANYOF_LOCALE;
3387 if (flags & SCF_DO_STCLASS_AND) {
3388 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3389 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3390 for (value = 0; value < 256; value++)
3392 ANYOF_BITMAP_CLEAR(data->start_class, value);
3396 if (data->start_class->flags & ANYOF_LOCALE)
3397 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3399 for (value = 0; value < 256; value++)
3400 if (!isALNUM(value))
3401 ANYOF_BITMAP_SET(data->start_class, value);
3406 if (flags & SCF_DO_STCLASS_AND) {
3407 if (data->start_class->flags & ANYOF_LOCALE)
3408 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3411 data->start_class->flags |= ANYOF_LOCALE;
3412 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3416 if (flags & SCF_DO_STCLASS_AND) {
3417 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3418 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3419 for (value = 0; value < 256; value++)
3420 if (!isSPACE(value))
3421 ANYOF_BITMAP_CLEAR(data->start_class, value);
3425 if (data->start_class->flags & ANYOF_LOCALE)
3426 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3428 for (value = 0; value < 256; value++)
3430 ANYOF_BITMAP_SET(data->start_class, value);
3435 if (flags & SCF_DO_STCLASS_AND) {
3436 if (data->start_class->flags & ANYOF_LOCALE)
3437 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3440 data->start_class->flags |= ANYOF_LOCALE;
3441 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3445 if (flags & SCF_DO_STCLASS_AND) {
3446 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3447 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3448 for (value = 0; value < 256; value++)
3450 ANYOF_BITMAP_CLEAR(data->start_class, value);
3454 if (data->start_class->flags & ANYOF_LOCALE)
3455 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3457 for (value = 0; value < 256; value++)
3458 if (!isSPACE(value))
3459 ANYOF_BITMAP_SET(data->start_class, value);
3464 if (flags & SCF_DO_STCLASS_AND) {
3465 if (data->start_class->flags & ANYOF_LOCALE) {
3466 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3467 for (value = 0; value < 256; value++)
3468 if (!isSPACE(value))
3469 ANYOF_BITMAP_CLEAR(data->start_class, value);
3473 data->start_class->flags |= ANYOF_LOCALE;
3474 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3478 if (flags & SCF_DO_STCLASS_AND) {
3479 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3480 for (value = 0; value < 256; value++)
3481 if (!isDIGIT(value))
3482 ANYOF_BITMAP_CLEAR(data->start_class, value);
3485 if (data->start_class->flags & ANYOF_LOCALE)
3486 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3488 for (value = 0; value < 256; value++)
3490 ANYOF_BITMAP_SET(data->start_class, value);
3495 if (flags & SCF_DO_STCLASS_AND) {
3496 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3497 for (value = 0; value < 256; value++)
3499 ANYOF_BITMAP_CLEAR(data->start_class, value);
3502 if (data->start_class->flags & ANYOF_LOCALE)
3503 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3505 for (value = 0; value < 256; value++)
3506 if (!isDIGIT(value))
3507 ANYOF_BITMAP_SET(data->start_class, value);
3512 if (flags & SCF_DO_STCLASS_OR)
3513 cl_and(data->start_class, and_withp);
3514 flags &= ~SCF_DO_STCLASS;
3517 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3518 data->flags |= (OP(scan) == MEOL
3522 else if ( PL_regkind[OP(scan)] == BRANCHJ
3523 /* Lookbehind, or need to calculate parens/evals/stclass: */
3524 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3525 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3526 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3527 || OP(scan) == UNLESSM )
3529 /* Negative Lookahead/lookbehind
3530 In this case we can't do fixed string optimisation.
3533 I32 deltanext, minnext, fake = 0;
3535 struct regnode_charclass_class intrnl;
3538 data_fake.flags = 0;
3540 data_fake.whilem_c = data->whilem_c;
3541 data_fake.last_closep = data->last_closep;
3544 data_fake.last_closep = &fake;
3545 data_fake.pos_delta = delta;
3546 if ( flags & SCF_DO_STCLASS && !scan->flags
3547 && OP(scan) == IFMATCH ) { /* Lookahead */
3548 cl_init(pRExC_state, &intrnl);
3549 data_fake.start_class = &intrnl;
3550 f |= SCF_DO_STCLASS_AND;
3552 if (flags & SCF_WHILEM_VISITED_POS)
3553 f |= SCF_WHILEM_VISITED_POS;
3554 next = regnext(scan);
3555 nscan = NEXTOPER(NEXTOPER(scan));
3556 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3557 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3560 FAIL("Variable length lookbehind not implemented");
3562 else if (minnext > (I32)U8_MAX) {
3563 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3565 scan->flags = (U8)minnext;
3568 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3570 if (data_fake.flags & SF_HAS_EVAL)
3571 data->flags |= SF_HAS_EVAL;
3572 data->whilem_c = data_fake.whilem_c;
3574 if (f & SCF_DO_STCLASS_AND) {
3575 const int was = (data->start_class->flags & ANYOF_EOS);
3577 cl_and(data->start_class, &intrnl);
3579 data->start_class->flags |= ANYOF_EOS;
3582 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3584 /* Positive Lookahead/lookbehind
3585 In this case we can do fixed string optimisation,
3586 but we must be careful about it. Note in the case of
3587 lookbehind the positions will be offset by the minimum
3588 length of the pattern, something we won't know about
3589 until after the recurse.
3591 I32 deltanext, fake = 0;
3593 struct regnode_charclass_class intrnl;
3595 /* We use SAVEFREEPV so that when the full compile
3596 is finished perl will clean up the allocated
3597 minlens when its all done. This was we don't
3598 have to worry about freeing them when we know
3599 they wont be used, which would be a pain.
3602 Newx( minnextp, 1, I32 );
3603 SAVEFREEPV(minnextp);
3606 StructCopy(data, &data_fake, scan_data_t);
3607 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3610 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3611 data_fake.last_found=newSVsv(data->last_found);
3615 data_fake.last_closep = &fake;
3616 data_fake.flags = 0;
3617 data_fake.pos_delta = delta;
3619 data_fake.flags |= SF_IS_INF;
3620 if ( flags & SCF_DO_STCLASS && !scan->flags
3621 && OP(scan) == IFMATCH ) { /* Lookahead */
3622 cl_init(pRExC_state, &intrnl);
3623 data_fake.start_class = &intrnl;
3624 f |= SCF_DO_STCLASS_AND;
3626 if (flags & SCF_WHILEM_VISITED_POS)
3627 f |= SCF_WHILEM_VISITED_POS;
3628 next = regnext(scan);
3629 nscan = NEXTOPER(NEXTOPER(scan));
3631 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3632 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3635 FAIL("Variable length lookbehind not implemented");
3637 else if (*minnextp > (I32)U8_MAX) {
3638 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3640 scan->flags = (U8)*minnextp;
3645 if (f & SCF_DO_STCLASS_AND) {
3646 const int was = (data->start_class->flags & ANYOF_EOS);
3648 cl_and(data->start_class, &intrnl);
3650 data->start_class->flags |= ANYOF_EOS;
3653 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3655 if (data_fake.flags & SF_HAS_EVAL)
3656 data->flags |= SF_HAS_EVAL;
3657 data->whilem_c = data_fake.whilem_c;
3658 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3659 if (RExC_rx->minlen<*minnextp)
3660 RExC_rx->minlen=*minnextp;
3661 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3662 SvREFCNT_dec(data_fake.last_found);
3664 if ( data_fake.minlen_fixed != minlenp )
3666 data->offset_fixed= data_fake.offset_fixed;
3667 data->minlen_fixed= data_fake.minlen_fixed;
3668 data->lookbehind_fixed+= scan->flags;
3670 if ( data_fake.minlen_float != minlenp )
3672 data->minlen_float= data_fake.minlen_float;
3673 data->offset_float_min=data_fake.offset_float_min;
3674 data->offset_float_max=data_fake.offset_float_max;
3675 data->lookbehind_float+= scan->flags;
3684 else if (OP(scan) == OPEN) {
3685 if (stopparen != (I32)ARG(scan))
3688 else if (OP(scan) == CLOSE) {
3689 if (stopparen == (I32)ARG(scan)) {
3692 if ((I32)ARG(scan) == is_par) {
3693 next = regnext(scan);
3695 if ( next && (OP(next) != WHILEM) && next < last)
3696 is_par = 0; /* Disable optimization */
3699 *(data->last_closep) = ARG(scan);
3701 else if (OP(scan) == EVAL) {
3703 data->flags |= SF_HAS_EVAL;
3705 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3706 if (flags & SCF_DO_SUBSTR) {
3707 SCAN_COMMIT(pRExC_state,data,minlenp);
3708 flags &= ~SCF_DO_SUBSTR;
3710 if (data && OP(scan)==ACCEPT) {
3711 data->flags |= SCF_SEEN_ACCEPT;
3716 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3718 if (flags & SCF_DO_SUBSTR) {
3719 SCAN_COMMIT(pRExC_state,data,minlenp);
3720 data->longest = &(data->longest_float);
3722 is_inf = is_inf_internal = 1;
3723 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3724 cl_anything(pRExC_state, data->start_class);
3725 flags &= ~SCF_DO_STCLASS;
3727 else if (OP(scan) == GPOS) {
3728 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3729 !(delta || is_inf || (data && data->pos_delta)))
3731 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3732 RExC_rx->extflags |= RXf_ANCH_GPOS;
3733 if (RExC_rx->gofs < (U32)min)
3734 RExC_rx->gofs = min;
3736 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3740 #ifdef TRIE_STUDY_OPT
3741 #ifdef FULL_TRIE_STUDY
3742 else if (PL_regkind[OP(scan)] == TRIE) {
3743 /* NOTE - There is similar code to this block above for handling
3744 BRANCH nodes on the initial study. If you change stuff here
3746 regnode *trie_node= scan;
3747 regnode *tail= regnext(scan);
3748 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3749 I32 max1 = 0, min1 = I32_MAX;
3750 struct regnode_charclass_class accum;
3752 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3753 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3754 if (flags & SCF_DO_STCLASS)
3755 cl_init_zero(pRExC_state, &accum);
3761 const regnode *nextbranch= NULL;
3764 for ( word=1 ; word <= trie->wordcount ; word++)
3766 I32 deltanext=0, minnext=0, f = 0, fake;
3767 struct regnode_charclass_class this_class;
3769 data_fake.flags = 0;
3771 data_fake.whilem_c = data->whilem_c;
3772 data_fake.last_closep = data->last_closep;
3775 data_fake.last_closep = &fake;
3776 data_fake.pos_delta = delta;
3777 if (flags & SCF_DO_STCLASS) {
3778 cl_init(pRExC_state, &this_class);
3779 data_fake.start_class = &this_class;
3780 f = SCF_DO_STCLASS_AND;
3782 if (flags & SCF_WHILEM_VISITED_POS)
3783 f |= SCF_WHILEM_VISITED_POS;
3785 if (trie->jump[word]) {
3787 nextbranch = trie_node + trie->jump[0];
3788 scan= trie_node + trie->jump[word];
3789 /* We go from the jump point to the branch that follows
3790 it. Note this means we need the vestigal unused branches
3791 even though they arent otherwise used.
3793 minnext = study_chunk(pRExC_state, &scan, minlenp,
3794 &deltanext, (regnode *)nextbranch, &data_fake,
3795 stopparen, recursed, NULL, f,depth+1);
3797 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3798 nextbranch= regnext((regnode*)nextbranch);
3800 if (min1 > (I32)(minnext + trie->minlen))
3801 min1 = minnext + trie->minlen;
3802 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3803 max1 = minnext + deltanext + trie->maxlen;
3804 if (deltanext == I32_MAX)
3805 is_inf = is_inf_internal = 1;
3807 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3809 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3810 if ( stopmin > min + min1)
3811 stopmin = min + min1;
3812 flags &= ~SCF_DO_SUBSTR;
3814 data->flags |= SCF_SEEN_ACCEPT;
3817 if (data_fake.flags & SF_HAS_EVAL)
3818 data->flags |= SF_HAS_EVAL;
3819 data->whilem_c = data_fake.whilem_c;
3821 if (flags & SCF_DO_STCLASS)
3822 cl_or(pRExC_state, &accum, &this_class);
3825 if (flags & SCF_DO_SUBSTR) {
3826 data->pos_min += min1;
3827 data->pos_delta += max1 - min1;
3828 if (max1 != min1 || is_inf)
3829 data->longest = &(data->longest_float);
3832 delta += max1 - min1;
3833 if (flags & SCF_DO_STCLASS_OR) {
3834 cl_or(pRExC_state, data->start_class, &accum);
3836 cl_and(data->start_class, and_withp);
3837 flags &= ~SCF_DO_STCLASS;
3840 else if (flags & SCF_DO_STCLASS_AND) {
3842 cl_and(data->start_class, &accum);
3843 flags &= ~SCF_DO_STCLASS;
3846 /* Switch to OR mode: cache the old value of
3847 * data->start_class */
3849 StructCopy(data->start_class, and_withp,
3850 struct regnode_charclass_class);
3851 flags &= ~SCF_DO_STCLASS_AND;
3852 StructCopy(&accum, data->start_class,
3853 struct regnode_charclass_class);
3854 flags |= SCF_DO_STCLASS_OR;
3855 data->start_class->flags |= ANYOF_EOS;
3862 else if (PL_regkind[OP(scan)] == TRIE) {
3863 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3866 min += trie->minlen;
3867 delta += (trie->maxlen - trie->minlen);
3868 flags &= ~SCF_DO_STCLASS; /* xxx */
3869 if (flags & SCF_DO_SUBSTR) {
3870 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3871 data->pos_min += trie->minlen;
3872 data->pos_delta += (trie->maxlen - trie->minlen);
3873 if (trie->maxlen != trie->minlen)
3874 data->longest = &(data->longest_float);
3876 if (trie->jump) /* no more substrings -- for now /grr*/
3877 flags &= ~SCF_DO_SUBSTR;
3879 #endif /* old or new */
3880 #endif /* TRIE_STUDY_OPT */
3881 /* Else: zero-length, ignore. */
3882 scan = regnext(scan);
3887 stopparen = frame->stop;
3888 frame = frame->prev;
3889 goto fake_study_recurse;
3894 DEBUG_STUDYDATA("pre-fin:",data,depth);
3897 *deltap = is_inf_internal ? I32_MAX : delta;
3898 if (flags & SCF_DO_SUBSTR && is_inf)
3899 data->pos_delta = I32_MAX - data->pos_min;
3900 if (is_par > (I32)U8_MAX)
3902 if (is_par && pars==1 && data) {
3903 data->flags |= SF_IN_PAR;
3904 data->flags &= ~SF_HAS_PAR;
3906 else if (pars && data) {
3907 data->flags |= SF_HAS_PAR;
3908 data->flags &= ~SF_IN_PAR;
3910 if (flags & SCF_DO_STCLASS_OR)
3911 cl_and(data->start_class, and_withp);
3912 if (flags & SCF_TRIE_RESTUDY)
3913 data->flags |= SCF_TRIE_RESTUDY;
3915 DEBUG_STUDYDATA("post-fin:",data,depth);
3917 return min < stopmin ? min : stopmin;
3921 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3923 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3925 Renewc(RExC_rxi->data,
3926 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3927 char, struct reg_data);
3929 Renew(RExC_rxi->data->what, count + n, U8);
3931 Newx(RExC_rxi->data->what, n, U8);
3932 RExC_rxi->data->count = count + n;
3933 Copy(s, RExC_rxi->data->what + count, n, U8);
3937 /*XXX: todo make this not included in a non debugging perl */
3938 #ifndef PERL_IN_XSUB_RE
3940 Perl_reginitcolors(pTHX)
3943 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3945 char *t = savepv(s);
3949 t = strchr(t, '\t');
3955 PL_colors[i] = t = (char *)"";
3960 PL_colors[i++] = (char *)"";
3967 #ifdef TRIE_STUDY_OPT
3968 #define CHECK_RESTUDY_GOTO \
3970 (data.flags & SCF_TRIE_RESTUDY) \
3974 #define CHECK_RESTUDY_GOTO
3978 - pregcomp - compile a regular expression into internal code
3980 * We can't allocate space until we know how big the compiled form will be,
3981 * but we can't compile it (and thus know how big it is) until we've got a
3982 * place to put the code. So we cheat: we compile it twice, once with code
3983 * generation turned off and size counting turned on, and once "for real".
3984 * This also means that we don't allocate space until we are sure that the
3985 * thing really will compile successfully, and we never have to move the
3986 * code and thus invalidate pointers into it. (Note that it has to be in
3987 * one piece because free() must be able to free it all.) [NB: not true in perl]
3989 * Beware that the optimization-preparation code in here knows about some
3990 * of the structure of the compiled regexp. [I'll say.]
