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_bound; /* First regnode outside of the allocated space */
113 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
114 I32 naughty; /* How bad is this pattern? */
115 I32 sawback; /* Did we see \1, ...? */
117 I32 size; /* Code size. */
118 I32 npar; /* Capture buffer count, (OPEN). */
119 I32 cpar; /* Capture buffer count, (CLOSE). */
120 I32 nestroot; /* root parens we are in - used by accept */
124 regnode **open_parens; /* pointers to open parens */
125 regnode **close_parens; /* pointers to close parens */
126 regnode *opend; /* END node in program */
127 I32 utf8; /* whether the pattern is utf8 or not */
128 I32 orig_utf8; /* whether the pattern was originally in utf8 */
129 /* XXX use this for future optimisation of case
130 * where pattern must be upgraded to utf8. */
131 HV *charnames; /* cache of named sequences */
132 HV *paren_names; /* Paren names */
134 regnode **recurse; /* Recurse regops */
135 I32 recurse_count; /* Number of recurse regops */
137 char *starttry; /* -Dr: where regtry was called. */
138 #define RExC_starttry (pRExC_state->starttry)
141 const char *lastparse;
143 AV *paren_name_list; /* idx -> name */
144 #define RExC_lastparse (pRExC_state->lastparse)
145 #define RExC_lastnum (pRExC_state->lastnum)
146 #define RExC_paren_name_list (pRExC_state->paren_name_list)
150 #define RExC_flags (pRExC_state->flags)
151 #define RExC_precomp (pRExC_state->precomp)
152 #define RExC_rx (pRExC_state->rx)
153 #define RExC_rxi (pRExC_state->rxi)
154 #define RExC_start (pRExC_state->start)
155 #define RExC_end (pRExC_state->end)
156 #define RExC_parse (pRExC_state->parse)
157 #define RExC_whilem_seen (pRExC_state->whilem_seen)
158 #ifdef RE_TRACK_PATTERN_OFFSETS
159 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
161 #define RExC_emit (pRExC_state->emit)
162 #define RExC_emit_start (pRExC_state->emit_start)
163 #define RExC_emit_bound (pRExC_state->emit_bound)
164 #define RExC_naughty (pRExC_state->naughty)
165 #define RExC_sawback (pRExC_state->sawback)
166 #define RExC_seen (pRExC_state->seen)
167 #define RExC_size (pRExC_state->size)
168 #define RExC_npar (pRExC_state->npar)
169 #define RExC_nestroot (pRExC_state->nestroot)
170 #define RExC_extralen (pRExC_state->extralen)
171 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
172 #define RExC_seen_evals (pRExC_state->seen_evals)
173 #define RExC_utf8 (pRExC_state->utf8)
174 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
175 #define RExC_charnames (pRExC_state->charnames)
176 #define RExC_open_parens (pRExC_state->open_parens)
177 #define RExC_close_parens (pRExC_state->close_parens)
178 #define RExC_opend (pRExC_state->opend)
179 #define RExC_paren_names (pRExC_state->paren_names)
180 #define RExC_recurse (pRExC_state->recurse)
181 #define RExC_recurse_count (pRExC_state->recurse_count)
184 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
185 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
186 ((*s) == '{' && regcurly(s)))
189 #undef SPSTART /* dratted cpp namespace... */
192 * Flags to be passed up and down.
194 #define WORST 0 /* Worst case. */
195 #define HASWIDTH 0x01 /* Known to match non-null strings. */
196 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
197 #define SPSTART 0x04 /* Starts with * or +. */
198 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
199 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
201 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
203 /* whether trie related optimizations are enabled */
204 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
205 #define TRIE_STUDY_OPT
206 #define FULL_TRIE_STUDY
212 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
213 #define PBITVAL(paren) (1 << ((paren) & 7))
214 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
215 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
216 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
219 /* About scan_data_t.
221 During optimisation we recurse through the regexp program performing
222 various inplace (keyhole style) optimisations. In addition study_chunk
223 and scan_commit populate this data structure with information about
224 what strings MUST appear in the pattern. We look for the longest
225 string that must appear for at a fixed location, and we look for the
226 longest string that may appear at a floating location. So for instance
231 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
232 strings (because they follow a .* construct). study_chunk will identify
233 both FOO and BAR as being the longest fixed and floating strings respectively.
235 The strings can be composites, for instance
239 will result in a composite fixed substring 'foo'.
241 For each string some basic information is maintained:
243 - offset or min_offset
244 This is the position the string must appear at, or not before.
245 It also implicitly (when combined with minlenp) tells us how many
246 character must match before the string we are searching.
247 Likewise when combined with minlenp and the length of the string
248 tells us how many characters must appear after the string we have
252 Only used for floating strings. This is the rightmost point that
253 the string can appear at. Ifset to I32 max it indicates that the
254 string can occur infinitely far to the right.
257 A pointer to the minimum length of the pattern that the string
258 was found inside. This is important as in the case of positive
259 lookahead or positive lookbehind we can have multiple patterns
264 The minimum length of the pattern overall is 3, the minimum length
265 of the lookahead part is 3, but the minimum length of the part that
266 will actually match is 1. So 'FOO's minimum length is 3, but the
267 minimum length for the F is 1. This is important as the minimum length
268 is used to determine offsets in front of and behind the string being
269 looked for. Since strings can be composites this is the length of the
270 pattern at the time it was commited with a scan_commit. Note that
271 the length is calculated by study_chunk, so that the minimum lengths
272 are not known until the full pattern has been compiled, thus the
273 pointer to the value.
277 In the case of lookbehind the string being searched for can be
278 offset past the start point of the final matching string.
279 If this value was just blithely removed from the min_offset it would
280 invalidate some of the calculations for how many chars must match
281 before or after (as they are derived from min_offset and minlen and
282 the length of the string being searched for).
283 When the final pattern is compiled and the data is moved from the
284 scan_data_t structure into the regexp structure the information
285 about lookbehind is factored in, with the information that would
286 have been lost precalculated in the end_shift field for the
289 The fields pos_min and pos_delta are used to store the minimum offset
290 and the delta to the maximum offset at the current point in the pattern.