3995 #ifndef PERL_IN_XSUB_RE
3996 #define RE_ENGINE_PTR &PL_core_reg_engine
3998 extern const struct regexp_engine my_reg_engine;
3999 #define RE_ENGINE_PTR &my_reg_engine
4002 #ifndef PERL_IN_XSUB_RE
4004 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
4007 HV * const table = GvHV(PL_hintgv);
4008 /* Dispatch a request to compile a regexp to correct
4011 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4012 GET_RE_DEBUG_FLAGS_DECL;
4013 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4014 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4016 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4019 return CALLREGCOMP_ENG(eng, exp, xend, pm);
4022 return Perl_re_compile(aTHX_ exp, xend, pm);
4027 Perl_re_compile(pTHX_ char *exp, char *xend, PMOP *pm)
4031 register regexp_internal *ri;
4039 RExC_state_t RExC_state;
4040 RExC_state_t * const pRExC_state = &RExC_state;
4041 #ifdef TRIE_STUDY_OPT
4043 RExC_state_t copyRExC_state;
4045 GET_RE_DEBUG_FLAGS_DECL;
4046 DEBUG_r(if (!PL_colorset) reginitcolors());
4049 FAIL("NULL regexp argument");
4051 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4055 SV *dsv= sv_newmortal();
4056 RE_PV_QUOTED_DECL(s, RExC_utf8,
4057 dsv, RExC_precomp, (xend - exp), 60);
4058 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4059 PL_colors[4],PL_colors[5],s);
4061 RExC_flags = pm->op_pmflags;
4065 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4066 RExC_seen_evals = 0;
4069 /* First pass: determine size, legality. */
4077 RExC_emit = &PL_regdummy;
4078 RExC_whilem_seen = 0;
4079 RExC_charnames = NULL;
4080 RExC_open_parens = NULL;
4081 RExC_close_parens = NULL;
4083 RExC_paren_names = NULL;
4085 RExC_paren_name_list = NULL;
4087 RExC_recurse = NULL;
4088 RExC_recurse_count = 0;
4090 #if 0 /* REGC() is (currently) a NOP at the first pass.
4091 * Clever compilers notice this and complain. --jhi */
4092 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4094 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4095 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4096 RExC_precomp = NULL;
4100 PerlIO_printf(Perl_debug_log,
4101 "Required size %"IVdf" nodes\n"
4102 "Starting second pass (creation)\n",
4105 RExC_lastparse=NULL;
4107 /* Small enough for pointer-storage convention?
4108 If extralen==0, this means that we will not need long jumps. */
4109 if (RExC_size >= 0x10000L && RExC_extralen)
4110 RExC_size += RExC_extralen;
4113 if (RExC_whilem_seen > 15)
4114 RExC_whilem_seen = 15;
4117 /* Make room for a sentinel value at the end of the program */
4121 /* Allocate space and zero-initialize. Note, the two step process
4122 of zeroing when in debug mode, thus anything assigned has to
4123 happen after that */
4124 Newxz(r, 1, regexp);
4125 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4126 char, regexp_internal);
4127 if ( r == NULL || ri == NULL )
4128 FAIL("Regexp out of space");
4130 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4131 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4133 /* bulk initialize base fields with 0. */
4134 Zero(ri, sizeof(regexp_internal), char);
4137 /* non-zero initialization begins here */
4139 r->engine= RE_ENGINE_PTR;
4141 r->prelen = xend - exp;
4142 r->precomp = savepvn(RExC_precomp, r->prelen);
4143 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4145 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4147 if (RExC_seen & REG_SEEN_RECURSE) {
4148 Newxz(RExC_open_parens, RExC_npar,regnode *);
4149 SAVEFREEPV(RExC_open_parens);
4150 Newxz(RExC_close_parens,RExC_npar,regnode *);
4151 SAVEFREEPV(RExC_close_parens);
4154 /* Useful during FAIL. */
4155 #ifdef RE_TRACK_PATTERN_OFFSETS
4156 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4157 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4158 "%s %"UVuf" bytes for offset annotations.\n",
4159 ri->u.offsets ? "Got" : "Couldn't get",
4160 (UV)((2*RExC_size+1) * sizeof(U32))));
4162 SetProgLen(ri,RExC_size);
4166 /* Second pass: emit code. */
4167 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4172 RExC_emit_start = ri->program;
4173 RExC_emit = ri->program;
4175 /* put a sentinal on the end of the program so we can check for
4177 ri->program[RExC_size].type = 255;
4179 /* Store the count of eval-groups for security checks: */
4180 RExC_rx->seen_evals = RExC_seen_evals;
4181 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4182 if (reg(pRExC_state, 0, &flags,1) == NULL)
4185 /* XXXX To minimize changes to RE engine we always allocate
4186 3-units-long substrs field. */
4187 Newx(r->substrs, 1, struct reg_substr_data);
4188 if (RExC_recurse_count) {
4189 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4190 SAVEFREEPV(RExC_recurse);
4194 r->minlen = minlen = sawplus = sawopen = 0;
4195 Zero(r->substrs, 1, struct reg_substr_data);
4197 #ifdef TRIE_STUDY_OPT
4200 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4202 RExC_state = copyRExC_state;
4203 if (seen & REG_TOP_LEVEL_BRANCHES)
4204 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4206 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4207 if (data.last_found) {
4208 SvREFCNT_dec(data.longest_fixed);
4209 SvREFCNT_dec(data.longest_float);
4210 SvREFCNT_dec(data.last_found);
4212 StructCopy(&zero_scan_data, &data, scan_data_t);
4214 StructCopy(&zero_scan_data, &data, scan_data_t);
4215 copyRExC_state = RExC_state;
4218 StructCopy(&zero_scan_data, &data, scan_data_t);
4221 /* Dig out information for optimizations. */
4222 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4223 pm->op_pmflags = RExC_flags;
4225 r->extflags |= RXf_UTF8; /* Unicode in it? */
4226 ri->regstclass = NULL;
4227 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4228 r->intflags |= PREGf_NAUGHTY;
4229 scan = ri->program + 1; /* First BRANCH. */
4231 /* testing for BRANCH here tells us whether there is "must appear"
4232 data in the pattern. If there is then we can use it for optimisations */
4233 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4235 STRLEN longest_float_length, longest_fixed_length;
4236 struct regnode_charclass_class ch_class; /* pointed to by data */
4238 I32 last_close = 0; /* pointed to by data */
4241 /* Skip introductions and multiplicators >= 1. */
4242 while ((OP(first) == OPEN && (sawopen = 1)) ||
4243 /* An OR of *one* alternative - should not happen now. */
4244 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4245 /* for now we can't handle lookbehind IFMATCH*/
4246 (OP(first) == IFMATCH && !first->flags) ||
4247 (OP(first) == PLUS) ||
4248 (OP(first) == MINMOD) ||
4249 /* An {n,m} with n>0 */
4250 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4253 if (OP(first) == PLUS)
4256 first += regarglen[OP(first)];
4257 if (OP(first) == IFMATCH) {
4258 first = NEXTOPER(first);
4259 first += EXTRA_STEP_2ARGS;
4260 } else /* XXX possible optimisation for /(?=)/ */
4261 first = NEXTOPER(first);
4264 /* Starting-point info. */
4266 DEBUG_PEEP("first:",first,0);
4267 /* Ignore EXACT as we deal with it later. */
4268 if (PL_regkind[OP(first)] == EXACT) {
4269 if (OP(first) == EXACT)
4270 NOOP; /* Empty, get anchored substr later. */
4271 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4272 ri->regstclass = first;
4275 else if (PL_regkind[OP(first)] == TRIE &&
4276 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4279 /* this can happen only on restudy */
4280 if ( OP(first) == TRIE ) {
4281 struct regnode_1 *trieop = (struct regnode_1 *)
4282 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4283 StructCopy(first,trieop,struct regnode_1);
4284 trie_op=(regnode *)trieop;
4286 struct regnode_charclass *trieop = (struct regnode_charclass *)
4287 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4288 StructCopy(first,trieop,struct regnode_charclass);
4289 trie_op=(regnode *)trieop;
4292 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4293 ri->regstclass = trie_op;
4296 else if (strchr((const char*)PL_simple,OP(first)))
4297 ri->regstclass = first;
4298 else if (PL_regkind[OP(first)] == BOUND ||
4299 PL_regkind[OP(first)] == NBOUND)
4300 ri->regstclass = first;
4301 else if (PL_regkind[OP(first)] == BOL) {
4302 r->extflags |= (OP(first) == MBOL
4304 : (OP(first) == SBOL
4307 first = NEXTOPER(first);
4310 else if (OP(first) == GPOS) {
4311 r->extflags |= RXf_ANCH_GPOS;
4312 first = NEXTOPER(first);
4315 else if ((!sawopen || !RExC_sawback) &&
4316 (OP(first) == STAR &&
4317 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4318 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4320 /* turn .* into ^.* with an implied $*=1 */
4322 (OP(NEXTOPER(first)) == REG_ANY)
4325 r->extflags |= type;
4326 r->intflags |= PREGf_IMPLICIT;
4327 first = NEXTOPER(first);
4330 if (sawplus && (!sawopen || !RExC_sawback)
4331 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4332 /* x+ must match at the 1st pos of run of x's */
4333 r->intflags |= PREGf_SKIP;
4335 /* Scan is after the zeroth branch, first is atomic matcher. */
4336 #ifdef TRIE_STUDY_OPT
4339 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4340 (IV)(first - scan + 1))
4344 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4345 (IV)(first - scan + 1))
4351 * If there's something expensive in the r.e., find the
4352 * longest literal string that must appear and make it the
4353 * regmust. Resolve ties in favor of later strings, since
4354 * the regstart check works with the beginning of the r.e.
4355 * and avoiding duplication strengthens checking. Not a
4356 * strong reason, but sufficient in the absence of others.
4357 * [Now we resolve ties in favor of the earlier string if
4358 * it happens that c_offset_min has been invalidated, since the
4359 * earlier string may buy us something the later one won't.]
4362 data.longest_fixed = newSVpvs("");
4363 data.longest_float = newSVpvs("");
4364 data.last_found = newSVpvs("");
4365 data.longest = &(data.longest_fixed);
4367 if (!ri->regstclass) {
4368 cl_init(pRExC_state, &ch_class);
4369 data.start_class = &ch_class;
4370 stclass_flag = SCF_DO_STCLASS_AND;
4371 } else /* XXXX Check for BOUND? */
4373 data.last_closep = &last_close;
4375 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4376 &data, -1, NULL, NULL,
4377 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4383 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4384 && data.last_start_min == 0 && data.last_end > 0
4385 && !RExC_seen_zerolen
4386 && !(RExC_seen & REG_SEEN_VERBARG)
4387 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4388 r->extflags |= RXf_CHECK_ALL;
4389 scan_commit(pRExC_state, &data,&minlen,0);
4390 SvREFCNT_dec(data.last_found);
4392 /* Note that code very similar to this but for anchored string
4393 follows immediately below, changes may need to be made to both.
4396 longest_float_length = CHR_SVLEN(data.longest_float);
4397 if (longest_float_length
4398 || (data.flags & SF_FL_BEFORE_EOL
4399 && (!(data.flags & SF_FL_BEFORE_MEOL)
4400 || (RExC_flags & RXf_PMf_MULTILINE))))
4404 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4405 && data.offset_fixed == data.offset_float_min
4406 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4407 goto remove_float; /* As in (a)+. */
4409 /* copy the information about the longest float from the reg_scan_data
4410 over to the program. */
4411 if (SvUTF8(data.longest_float)) {
4412 r->float_utf8 = data.longest_float;
4413 r->float_substr = NULL;
4415 r->float_substr = data.longest_float;
4416 r->float_utf8 = NULL;
4418 /* float_end_shift is how many chars that must be matched that
4419 follow this item. We calculate it ahead of time as once the
4420 lookbehind offset is added in we lose the ability to correctly
4422 ml = data.minlen_float ? *(data.minlen_float)
4423 : (I32)longest_float_length;
4424 r->float_end_shift = ml - data.offset_float_min
4425 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4426 + data.lookbehind_float;
4427 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4428 r->float_max_offset = data.offset_float_max;
4429 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4430 r->float_max_offset -= data.lookbehind_float;
4432 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4433 && (!(data.flags & SF_FL_BEFORE_MEOL)
4434 || (RExC_flags & RXf_PMf_MULTILINE)));
4435 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4439 r->float_substr = r->float_utf8 = NULL;
4440 SvREFCNT_dec(data.longest_float);
4441 longest_float_length = 0;
4444 /* Note that code very similar to this but for floating string
4445 is immediately above, changes may need to be made to both.
4448 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4449 if (longest_fixed_length
4450 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4451 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4452 || (RExC_flags & RXf_PMf_MULTILINE))))
4456 /* copy the information about the longest fixed
4457 from the reg_scan_data over to the program. */
4458 if (SvUTF8(data.longest_fixed)) {
4459 r->anchored_utf8 = data.longest_fixed;
4460 r->anchored_substr = NULL;
4462 r->anchored_substr = data.longest_fixed;
4463 r->anchored_utf8 = NULL;
4465 /* fixed_end_shift is how many chars that must be matched that
4466 follow this item. We calculate it ahead of time as once the
4467 lookbehind offset is added in we lose the ability to correctly
4469 ml = data.minlen_fixed ? *(data.minlen_fixed)
4470 : (I32)longest_fixed_length;
4471 r->anchored_end_shift = ml - data.offset_fixed
4472 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4473 + data.lookbehind_fixed;
4474 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4476 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4477 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4478 || (RExC_flags & RXf_PMf_MULTILINE)));
4479 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4482 r->anchored_substr = r->anchored_utf8 = NULL;
4483 SvREFCNT_dec(data.longest_fixed);
4484 longest_fixed_length = 0;
4487 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4488 ri->regstclass = NULL;
4489 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4491 && !(data.start_class->flags & ANYOF_EOS)
4492 && !cl_is_anything(data.start_class))
4494 const U32 n = add_data(pRExC_state, 1, "f");
4496 Newx(RExC_rxi->data->data[n], 1,
4497 struct regnode_charclass_class);
4498 StructCopy(data.start_class,
4499 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4500 struct regnode_charclass_class);
4501 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4502 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4503 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4504 regprop(r, sv, (regnode*)data.start_class);
4505 PerlIO_printf(Perl_debug_log,
4506 "synthetic stclass \"%s\".\n",
4507 SvPVX_const(sv));});
4510 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4511 if (longest_fixed_length > longest_float_length) {
4512 r->check_end_shift = r->anchored_end_shift;
4513 r->check_substr = r->anchored_substr;
4514 r->check_utf8 = r->anchored_utf8;
4515 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4516 if (r->extflags & RXf_ANCH_SINGLE)
4517 r->extflags |= RXf_NOSCAN;
4520 r->check_end_shift = r->float_end_shift;
4521 r->check_substr = r->float_substr;
4522 r->check_utf8 = r->float_utf8;
4523 r->check_offset_min = r->float_min_offset;
4524 r->check_offset_max = r->float_max_offset;
4526 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4527 This should be changed ASAP! */
4528 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4529 r->extflags |= RXf_USE_INTUIT;
4530 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4531 r->extflags |= RXf_INTUIT_TAIL;
4533 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4534 if ( (STRLEN)minlen < longest_float_length )
4535 minlen= longest_float_length;
4536 if ( (STRLEN)minlen < longest_fixed_length )
4537 minlen= longest_fixed_length;
4541 /* Several toplevels. Best we can is to set minlen. */
4543 struct regnode_charclass_class ch_class;
4546 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4548 scan = ri->program + 1;
4549 cl_init(pRExC_state, &ch_class);
4550 data.start_class = &ch_class;
4551 data.last_closep = &last_close;
4554 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4555 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4559 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4560 = r->float_substr = r->float_utf8 = NULL;
4561 if (!(data.start_class->flags & ANYOF_EOS)
4562 && !cl_is_anything(data.start_class))
4564 const U32 n = add_data(pRExC_state, 1, "f");
4566 Newx(RExC_rxi->data->data[n], 1,
4567 struct regnode_charclass_class);
4568 StructCopy(data.start_class,
4569 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4570 struct regnode_charclass_class);
4571 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4572 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4573 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4574 regprop(r, sv, (regnode*)data.start_class);
4575 PerlIO_printf(Perl_debug_log,
4576 "synthetic stclass \"%s\".\n",
4577 SvPVX_const(sv));});
4581 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4582 the "real" pattern. */
4584 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4585 (IV)minlen, (IV)r->minlen);
4587 r->minlenret = minlen;
4588 if (r->minlen < minlen)
4591 if (RExC_seen & REG_SEEN_GPOS)
4592 r->extflags |= RXf_GPOS_SEEN;
4593 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4594 r->extflags |= RXf_LOOKBEHIND_SEEN;
4595 if (RExC_seen & REG_SEEN_EVAL)
4596 r->extflags |= RXf_EVAL_SEEN;
4597 if (RExC_seen & REG_SEEN_CANY)
4598 r->extflags |= RXf_CANY_SEEN;
4599 if (RExC_seen & REG_SEEN_VERBARG)
4600 r->intflags |= PREGf_VERBARG_SEEN;
4601 if (RExC_seen & REG_SEEN_CUTGROUP)
4602 r->intflags |= PREGf_CUTGROUP_SEEN;
4603 if (RExC_paren_names)
4604 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4606 r->paren_names = NULL;
4607 if (r->prelen == 3 && strEQ("\\s+", r->precomp))
4608 r->extflags |= RXf_WHITE;
4609 else if (r->prelen == 1 && r->precomp[0] == '^')
4610 r->extflags |= RXf_START_ONLY;
4613 if (RExC_paren_names) {
4614 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4615 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4618 ri->name_list_idx = 0;
4620 if (RExC_recurse_count) {
4621 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4622 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4623 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4626 Newxz(r->startp, RExC_npar, I32);
4627 Newxz(r->endp, RExC_npar, I32);
4628 /* assume we don't need to swap parens around before we match */
4631 PerlIO_printf(Perl_debug_log,"Final program:\n");
4634 #ifdef RE_TRACK_PATTERN_OFFSETS
4635 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4636 const U32 len = ri->u.offsets[0];
4638 GET_RE_DEBUG_FLAGS_DECL;
4639 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4640 for (i = 1; i <= len; i++) {
4641 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4642 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4643 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4645 PerlIO_printf(Perl_debug_log, "\n");
4651 #undef CORE_ONLY_BLOCK
4652 #undef RE_ENGINE_PTR
4654 #ifndef PERL_IN_XSUB_RE
4656 Perl_reg_named_buff_get(pTHX_ SV* namesv, const REGEXP * const from_re, U32 flags)
4658 AV *retarray = NULL;
4663 if (from_re || PL_curpm) {
4664 const REGEXP * const rx = from_re ? from_re : PM_GETRE(PL_curpm);
4665 if (rx && rx->paren_names) {
4666 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4669 SV* sv_dat=HeVAL(he_str);
4670 I32 *nums=(I32*)SvPVX(sv_dat);
4671 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4672 if ((I32)(rx->lastparen) >= nums[i] &&
4673 rx->endp[nums[i]] != -1)
4675 ret = reg_numbered_buff_get(nums[i],rx,NULL,0);
4679 ret = newSVsv(&PL_sv_undef);
4683 av_push(retarray, ret);
4687 return (SV*)retarray;
4695 Perl_reg_numbered_buff_get(pTHX_ I32 paren, const REGEXP * const rx, SV* usesv, U32 flags)
4700 SV *sv = usesv ? usesv : newSVpvs("");
4701 PERL_UNUSED_ARG(flags);
4704 sv_setsv(sv,&PL_sv_undef);
4708 if (paren == -2 && rx->startp[0] != -1) {
4714 if (paren == -1 && rx->endp[0] != -1) {
4716 s = rx->subbeg + rx->endp[0];
4717 i = rx->sublen - rx->endp[0];
4720 if ( 0 <= paren && paren <= (I32)rx->nparens &&
4721 (s1 = rx->startp[paren]) != -1 &&
4722 (t1 = rx->endp[paren]) != -1)
4726 s = rx->subbeg + s1;
4728 sv_setsv(sv,&PL_sv_undef);
4731 assert(rx->sublen >= (s - rx->subbeg) + i );
4733 const int oldtainted = PL_tainted;
4735 sv_setpvn(sv, s, i);
4736 PL_tainted = oldtainted;
4737 if ( (rx->extflags & RXf_CANY_SEEN)
4738 ? (RX_MATCH_UTF8(rx)
4739 && (!i || is_utf8_string((U8*)s, i)))
4740 : (RX_MATCH_UTF8(rx)) )
4747 if (RX_MATCH_TAINTED(rx)) {
4748 if (SvTYPE(sv) >= SVt_PVMG) {
4749 MAGIC* const mg = SvMAGIC(sv);
4752 SvMAGIC_set(sv, mg->mg_moremagic);
4754 if ((mgt = SvMAGIC(sv))) {
4755 mg->mg_moremagic = mgt;
4756 SvMAGIC_set(sv, mg);
4766 sv_setsv(sv,&PL_sv_undef);
4772 /* Scans the name of a named buffer from the pattern.
4773 * If flags is REG_RSN_RETURN_NULL returns null.
4774 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4775 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4776 * to the parsed name as looked up in the RExC_paren_names hash.
4777 * If there is an error throws a vFAIL().. type exception.
4780 #define REG_RSN_RETURN_NULL 0
4781 #define REG_RSN_RETURN_NAME 1
4782 #define REG_RSN_RETURN_DATA 2
4785 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4786 char *name_start = RExC_parse;
4788 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
4789 /* skip IDFIRST by using do...while */
4792 RExC_parse += UTF8SKIP(RExC_parse);
4793 } while (isALNUM_utf8((U8*)RExC_parse));
4797 } while (isALNUM(*RExC_parse));
4801 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4802 (int)(RExC_parse - name_start)));
4805 if ( flags == REG_RSN_RETURN_NAME)
4807 else if (flags==REG_RSN_RETURN_DATA) {
4810 if ( ! sv_name ) /* should not happen*/
4811 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4812 if (RExC_paren_names)
4813 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4815 sv_dat = HeVAL(he_str);
4817 vFAIL("Reference to nonexistent named group");
4821 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4828 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4829 int rem=(int)(RExC_end - RExC_parse); \
4838 if (RExC_lastparse!=RExC_parse) \
4839 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4842 iscut ? "..." : "<" \
4845 PerlIO_printf(Perl_debug_log,"%16s",""); \
4850 num=REG_NODE_NUM(RExC_emit); \
4851 if (RExC_lastnum!=num) \
4852 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4854 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4855 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4856 (int)((depth*2)), "", \
4860 RExC_lastparse=RExC_parse; \
4865 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4866 DEBUG_PARSE_MSG((funcname)); \
4867 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4869 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4870 DEBUG_PARSE_MSG((funcname)); \
4871 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4874 - reg - regular expression, i.e. main body or parenthesized thing
4876 * Caller must absorb opening parenthesis.
4878 * Combining parenthesis handling with the base level of regular expression
4879 * is a trifle forced, but the need to tie the tails of the branches to what
4880 * follows makes it hard to avoid.
4882 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4884 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4886 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4889 /* this idea is borrowed from STR_WITH_LEN in handy.h */
4890 #define CHECK_WORD(s,v,l) \
4891 (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
4894 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4895 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4898 register regnode *ret; /* Will be the head of the group. */
4899 register regnode *br;
4900 register regnode *lastbr;
4901 register regnode *ender = NULL;
4902 register I32 parno = 0;
4904 const I32 oregflags = RExC_flags;
4905 bool have_branch = 0;
4908 /* for (?g), (?gc), and (?o) warnings; warning
4909 about (?c) will warn about (?g) -- japhy */
4911 #define WASTED_O 0x01
4912 #define WASTED_G 0x02
4913 #define WASTED_C 0x04
4914 #define WASTED_GC (0x02|0x04)
4915 I32 wastedflags = 0x00;
4917 char * parse_start = RExC_parse; /* MJD */
4918 char * const oregcomp_parse = RExC_parse;
4920 GET_RE_DEBUG_FLAGS_DECL;
4921 DEBUG_PARSE("reg ");
4924 *flagp = 0; /* Tentatively. */
4927 /* Make an OPEN node, if parenthesized. */
4929 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4930 char *start_verb = RExC_parse;
4931 STRLEN verb_len = 0;
4932 char *start_arg = NULL;
4933 unsigned char op = 0;
4935 int internal_argval = 0; /* internal_argval is only useful if !argok */
4936 while ( *RExC_parse && *RExC_parse != ')' ) {
4937 if ( *RExC_parse == ':' ) {
4938 start_arg = RExC_parse + 1;
4944 verb_len = RExC_parse - start_verb;
4947 while ( *RExC_parse && *RExC_parse != ')' )
4949 if ( *RExC_parse != ')' )
4950 vFAIL("Unterminated verb pattern argument");
4951 if ( RExC_parse == start_arg )
4954 if ( *RExC_parse != ')' )
4955 vFAIL("Unterminated verb pattern");
4958 switch ( *start_verb ) {
4959 case 'A': /* (*ACCEPT) */
4960 if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
4962 internal_argval = RExC_nestroot;
4965 case 'C': /* (*COMMIT) */
4966 if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
4969 case 'F': /* (*FAIL) */
4970 if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
4975 case ':': /* (*:NAME) */
4976 case 'M': /* (*MARK:NAME) */
4977 if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
4982 case 'P': /* (*PRUNE) */
4983 if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
4986 case 'S': /* (*SKIP) */
4987 if ( CHECK_WORD("SKIP",start_verb,verb_len) )
4990 case 'T': /* (*THEN) */
4991 /* [19:06] <TimToady> :: is then */
4992 if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
4994 RExC_seen |= REG_SEEN_CUTGROUP;
5000 vFAIL3("Unknown verb pattern '%.*s'",
5001 verb_len, start_verb);
5004 if ( start_arg && internal_argval ) {
5005 vFAIL3("Verb pattern '%.*s' may not have an argument",
5006 verb_len, start_verb);
5007 } else if ( argok < 0 && !start_arg ) {
5008 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5009 verb_len, start_verb);
5011 ret = reganode(pRExC_state, op, internal_argval);
5012 if ( ! internal_argval && ! SIZE_ONLY ) {
5014 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5015 ARG(ret) = add_data( pRExC_state, 1, "S" );
5016 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5023 if (!internal_argval)
5024 RExC_seen |= REG_SEEN_VERBARG;
5025 } else if ( start_arg ) {
5026 vFAIL3("Verb pattern '%.*s' may not have an argument",
5027 verb_len, start_verb);
5029 ret = reg_node(pRExC_state, op);
5031 nextchar(pRExC_state);
5034 if (*RExC_parse == '?') { /* (?...) */
5035 bool is_logical = 0;
5036 const char * const seqstart = RExC_parse;
5039 paren = *RExC_parse++;
5040 ret = NULL; /* For look-ahead/behind. */
5043 case 'P': /* (?P...) variants for those used to PCRE/Python */
5044 paren = *RExC_parse++;
5045 if ( paren == '<') /* (?P<...>) named capture */
5047 else if (paren == '>') { /* (?P>name) named recursion */
5048 goto named_recursion;
5050 else if (paren == '=') { /* (?P=...) named backref */
5051 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5052 you change this make sure you change that */
5053 char* name_start = RExC_parse;
5055 SV *sv_dat = reg_scan_name(pRExC_state,
5056 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5057 if (RExC_parse == name_start || *RExC_parse != ')')
5058 vFAIL2("Sequence %.3s... not terminated",parse_start);
5061 num = add_data( pRExC_state, 1, "S" );
5062 RExC_rxi->data->data[num]=(void*)sv_dat;
5063 SvREFCNT_inc(sv_dat);
5066 ret = reganode(pRExC_state,
5067 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5071 Set_Node_Offset(ret, parse_start+1);
5072 Set_Node_Cur_Length(ret); /* MJD */
5074 nextchar(pRExC_state);
5078 case '<': /* (?<...) */
5079 if (*RExC_parse == '!')
5081 else if (*RExC_parse != '=')
5087 case '\'': /* (?'...') */
5088 name_start= RExC_parse;
5089 svname = reg_scan_name(pRExC_state,
5090 SIZE_ONLY ? /* reverse test from the others */
5091 REG_RSN_RETURN_NAME :
5092 REG_RSN_RETURN_NULL);
5093 if (RExC_parse == name_start)
5095 if (*RExC_parse != paren)
5096 vFAIL2("Sequence (?%c... not terminated",
5097 paren=='>' ? '<' : paren);
5101 if (!svname) /* shouldnt happen */
5103 "panic: reg_scan_name returned NULL");
5104 if (!RExC_paren_names) {
5105 RExC_paren_names= newHV();
5106 sv_2mortal((SV*)RExC_paren_names);
5108 RExC_paren_name_list= newAV();
5109 sv_2mortal((SV*)RExC_paren_name_list);
5112 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5114 sv_dat = HeVAL(he_str);
5116 /* croak baby croak */
5118 "panic: paren_name hash element allocation failed");
5119 } else if ( SvPOK(sv_dat) ) {
5120 IV count=SvIV(sv_dat);
5121 I32 *pv=(I32*)SvGROW(sv_dat,SvCUR(sv_dat)+sizeof(I32)+1);
5122 SvCUR_set(sv_dat,SvCUR(sv_dat)+sizeof(I32));
5123 pv[count]=RExC_npar;
5126 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5127 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5132 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5133 SvREFCNT_dec(svname);
5136 /*sv_dump(sv_dat);*/
5138 nextchar(pRExC_state);
5140 goto capturing_parens;
5142 RExC_seen |= REG_SEEN_LOOKBEHIND;
5144 case '=': /* (?=...) */
5145 case '!': /* (?!...) */
5146 RExC_seen_zerolen++;
5147 if (*RExC_parse == ')') {
5148 ret=reg_node(pRExC_state, OPFAIL);
5149 nextchar(pRExC_state);
5152 case ':': /* (?:...) */
5153 case '>': /* (?>...) */
5155 case '$': /* (?$...) */
5156 case '@': /* (?@...) */
5157 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5159 case '#': /* (?#...) */
5160 while (*RExC_parse && *RExC_parse != ')')
5162 if (*RExC_parse != ')')
5163 FAIL("Sequence (?#... not terminated");
5164 nextchar(pRExC_state);
5167 case '0' : /* (?0) */
5168 case 'R' : /* (?R) */
5169 if (*RExC_parse != ')')
5170 FAIL("Sequence (?R) not terminated");
5171 ret = reg_node(pRExC_state, GOSTART);
5172 nextchar(pRExC_state);
5175 { /* named and numeric backreferences */
5177 case '&': /* (?&NAME) */
5178 parse_start = RExC_parse - 1;
5181 SV *sv_dat = reg_scan_name(pRExC_state,
5182 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5183 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5185 goto gen_recurse_regop;
5188 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5190 vFAIL("Illegal pattern");
5192 goto parse_recursion;
5194 case '-': /* (?-1) */
5195 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5196 RExC_parse--; /* rewind to let it be handled later */
5200 case '1': case '2': case '3': case '4': /* (?1) */
5201 case '5': case '6': case '7': case '8': case '9':
5204 num = atoi(RExC_parse);
5205 parse_start = RExC_parse - 1; /* MJD */
5206 if (*RExC_parse == '-')
5208 while (isDIGIT(*RExC_parse))
5210 if (*RExC_parse!=')')
5211 vFAIL("Expecting close bracket");