294 typedef struct scan_data_t {
295 /*I32 len_min; unused */
296 /*I32 len_delta; unused */
300 I32 last_end; /* min value, <0 unless valid. */
303 SV **longest; /* Either &l_fixed, or &l_float. */
304 SV *longest_fixed; /* longest fixed string found in pattern */
305 I32 offset_fixed; /* offset where it starts */
306 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
307 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
308 SV *longest_float; /* longest floating string found in pattern */
309 I32 offset_float_min; /* earliest point in string it can appear */
310 I32 offset_float_max; /* latest point in string it can appear */
311 I32 *minlen_float; /* pointer to the minlen relevent to the string */
312 I32 lookbehind_float; /* is the position of the string modified by LB */
316 struct regnode_charclass_class *start_class;
320 * Forward declarations for pregcomp()'s friends.
323 static const scan_data_t zero_scan_data =
324 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
326 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
327 #define SF_BEFORE_SEOL 0x0001
328 #define SF_BEFORE_MEOL 0x0002
329 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
330 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
333 # define SF_FIX_SHIFT_EOL (0+2)
334 # define SF_FL_SHIFT_EOL (0+4)
336 # define SF_FIX_SHIFT_EOL (+2)
337 # define SF_FL_SHIFT_EOL (+4)
340 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
341 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
343 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
344 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
345 #define SF_IS_INF 0x0040
346 #define SF_HAS_PAR 0x0080
347 #define SF_IN_PAR 0x0100
348 #define SF_HAS_EVAL 0x0200
349 #define SCF_DO_SUBSTR 0x0400
350 #define SCF_DO_STCLASS_AND 0x0800
351 #define SCF_DO_STCLASS_OR 0x1000
352 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
353 #define SCF_WHILEM_VISITED_POS 0x2000
355 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
356 #define SCF_SEEN_ACCEPT 0x8000
358 #define UTF (RExC_utf8 != 0)
359 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
360 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
362 #define OOB_UNICODE 12345678
363 #define OOB_NAMEDCLASS -1
365 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
366 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
369 /* length of regex to show in messages that don't mark a position within */
370 #define RegexLengthToShowInErrorMessages 127
373 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
374 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
375 * op/pragma/warn/regcomp.
377 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
378 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
380 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
383 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
384 * arg. Show regex, up to a maximum length. If it's too long, chop and add
387 #define _FAIL(code) STMT_START { \
388 const char *ellipses = ""; \
389 IV len = RExC_end - RExC_precomp; \
392 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
393 if (len > RegexLengthToShowInErrorMessages) { \
394 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
395 len = RegexLengthToShowInErrorMessages - 10; \
401 #define FAIL(msg) _FAIL( \
402 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
403 msg, (int)len, RExC_precomp, ellipses))
405 #define FAIL2(msg,arg) _FAIL( \
406 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
407 arg, (int)len, RExC_precomp, ellipses))
410 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
412 #define Simple_vFAIL(m) STMT_START { \
413 const IV offset = RExC_parse - RExC_precomp; \
414 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
415 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
419 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
421 #define vFAIL(m) STMT_START { \
423 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
428 * Like Simple_vFAIL(), but accepts two arguments.
430 #define Simple_vFAIL2(m,a1) STMT_START { \
431 const IV offset = RExC_parse - RExC_precomp; \
432 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
433 (int)offset, RExC_precomp, RExC_precomp + offset); \
437 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
439 #define vFAIL2(m,a1) STMT_START { \
441 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
442 Simple_vFAIL2(m, a1); \
447 * Like Simple_vFAIL(), but accepts three arguments.
449 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
450 const IV offset = RExC_parse - RExC_precomp; \
451 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
452 (int)offset, RExC_precomp, RExC_precomp + offset); \
456 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
458 #define vFAIL3(m,a1,a2) STMT_START { \
460 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
461 Simple_vFAIL3(m, a1, a2); \
465 * Like Simple_vFAIL(), but accepts four arguments.
467 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
468 const IV offset = RExC_parse - RExC_precomp; \
469 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
470 (int)offset, RExC_precomp, RExC_precomp + offset); \
473 #define vWARN(loc,m) STMT_START { \
474 const IV offset = loc - RExC_precomp; \
475 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
476 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
479 #define vWARNdep(loc,m) STMT_START { \
480 const IV offset = loc - RExC_precomp; \
481 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
482 "%s" REPORT_LOCATION, \
483 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
487 #define vWARN2(loc, m, a1) STMT_START { \
488 const IV offset = loc - RExC_precomp; \
489 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
493 #define vWARN3(loc, m, a1, a2) STMT_START { \
494 const IV offset = loc - RExC_precomp; \
495 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
496 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
499 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
500 const IV offset = loc - RExC_precomp; \
501 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
502 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
505 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
506 const IV offset = loc - RExC_precomp; \
507 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
508 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
512 /* Allow for side effects in s */
513 #define REGC(c,s) STMT_START { \
514 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
517 /* Macros for recording node offsets. 20001227 mjd@plover.com
518 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
519 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
520 * Element 0 holds the number n.
521 * Position is 1 indexed.