5214 if ( paren == '-' ) {
5216 Diagram of capture buffer numbering.
5217 Top line is the normal capture buffer numbers
5218 Botton line is the negative indexing as from
5222 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5226 num = RExC_npar + num;
5229 vFAIL("Reference to nonexistent group");
5231 } else if ( paren == '+' ) {
5232 num = RExC_npar + num - 1;
5235 ret = reganode(pRExC_state, GOSUB, num);
5237 if (num > (I32)RExC_rx->nparens) {
5239 vFAIL("Reference to nonexistent group");
5241 ARG2L_SET( ret, RExC_recurse_count++);
5243 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5244 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5248 RExC_seen |= REG_SEEN_RECURSE;
5249 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5250 Set_Node_Offset(ret, parse_start); /* MJD */
5252 nextchar(pRExC_state);
5254 } /* named and numeric backreferences */
5257 case 'p': /* (?p...) */
5258 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
5259 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
5261 case '?': /* (??...) */
5263 if (*RExC_parse != '{')
5265 paren = *RExC_parse++;
5267 case '{': /* (?{...}) */
5272 char *s = RExC_parse;
5274 RExC_seen_zerolen++;
5275 RExC_seen |= REG_SEEN_EVAL;
5276 while (count && (c = *RExC_parse)) {
5287 if (*RExC_parse != ')') {
5289 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5293 OP_4tree *sop, *rop;
5294 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5297 Perl_save_re_context(aTHX);
5298 rop = sv_compile_2op(sv, &sop, "re", &pad);
5299 sop->op_private |= OPpREFCOUNTED;
5300 /* re_dup will OpREFCNT_inc */
5301 OpREFCNT_set(sop, 1);
5304 n = add_data(pRExC_state, 3, "nop");
5305 RExC_rxi->data->data[n] = (void*)rop;
5306 RExC_rxi->data->data[n+1] = (void*)sop;
5307 RExC_rxi->data->data[n+2] = (void*)pad;
5310 else { /* First pass */
5311 if (PL_reginterp_cnt < ++RExC_seen_evals
5313 /* No compiled RE interpolated, has runtime
5314 components ===> unsafe. */
5315 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5316 if (PL_tainting && PL_tainted)
5317 FAIL("Eval-group in insecure regular expression");
5318 #if PERL_VERSION > 8
5319 if (IN_PERL_COMPILETIME)
5324 nextchar(pRExC_state);
5326 ret = reg_node(pRExC_state, LOGICAL);
5329 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5330 /* deal with the length of this later - MJD */
5333 ret = reganode(pRExC_state, EVAL, n);
5334 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5335 Set_Node_Offset(ret, parse_start);
5338 case '(': /* (?(?{...})...) and (?(?=...)...) */
5341 if (RExC_parse[0] == '?') { /* (?(?...)) */
5342 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5343 || RExC_parse[1] == '<'
5344 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5347 ret = reg_node(pRExC_state, LOGICAL);
5350 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5354 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5355 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5357 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5358 char *name_start= RExC_parse++;
5360 SV *sv_dat=reg_scan_name(pRExC_state,
5361 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5362 if (RExC_parse == name_start || *RExC_parse != ch)
5363 vFAIL2("Sequence (?(%c... not terminated",
5364 (ch == '>' ? '<' : ch));
5367 num = add_data( pRExC_state, 1, "S" );
5368 RExC_rxi->data->data[num]=(void*)sv_dat;
5369 SvREFCNT_inc(sv_dat);
5371 ret = reganode(pRExC_state,NGROUPP,num);
5372 goto insert_if_check_paren;
5374 else if (RExC_parse[0] == 'D' &&
5375 RExC_parse[1] == 'E' &&
5376 RExC_parse[2] == 'F' &&
5377 RExC_parse[3] == 'I' &&
5378 RExC_parse[4] == 'N' &&
5379 RExC_parse[5] == 'E')
5381 ret = reganode(pRExC_state,DEFINEP,0);
5384 goto insert_if_check_paren;
5386 else if (RExC_parse[0] == 'R') {
5389 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5390 parno = atoi(RExC_parse++);
5391 while (isDIGIT(*RExC_parse))
5393 } else if (RExC_parse[0] == '&') {
5396 sv_dat = reg_scan_name(pRExC_state,
5397 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5398 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5400 ret = reganode(pRExC_state,INSUBP,parno);
5401 goto insert_if_check_paren;
5403 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5406 parno = atoi(RExC_parse++);
5408 while (isDIGIT(*RExC_parse))
5410 ret = reganode(pRExC_state, GROUPP, parno);
5412 insert_if_check_paren:
5413 if ((c = *nextchar(pRExC_state)) != ')')
5414 vFAIL("Switch condition not recognized");
5416 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5417 br = regbranch(pRExC_state, &flags, 1,depth+1);
5419 br = reganode(pRExC_state, LONGJMP, 0);
5421 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5422 c = *nextchar(pRExC_state);
5427 vFAIL("(?(DEFINE)....) does not allow branches");
5428 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5429 regbranch(pRExC_state, &flags, 1,depth+1);
5430 REGTAIL(pRExC_state, ret, lastbr);
5433 c = *nextchar(pRExC_state);
5438 vFAIL("Switch (?(condition)... contains too many branches");
5439 ender = reg_node(pRExC_state, TAIL);
5440 REGTAIL(pRExC_state, br, ender);
5442 REGTAIL(pRExC_state, lastbr, ender);
5443 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5446 REGTAIL(pRExC_state, ret, ender);
5450 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5454 RExC_parse--; /* for vFAIL to print correctly */
5455 vFAIL("Sequence (? incomplete");
5459 parse_flags: /* (?i) */
5461 U32 posflags = 0, negflags = 0;
5462 U32 *flagsp = &posflags;
5464 while (*RExC_parse) {
5465 /* && strchr("iogcmsx", *RExC_parse) */
5466 /* (?g), (?gc) and (?o) are useless here
5467 and must be globally applied -- japhy */
5468 switch (*RExC_parse) {
5469 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5472 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5473 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5474 if (! (wastedflags & wflagbit) ) {
5475 wastedflags |= wflagbit;
5478 "Useless (%s%c) - %suse /%c modifier",
5479 flagsp == &negflags ? "?-" : "?",
5481 flagsp == &negflags ? "don't " : "",
5489 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5490 if (! (wastedflags & WASTED_C) ) {
5491 wastedflags |= WASTED_GC;
5494 "Useless (%sc) - %suse /gc modifier",
5495 flagsp == &negflags ? "?-" : "?",
5496 flagsp == &negflags ? "don't " : ""
5502 if (flagsp == &negflags) {
5503 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5504 vWARN(RExC_parse + 1,"Useless use of (?-k)");
5506 *flagsp |= RXf_PMf_KEEPCOPY;
5510 if (flagsp == &negflags)
5513 wastedflags = 0; /* reset so (?g-c) warns twice */
5519 RExC_flags |= posflags;
5520 RExC_flags &= ~negflags;
5521 nextchar(pRExC_state);
5533 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5538 }} /* one for the default block, one for the switch */
5545 ret = reganode(pRExC_state, OPEN, parno);
5548 RExC_nestroot = parno;
5549 if (RExC_seen & REG_SEEN_RECURSE) {
5550 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5551 "Setting open paren #%"IVdf" to %d\n",
5552 (IV)parno, REG_NODE_NUM(ret)));
5553 RExC_open_parens[parno-1]= ret;
5556 Set_Node_Length(ret, 1); /* MJD */
5557 Set_Node_Offset(ret, RExC_parse); /* MJD */
5565 /* Pick up the branches, linking them together. */
5566 parse_start = RExC_parse; /* MJD */
5567 br = regbranch(pRExC_state, &flags, 1,depth+1);
5568 /* branch_len = (paren != 0); */
5572 if (*RExC_parse == '|') {
5573 if (!SIZE_ONLY && RExC_extralen) {
5574 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5577 reginsert(pRExC_state, BRANCH, br, depth+1);
5578 Set_Node_Length(br, paren != 0);
5579 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5583 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5585 else if (paren == ':') {
5586 *flagp |= flags&SIMPLE;
5588 if (is_open) { /* Starts with OPEN. */
5589 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5591 else if (paren != '?') /* Not Conditional */
5593 *flagp |= flags & (SPSTART | HASWIDTH);
5595 while (*RExC_parse == '|') {
5596 if (!SIZE_ONLY && RExC_extralen) {
5597 ender = reganode(pRExC_state, LONGJMP,0);
5598 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5601 RExC_extralen += 2; /* Account for LONGJMP. */
5602 nextchar(pRExC_state);
5603 br = regbranch(pRExC_state, &flags, 0, depth+1);
5607 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5611 *flagp |= flags&SPSTART;
5614 if (have_branch || paren != ':') {
5615 /* Make a closing node, and hook it on the end. */
5618 ender = reg_node(pRExC_state, TAIL);
5621 ender = reganode(pRExC_state, CLOSE, parno);
5622 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5623 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5624 "Setting close paren #%"IVdf" to %d\n",
5625 (IV)parno, REG_NODE_NUM(ender)));
5626 RExC_close_parens[parno-1]= ender;
5627 if (RExC_nestroot == parno)
5630 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5631 Set_Node_Length(ender,1); /* MJD */
5637 *flagp &= ~HASWIDTH;
5640 ender = reg_node(pRExC_state, SUCCEED);
5643 ender = reg_node(pRExC_state, END);
5645 assert(!RExC_opend); /* there can only be one! */
5650 REGTAIL(pRExC_state, lastbr, ender);
5652 if (have_branch && !SIZE_ONLY) {
5654 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5656 /* Hook the tails of the branches to the closing node. */
5657 for (br = ret; br; br = regnext(br)) {
5658 const U8 op = PL_regkind[OP(br)];
5660 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5662 else if (op == BRANCHJ) {
5663 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5671 static const char parens[] = "=!<,>";
5673 if (paren && (p = strchr(parens, paren))) {
5674 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5675 int flag = (p - parens) > 1;
5678 node = SUSPEND, flag = 0;
5679 reginsert(pRExC_state, node,ret, depth+1);
5680 Set_Node_Cur_Length(ret);
5681 Set_Node_Offset(ret, parse_start + 1);
5683 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5687 /* Check for proper termination. */
5689 RExC_flags = oregflags;
5690 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5691 RExC_parse = oregcomp_parse;
5692 vFAIL("Unmatched (");
5695 else if (!paren && RExC_parse < RExC_end) {
5696 if (*RExC_parse == ')') {
5698 vFAIL("Unmatched )");
5701 FAIL("Junk on end of regexp"); /* "Can't happen". */
5709 - regbranch - one alternative of an | operator
5711 * Implements the concatenation operator.
5714 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5717 register regnode *ret;
5718 register regnode *chain = NULL;
5719 register regnode *latest;
5720 I32 flags = 0, c = 0;
5721 GET_RE_DEBUG_FLAGS_DECL;
5722 DEBUG_PARSE("brnc");
5726 if (!SIZE_ONLY && RExC_extralen)
5727 ret = reganode(pRExC_state, BRANCHJ,0);
5729 ret = reg_node(pRExC_state, BRANCH);
5730 Set_Node_Length(ret, 1);
5734 if (!first && SIZE_ONLY)
5735 RExC_extralen += 1; /* BRANCHJ */
5737 *flagp = WORST; /* Tentatively. */
5740 nextchar(pRExC_state);
5741 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5743 latest = regpiece(pRExC_state, &flags,depth+1);
5744 if (latest == NULL) {
5745 if (flags & TRYAGAIN)
5749 else if (ret == NULL)
5751 *flagp |= flags&HASWIDTH;
5752 if (chain == NULL) /* First piece. */
5753 *flagp |= flags&SPSTART;
5756 REGTAIL(pRExC_state, chain, latest);
5761 if (chain == NULL) { /* Loop ran zero times. */
5762 chain = reg_node(pRExC_state, NOTHING);
5767 *flagp |= flags&SIMPLE;
5774 - regpiece - something followed by possible [*+?]
5776 * Note that the branching code sequences used for ? and the general cases
5777 * of * and + are somewhat optimized: they use the same NOTHING node as
5778 * both the endmarker for their branch list and the body of the last branch.
5779 * It might seem that this node could be dispensed with entirely, but the
5780 * endmarker role is not redundant.
5783 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5786 register regnode *ret;
5788 register char *next;
5790 const char * const origparse = RExC_parse;
5792 I32 max = REG_INFTY;
5794 const char *maxpos = NULL;
5795 GET_RE_DEBUG_FLAGS_DECL;
5796 DEBUG_PARSE("piec");
5798 ret = regatom(pRExC_state, &flags,depth+1);
5800 if (flags & TRYAGAIN)
5807 if (op == '{' && regcurly(RExC_parse)) {
5809 parse_start = RExC_parse; /* MJD */
5810 next = RExC_parse + 1;
5811 while (isDIGIT(*next) || *next == ',') {
5820 if (*next == '}') { /* got one */
5824 min = atoi(RExC_parse);
5828 maxpos = RExC_parse;
5830 if (!max && *maxpos != '0')
5831 max = REG_INFTY; /* meaning "infinity" */
5832 else if (max >= REG_INFTY)
5833 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5835 nextchar(pRExC_state);
5838 if ((flags&SIMPLE)) {
5839 RExC_naughty += 2 + RExC_naughty / 2;
5840 reginsert(pRExC_state, CURLY, ret, depth+1);
5841 Set_Node_Offset(ret, parse_start+1); /* MJD */
5842 Set_Node_Cur_Length(ret);
5845 regnode * const w = reg_node(pRExC_state, WHILEM);
5848 REGTAIL(pRExC_state, ret, w);
5849 if (!SIZE_ONLY && RExC_extralen) {
5850 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5851 reginsert(pRExC_state, NOTHING,ret, depth+1);
5852 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5854 reginsert(pRExC_state, CURLYX,ret, depth+1);
5856 Set_Node_Offset(ret, parse_start+1);
5857 Set_Node_Length(ret,
5858 op == '{' ? (RExC_parse - parse_start) : 1);
5860 if (!SIZE_ONLY && RExC_extralen)
5861 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5862 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5864 RExC_whilem_seen++, RExC_extralen += 3;
5865 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5873 if (max && max < min)
5874 vFAIL("Can't do {n,m} with n > m");
5876 ARG1_SET(ret, (U16)min);
5877 ARG2_SET(ret, (U16)max);
5889 #if 0 /* Now runtime fix should be reliable. */
5891 /* if this is reinstated, don't forget to put this back into perldiag:
5893 =item Regexp *+ operand could be empty at {#} in regex m/%s/
5895 (F) The part of the regexp subject to either the * or + quantifier
5896 could match an empty string. The {#} shows in the regular
5897 expression about where the problem was discovered.
5901 if (!(flags&HASWIDTH) && op != '?')
5902 vFAIL("Regexp *+ operand could be empty");
5905 parse_start = RExC_parse;
5906 nextchar(pRExC_state);
5908 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
5910 if (op == '*' && (flags&SIMPLE)) {
5911 reginsert(pRExC_state, STAR, ret, depth+1);
5915 else if (op == '*') {
5919 else if (op == '+' && (flags&SIMPLE)) {
5920 reginsert(pRExC_state, PLUS, ret, depth+1);
5924 else if (op == '+') {
5928 else if (op == '?') {
5933 if (!SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
5935 "%.*s matches null string many times",
5936 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
5940 if (RExC_parse < RExC_end && *RExC_parse == '?') {
5941 nextchar(pRExC_state);
5942 reginsert(pRExC_state, MINMOD, ret, depth+1);
5943 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
5945 #ifndef REG_ALLOW_MINMOD_SUSPEND
5948 if (RExC_parse < RExC_end && *RExC_parse == '+') {
5950 nextchar(pRExC_state);
5951 ender = reg_node(pRExC_state, SUCCEED);
5952 REGTAIL(pRExC_state, ret, ender);
5953 reginsert(pRExC_state, SUSPEND, ret, depth+1);
5955 ender = reg_node(pRExC_state, TAIL);
5956 REGTAIL(pRExC_state, ret, ender);
5960 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
5962 vFAIL("Nested quantifiers");
5969 /* reg_namedseq(pRExC_state,UVp)
5971 This is expected to be called by a parser routine that has
5972 recognized'\N' and needs to handle the rest. RExC_parse is
5973 expected to point at the first char following the N at the time
5976 If valuep is non-null then it is assumed that we are parsing inside
5977 of a charclass definition and the first codepoint in the resolved
5978 string is returned via *valuep and the routine will return NULL.
5979 In this mode if a multichar string is returned from the charnames
5980 handler a warning will be issued, and only the first char in the
5981 sequence will be examined. If the string returned is zero length
5982 then the value of *valuep is undefined and NON-NULL will
5983 be returned to indicate failure. (This will NOT be a valid pointer
5986 If value is null then it is assumed that we are parsing normal text
5987 and inserts a new EXACT node into the program containing the resolved
5988 string and returns a pointer to the new node. If the string is
5989 zerolength a NOTHING node is emitted.
5991 On success RExC_parse is set to the char following the endbrace.
5992 Parsing failures will generate a fatal errorvia vFAIL(...)
5994 NOTE: We cache all results from the charnames handler locally in
5995 the RExC_charnames hash (created on first use) to prevent a charnames
5996 handler from playing silly-buggers and returning a short string and
5997 then a long string for a given pattern. Since the regexp program
5998 size is calculated during an initial parse this would result
5999 in a buffer overrun so we cache to prevent the charname result from
6000 changing during the course of the parse.
6004 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6006 char * name; /* start of the content of the name */
6007 char * endbrace; /* endbrace following the name */
6010 STRLEN len; /* this has various purposes throughout the code */
6011 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6012 regnode *ret = NULL;
6014 if (*RExC_parse != '{') {
6015 vFAIL("Missing braces on \\N{}");
6017 name = RExC_parse+1;
6018 endbrace = strchr(RExC_parse, '}');
6021 vFAIL("Missing right brace on \\N{}");
6023 RExC_parse = endbrace + 1;
6026 /* RExC_parse points at the beginning brace,
6027 endbrace points at the last */
6028 if ( name[0]=='U' && name[1]=='+' ) {
6029 /* its a "unicode hex" notation {U+89AB} */
6030 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6031 | PERL_SCAN_DISALLOW_PREFIX
6032 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6034 len = (STRLEN)(endbrace - name - 2);
6035 cp = grok_hex(name + 2, &len, &fl, NULL);
6036 if ( len != (STRLEN)(endbrace - name - 2) ) {
6045 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6047 /* fetch the charnames handler for this scope */
6048 HV * const table = GvHV(PL_hintgv);
6050 hv_fetchs(table, "charnames", FALSE) :
6052 SV *cv= cvp ? *cvp : NULL;
6055 /* create an SV with the name as argument */
6056 sv_name = newSVpvn(name, endbrace - name);
6058 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6059 vFAIL2("Constant(\\N{%s}) unknown: "
6060 "(possibly a missing \"use charnames ...\")",
6063 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6064 vFAIL2("Constant(\\N{%s}): "
6065 "$^H{charnames} is not defined",SvPVX(sv_name));
6070 if (!RExC_charnames) {
6071 /* make sure our cache is allocated */
6072 RExC_charnames = newHV();
6073 sv_2mortal((SV*)RExC_charnames);
6075 /* see if we have looked this one up before */
6076 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6078 sv_str = HeVAL(he_str);
6091 count= call_sv(cv, G_SCALAR);
6093 if (count == 1) { /* XXXX is this right? dmq */
6095 SvREFCNT_inc_simple_void(sv_str);
6103 if ( !sv_str || !SvOK(sv_str) ) {
6104 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6105 "did not return a defined value",SvPVX(sv_name));
6107 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6112 char *p = SvPV(sv_str, len);
6115 if ( SvUTF8(sv_str) ) {
6116 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6120 We have to turn on utf8 for high bit chars otherwise
6121 we get failures with
6123 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6124 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6126 This is different from what \x{} would do with the same
6127 codepoint, where the condition is > 0xFF.