523 #ifndef RE_TRACK_PATTERN_OFFSETS
524 #define Set_Node_Offset_To_R(node,byte)
525 #define Set_Node_Offset(node,byte)
526 #define Set_Cur_Node_Offset
527 #define Set_Node_Length_To_R(node,len)
528 #define Set_Node_Length(node,len)
529 #define Set_Node_Cur_Length(node)
530 #define Node_Offset(n)
531 #define Node_Length(n)
532 #define Set_Node_Offset_Length(node,offset,len)
533 #define ProgLen(ri) ri->u.proglen
534 #define SetProgLen(ri,x) ri->u.proglen = x
536 #define ProgLen(ri) ri->u.offsets[0]
537 #define SetProgLen(ri,x) ri->u.offsets[0] = x
538 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
540 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
541 __LINE__, (int)(node), (int)(byte))); \
543 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
545 RExC_offsets[2*(node)-1] = (byte); \
550 #define Set_Node_Offset(node,byte) \
551 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
552 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
554 #define Set_Node_Length_To_R(node,len) STMT_START { \
556 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
557 __LINE__, (int)(node), (int)(len))); \
559 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
561 RExC_offsets[2*(node)] = (len); \
566 #define Set_Node_Length(node,len) \
567 Set_Node_Length_To_R((node)-RExC_emit_start, len)
568 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
569 #define Set_Node_Cur_Length(node) \
570 Set_Node_Length(node, RExC_parse - parse_start)
572 /* Get offsets and lengths */
573 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
574 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
576 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
577 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
578 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
582 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
583 #define EXPERIMENTAL_INPLACESCAN
584 #endif /*RE_TRACK_PATTERN_OFFSETS*/
586 #define DEBUG_STUDYDATA(str,data,depth) \
587 DEBUG_OPTIMISE_MORE_r(if(data){ \
588 PerlIO_printf(Perl_debug_log, \
589 "%*s" str "Pos:%"IVdf"/%"IVdf \
590 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
591 (int)(depth)*2, "", \
592 (IV)((data)->pos_min), \
593 (IV)((data)->pos_delta), \
594 (UV)((data)->flags), \
595 (IV)((data)->whilem_c), \
596 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
597 is_inf ? "INF " : "" \
599 if ((data)->last_found) \
600 PerlIO_printf(Perl_debug_log, \
601 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
602 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
603 SvPVX_const((data)->last_found), \
604 (IV)((data)->last_end), \
605 (IV)((data)->last_start_min), \
606 (IV)((data)->last_start_max), \
607 ((data)->longest && \
608 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
609 SvPVX_const((data)->longest_fixed), \
610 (IV)((data)->offset_fixed), \
611 ((data)->longest && \
612 (data)->longest==&((data)->longest_float)) ? "*" : "", \
613 SvPVX_const((data)->longest_float), \
614 (IV)((data)->offset_float_min), \
615 (IV)((data)->offset_float_max) \
617 PerlIO_printf(Perl_debug_log,"\n"); \
620 static void clear_re(pTHX_ void *r);
622 /* Mark that we cannot extend a found fixed substring at this point.
623 Update the longest found anchored substring and the longest found
624 floating substrings if needed. */
627 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
629 const STRLEN l = CHR_SVLEN(data->last_found);
630 const STRLEN old_l = CHR_SVLEN(*data->longest);
631 GET_RE_DEBUG_FLAGS_DECL;
633 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
634 SvSetMagicSV(*data->longest, data->last_found);
635 if (*data->longest == data->longest_fixed) {
636 data->offset_fixed = l ? data->last_start_min : data->pos_min;
637 if (data->flags & SF_BEFORE_EOL)
639 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
641 data->flags &= ~SF_FIX_BEFORE_EOL;
642 data->minlen_fixed=minlenp;
643 data->lookbehind_fixed=0;
645 else { /* *data->longest == data->longest_float */
646 data->offset_float_min = l ? data->last_start_min : data->pos_min;
647 data->offset_float_max = (l
648 ? data->last_start_max
649 : data->pos_min + data->pos_delta);
650 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
651 data->offset_float_max = I32_MAX;
652 if (data->flags & SF_BEFORE_EOL)
654 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
656 data->flags &= ~SF_FL_BEFORE_EOL;
657 data->minlen_float=minlenp;
658 data->lookbehind_float=0;
661 SvCUR_set(data->last_found, 0);
663 SV * const sv = data->last_found;
664 if (SvUTF8(sv) && SvMAGICAL(sv)) {
665 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
671 data->flags &= ~SF_BEFORE_EOL;
672 DEBUG_STUDYDATA("commit: ",data,0);
675 /* Can match anything (initialization) */
677 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
679 ANYOF_CLASS_ZERO(cl);
680 ANYOF_BITMAP_SETALL(cl);
681 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
683 cl->flags |= ANYOF_LOCALE;
686 /* Can match anything (initialization) */
688 S_cl_is_anything(const struct regnode_charclass_class *cl)
692 for (value = 0; value <= ANYOF_MAX; value += 2)
693 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
695 if (!(cl->flags & ANYOF_UNICODE_ALL))
697 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
702 /* Can match anything (initialization) */
704 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
706 Zero(cl, 1, struct regnode_charclass_class);
708 cl_anything(pRExC_state, cl);
712 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
714 Zero(cl, 1, struct regnode_charclass_class);
716 cl_anything(pRExC_state, cl);
718 cl->flags |= ANYOF_LOCALE;
721 /* 'And' a given class with another one. Can create false positives */
722 /* We assume that cl is not inverted */
724 S_cl_and(struct regnode_charclass_class *cl,
725 const struct regnode_charclass_class *and_with)
728 assert(and_with->type == ANYOF);
729 if (!(and_with->flags & ANYOF_CLASS)
730 && !(cl->flags & ANYOF_CLASS)
731 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
732 && !(and_with->flags & ANYOF_FOLD)
733 && !(cl->flags & ANYOF_FOLD)) {
736 if (and_with->flags & ANYOF_INVERT)
737 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
738 cl->bitmap[i] &= ~and_with->bitmap[i];
740 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
741 cl->bitmap[i] &= and_with->bitmap[i];
742 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
743 if (!(and_with->flags & ANYOF_EOS))
744 cl->flags &= ~ANYOF_EOS;
746 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
747 !(and_with->flags & ANYOF_INVERT)) {
748 cl->flags &= ~ANYOF_UNICODE_ALL;
749 cl->flags |= ANYOF_UNICODE;
750 ARG_SET(cl, ARG(and_with));
752 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
753 !(and_with->flags & ANYOF_INVERT))
754 cl->flags &= ~ANYOF_UNICODE_ALL;
755 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
756 !(and_with->flags & ANYOF_INVERT))
757 cl->flags &= ~ANYOF_UNICODE;
760 /* 'OR' a given class with another one. Can create false positives */
761 /* We assume that cl is not inverted */
763 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
765 if (or_with->flags & ANYOF_INVERT) {
767 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
768 * <= (B1 | !B2) | (CL1 | !CL2)
769 * which is wasteful if CL2 is small, but we ignore CL2:
770 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
771 * XXXX Can we handle case-fold? Unclear:
772 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
773 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
775 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
776 && !(or_with->flags & ANYOF_FOLD)
777 && !(cl->flags & ANYOF_FOLD) ) {
780 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
781 cl->bitmap[i] |= ~or_with->bitmap[i];
782 } /* XXXX: logic is complicated otherwise */
784 cl_anything(pRExC_state, cl);
787 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
788 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
789 && (!(or_with->flags & ANYOF_FOLD)
790 || (cl->flags & ANYOF_FOLD)) ) {
793 /* OR char bitmap and class bitmap separately */
794 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
795 cl->bitmap[i] |= or_with->bitmap[i];
796 if (or_with->flags & ANYOF_CLASS) {
797 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
798 cl->classflags[i] |= or_with->classflags[i];
799 cl->flags |= ANYOF_CLASS;
802 else { /* XXXX: logic is complicated, leave it along for a moment. */
803 cl_anything(pRExC_state, cl);
806 if (or_with->flags & ANYOF_EOS)
807 cl->flags |= ANYOF_EOS;
809 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
810 ARG(cl) != ARG(or_with)) {
811 cl->flags |= ANYOF_UNICODE_ALL;
812 cl->flags &= ~ANYOF_UNICODE;
814 if (or_with->flags & ANYOF_UNICODE_ALL) {
815 cl->flags |= ANYOF_UNICODE_ALL;
816 cl->flags &= ~ANYOF_UNICODE;
820 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
821 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
822 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
823 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
828 dump_trie(trie,widecharmap,revcharmap)
829 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
830 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
832 These routines dump out a trie in a somewhat readable format.
833 The _interim_ variants are used for debugging the interim
834 tables that are used to generate the final compressed
835 representation which is what dump_trie expects.
837 Part of the reason for their existance is to provide a form
838 of documentation as to how the different representations function.
843 Dumps the final compressed table form of the trie to Perl_debug_log.
844 Used for debugging make_trie().
848 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
849 AV *revcharmap, U32 depth)
852 SV *sv=sv_newmortal();
853 int colwidth= widecharmap ? 6 : 4;
854 GET_RE_DEBUG_FLAGS_DECL;
857 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
858 (int)depth * 2 + 2,"",
859 "Match","Base","Ofs" );
861 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
862 SV ** const tmp = av_fetch( revcharmap, state, 0);
864 PerlIO_printf( Perl_debug_log, "%*s",
866 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
867 PL_colors[0], PL_colors[1],
868 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
869 PERL_PV_ESCAPE_FIRSTCHAR
874 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
875 (int)depth * 2 + 2,"");
877 for( state = 0 ; state < trie->uniquecharcount ; state++ )
878 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
879 PerlIO_printf( Perl_debug_log, "\n");
881 for( state = 1 ; state < trie->statecount ; state++ ) {
882 const U32 base = trie->states[ state ].trans.base;
884 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
886 if ( trie->states[ state ].wordnum ) {
887 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
889 PerlIO_printf( Perl_debug_log, "%6s", "" );
892 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
897 while( ( base + ofs < trie->uniquecharcount ) ||
898 ( base + ofs - trie->uniquecharcount < trie->lasttrans
899 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
902 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
904 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
905 if ( ( base + ofs >= trie->uniquecharcount ) &&
906 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
907 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
909 PerlIO_printf( Perl_debug_log, "%*"UVXf,
911 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
913 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
917 PerlIO_printf( Perl_debug_log, "]");
920 PerlIO_printf( Perl_debug_log, "\n" );
924 Dumps a fully constructed but uncompressed trie in list form.
925 List tries normally only are used for construction when the number of
926 possible chars (trie->uniquecharcount) is very high.
927 Used for debugging make_trie().
930 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
931 HV *widecharmap, AV *revcharmap, U32 next_alloc,
935 SV *sv=sv_newmortal();
936 int colwidth= widecharmap ? 6 : 4;
937 GET_RE_DEBUG_FLAGS_DECL;
938 /* print out the table precompression. */
939 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
940 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
941 "------:-----+-----------------\n" );
943 for( state=1 ; state < next_alloc ; state ++ ) {
946 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
947 (int)depth * 2 + 2,"", (UV)state );
948 if ( ! trie->states[ state ].wordnum ) {
949 PerlIO_printf( Perl_debug_log, "%5s| ","");
951 PerlIO_printf( Perl_debug_log, "W%4x| ",
952 trie->states[ state ].wordnum
955 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
956 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
958 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
960 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
961 PL_colors[0], PL_colors[1],
962 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
963 PERL_PV_ESCAPE_FIRSTCHAR
965 TRIE_LIST_ITEM(state,charid).forid,
966 (UV)TRIE_LIST_ITEM(state,charid).newstate
969 PerlIO_printf(Perl_debug_log, "\n%*s| ",
970 (int)((depth * 2) + 14), "");
973 PerlIO_printf( Perl_debug_log, "\n");
978 Dumps a fully constructed but uncompressed trie in table form.
979 This is the normal DFA style state transition table, with a few
980 twists to facilitate compression later.
981 Used for debugging make_trie().
984 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
985 HV *widecharmap, AV *revcharmap, U32 next_alloc,
990 SV *sv=sv_newmortal();
991 int colwidth= widecharmap ? 6 : 4;
992 GET_RE_DEBUG_FLAGS_DECL;
995 print out the table precompression so that we can do a visual check
996 that they are identical.
999 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1001 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1002 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1004 PerlIO_printf( Perl_debug_log, "%*s",
1006 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1007 PL_colors[0], PL_colors[1],
1008 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1009 PERL_PV_ESCAPE_FIRSTCHAR
1015 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1017 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1018 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1021 PerlIO_printf( Perl_debug_log, "\n" );
1023 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1025 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1026 (int)depth * 2 + 2,"",
1027 (UV)TRIE_NODENUM( state ) );
1029 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1030 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1032 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1034 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1036 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1037 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1039 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1040 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1047 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1048 startbranch: the first branch in the whole branch sequence
1049 first : start branch of sequence of branch-exact nodes.
1050 May be the same as startbranch
1051 last : Thing following the last branch.
1052 May be the same as tail.
1053 tail : item following the branch sequence
1054 count : words in the sequence
1055 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1056 depth : indent depth
1058 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1060 A trie is an N'ary tree where the branches are determined by digital
1061 decomposition of the key. IE, at the root node you look up the 1st character and
1062 follow that branch repeat until you find the end of the branches. Nodes can be
1063 marked as "accepting" meaning they represent a complete word. Eg:
1067 would convert into the following structure. Numbers represent states, letters
1068 following numbers represent valid transitions on the letter from that state, if
1069 the number is in square brackets it represents an accepting state, otherwise it
1070 will be in parenthesis.
1072 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1076 (1) +-i->(6)-+-s->[7]
1078 +-s->(3)-+-h->(4)-+-e->[5]
1080 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1082 This shows that when matching against the string 'hers' we will begin at state 1
1083 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1084 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1085 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1086 single traverse. We store a mapping from accepting to state to which word was
1087 matched, and then when we have multiple possibilities we try to complete the
1088 rest of the regex in the order in which they occured in the alternation.