6134 /* warn if we havent used the whole string? */
6136 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6138 "Ignoring excess chars from \\N{%s} in character class",
6142 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6144 "Ignoring zero length \\N{%s} in character class",
6149 SvREFCNT_dec(sv_name);
6151 SvREFCNT_dec(sv_str);
6152 return len ? NULL : (regnode *)&len;
6153 } else if(SvCUR(sv_str)) {
6158 char * parse_start = name-3; /* needed for the offsets */
6159 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6161 ret = reg_node(pRExC_state,
6162 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6165 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6166 sv_utf8_upgrade(sv_str);
6167 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6171 p = SvPV(sv_str, len);
6173 /* len is the length written, charlen is the size the char read */
6174 for ( len = 0; p < pend; p += charlen ) {
6176 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6178 STRLEN foldlen,numlen;
6179 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6180 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6181 /* Emit all the Unicode characters. */
6183 for (foldbuf = tmpbuf;
6187 uvc = utf8_to_uvchr(foldbuf, &numlen);
6189 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6192 /* In EBCDIC the numlen
6193 * and unilen can differ. */
6195 if (numlen >= foldlen)
6199 break; /* "Can't happen." */
6202 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6214 RExC_size += STR_SZ(len);
6217 RExC_emit += STR_SZ(len);
6219 Set_Node_Cur_Length(ret); /* MJD */
6221 nextchar(pRExC_state);
6223 ret = reg_node(pRExC_state,NOTHING);
6226 SvREFCNT_dec(sv_str);
6229 SvREFCNT_dec(sv_name);
6239 * It returns the code point in utf8 for the value in *encp.
6240 * value: a code value in the source encoding
6241 * encp: a pointer to an Encode object
6243 * If the result from Encode is not a single character,
6244 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6247 S_reg_recode(pTHX_ const char value, SV **encp)
6250 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6251 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6253 const STRLEN newlen = SvCUR(sv);
6254 UV uv = UNICODE_REPLACEMENT;
6258 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6261 if (!newlen || numlen != newlen) {
6262 uv = UNICODE_REPLACEMENT;
6271 - regatom - the lowest level
6273 Try to identify anything special at the start of the pattern. If there
6274 is, then handle it as required. This may involve generating a single regop,
6275 such as for an assertion; or it may involve recursing, such as to
6276 handle a () structure.
6278 If the string doesn't start with something special then we gobble up
6279 as much literal text as we can.
6281 Once we have been able to handle whatever type of thing started the
6282 sequence, we return.
6284 Note: we have to be careful with escapes, as they can be both literal
6285 and special, and in the case of \10 and friends can either, depending
6286 on context. Specifically there are two seperate switches for handling
6287 escape sequences, with the one for handling literal escapes requiring
6288 a dummy entry for all of the special escapes that are actually handled
6293 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6296 register regnode *ret = NULL;
6298 char *parse_start = RExC_parse;
6299 GET_RE_DEBUG_FLAGS_DECL;
6300 DEBUG_PARSE("atom");
6301 *flagp = WORST; /* Tentatively. */
6305 switch (*RExC_parse) {
6307 RExC_seen_zerolen++;
6308 nextchar(pRExC_state);
6309 if (RExC_flags & RXf_PMf_MULTILINE)
6310 ret = reg_node(pRExC_state, MBOL);
6311 else if (RExC_flags & RXf_PMf_SINGLELINE)
6312 ret = reg_node(pRExC_state, SBOL);
6314 ret = reg_node(pRExC_state, BOL);
6315 Set_Node_Length(ret, 1); /* MJD */
6318 nextchar(pRExC_state);
6320 RExC_seen_zerolen++;
6321 if (RExC_flags & RXf_PMf_MULTILINE)
6322 ret = reg_node(pRExC_state, MEOL);
6323 else if (RExC_flags & RXf_PMf_SINGLELINE)
6324 ret = reg_node(pRExC_state, SEOL);
6326 ret = reg_node(pRExC_state, EOL);
6327 Set_Node_Length(ret, 1); /* MJD */
6330 nextchar(pRExC_state);
6331 if (RExC_flags & RXf_PMf_SINGLELINE)
6332 ret = reg_node(pRExC_state, SANY);
6334 ret = reg_node(pRExC_state, REG_ANY);
6335 *flagp |= HASWIDTH|SIMPLE;
6337 Set_Node_Length(ret, 1); /* MJD */
6341 char * const oregcomp_parse = ++RExC_parse;
6342 ret = regclass(pRExC_state,depth+1);
6343 if (*RExC_parse != ']') {
6344 RExC_parse = oregcomp_parse;
6345 vFAIL("Unmatched [");
6347 nextchar(pRExC_state);
6348 *flagp |= HASWIDTH|SIMPLE;
6349 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6353 nextchar(pRExC_state);
6354 ret = reg(pRExC_state, 1, &flags,depth+1);
6356 if (flags & TRYAGAIN) {
6357 if (RExC_parse == RExC_end) {
6358 /* Make parent create an empty node if needed. */
6366 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
6370 if (flags & TRYAGAIN) {
6374 vFAIL("Internal urp");
6375 /* Supposed to be caught earlier. */
6378 if (!regcurly(RExC_parse)) {
6387 vFAIL("Quantifier follows nothing");
6392 This switch handles escape sequences that resolve to some kind
6393 of special regop and not to literal text. Escape sequnces that
6394 resolve to literal text are handled below in the switch marked
6397 Every entry in this switch *must* have a corresponding entry
6398 in the literal escape switch. However, the opposite is not
6399 required, as the default for this switch is to jump to the
6400 literal text handling code.
6402 switch (*++RExC_parse) {
6403 /* Special Escapes */
6405 RExC_seen_zerolen++;
6406 ret = reg_node(pRExC_state, SBOL);
6408 goto finish_meta_pat;
6410 ret = reg_node(pRExC_state, GPOS);
6411 RExC_seen |= REG_SEEN_GPOS;
6413 goto finish_meta_pat;
6415 RExC_seen_zerolen++;
6416 ret = reg_node(pRExC_state, KEEPS);
6418 goto finish_meta_pat;
6420 ret = reg_node(pRExC_state, SEOL);
6422 RExC_seen_zerolen++; /* Do not optimize RE away */
6423 goto finish_meta_pat;
6425 ret = reg_node(pRExC_state, EOS);
6427 RExC_seen_zerolen++; /* Do not optimize RE away */
6428 goto finish_meta_pat;
6430 ret = reg_node(pRExC_state, CANY);
6431 RExC_seen |= REG_SEEN_CANY;
6432 *flagp |= HASWIDTH|SIMPLE;
6433 goto finish_meta_pat;
6435 ret = reg_node(pRExC_state, CLUMP);
6437 goto finish_meta_pat;
6439 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6440 *flagp |= HASWIDTH|SIMPLE;
6441 goto finish_meta_pat;
6443 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6444 *flagp |= HASWIDTH|SIMPLE;
6445 goto finish_meta_pat;
6447 RExC_seen_zerolen++;
6448 RExC_seen |= REG_SEEN_LOOKBEHIND;
6449 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6451 goto finish_meta_pat;
6453 RExC_seen_zerolen++;
6454 RExC_seen |= REG_SEEN_LOOKBEHIND;
6455 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6457 goto finish_meta_pat;
6459 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6460 *flagp |= HASWIDTH|SIMPLE;
6461 goto finish_meta_pat;
6463 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6464 *flagp |= HASWIDTH|SIMPLE;
6465 goto finish_meta_pat;
6467 ret = reg_node(pRExC_state, DIGIT);
6468 *flagp |= HASWIDTH|SIMPLE;
6469 goto finish_meta_pat;
6471 ret = reg_node(pRExC_state, NDIGIT);
6472 *flagp |= HASWIDTH|SIMPLE;
6473 goto finish_meta_pat;
6475 ret = reganode(pRExC_state, PRUNE, 0);
6478 goto finish_meta_pat;
6480 ret = reganode(pRExC_state, SKIP, 0);
6484 nextchar(pRExC_state);
6485 Set_Node_Length(ret, 2); /* MJD */
6490 char* const oldregxend = RExC_end;
6491 char* parse_start = RExC_parse - 2;
6493 if (RExC_parse[1] == '{') {
6494 /* a lovely hack--pretend we saw [\pX] instead */
6495 RExC_end = strchr(RExC_parse, '}');
6497 const U8 c = (U8)*RExC_parse;
6499 RExC_end = oldregxend;
6500 vFAIL2("Missing right brace on \\%c{}", c);
6505 RExC_end = RExC_parse + 2;
6506 if (RExC_end > oldregxend)
6507 RExC_end = oldregxend;
6511 ret = regclass(pRExC_state,depth+1);
6513 RExC_end = oldregxend;
6516 Set_Node_Offset(ret, parse_start + 2);
6517 Set_Node_Cur_Length(ret);
6518 nextchar(pRExC_state);
6519 *flagp |= HASWIDTH|SIMPLE;
6523 /* Handle \N{NAME} here and not below because it can be
6524 multicharacter. join_exact() will join them up later on.
6525 Also this makes sure that things like /\N{BLAH}+/ and
6526 \N{BLAH} being multi char Just Happen. dmq*/
6528 ret= reg_namedseq(pRExC_state, NULL);
6530 case 'k': /* Handle \k<NAME> and \k'NAME' */
6533 char ch= RExC_parse[1];
6534 if (ch != '<' && ch != '\'' && ch != '{') {
6536 vFAIL2("Sequence %.2s... not terminated",parse_start);
6538 /* this pretty much dupes the code for (?P=...) in reg(), if
6539 you change this make sure you change that */
6540 char* name_start = (RExC_parse += 2);
6542 SV *sv_dat = reg_scan_name(pRExC_state,
6543 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6544 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
6545 if (RExC_parse == name_start || *RExC_parse != ch)
6546 vFAIL2("Sequence %.3s... not terminated",parse_start);
6549 num = add_data( pRExC_state, 1, "S" );
6550 RExC_rxi->data->data[num]=(void*)sv_dat;
6551 SvREFCNT_inc(sv_dat);
6555 ret = reganode(pRExC_state,
6556 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6560 /* override incorrect value set in reganode MJD */
6561 Set_Node_Offset(ret, parse_start+1);
6562 Set_Node_Cur_Length(ret); /* MJD */
6563 nextchar(pRExC_state);
6569 case '1': case '2': case '3': case '4':
6570 case '5': case '6': case '7': case '8': case '9':
6573 bool isg = *RExC_parse == 'g';
6578 if (*RExC_parse == '{') {
6582 if (*RExC_parse == '-') {
6586 if (hasbrace && !isDIGIT(*RExC_parse)) {
6587 if (isrel) RExC_parse--;
6589 goto parse_named_seq;
6591 num = atoi(RExC_parse);
6593 num = RExC_npar - num;
6595 vFAIL("Reference to nonexistent or unclosed group");
6597 if (!isg && num > 9 && num >= RExC_npar)
6600 char * const parse_start = RExC_parse - 1; /* MJD */
6601 while (isDIGIT(*RExC_parse))
6603 if (parse_start == RExC_parse - 1)
6604 vFAIL("Unterminated \\g... pattern");
6606 if (*RExC_parse != '}')
6607 vFAIL("Unterminated \\g{...} pattern");
6611 if (num > (I32)RExC_rx->nparens)
6612 vFAIL("Reference to nonexistent group");
6615 ret = reganode(pRExC_state,
6616 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6620 /* override incorrect value set in reganode MJD */
6621 Set_Node_Offset(ret, parse_start+1);
6622 Set_Node_Cur_Length(ret); /* MJD */
6624 nextchar(pRExC_state);
6629 if (RExC_parse >= RExC_end)
6630 FAIL("Trailing \\");
6633 /* Do not generate "unrecognized" warnings here, we fall
6634 back into the quick-grab loop below */
6641 if (RExC_flags & RXf_PMf_EXTENDED) {
6642 while (RExC_parse < RExC_end && *RExC_parse != '\n')
6644 if (RExC_parse < RExC_end)
6650 register STRLEN len;
6655 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6657 parse_start = RExC_parse - 1;
6663 ret = reg_node(pRExC_state,
6664 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6666 for (len = 0, p = RExC_parse - 1;
6667 len < 127 && p < RExC_end;
6670 char * const oldp = p;
6672 if (RExC_flags & RXf_PMf_EXTENDED)
6673 p = regwhite(p, RExC_end);
6684 /* Literal Escapes Switch
6686 This switch is meant to handle escape sequences that
6687 resolve to a literal character.
6689 Every escape sequence that represents something
6690 else, like an assertion or a char class, is handled
6691 in the switch marked 'Special Escapes' above in this
6692 routine, but also has an entry here as anything that
6693 isn't explicitly mentioned here will be treated as
6694 an unescaped equivalent literal.
6698 /* These are all the special escapes. */
6699 case 'A': /* Start assertion */
6700 case 'b': case 'B': /* Word-boundary assertion*/
6701 case 'C': /* Single char !DANGEROUS! */
6702 case 'd': case 'D': /* digit class */
6703 case 'g': case 'G': /* generic-backref, pos assertion */
6704 case 'k': case 'K': /* named backref, keep marker */
6705 case 'N': /* named char sequence */
6706 case 'p': case 'P': /* unicode property */
6707 case 's': case 'S': /* space class */
6708 case 'v': case 'V': /* (*PRUNE) and (*SKIP) */
6709 case 'w': case 'W': /* word class */
6710 case 'X': /* eXtended Unicode "combining character sequence" */
6711 case 'z': case 'Z': /* End of line/string assertion */
6715 /* Anything after here is an escape that resolves to a
6716 literal. (Except digits, which may or may not)
6735 ender = ASCII_TO_NATIVE('\033');
6739 ender = ASCII_TO_NATIVE('\007');
6744 char* const e = strchr(p, '}');
6748 vFAIL("Missing right brace on \\x{}");
6751 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6752 | PERL_SCAN_DISALLOW_PREFIX;
6753 STRLEN numlen = e - p - 1;
6754 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6761 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6763 ender = grok_hex(p, &numlen, &flags, NULL);
6766 if (PL_encoding && ender < 0x100)
6767 goto recode_encoding;
6771 ender = UCHARAT(p++);
6772 ender = toCTRL(ender);
6774 case '0': case '1': case '2': case '3':case '4':
6775 case '5': case '6': case '7': case '8':case '9':
6777 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6780 ender = grok_oct(p, &numlen, &flags, NULL);
6787 if (PL_encoding && ender < 0x100)
6788 goto recode_encoding;
6792 SV* enc = PL_encoding;
6793 ender = reg_recode((const char)(U8)ender, &enc);
6794 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6795 vWARN(p, "Invalid escape in the specified encoding");
6801 FAIL("Trailing \\");
6804 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6805 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6806 goto normal_default;
6811 if (UTF8_IS_START(*p) && UTF) {
6813 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6814 &numlen, UTF8_ALLOW_DEFAULT);
6821 if (RExC_flags & RXf_PMf_EXTENDED)
6822 p = regwhite(p, RExC_end);
6824 /* Prime the casefolded buffer. */
6825 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6827 if (ISMULT2(p)) { /* Back off on ?+*. */
6832 /* Emit all the Unicode characters. */
6834 for (foldbuf = tmpbuf;
6836 foldlen -= numlen) {
6837 ender = utf8_to_uvchr(foldbuf, &numlen);
6839 const STRLEN unilen = reguni(pRExC_state, ender, s);
6842 /* In EBCDIC the numlen
6843 * and unilen can differ. */
6845 if (numlen >= foldlen)
6849 break; /* "Can't happen." */
6853 const STRLEN unilen = reguni(pRExC_state, ender, s);
6862 REGC((char)ender, s++);
6868 /* Emit all the Unicode characters. */
6870 for (foldbuf = tmpbuf;
6872 foldlen -= numlen) {
6873 ender = utf8_to_uvchr(foldbuf, &numlen);
6875 const STRLEN unilen = reguni(pRExC_state, ender, s);
6878 /* In EBCDIC the numlen
6879 * and unilen can differ. */
6881 if (numlen >= foldlen)
6889 const STRLEN unilen = reguni(pRExC_state, ender, s);
6898 REGC((char)ender, s++);
6902 Set_Node_Cur_Length(ret); /* MJD */
6903 nextchar(pRExC_state);
6905 /* len is STRLEN which is unsigned, need to copy to signed */
6908 vFAIL("Internal disaster");
6912 if (len == 1 && UNI_IS_INVARIANT(ender))
6916 RExC_size += STR_SZ(len);
6919 RExC_emit += STR_SZ(len);
6929 S_regwhite(char *p, const char *e)
6934 else if (*p == '#') {
6937 } while (p < e && *p != '\n');