1090 The only prior NFA like behaviour that would be changed by the TRIE support is
1091 the silent ignoring of duplicate alternations which are of the form:
1093 / (DUPE|DUPE) X? (?{ ... }) Y /x
1095 Thus EVAL blocks follwing a trie may be called a different number of times with
1096 and without the optimisation. With the optimisations dupes will be silently
1097 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1098 the following demonstrates:
1100 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1102 which prints out 'word' three times, but
1104 'words'=~/(word|word|word)(?{ print $1 })S/
1106 which doesnt print it out at all. This is due to other optimisations kicking in.
1108 Example of what happens on a structural level:
1110 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1112 1: CURLYM[1] {1,32767}(18)
1123 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1124 and should turn into:
1126 1: CURLYM[1] {1,32767}(18)
1128 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1136 Cases where tail != last would be like /(?foo|bar)baz/:
1146 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1147 and would end up looking like:
1150 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1157 d = uvuni_to_utf8_flags(d, uv, 0);
1159 is the recommended Unicode-aware way of saying
1164 #define TRIE_STORE_REVCHAR \
1166 SV *tmp = newSVpvs(""); \
1167 if (UTF) SvUTF8_on(tmp); \
1168 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1169 av_push( revcharmap, tmp ); \
1172 #define TRIE_READ_CHAR STMT_START { \
1176 if ( foldlen > 0 ) { \
1177 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1182 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1183 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1184 foldlen -= UNISKIP( uvc ); \
1185 scan = foldbuf + UNISKIP( uvc ); \
1188 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1198 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1199 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1200 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1201 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1203 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1204 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1205 TRIE_LIST_CUR( state )++; \
1208 #define TRIE_LIST_NEW(state) STMT_START { \
1209 Newxz( trie->states[ state ].trans.list, \
1210 4, reg_trie_trans_le ); \
1211 TRIE_LIST_CUR( state ) = 1; \
1212 TRIE_LIST_LEN( state ) = 4; \
1215 #define TRIE_HANDLE_WORD(state) STMT_START { \
1216 U16 dupe= trie->states[ state ].wordnum; \
1217 regnode * const noper_next = regnext( noper ); \
1219 if (trie->wordlen) \
1220 trie->wordlen[ curword ] = wordlen; \
1222 /* store the word for dumping */ \
1224 if (OP(noper) != NOTHING) \
1225 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1227 tmp = newSVpvn( "", 0 ); \
1228 if ( UTF ) SvUTF8_on( tmp ); \
1229 av_push( trie_words, tmp ); \
1234 if ( noper_next < tail ) { \
1236 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1237 trie->jump[curword] = (U16)(noper_next - convert); \
1239 jumper = noper_next; \
1241 nextbranch= regnext(cur); \
1245 /* So it's a dupe. This means we need to maintain a */\
1246 /* linked-list from the first to the next. */\
1247 /* we only allocate the nextword buffer when there */\
1248 /* a dupe, so first time we have to do the allocation */\
1249 if (!trie->nextword) \
1250 trie->nextword = (U16 *) \
1251 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1252 while ( trie->nextword[dupe] ) \
1253 dupe= trie->nextword[dupe]; \
1254 trie->nextword[dupe]= curword; \
1256 /* we haven't inserted this word yet. */ \
1257 trie->states[ state ].wordnum = curword; \
1262 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1263 ( ( base + charid >= ucharcount \
1264 && base + charid < ubound \
1265 && state == trie->trans[ base - ucharcount + charid ].check \
1266 && trie->trans[ base - ucharcount + charid ].next ) \
1267 ? trie->trans[ base - ucharcount + charid ].next \
1268 : ( state==1 ? special : 0 ) \
1272 #define MADE_JUMP_TRIE 2
1273 #define MADE_EXACT_TRIE 4
1276 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1279 /* first pass, loop through and scan words */
1280 reg_trie_data *trie;
1281 HV *widecharmap = NULL;
1282 AV *revcharmap = newAV();
1284 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1289 regnode *jumper = NULL;
1290 regnode *nextbranch = NULL;
1291 regnode *convert = NULL;
1292 /* we just use folder as a flag in utf8 */
1293 const U8 * const folder = ( flags == EXACTF
1295 : ( flags == EXACTFL
1302 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1303 AV *trie_words = NULL;
1304 /* along with revcharmap, this only used during construction but both are
1305 * useful during debugging so we store them in the struct when debugging.
1308 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1309 STRLEN trie_charcount=0;
1311 SV *re_trie_maxbuff;
1312 GET_RE_DEBUG_FLAGS_DECL;
1314 PERL_UNUSED_ARG(depth);
1317 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1319 trie->startstate = 1;
1320 trie->wordcount = word_count;
1321 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1322 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1323 if (!(UTF && folder))
1324 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1326 trie_words = newAV();
1329 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1330 if (!SvIOK(re_trie_maxbuff)) {
1331 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1334 PerlIO_printf( Perl_debug_log,
1335 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1336 (int)depth * 2 + 2, "",
1337 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1338 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1342 /* Find the node we are going to overwrite */
1343 if ( first == startbranch && OP( last ) != BRANCH ) {
1344 /* whole branch chain */
1347 /* branch sub-chain */
1348 convert = NEXTOPER( first );
1351 /* -- First loop and Setup --
1353 We first traverse the branches and scan each word to determine if it
1354 contains widechars, and how many unique chars there are, this is
1355 important as we have to build a table with at least as many columns as we
1358 We use an array of integers to represent the character codes 0..255
1359 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1360 native representation of the character value as the key and IV's for the
1363 *TODO* If we keep track of how many times each character is used we can
1364 remap the columns so that the table compression later on is more
1365 efficient in terms of memory by ensuring most common value is in the
1366 middle and the least common are on the outside. IMO this would be better
1367 than a most to least common mapping as theres a decent chance the most
1368 common letter will share a node with the least common, meaning the node
1369 will not be compressable. With a middle is most common approach the worst
1370 case is when we have the least common nodes twice.
1374 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1375 regnode * const noper = NEXTOPER( cur );
1376 const U8 *uc = (U8*)STRING( noper );
1377 const U8 * const e = uc + STR_LEN( noper );
1379 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1380 const U8 *scan = (U8*)NULL;
1381 U32 wordlen = 0; /* required init */
1383 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1385 if (OP(noper) == NOTHING) {
1389 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1390 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1391 regardless of encoding */
1393 for ( ; uc < e ; uc += len ) {
1394 TRIE_CHARCOUNT(trie)++;
1398 if ( !trie->charmap[ uvc ] ) {
1399 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1401 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1405 /* store the codepoint in the bitmap, and if its ascii
1406 also store its folded equivelent. */
1407 TRIE_BITMAP_SET(trie,uvc);
1408 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1409 set_bit = 0; /* We've done our bit :-) */
1414 widecharmap = newHV();
1416 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1419 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1421 if ( !SvTRUE( *svpp ) ) {
1422 sv_setiv( *svpp, ++trie->uniquecharcount );
1427 if( cur == first ) {
1430 } else if (chars < trie->minlen) {
1432 } else if (chars > trie->maxlen) {
1436 } /* end first pass */
1437 DEBUG_TRIE_COMPILE_r(
1438 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1439 (int)depth * 2 + 2,"",
1440 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1441 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1442 (int)trie->minlen, (int)trie->maxlen )
1444 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1447 We now know what we are dealing with in terms of unique chars and
1448 string sizes so we can calculate how much memory a naive
1449 representation using a flat table will take. If it's over a reasonable
1450 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1451 conservative but potentially much slower representation using an array
1454 At the end we convert both representations into the same compressed
1455 form that will be used in regexec.c for matching with. The latter
1456 is a form that cannot be used to construct with but has memory
1457 properties similar to the list form and access properties similar
1458 to the table form making it both suitable for fast searches and
1459 small enough that its feasable to store for the duration of a program.
1461 See the comment in the code where the compressed table is produced
1462 inplace from the flat tabe representation for an explanation of how
1463 the compression works.
1468 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1470 Second Pass -- Array Of Lists Representation
1472 Each state will be represented by a list of charid:state records
1473 (reg_trie_trans_le) the first such element holds the CUR and LEN
1474 points of the allocated array. (See defines above).
1476 We build the initial structure using the lists, and then convert
1477 it into the compressed table form which allows faster lookups
1478 (but cant be modified once converted).
1481 STRLEN transcount = 1;
1483 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1484 "%*sCompiling trie using list compiler\n",
1485 (int)depth * 2 + 2, ""));
1487 trie->states = (reg_trie_state *)
1488 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1489 sizeof(reg_trie_state) );
1493 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1495 regnode * const noper = NEXTOPER( cur );
1496 U8 *uc = (U8*)STRING( noper );
1497 const U8 * const e = uc + STR_LEN( noper );
1498 U32 state = 1; /* required init */
1499 U16 charid = 0; /* sanity init */
1500 U8 *scan = (U8*)NULL; /* sanity init */
1501 STRLEN foldlen = 0; /* required init */
1502 U32 wordlen = 0; /* required init */
1503 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1505 if (OP(noper) != NOTHING) {
1506 for ( ; uc < e ; uc += len ) {
1511 charid = trie->charmap[ uvc ];
1513 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1517 charid=(U16)SvIV( *svpp );
1520 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1527 if ( !trie->states[ state ].trans.list ) {
1528 TRIE_LIST_NEW( state );
1530 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1531 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1532 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1537 newstate = next_alloc++;
1538 TRIE_LIST_PUSH( state, charid, newstate );
1543 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1547 TRIE_HANDLE_WORD(state);
1549 } /* end second pass */
1551 /* next alloc is the NEXT state to be allocated */
1552 trie->statecount = next_alloc;
1553 trie->states = (reg_trie_state *)
1554 PerlMemShared_realloc( trie->states,
1556 * sizeof(reg_trie_state) );
1558 /* and now dump it out before we compress it */
1559 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1560 revcharmap, next_alloc,
1564 trie->trans = (reg_trie_trans *)
1565 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1572 for( state=1 ; state < next_alloc ; state ++ ) {
1576 DEBUG_TRIE_COMPILE_MORE_r(
1577 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1581 if (trie->states[state].trans.list) {
1582 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1586 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1587 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1588 if ( forid < minid ) {
1590 } else if ( forid > maxid ) {
1594 if ( transcount < tp + maxid - minid + 1) {
1596 trie->trans = (reg_trie_trans *)
1597 PerlMemShared_realloc( trie->trans,
1599 * sizeof(reg_trie_trans) );
1600 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1602 base = trie->uniquecharcount + tp - minid;
1603 if ( maxid == minid ) {
1605 for ( ; zp < tp ; zp++ ) {
1606 if ( ! trie->trans[ zp ].next ) {
1607 base = trie->uniquecharcount + zp - minid;
1608 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1609 trie->trans[ zp ].check = state;
1615 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1616 trie->trans[ tp ].check = state;
1621 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1622 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1623 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1624 trie->trans[ tid ].check = state;
1626 tp += ( maxid - minid + 1 );
1628 Safefree(trie->states[ state ].trans.list);
1631 DEBUG_TRIE_COMPILE_MORE_r(
1632 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1635 trie->states[ state ].trans.base=base;
1637 trie->lasttrans = tp + 1;
1641 Second Pass -- Flat Table Representation.
1643 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1644 We know that we will need Charcount+1 trans at most to store the data
1645 (one row per char at worst case) So we preallocate both structures
1646 assuming worst case.
1648 We then construct the trie using only the .next slots of the entry
1651 We use the .check field of the first entry of the node temporarily to
1652 make compression both faster and easier by keeping track of how many non
1653 zero fields are in the node.
1655 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1658 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1659 number representing the first entry of the node, and state as a
1660 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1661 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1662 are 2 entrys per node. eg:
1670 The table is internally in the right hand, idx form. However as we also
1671 have to deal with the states array which is indexed by nodenum we have to
1672 use TRIE_NODENUM() to convert.