6945 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
6946 Character classes ([:foo:]) can also be negated ([:^foo:]).
6947 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
6948 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
6949 but trigger failures because they are currently unimplemented. */
6951 #define POSIXCC_DONE(c) ((c) == ':')
6952 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
6953 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
6956 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
6959 I32 namedclass = OOB_NAMEDCLASS;
6961 if (value == '[' && RExC_parse + 1 < RExC_end &&
6962 /* I smell either [: or [= or [. -- POSIX has been here, right? */
6963 POSIXCC(UCHARAT(RExC_parse))) {
6964 const char c = UCHARAT(RExC_parse);
6965 char* const s = RExC_parse++;
6967 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
6969 if (RExC_parse == RExC_end)
6970 /* Grandfather lone [:, [=, [. */
6973 const char* const t = RExC_parse++; /* skip over the c */
6976 if (UCHARAT(RExC_parse) == ']') {
6977 const char *posixcc = s + 1;
6978 RExC_parse++; /* skip over the ending ] */
6981 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
6982 const I32 skip = t - posixcc;
6984 /* Initially switch on the length of the name. */
6987 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
6988 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
6991 /* Names all of length 5. */
6992 /* alnum alpha ascii blank cntrl digit graph lower
6993 print punct space upper */
6994 /* Offset 4 gives the best switch position. */
6995 switch (posixcc[4]) {
6997 if (memEQ(posixcc, "alph", 4)) /* alpha */
6998 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7001 if (memEQ(posixcc, "spac", 4)) /* space */
7002 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7005 if (memEQ(posixcc, "grap", 4)) /* graph */
7006 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7009 if (memEQ(posixcc, "asci", 4)) /* ascii */
7010 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7013 if (memEQ(posixcc, "blan", 4)) /* blank */
7014 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7017 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7018 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7021 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7022 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7025 if (memEQ(posixcc, "lowe", 4)) /* lower */
7026 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7027 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7028 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7031 if (memEQ(posixcc, "digi", 4)) /* digit */
7032 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7033 else if (memEQ(posixcc, "prin", 4)) /* print */
7034 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7035 else if (memEQ(posixcc, "punc", 4)) /* punct */
7036 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7041 if (memEQ(posixcc, "xdigit", 6))
7042 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7046 if (namedclass == OOB_NAMEDCLASS)
7047 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7049 assert (posixcc[skip] == ':');
7050 assert (posixcc[skip+1] == ']');
7051 } else if (!SIZE_ONLY) {
7052 /* [[=foo=]] and [[.foo.]] are still future. */
7054 /* adjust RExC_parse so the warning shows after
7056 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7058 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7061 /* Maternal grandfather:
7062 * "[:" ending in ":" but not in ":]" */
7072 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7075 if (POSIXCC(UCHARAT(RExC_parse))) {
7076 const char *s = RExC_parse;
7077 const char c = *s++;
7081 if (*s && c == *s && s[1] == ']') {
7082 if (ckWARN(WARN_REGEXP))
7084 "POSIX syntax [%c %c] belongs inside character classes",
7087 /* [[=foo=]] and [[.foo.]] are still future. */
7088 if (POSIXCC_NOTYET(c)) {
7089 /* adjust RExC_parse so the error shows after
7091 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7093 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7100 #define _C_C_T_(NAME,TEST,WORD) \
7103 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7105 for (value = 0; value < 256; value++) \
7107 ANYOF_BITMAP_SET(ret, value); \
7112 case ANYOF_N##NAME: \
7114 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7116 for (value = 0; value < 256; value++) \
7118 ANYOF_BITMAP_SET(ret, value); \
7126 parse a class specification and produce either an ANYOF node that
7127 matches the pattern or if the pattern matches a single char only and
7128 that char is < 256 and we are case insensitive then we produce an
7133 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7136 register UV value = 0;
7137 register UV nextvalue;
7138 register IV prevvalue = OOB_UNICODE;
7139 register IV range = 0;
7140 register regnode *ret;
7143 char *rangebegin = NULL;
7144 bool need_class = 0;
7147 bool optimize_invert = TRUE;
7148 AV* unicode_alternate = NULL;
7150 UV literal_endpoint = 0;
7152 UV stored = 0; /* number of chars stored in the class */
7154 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7155 case we need to change the emitted regop to an EXACT. */
7156 const char * orig_parse = RExC_parse;
7157 GET_RE_DEBUG_FLAGS_DECL;
7159 PERL_UNUSED_ARG(depth);
7162 DEBUG_PARSE("clas");
7164 /* Assume we are going to generate an ANYOF node. */
7165 ret = reganode(pRExC_state, ANYOF, 0);
7168 ANYOF_FLAGS(ret) = 0;
7170 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7174 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7178 RExC_size += ANYOF_SKIP;
7179 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7182 RExC_emit += ANYOF_SKIP;
7184 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7186 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7187 ANYOF_BITMAP_ZERO(ret);
7188 listsv = newSVpvs("# comment\n");
7191 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7193 if (!SIZE_ONLY && POSIXCC(nextvalue))
7194 checkposixcc(pRExC_state);
7196 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7197 if (UCHARAT(RExC_parse) == ']')
7201 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7205 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7208 rangebegin = RExC_parse;
7210 value = utf8n_to_uvchr((U8*)RExC_parse,
7211 RExC_end - RExC_parse,
7212 &numlen, UTF8_ALLOW_DEFAULT);
7213 RExC_parse += numlen;
7216 value = UCHARAT(RExC_parse++);
7218 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7219 if (value == '[' && POSIXCC(nextvalue))
7220 namedclass = regpposixcc(pRExC_state, value);
7221 else if (value == '\\') {
7223 value = utf8n_to_uvchr((U8*)RExC_parse,
7224 RExC_end - RExC_parse,
7225 &numlen, UTF8_ALLOW_DEFAULT);
7226 RExC_parse += numlen;
7229 value = UCHARAT(RExC_parse++);
7230 /* Some compilers cannot handle switching on 64-bit integer
7231 * values, therefore value cannot be an UV. Yes, this will
7232 * be a problem later if we want switch on Unicode.
7233 * A similar issue a little bit later when switching on
7234 * namedclass. --jhi */
7235 switch ((I32)value) {
7236 case 'w': namedclass = ANYOF_ALNUM; break;
7237 case 'W': namedclass = ANYOF_NALNUM; break;
7238 case 's': namedclass = ANYOF_SPACE; break;
7239 case 'S': namedclass = ANYOF_NSPACE; break;
7240 case 'd': namedclass = ANYOF_DIGIT; break;
7241 case 'D': namedclass = ANYOF_NDIGIT; break;
7242 case 'N': /* Handle \N{NAME} in class */
7244 /* We only pay attention to the first char of
7245 multichar strings being returned. I kinda wonder
7246 if this makes sense as it does change the behaviour
7247 from earlier versions, OTOH that behaviour was broken
7249 UV v; /* value is register so we cant & it /grrr */
7250 if (reg_namedseq(pRExC_state, &v)) {
7260 if (RExC_parse >= RExC_end)
7261 vFAIL2("Empty \\%c{}", (U8)value);
7262 if (*RExC_parse == '{') {
7263 const U8 c = (U8)value;
7264 e = strchr(RExC_parse++, '}');
7266 vFAIL2("Missing right brace on \\%c{}", c);
7267 while (isSPACE(UCHARAT(RExC_parse)))
7269 if (e == RExC_parse)
7270 vFAIL2("Empty \\%c{}", c);
7272 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7280 if (UCHARAT(RExC_parse) == '^') {
7283 value = value == 'p' ? 'P' : 'p'; /* toggle */
7284 while (isSPACE(UCHARAT(RExC_parse))) {
7289 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7290 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7293 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7294 namedclass = ANYOF_MAX; /* no official name, but it's named */
7297 case 'n': value = '\n'; break;
7298 case 'r': value = '\r'; break;
7299 case 't': value = '\t'; break;
7300 case 'f': value = '\f'; break;
7301 case 'b': value = '\b'; break;
7302 case 'e': value = ASCII_TO_NATIVE('\033');break;
7303 case 'a': value = ASCII_TO_NATIVE('\007');break;
7305 if (*RExC_parse == '{') {
7306 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7307 | PERL_SCAN_DISALLOW_PREFIX;
7308 char * const e = strchr(RExC_parse++, '}');
7310 vFAIL("Missing right brace on \\x{}");
7312 numlen = e - RExC_parse;
7313 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7317 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7319 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7320 RExC_parse += numlen;
7322 if (PL_encoding && value < 0x100)
7323 goto recode_encoding;
7326 value = UCHARAT(RExC_parse++);
7327 value = toCTRL(value);
7329 case '0': case '1': case '2': case '3': case '4':
7330 case '5': case '6': case '7': case '8': case '9':
7334 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7335 RExC_parse += numlen;
7336 if (PL_encoding && value < 0x100)
7337 goto recode_encoding;
7342 SV* enc = PL_encoding;
7343 value = reg_recode((const char)(U8)value, &enc);
7344 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7346 "Invalid escape in the specified encoding");
7350 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7352 "Unrecognized escape \\%c in character class passed through",
7356 } /* end of \blah */
7362 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7364 if (!SIZE_ONLY && !need_class)
7365 ANYOF_CLASS_ZERO(ret);
7369 /* a bad range like a-\d, a-[:digit:] ? */
7372 if (ckWARN(WARN_REGEXP)) {
7374 RExC_parse >= rangebegin ?
7375 RExC_parse - rangebegin : 0;
7377 "False [] range \"%*.*s\"",
7380 if (prevvalue < 256) {
7381 ANYOF_BITMAP_SET(ret, prevvalue);
7382 ANYOF_BITMAP_SET(ret, '-');
7385 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7386 Perl_sv_catpvf(aTHX_ listsv,
7387 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7391 range = 0; /* this was not a true range */
7397 const char *what = NULL;
7400 if (namedclass > OOB_NAMEDCLASS)
7401 optimize_invert = FALSE;
7402 /* Possible truncation here but in some 64-bit environments
7403 * the compiler gets heartburn about switch on 64-bit values.
7404 * A similar issue a little earlier when switching on value.
7406 switch ((I32)namedclass) {
7407 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7408 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7409 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7410 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7411 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7412 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7413 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7414 case _C_C_T_(PRINT, isPRINT(value), "Print");
7415 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7416 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7417 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7418 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7419 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7422 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7425 for (value = 0; value < 128; value++)
7426 ANYOF_BITMAP_SET(ret, value);
7428 for (value = 0; value < 256; value++) {
7430 ANYOF_BITMAP_SET(ret, value);
7439 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7442 for (value = 128; value < 256; value++)
7443 ANYOF_BITMAP_SET(ret, value);
7445 for (value = 0; value < 256; value++) {
7446 if (!isASCII(value))
7447 ANYOF_BITMAP_SET(ret, value);
7456 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7458 /* consecutive digits assumed */
7459 for (value = '0'; value <= '9'; value++)
7460 ANYOF_BITMAP_SET(ret, value);
7467 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7469 /* consecutive digits assumed */
7470 for (value = 0; value < '0'; value++)
7471 ANYOF_BITMAP_SET(ret, value);
7472 for (value = '9' + 1; value < 256; value++)
7473 ANYOF_BITMAP_SET(ret, value);
7479 /* this is to handle \p and \P */
7482 vFAIL("Invalid [::] class");
7486 /* Strings such as "+utf8::isWord\n" */
7487 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7490 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7493 } /* end of namedclass \blah */
7496 if (prevvalue > (IV)value) /* b-a */ {
7497 const int w = RExC_parse - rangebegin;
7498 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7499 range = 0; /* not a valid range */
7503 prevvalue = value; /* save the beginning of the range */
7504 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7505 RExC_parse[1] != ']') {
7508 /* a bad range like \w-, [:word:]- ? */
7509 if (namedclass > OOB_NAMEDCLASS) {
7510 if (ckWARN(WARN_REGEXP)) {
7512 RExC_parse >= rangebegin ?
7513 RExC_parse - rangebegin : 0;
7515 "False [] range \"%*.*s\"",
7519 ANYOF_BITMAP_SET(ret, '-');
7521 range = 1; /* yeah, it's a range! */
7522 continue; /* but do it the next time */
7526 /* now is the next time */
7527 /*stored += (value - prevvalue + 1);*/
7529 if (prevvalue < 256) {
7530 const IV ceilvalue = value < 256 ? value : 255;
7533 /* In EBCDIC [\x89-\x91] should include
7534 * the \x8e but [i-j] should not. */
7535 if (literal_endpoint == 2 &&
7536 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7537 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7539 if (isLOWER(prevvalue)) {
7540 for (i = prevvalue; i <= ceilvalue; i++)
7542 ANYOF_BITMAP_SET(ret, i);
7544 for (i = prevvalue; i <= ceilvalue; i++)
7546 ANYOF_BITMAP_SET(ret, i);
7551 for (i = prevvalue; i <= ceilvalue; i++) {
7552 if (!ANYOF_BITMAP_TEST(ret,i)) {
7554 ANYOF_BITMAP_SET(ret, i);
7558 if (value > 255 || UTF) {
7559 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7560 const UV natvalue = NATIVE_TO_UNI(value);
7561 stored+=2; /* can't optimize this class */
7562 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7563 if (prevnatvalue < natvalue) { /* what about > ? */
7564 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7565 prevnatvalue, natvalue);
7567 else if (prevnatvalue == natvalue) {
7568 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7570 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7572 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7574 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7575 if (RExC_precomp[0] == ':' &&
7576 RExC_precomp[1] == '[' &&
7577 (f == 0xDF || f == 0x92)) {
7578 f = NATIVE_TO_UNI(f);
7581 /* If folding and foldable and a single
7582 * character, insert also the folded version
7583 * to the charclass. */
7585 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7586 if ((RExC_precomp[0] == ':' &&
7587 RExC_precomp[1] == '[' &&
7589 (value == 0xFB05 || value == 0xFB06))) ?
7590 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7591 foldlen == (STRLEN)UNISKIP(f) )
7593 if (foldlen == (STRLEN)UNISKIP(f))
7595 Perl_sv_catpvf(aTHX_ listsv,
7598 /* Any multicharacter foldings
7599 * require the following transform:
7600 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7601 * where E folds into "pq" and F folds
7602 * into "rst", all other characters
7603 * fold to single characters. We save
7604 * away these multicharacter foldings,
7605 * to be later saved as part of the
7606 * additional "s" data. */
7609 if (!unicode_alternate)
7610 unicode_alternate = newAV();
7611 sv = newSVpvn((char*)foldbuf, foldlen);
7613 av_push(unicode_alternate, sv);
7617 /* If folding and the value is one of the Greek
7618 * sigmas insert a few more sigmas to make the
7619 * folding rules of the sigmas to work right.
7620 * Note that not all the possible combinations
7621 * are handled here: some of them are handled
7622 * by the standard folding rules, and some of
7623 * them (literal or EXACTF cases) are handled
7624 * during runtime in regexec.c:S_find_byclass(). */
7625 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7626 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7627 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7628 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7629 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7631 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7632 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7633 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7638 literal_endpoint = 0;
7642 range = 0; /* this range (if it was one) is done now */
7646 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7648 RExC_size += ANYOF_CLASS_ADD_SKIP;
7650 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7656 /****** !SIZE_ONLY AFTER HERE *********/
7658 if( stored == 1 && value < 256
7659 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7661 /* optimize single char class to an EXACT node
7662 but *only* when its not a UTF/high char */
7663 const char * cur_parse= RExC_parse;
7664 RExC_emit = (regnode *)orig_emit;
7665 RExC_parse = (char *)orig_parse;
7666 ret = reg_node(pRExC_state,
7667 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7668 RExC_parse = (char *)cur_parse;
7669 *STRING(ret)= (char)value;
7671 RExC_emit += STR_SZ(1);
7674 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7675 if ( /* If the only flag is folding (plus possibly inversion). */
7676 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7678 for (value = 0; value < 256; ++value) {
7679 if (ANYOF_BITMAP_TEST(ret, value)) {
7680 UV fold = PL_fold[value];
7683 ANYOF_BITMAP_SET(ret, fold);
7686 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7689 /* optimize inverted simple patterns (e.g. [^a-z]) */
7690 if (optimize_invert &&
7691 /* If the only flag is inversion. */
7692 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7693 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7694 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7695 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7698 AV * const av = newAV();
7700 /* The 0th element stores the character class description
7701 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7702 * to initialize the appropriate swash (which gets stored in
7703 * the 1st element), and also useful for dumping the regnode.