1675 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1676 "%*sCompiling trie using table compiler\n",
1677 (int)depth * 2 + 2, ""));
1679 trie->trans = (reg_trie_trans *)
1680 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1681 * trie->uniquecharcount + 1,
1682 sizeof(reg_trie_trans) );
1683 trie->states = (reg_trie_state *)
1684 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1685 sizeof(reg_trie_state) );
1686 next_alloc = trie->uniquecharcount + 1;
1689 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1691 regnode * const noper = NEXTOPER( cur );
1692 const U8 *uc = (U8*)STRING( noper );
1693 const U8 * const e = uc + STR_LEN( noper );
1695 U32 state = 1; /* required init */
1697 U16 charid = 0; /* sanity init */
1698 U32 accept_state = 0; /* sanity init */
1699 U8 *scan = (U8*)NULL; /* sanity init */
1701 STRLEN foldlen = 0; /* required init */
1702 U32 wordlen = 0; /* required init */
1703 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1705 if ( OP(noper) != NOTHING ) {
1706 for ( ; uc < e ; uc += len ) {
1711 charid = trie->charmap[ uvc ];
1713 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1714 charid = svpp ? (U16)SvIV(*svpp) : 0;
1718 if ( !trie->trans[ state + charid ].next ) {
1719 trie->trans[ state + charid ].next = next_alloc;
1720 trie->trans[ state ].check++;
1721 next_alloc += trie->uniquecharcount;
1723 state = trie->trans[ state + charid ].next;
1725 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1727 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1730 accept_state = TRIE_NODENUM( state );
1731 TRIE_HANDLE_WORD(accept_state);
1733 } /* end second pass */
1735 /* and now dump it out before we compress it */
1736 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1738 next_alloc, depth+1));
1742 * Inplace compress the table.*
1744 For sparse data sets the table constructed by the trie algorithm will
1745 be mostly 0/FAIL transitions or to put it another way mostly empty.
1746 (Note that leaf nodes will not contain any transitions.)
1748 This algorithm compresses the tables by eliminating most such
1749 transitions, at the cost of a modest bit of extra work during lookup:
1751 - Each states[] entry contains a .base field which indicates the
1752 index in the state[] array wheres its transition data is stored.
1754 - If .base is 0 there are no valid transitions from that node.
1756 - If .base is nonzero then charid is added to it to find an entry in
1759 -If trans[states[state].base+charid].check!=state then the
1760 transition is taken to be a 0/Fail transition. Thus if there are fail
1761 transitions at the front of the node then the .base offset will point
1762 somewhere inside the previous nodes data (or maybe even into a node
1763 even earlier), but the .check field determines if the transition is
1767 The following process inplace converts the table to the compressed
1768 table: We first do not compress the root node 1,and mark its all its
1769 .check pointers as 1 and set its .base pointer as 1 as well. This
1770 allows to do a DFA construction from the compressed table later, and
1771 ensures that any .base pointers we calculate later are greater than
1774 - We set 'pos' to indicate the first entry of the second node.
1776 - We then iterate over the columns of the node, finding the first and
1777 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1778 and set the .check pointers accordingly, and advance pos
1779 appropriately and repreat for the next node. Note that when we copy
1780 the next pointers we have to convert them from the original
1781 NODEIDX form to NODENUM form as the former is not valid post
1784 - If a node has no transitions used we mark its base as 0 and do not
1785 advance the pos pointer.
1787 - If a node only has one transition we use a second pointer into the
1788 structure to fill in allocated fail transitions from other states.
1789 This pointer is independent of the main pointer and scans forward
1790 looking for null transitions that are allocated to a state. When it
1791 finds one it writes the single transition into the "hole". If the
1792 pointer doesnt find one the single transition is appended as normal.
1794 - Once compressed we can Renew/realloc the structures to release the
1797 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1798 specifically Fig 3.47 and the associated pseudocode.
1802 const U32 laststate = TRIE_NODENUM( next_alloc );
1805 trie->statecount = laststate;
1807 for ( state = 1 ; state < laststate ; state++ ) {
1809 const U32 stateidx = TRIE_NODEIDX( state );
1810 const U32 o_used = trie->trans[ stateidx ].check;
1811 U32 used = trie->trans[ stateidx ].check;
1812 trie->trans[ stateidx ].check = 0;
1814 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1815 if ( flag || trie->trans[ stateidx + charid ].next ) {
1816 if ( trie->trans[ stateidx + charid ].next ) {
1818 for ( ; zp < pos ; zp++ ) {
1819 if ( ! trie->trans[ zp ].next ) {
1823 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1824 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1825 trie->trans[ zp ].check = state;
1826 if ( ++zp > pos ) pos = zp;
1833 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1835 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1836 trie->trans[ pos ].check = state;
1841 trie->lasttrans = pos + 1;
1842 trie->states = (reg_trie_state *)
1843 PerlMemShared_realloc( trie->states, laststate
1844 * sizeof(reg_trie_state) );
1845 DEBUG_TRIE_COMPILE_MORE_r(
1846 PerlIO_printf( Perl_debug_log,
1847 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1848 (int)depth * 2 + 2,"",
1849 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1852 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1855 } /* end table compress */
1857 DEBUG_TRIE_COMPILE_MORE_r(
1858 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1859 (int)depth * 2 + 2, "",
1860 (UV)trie->statecount,
1861 (UV)trie->lasttrans)
1863 /* resize the trans array to remove unused space */
1864 trie->trans = (reg_trie_trans *)
1865 PerlMemShared_realloc( trie->trans, trie->lasttrans
1866 * sizeof(reg_trie_trans) );
1868 /* and now dump out the compressed format */
1869 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1871 { /* Modify the program and insert the new TRIE node*/
1872 U8 nodetype =(U8)(flags & 0xFF);
1876 regnode *optimize = NULL;
1877 #ifdef RE_TRACK_PATTERN_OFFSETS
1880 U32 mjd_nodelen = 0;
1881 #endif /* RE_TRACK_PATTERN_OFFSETS */
1882 #endif /* DEBUGGING */
1884 This means we convert either the first branch or the first Exact,
1885 depending on whether the thing following (in 'last') is a branch
1886 or not and whther first is the startbranch (ie is it a sub part of
1887 the alternation or is it the whole thing.)
1888 Assuming its a sub part we conver the EXACT otherwise we convert
1889 the whole branch sequence, including the first.
1891 /* Find the node we are going to overwrite */
1892 if ( first != startbranch || OP( last ) == BRANCH ) {
1893 /* branch sub-chain */
1894 NEXT_OFF( first ) = (U16)(last - first);
1895 #ifdef RE_TRACK_PATTERN_OFFSETS
1897 mjd_offset= Node_Offset((convert));
1898 mjd_nodelen= Node_Length((convert));
1901 /* whole branch chain */
1903 #ifdef RE_TRACK_PATTERN_OFFSETS
1906 const regnode *nop = NEXTOPER( convert );
1907 mjd_offset= Node_Offset((nop));
1908 mjd_nodelen= Node_Length((nop));
1912 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1913 (int)depth * 2 + 2, "",
1914 (UV)mjd_offset, (UV)mjd_nodelen)
1917 /* But first we check to see if there is a common prefix we can
1918 split out as an EXACT and put in front of the TRIE node. */
1919 trie->startstate= 1;
1920 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1922 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1926 const U32 base = trie->states[ state ].trans.base;
1928 if ( trie->states[state].wordnum )
1931 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1932 if ( ( base + ofs >= trie->uniquecharcount ) &&
1933 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1934 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1936 if ( ++count > 1 ) {
1937 SV **tmp = av_fetch( revcharmap, ofs, 0);
1938 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1939 if ( state == 1 ) break;
1941 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1943 PerlIO_printf(Perl_debug_log,
1944 "%*sNew Start State=%"UVuf" Class: [",
1945 (int)depth * 2 + 2, "",
1948 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1949 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1951 TRIE_BITMAP_SET(trie,*ch);
1953 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1955 PerlIO_printf(Perl_debug_log, (char*)ch)
1959 TRIE_BITMAP_SET(trie,*ch);
1961 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1962 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1968 SV **tmp = av_fetch( revcharmap, idx, 0);
1969 char *ch = SvPV_nolen( *tmp );
1971 SV *sv=sv_newmortal();
1972 PerlIO_printf( Perl_debug_log,
1973 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1974 (int)depth * 2 + 2, "",
1976 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1977 PL_colors[0], PL_colors[1],
1978 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1979 PERL_PV_ESCAPE_FIRSTCHAR
1984 OP( convert ) = nodetype;
1985 str=STRING(convert);
1996 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2002 regnode *n = convert+NODE_SZ_STR(convert);
2003 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2004 trie->startstate = state;
2005 trie->minlen -= (state - 1);
2006 trie->maxlen -= (state - 1);
2008 regnode *fix = convert;
2009 U32 word = trie->wordcount;
2011 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2012 while( ++fix < n ) {
2013 Set_Node_Offset_Length(fix, 0, 0);
2016 SV ** const tmp = av_fetch( trie_words, word, 0 );
2018 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2019 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2021 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2028 NEXT_OFF(convert) = (U16)(tail - convert);
2029 DEBUG_r(optimize= n);
2035 if ( trie->maxlen ) {
2036 NEXT_OFF( convert ) = (U16)(tail - convert);
2037 ARG_SET( convert, data_slot );
2038 /* Store the offset to the first unabsorbed branch in
2039 jump[0], which is otherwise unused by the jump logic.
2040 We use this when dumping a trie and during optimisation. */
2042 trie->jump[0] = (U16)(nextbranch - convert);
2045 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2046 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2048 OP( convert ) = TRIEC;
2049 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2050 PerlMemShared_free(trie->bitmap);
2053 OP( convert ) = TRIE;
2055 /* store the type in the flags */
2056 convert->flags = nodetype;
2060 + regarglen[ OP( convert ) ];
2062 /* XXX We really should free up the resource in trie now,
2063 as we won't use them - (which resources?) dmq */
2065 /* needed for dumping*/
2066 DEBUG_r(if (optimize) {
2067 regnode *opt = convert;
2069 while ( ++opt < optimize) {
2070 Set_Node_Offset_Length(opt,0,0);
2073 Try to clean up some of the debris left after the
2076 while( optimize < jumper ) {
2077 mjd_nodelen += Node_Length((optimize));
2078 OP( optimize ) = OPTIMIZED;
2079 Set_Node_Offset_Length(optimize,0,0);
2082 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2084 } /* end node insert */
2085 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2087 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2088 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2090 SvREFCNT_dec(revcharmap);
2094 : trie->startstate>1
2100 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2102 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2104 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2105 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2108 We find the fail state for each state in the trie, this state is the longest proper
2109 suffix of the current states 'word' that is also a proper prefix of another word in our
2110 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2111 the DFA not to have to restart after its tried and failed a word at a given point, it
2112 simply continues as though it had been matching the other word in the first place.
2114 'abcdgu'=~/abcdefg|cdgu/
2115 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2116 fail, which would bring use to the state representing 'd' in the second word where we would
2117 try 'g' and succeed, prodceding to match 'cdgu'.
2119 /* add a fail transition */
2120 const U32 trie_offset = ARG(source);
2121 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2123 const U32 ucharcount = trie->uniquecharcount;
2124 const U32 numstates = trie->statecount;
2125 const U32 ubound = trie->lasttrans + ucharcount;
2129 U32 base = trie->states[ 1 ].trans.base;
2132 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2133 GET_RE_DEBUG_FLAGS_DECL;
2135 PERL_UNUSED_ARG(depth);
2139 ARG_SET( stclass, data_slot );
2140 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2141 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2142 aho->trie=trie_offset;
2143 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2144 Copy( trie->states, aho->states, numstates, reg_trie_state );
2145 Newxz( q, numstates, U32);
2146 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2149 /* initialize fail[0..1] to be 1 so that we always have
2150 a valid final fail state */
2151 fail[ 0 ] = fail[ 1 ] = 1;
2153 for ( charid = 0; charid < ucharcount ; charid++ ) {
2154 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2156 q[ q_write ] = newstate;
2157 /* set to point at the root */
2158 fail[ q[ q_write++ ] ]=1;
2161 while ( q_read < q_write) {
2162 const U32 cur = q[ q_read++ % numstates ];
2163 base = trie->states[ cur ].trans.base;
2165 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2166 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2168 U32 fail_state = cur;
2171 fail_state = fail[ fail_state ];
2172 fail_base = aho->states[ fail_state ].trans.base;
2173 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2175 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2176 fail[ ch_state ] = fail_state;
2177 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2179 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2181 q[ q_write++ % numstates] = ch_state;
2185 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2186 when we fail in state 1, this allows us to use the
2187 charclass scan to find a valid start char. This is based on the principle
2188 that theres a good chance the string being searched contains lots of stuff
2189 that cant be a start char.
2191 fail[ 0 ] = fail[ 1 ] = 0;
2192 DEBUG_TRIE_COMPILE_r({