7704 * The 2nd element stores the multicharacter foldings,
7705 * used later (regexec.c:S_reginclass()). */
7706 av_store(av, 0, listsv);
7707 av_store(av, 1, NULL);
7708 av_store(av, 2, (SV*)unicode_alternate);
7709 rv = newRV_noinc((SV*)av);
7710 n = add_data(pRExC_state, 1, "s");
7711 RExC_rxi->data->data[n] = (void*)rv;
7720 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7722 char* const retval = RExC_parse++;
7725 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7726 RExC_parse[2] == '#') {
7727 while (*RExC_parse != ')') {
7728 if (RExC_parse == RExC_end)
7729 FAIL("Sequence (?#... not terminated");
7735 if (RExC_flags & RXf_PMf_EXTENDED) {
7736 if (isSPACE(*RExC_parse)) {
7740 else if (*RExC_parse == '#') {
7741 while (RExC_parse < RExC_end)
7742 if (*RExC_parse++ == '\n') break;
7751 - reg_node - emit a node
7753 STATIC regnode * /* Location. */
7754 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7757 register regnode *ptr;
7758 regnode * const ret = RExC_emit;
7759 GET_RE_DEBUG_FLAGS_DECL;
7762 SIZE_ALIGN(RExC_size);
7767 if (OP(RExC_emit) == 255)
7768 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %s: %d ",
7769 reg_name[op], OP(RExC_emit));
7771 NODE_ALIGN_FILL(ret);
7773 FILL_ADVANCE_NODE(ptr, op);
7774 #ifdef RE_TRACK_PATTERN_OFFSETS
7775 if (RExC_offsets) { /* MJD */
7776 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7777 "reg_node", __LINE__,
7779 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7780 ? "Overwriting end of array!\n" : "OK",
7781 (UV)(RExC_emit - RExC_emit_start),
7782 (UV)(RExC_parse - RExC_start),
7783 (UV)RExC_offsets[0]));
7784 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7792 - reganode - emit a node with an argument
7794 STATIC regnode * /* Location. */
7795 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7798 register regnode *ptr;
7799 regnode * const ret = RExC_emit;
7800 GET_RE_DEBUG_FLAGS_DECL;
7803 SIZE_ALIGN(RExC_size);
7808 assert(2==regarglen[op]+1);
7810 Anything larger than this has to allocate the extra amount.
7811 If we changed this to be:
7813 RExC_size += (1 + regarglen[op]);
7815 then it wouldn't matter. Its not clear what side effect
7816 might come from that so its not done so far.
7822 if (OP(RExC_emit) == 255)
7823 Perl_croak(aTHX_ "panic: reganode overwriting end of allocated program space");
7825 NODE_ALIGN_FILL(ret);
7827 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7828 #ifdef RE_TRACK_PATTERN_OFFSETS
7829 if (RExC_offsets) { /* MJD */
7830 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7834 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7835 "Overwriting end of array!\n" : "OK",
7836 (UV)(RExC_emit - RExC_emit_start),
7837 (UV)(RExC_parse - RExC_start),
7838 (UV)RExC_offsets[0]));
7839 Set_Cur_Node_Offset;
7847 - reguni - emit (if appropriate) a Unicode character
7850 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
7853 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
7857 - reginsert - insert an operator in front of already-emitted operand
7859 * Means relocating the operand.
7862 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
7865 register regnode *src;
7866 register regnode *dst;
7867 register regnode *place;
7868 const int offset = regarglen[(U8)op];
7869 const int size = NODE_STEP_REGNODE + offset;
7870 GET_RE_DEBUG_FLAGS_DECL;
7871 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
7872 DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
7881 if (RExC_open_parens) {
7883 DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);
7884 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
7885 if ( RExC_open_parens[paren] >= opnd ) {
7886 DEBUG_PARSE_FMT("open"," - %d",size);
7887 RExC_open_parens[paren] += size;
7889 DEBUG_PARSE_FMT("open"," - %s","ok");
7891 if ( RExC_close_parens[paren] >= opnd ) {
7892 DEBUG_PARSE_FMT("close"," - %d",size);
7893 RExC_close_parens[paren] += size;
7895 DEBUG_PARSE_FMT("close"," - %s","ok");
7900 while (src > opnd) {
7901 StructCopy(--src, --dst, regnode);
7902 #ifdef RE_TRACK_PATTERN_OFFSETS
7903 if (RExC_offsets) { /* MJD 20010112 */
7904 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
7908 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
7909 ? "Overwriting end of array!\n" : "OK",
7910 (UV)(src - RExC_emit_start),
7911 (UV)(dst - RExC_emit_start),
7912 (UV)RExC_offsets[0]));
7913 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
7914 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
7920 place = opnd; /* Op node, where operand used to be. */
7921 #ifdef RE_TRACK_PATTERN_OFFSETS
7922 if (RExC_offsets) { /* MJD */
7923 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7927 (UV)(place - RExC_emit_start) > RExC_offsets[0]
7928 ? "Overwriting end of array!\n" : "OK",
7929 (UV)(place - RExC_emit_start),
7930 (UV)(RExC_parse - RExC_start),
7931 (UV)RExC_offsets[0]));
7932 Set_Node_Offset(place, RExC_parse);
7933 Set_Node_Length(place, 1);
7936 src = NEXTOPER(place);
7937 FILL_ADVANCE_NODE(place, op);
7938 Zero(src, offset, regnode);
7942 - regtail - set the next-pointer at the end of a node chain of p to val.
7943 - SEE ALSO: regtail_study
7945 /* TODO: All three parms should be const */
7947 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7950 register regnode *scan;
7951 GET_RE_DEBUG_FLAGS_DECL;
7953 PERL_UNUSED_ARG(depth);
7959 /* Find last node. */
7962 regnode * const temp = regnext(scan);
7964 SV * const mysv=sv_newmortal();
7965 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
7966 regprop(RExC_rx, mysv, scan);
7967 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
7968 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
7969 (temp == NULL ? "->" : ""),
7970 (temp == NULL ? reg_name[OP(val)] : "")
7978 if (reg_off_by_arg[OP(scan)]) {
7979 ARG_SET(scan, val - scan);
7982 NEXT_OFF(scan) = val - scan;
7988 - regtail_study - set the next-pointer at the end of a node chain of p to val.
7989 - Look for optimizable sequences at the same time.
7990 - currently only looks for EXACT chains.
7992 This is expermental code. The idea is to use this routine to perform
7993 in place optimizations on branches and groups as they are constructed,
7994 with the long term intention of removing optimization from study_chunk so
7995 that it is purely analytical.
7997 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
7998 to control which is which.
8001 /* TODO: All four parms should be const */
8004 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8007 register regnode *scan;
8009 #ifdef EXPERIMENTAL_INPLACESCAN
8013 GET_RE_DEBUG_FLAGS_DECL;
8019 /* Find last node. */
8023 regnode * const temp = regnext(scan);
8024 #ifdef EXPERIMENTAL_INPLACESCAN
8025 if (PL_regkind[OP(scan)] == EXACT)
8026 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8034 if( exact == PSEUDO )
8036 else if ( exact != OP(scan) )
8045 SV * const mysv=sv_newmortal();
8046 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8047 regprop(RExC_rx, mysv, scan);
8048 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8049 SvPV_nolen_const(mysv),
8058 SV * const mysv_val=sv_newmortal();
8059 DEBUG_PARSE_MSG("");
8060 regprop(RExC_rx, mysv_val, val);
8061 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8062 SvPV_nolen_const(mysv_val),
8063 (IV)REG_NODE_NUM(val),
8067 if (reg_off_by_arg[OP(scan)]) {
8068 ARG_SET(scan, val - scan);
8071 NEXT_OFF(scan) = val - scan;
8079 - regcurly - a little FSA that accepts {\d+,?\d*}
8082 S_regcurly(register const char *s)
8101 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8104 Perl_regdump(pTHX_ const regexp *r)
8108 SV * const sv = sv_newmortal();
8109 SV *dsv= sv_newmortal();
8112 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8114 /* Header fields of interest. */
8115 if (r->anchored_substr) {
8116 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8117 RE_SV_DUMPLEN(r->anchored_substr), 30);
8118 PerlIO_printf(Perl_debug_log,
8119 "anchored %s%s at %"IVdf" ",
8120 s, RE_SV_TAIL(r->anchored_substr),
8121 (IV)r->anchored_offset);
8122 } else if (r->anchored_utf8) {
8123 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8124 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8125 PerlIO_printf(Perl_debug_log,
8126 "anchored utf8 %s%s at %"IVdf" ",
8127 s, RE_SV_TAIL(r->anchored_utf8),
8128 (IV)r->anchored_offset);
8130 if (r->float_substr) {
8131 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8132 RE_SV_DUMPLEN(r->float_substr), 30);
8133 PerlIO_printf(Perl_debug_log,
8134 "floating %s%s at %"IVdf"..%"UVuf" ",
8135 s, RE_SV_TAIL(r->float_substr),
8136 (IV)r->float_min_offset, (UV)r->float_max_offset);
8137 } else if (r->float_utf8) {
8138 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8139 RE_SV_DUMPLEN(r->float_utf8), 30);
8140 PerlIO_printf(Perl_debug_log,
8141 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8142 s, RE_SV_TAIL(r->float_utf8),
8143 (IV)r->float_min_offset, (UV)r->float_max_offset);
8145 if (r->check_substr || r->check_utf8)
8146 PerlIO_printf(Perl_debug_log,
8148 (r->check_substr == r->float_substr
8149 && r->check_utf8 == r->float_utf8
8150 ? "(checking floating" : "(checking anchored"));
8151 if (r->extflags & RXf_NOSCAN)
8152 PerlIO_printf(Perl_debug_log, " noscan");
8153 if (r->extflags & RXf_CHECK_ALL)
8154 PerlIO_printf(Perl_debug_log, " isall");
8155 if (r->check_substr || r->check_utf8)
8156 PerlIO_printf(Perl_debug_log, ") ");
8158 if (ri->regstclass) {
8159 regprop(r, sv, ri->regstclass);
8160 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8162 if (r->extflags & RXf_ANCH) {
8163 PerlIO_printf(Perl_debug_log, "anchored");
8164 if (r->extflags & RXf_ANCH_BOL)
8165 PerlIO_printf(Perl_debug_log, "(BOL)");
8166 if (r->extflags & RXf_ANCH_MBOL)
8167 PerlIO_printf(Perl_debug_log, "(MBOL)");
8168 if (r->extflags & RXf_ANCH_SBOL)
8169 PerlIO_printf(Perl_debug_log, "(SBOL)");
8170 if (r->extflags & RXf_ANCH_GPOS)
8171 PerlIO_printf(Perl_debug_log, "(GPOS)");
8172 PerlIO_putc(Perl_debug_log, ' ');
8174 if (r->extflags & RXf_GPOS_SEEN)
8175 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8176 if (r->intflags & PREGf_SKIP)
8177 PerlIO_printf(Perl_debug_log, "plus ");
8178 if (r->intflags & PREGf_IMPLICIT)
8179 PerlIO_printf(Perl_debug_log, "implicit ");
8180 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8181 if (r->extflags & RXf_EVAL_SEEN)
8182 PerlIO_printf(Perl_debug_log, "with eval ");
8183 PerlIO_printf(Perl_debug_log, "\n");
8185 PERL_UNUSED_CONTEXT;
8187 #endif /* DEBUGGING */
8191 - regprop - printable representation of opcode
8194 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8199 RXi_GET_DECL(prog,progi);
8200 GET_RE_DEBUG_FLAGS_DECL;
8203 sv_setpvn(sv, "", 0);
8205 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8206 /* It would be nice to FAIL() here, but this may be called from
8207 regexec.c, and it would be hard to supply pRExC_state. */
8208 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8209 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
8211 k = PL_regkind[OP(o)];
8214 SV * const dsv = sv_2mortal(newSVpvs(""));
8215 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8216 * is a crude hack but it may be the best for now since
8217 * we have no flag "this EXACTish node was UTF-8"
8219 const char * const s =
8220 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8221 PL_colors[0], PL_colors[1],
8222 PERL_PV_ESCAPE_UNI_DETECT |
8223 PERL_PV_PRETTY_ELIPSES |
8226 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8227 } else if (k == TRIE) {
8228 /* print the details of the trie in dumpuntil instead, as
8229 * progi->data isn't available here */
8230 const char op = OP(o);
8231 const U32 n = ARG(o);
8232 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8233 (reg_ac_data *)progi->data->data[n] :
8235 const reg_trie_data * const trie
8236 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8238 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
8239 DEBUG_TRIE_COMPILE_r(
8240 Perl_sv_catpvf(aTHX_ sv,
8241 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8242 (UV)trie->startstate,
8243 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8244 (UV)trie->wordcount,
8247 (UV)TRIE_CHARCOUNT(trie),
8248 (UV)trie->uniquecharcount
8251 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8253 int rangestart = -1;
8254 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8255 Perl_sv_catpvf(aTHX_ sv, "[");
8256 for (i = 0; i <= 256; i++) {
8257 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8258 if (rangestart == -1)
8260 } else if (rangestart != -1) {
8261 if (i <= rangestart + 3)
8262 for (; rangestart < i; rangestart++)
8263 put_byte(sv, rangestart);
8265 put_byte(sv, rangestart);
8267 put_byte(sv, i - 1);
8272 Perl_sv_catpvf(aTHX_ sv, "]");
8275 } else if (k == CURLY) {
8276 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8277 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8278 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8280 else if (k == WHILEM && o->flags) /* Ordinal/of */
8281 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8282 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8283 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8284 if ( prog->paren_names ) {
8285 if ( k != REF || OP(o) < NREF) {
8286 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8287 SV **name= av_fetch(list, ARG(o), 0 );
8289 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8292 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8293 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8294 I32 *nums=(I32*)SvPVX(sv_dat);
8295 SV **name= av_fetch(list, nums[0], 0 );
8298 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8299 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8300 (n ? "," : ""), (IV)nums[n]);
8302 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8306 } else if (k == GOSUB)
8307 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8308 else if (k == VERB) {
8310 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8311 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8312 } else if (k == LOGICAL)
8313 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8314 else if (k == ANYOF) {
8315 int i, rangestart = -1;
8316 const U8 flags = ANYOF_FLAGS(o);
8318 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8319 static const char * const anyofs[] = {
8352 if (flags & ANYOF_LOCALE)
8353 sv_catpvs(sv, "{loc}");
8354 if (flags & ANYOF_FOLD)
8355 sv_catpvs(sv, "{i}");
8356 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8357 if (flags & ANYOF_INVERT)
8359 for (i = 0; i <= 256; i++) {
8360 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8361 if (rangestart == -1)
8363 } else if (rangestart != -1) {
8364 if (i <= rangestart + 3)
8365 for (; rangestart < i; rangestart++)
8366 put_byte(sv, rangestart);
8368 put_byte(sv, rangestart);
8370 put_byte(sv, i - 1);
8376 if (o->flags & ANYOF_CLASS)
8377 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8378 if (ANYOF_CLASS_TEST(o,i))
8379 sv_catpv(sv, anyofs[i]);
8381 if (flags & ANYOF_UNICODE)
8382 sv_catpvs(sv, "{unicode}");
8383 else if (flags & ANYOF_UNICODE_ALL)
8384 sv_catpvs(sv, "{unicode_all}");
8388 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8392 U8 s[UTF8_MAXBYTES_CASE+1];
8394 for (i = 0; i <= 256; i++) { /* just the first 256 */
8395 uvchr_to_utf8(s, i);
8397 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8398 if (rangestart == -1)
8400 } else if (rangestart != -1) {
8401 if (i <= rangestart + 3)
8402 for (; rangestart < i; rangestart++) {
8403 const U8 * const e = uvchr_to_utf8(s,rangestart);
8405 for(p = s; p < e; p++)
8409 const U8 *e = uvchr_to_utf8(s,rangestart);
8411 for (p = s; p < e; p++)
8414 e = uvchr_to_utf8(s, i-1);
8415 for (p = s; p < e; p++)
8422 sv_catpvs(sv, "..."); /* et cetera */
8426 char *s = savesvpv(lv);
8427 char * const origs = s;
8429 while (*s && *s != '\n')
8433 const char * const t = ++s;
8451 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8453 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8454 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8456 PERL_UNUSED_CONTEXT;
8457 PERL_UNUSED_ARG(sv);
8459 PERL_UNUSED_ARG(prog);
8460 #endif /* DEBUGGING */
8464 Perl_re_intuit_string(pTHX_ regexp *prog)
8465 { /* Assume that RE_INTUIT is set */
8467 GET_RE_DEBUG_FLAGS_DECL;
8468 PERL_UNUSED_CONTEXT;
8472 const char * const s = SvPV_nolen_const(prog->check_substr
8473 ? prog->check_substr : prog->check_utf8);
8475 if (!PL_colorset) reginitcolors();
8476 PerlIO_printf(Perl_debug_log,
8477 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8479 prog->check_substr ? "" : "utf8 ",
8480 PL_colors[5],PL_colors[0],
8483 (strlen(s) > 60 ? "..." : ""));
8486 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8492 handles refcounting and freeing the perl core regexp structure. When
8493 it is necessary to actually free the structure the first thing it
8494 does is call the 'free' method of the regexp_engine associated to to
8495 the regexp, allowing the handling of the void *pprivate; member
8496 first. (This routine is not overridable by extensions, which is why
8497 the extensions free is called first.)
8499 See regdupe and regdupe_internal if you change anything here.
8501 #ifndef PERL_IN_XSUB_RE
8503 Perl_pregfree(pTHX_ struct regexp *r)
8506 GET_RE_DEBUG_FLAGS_DECL;
8508 if (!r || (--r->refcnt > 0))
8511 CALLREGFREE_PVT(r); /* free the private data */
8513 /* gcov results gave these as non-null 100% of the time, so there's no
8514 optimisation in checking them before calling Safefree */
8515 Safefree(r->precomp);
8516 RX_MATCH_COPY_FREE(r);
8517 #ifdef PERL_OLD_COPY_ON_WRITE
8519 SvREFCNT_dec(r->saved_copy);
8522 if (r->anchored_substr)
8523 SvREFCNT_dec(r->anchored_substr);
8524 if (r->anchored_utf8)
8525 SvREFCNT_dec(r->anchored_utf8);
8526 if (r->float_substr)
8527 SvREFCNT_dec(r->float_substr);
8529 SvREFCNT_dec(r->float_utf8);
8530 Safefree(r->substrs);
8533 SvREFCNT_dec(r->paren_names);
8535 Safefree(r->startp);
8541 /* regfree_internal()
8543 Free the private data in a regexp. This is overloadable by
8544 extensions. Perl takes care of the regexp structure in pregfree(),
8545 this covers the *pprivate pointer which technically perldoesnt
8546 know about, however of course we have to handle the
8547 regexp_internal structure when no extension is in use.
8549 Note this is called before freeing anything in the regexp
8554 Perl_regfree_internal(pTHX_ struct regexp *r)
8558 GET_RE_DEBUG_FLAGS_DECL;
8564 SV *dsv= sv_newmortal();
8565 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8566 dsv, r->precomp, r->prelen, 60);
8567 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8568 PL_colors[4],PL_colors[5],s);
8571 #ifdef RE_TRACK_PATTERN_OFFSETS
8573 Safefree(ri->u.offsets); /* 20010421 MJD */
8576 int n = ri->data->count;
8577 PAD* new_comppad = NULL;
8582 /* If you add a ->what type here, update the comment in regcomp.h */
8583 switch (ri->data->what[n]) {
8587 SvREFCNT_dec((SV*)ri->data->data[n]);
8590 Safefree(ri->data->data[n]);
8593 new_comppad = (AV*)ri->data->data[n];
8596 if (new_comppad == NULL)
8597 Perl_croak(aTHX_ "panic: pregfree comppad");
8598 PAD_SAVE_LOCAL(old_comppad,
8599 /* Watch out for global destruction's random ordering. */
8600 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8603 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8606 op_free((OP_4tree*)ri->data->data[n]);
8608 PAD_RESTORE_LOCAL(old_comppad);
8609 SvREFCNT_dec((SV*)new_comppad);
8615 { /* Aho Corasick add-on structure for a trie node.
8616 Used in stclass optimization only */
8618 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8620 refcount = --aho->refcount;
8623 PerlMemShared_free(aho->states);
8624 PerlMemShared_free(aho->fail);
8625 /* do this last!!!! */
8626 PerlMemShared_free(ri->data->data[n]);
8627 PerlMemShared_free(ri->regstclass);
8633 /* trie structure. */
8635 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8637 refcount = --trie->refcount;
8640 PerlMemShared_free(trie->charmap);
8641 PerlMemShared_free(trie->states);
8642 PerlMemShared_free(trie->trans);
8644 PerlMemShared_free(trie->bitmap);
8646 PerlMemShared_free(trie->wordlen);
8648 PerlMemShared_free(trie->jump);
8650 PerlMemShared_free(trie->nextword);
8651 /* do this last!!!! */
8652 PerlMemShared_free(ri->data->data[n]);
8657 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8660 Safefree(ri->data->what);
8664 Safefree(ri->swap->startp);
8665 Safefree(ri->swap->endp);
8671 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8672 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8673 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8674 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8677 regdupe - duplicate a regexp.
8679 This routine is called by sv.c's re_dup and is expected to clone a
8680 given regexp structure. It is a no-op when not under USE_ITHREADS.
8681 (Originally this *was* re_dup() for change history see sv.c)
8683 After all of the core data stored in struct regexp is duplicated
8684 the regexp_engine.dupe method is used to copy any private data
8685 stored in the *pprivate pointer. This allows extensions to handle
8686 any duplication it needs to do.
8688 See pregfree() and regfree_internal() if you change anything here.
8690 #if defined(USE_ITHREADS)
8691 #ifndef PERL_IN_XSUB_RE
8693 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
8698 struct reg_substr_datum *s;
8701 return (REGEXP *)NULL;
8703 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8707 npar = r->nparens+1;
8708 Newxz(ret, 1, regexp);
8709 Newx(ret->startp, npar, I32);
8710 Copy(r->startp, ret->startp, npar, I32);
8711 Newx(ret->endp, npar, I32);
8712 Copy(r->endp, ret->endp, npar, I32);
8715 Newx(ret->substrs, 1, struct reg_substr_data);
8716 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8717 s->min_offset = r->substrs->data[i].min_offset;
8718 s->max_offset = r->substrs->data[i].max_offset;
8719 s->end_shift = r->substrs->data[i].end_shift;
8720 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8721 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8724 ret->substrs = NULL;
8726 ret->precomp = SAVEPVN(r->precomp, r->prelen);
8727 ret->refcnt = r->refcnt;
8728 ret->minlen = r->minlen;
8729 ret->minlenret = r->minlenret;
8730 ret->prelen = r->prelen;
8731 ret->nparens = r->nparens;
8732 ret->lastparen = r->lastparen;
8733 ret->lastcloseparen = r->lastcloseparen;
8734 ret->intflags = r->intflags;
8735 ret->extflags = r->extflags;
8737 ret->sublen = r->sublen;
8739 ret->engine = r->engine;
8741 ret->paren_names = hv_dup_inc(r->paren_names, param);
8743 if (RX_MATCH_COPIED(ret))
8744 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8747 #ifdef PERL_OLD_COPY_ON_WRITE
8748 ret->saved_copy = NULL;
8751 ret->pprivate = r->pprivate;
8753 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
8755 ptr_table_store(PL_ptr_table, r, ret);
8758 #endif /* PERL_IN_XSUB_RE */
8763 This is the internal complement to regdupe() which is used to copy
8764 the structure pointed to by the *pprivate pointer in the regexp.
8765 This is the core version of the extension overridable cloning hook.
8766 The regexp structure being duplicated will be copied by perl prior
8767 to this and will be provided as the regexp *r argument, however
8768 with the /old/ structures pprivate pointer value. Thus this routine
8769 may override any copying normally done by perl.
8771 It returns a pointer to the new regexp_internal structure.
8775 Perl_regdupe_internal(pTHX_ const regexp *r, CLONE_PARAMS *param)
8778 regexp_internal *reti;
8782 npar = r->nparens+1;
8785 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8786 Copy(ri->program, reti->program, len+1, regnode);
8789 Newx(reti->swap, 1, regexp_paren_ofs);
8790 /* no need to copy these */
8791 Newx(reti->swap->startp, npar, I32);
8792 Newx(reti->swap->endp, npar, I32);
8798 reti->regstclass = NULL;
8801 const int count = ri->data->count;
8804 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8805 char, struct reg_data);
8806 Newx(d->what, count, U8);
8809 for (i = 0; i < count; i++) {
8810 d->what[i] = ri->data->what[i];
8811 switch (d->what[i]) {
8812 /* legal options are one of: sSfpontTu
8813 see also regcomp.h and pregfree() */
8816 case 'p': /* actually an AV, but the dup function is identical. */
8817 case 'u': /* actually an HV, but the dup function is identical. */
8818 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
8821 /* This is cheating. */
8822 Newx(d->data[i], 1, struct regnode_charclass_class);
8823 StructCopy(ri->data->data[i], d->data[i],
8824 struct regnode_charclass_class);
8825 reti->regstclass = (regnode*)d->data[i];
8828 /* Compiled op trees are readonly and in shared memory,
8829 and can thus be shared without duplication. */
8831 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
8835 /* Trie stclasses are readonly and can thus be shared
8836 * without duplication. We free the stclass in pregfree
8837 * when the corresponding reg_ac_data struct is freed.
8839 reti->regstclass= ri->regstclass;
8843 ((reg_trie_data*)ri->data->data[i])->refcount++;
8847 d->data[i] = ri->data->data[i];
8850 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
8859 reti->name_list_idx = ri->name_list_idx;
8861 #ifdef RE_TRACK_PATTERN_OFFSETS
8862 if (ri->u.offsets) {
8863 Newx(reti->u.offsets, 2*len+1, U32);
8864 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
8867 SetProgLen(reti,len);
8873 #endif /* USE_ITHREADS */
8878 converts a regexp embedded in a MAGIC struct to its stringified form,
8879 caching the converted form in the struct and returns the cached
8882 If lp is nonnull then it is used to return the length of the
8885 If flags is nonnull and the returned string contains UTF8 then
8886 (*flags & 1) will be true.
8888 If haseval is nonnull then it is used to return whether the pattern
8891 Normally called via macro:
8893 CALLREG_STRINGIFY(mg,&len,&utf8);
8897 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
8899 See sv_2pv_flags() in sv.c for an example of internal usage.
8902 #ifndef PERL_IN_XSUB_RE
8904 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
8906 const regexp * const re = (regexp *)mg->mg_obj;
8909 const char *fptr = STD_PAT_MODS; /*"msix"*/
8912 bool hask = ((re->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
8913 bool hasm = ((re->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
8914 U16 reganch = (U16)((re->extflags & RXf_PMf_STD_PMMOD) >> 12);
8915 bool need_newline = 0;
8917 int right = 4 + hask;
8919 reflags[left++] = KEEPCOPY_PAT_MOD; /*'k'*/
8920 while((ch = *fptr++)) {
8922 reflags[left++] = ch;
8925 reflags[right--] = ch;
8930 reflags[left] = '-';
8933 /* printf("[%*.7s]\n",left,reflags); */
8934 mg->mg_len = re->prelen + 4 + left;
8936 * If /x was used, we have to worry about a regex ending with a
8937 * comment later being embedded within another regex. If so, we don't
8938 * want this regex's "commentization" to leak out to the right part of
8939 * the enclosing regex, we must cap it with a newline.
8941 * So, if /x was used, we scan backwards from the end of the regex. If
8942 * we find a '#' before we find a newline, we need to add a newline
8943 * ourself. If we find a '\n' first (or if we don't find '#' or '\n'),
8944 * we don't need to add anything. -jfriedl
8946 if (PMf_EXTENDED & re->extflags) {
8947 const char *endptr = re->precomp + re->prelen;
8948 while (endptr >= re->precomp) {
8949 const char c = *(endptr--);
8951 break; /* don't need another */
8953 /* we end while in a comment, so we need a newline */
8954 mg->mg_len++; /* save space for it */
8955 need_newline = 1; /* note to add it */
8961 Newx(mg->mg_ptr, mg->mg_len + 1 + left, char);
8962 mg->mg_ptr[0] = '(';
8963 mg->mg_ptr[1] = '?';
8964 Copy(reflags, mg->mg_ptr+2, left, char);
8965 *(mg->mg_ptr+left+2) = ':';
8966 Copy(re->precomp, mg->mg_ptr+3+left, re->prelen, char);
8968 mg->mg_ptr[mg->mg_len - 2] = '\n';
8969 mg->mg_ptr[mg->mg_len - 1] = ')';
8970 mg->mg_ptr[mg->mg_len] = 0;
8973 *haseval = re->seen_evals;
8975 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
8983 - regnext - dig the "next" pointer out of a node
8986 Perl_regnext(pTHX_ register regnode *p)
8989 register I32 offset;
8994 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9003 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9006 STRLEN l1 = strlen(pat1);
9007 STRLEN l2 = strlen(pat2);
9010 const char *message;
9016 Copy(pat1, buf, l1 , char);
9017 Copy(pat2, buf + l1, l2 , char);
9018 buf[l1 + l2] = '\n';
9019 buf[l1 + l2 + 1] = '\0';
9021 /* ANSI variant takes additional second argument */
9022 va_start(args, pat2);
9026 msv = vmess(buf, &args);
9028 message = SvPV_const(msv,l1);
9031 Copy(message, buf, l1 , char);
9032 buf[l1-1] = '\0'; /* Overwrite \n */
9033 Perl_croak(aTHX_ "%s", buf);
9036 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9038 #ifndef PERL_IN_XSUB_RE
9040 Perl_save_re_context(pTHX)
9044 struct re_save_state *state;
9046 SAVEVPTR(PL_curcop);
9047 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9049 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9050 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9051 SSPUSHINT(SAVEt_RE_STATE);
9053 Copy(&PL_reg_state, state, 1, struct re_save_state);
9055 PL_reg_start_tmp = 0;
9056 PL_reg_start_tmpl = 0;
9057 PL_reg_oldsaved = NULL;
9058 PL_reg_oldsavedlen = 0;
9060 PL_reg_leftiter = 0;
9061 PL_reg_poscache = NULL;
9062 PL_reg_poscache_size = 0;
9063 #ifdef PERL_OLD_COPY_ON_WRITE
9067 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9069 const REGEXP * const rx = PM_GETRE(PL_curpm);
9072 for (i = 1; i <= rx->nparens; i++) {
9073 char digits[TYPE_CHARS(long)];
9074 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9075 GV *const *const gvp
9076 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9079 GV * const gv = *gvp;
9080 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9090 clear_re(pTHX_ void *r)
9093 ReREFCNT_dec((regexp *)r);
9099 S_put_byte(pTHX_ SV *sv, int c)
9101 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9102 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9103 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9104 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9106 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9110 #define CLEAR_OPTSTART \
9111 if (optstart) STMT_START { \
9112 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9116 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9118 STATIC const regnode *
9119 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9120 const regnode *last, const regnode *plast,
9121 SV* sv, I32 indent, U32 depth)
9124 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9125 register const regnode *next;
9126 const regnode *optstart= NULL;
9129 GET_RE_DEBUG_FLAGS_DECL;
9131 #ifdef DEBUG_DUMPUNTIL
9132 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9133 last ? last-start : 0,plast ? plast-start : 0);
9136 if (plast && plast < last)
9139 while (PL_regkind[op] != END && (!last || node < last)) {
9140 /* While that wasn't END last time... */
9143 if (op == CLOSE || op == WHILEM)
9145 next = regnext((regnode *)node);
9148 if (OP(node) == OPTIMIZED) {
9149 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9156 regprop(r, sv, node);
9157 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9158 (int)(2*indent + 1), "", SvPVX_const(sv));
9160 if (OP(node) != OPTIMIZED) {
9161 if (next == NULL) /* Next ptr. */
9162 PerlIO_printf(Perl_debug_log, " (0)");
9163 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9164 PerlIO_printf(Perl_debug_log, " (FAIL)");
9166 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9167 (void)PerlIO_putc(Perl_debug_log, '\n');
9171 if (PL_regkind[(U8)op] == BRANCHJ) {
9174 register const regnode *nnode = (OP(next) == LONGJMP
9175 ? regnext((regnode *)next)
9177 if (last && nnode > last)
9179 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9182 else if (PL_regkind[(U8)op] == BRANCH) {
9184 DUMPUNTIL(NEXTOPER(node), next);
9186 else if ( PL_regkind[(U8)op] == TRIE ) {
9187 const regnode *this_trie = node;
9188 const char op = OP(node);
9189 const U32 n = ARG(node);
9190 const reg_ac_data * const ac = op>=AHOCORASICK ?
9191 (reg_ac_data *)ri->data->data[n] :
9193 const reg_trie_data * const trie =
9194 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9196 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9198 const regnode *nextbranch= NULL;
9200 sv_setpvn(sv, "", 0);
9201 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9202 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9204 PerlIO_printf(Perl_debug_log, "%*s%s ",
9205 (int)(2*(indent+3)), "",
9206 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9207 PL_colors[0], PL_colors[1],
9208 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9209 PERL_PV_PRETTY_ELIPSES |
9215 U16 dist= trie->jump[word_idx+1];
9216 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9217 (UV)((dist ? this_trie + dist : next) - start));
9220 nextbranch= this_trie + trie->jump[0];
9221 DUMPUNTIL(this_trie + dist, nextbranch);
9223 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9224 nextbranch= regnext((regnode *)nextbranch);
9226 PerlIO_printf(Perl_debug_log, "\n");
9229 if (last && next > last)
9234 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9235 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9236 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9238 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9240 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9242 else if ( op == PLUS || op == STAR) {
9243 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9245 else if (op == ANYOF) {
9246 /* arglen 1 + class block */
9247 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9248 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9249 node = NEXTOPER(node);
9251 else if (PL_regkind[(U8)op] == EXACT) {
9252 /* Literal string, where present. */
9253 node += NODE_SZ_STR(node) - 1;
9254 node = NEXTOPER(node);
9257 node = NEXTOPER(node);
9258 node += regarglen[(U8)op];
9260 if (op == CURLYX || op == OPEN)
9264 #ifdef DEBUG_DUMPUNTIL
9265 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9270 #endif /* DEBUGGING */
9274 * c-indentation-style: bsd
9276 * indent-tabs-mode: t
9279 * ex: set ts=8 sts=4 sw=4 noet: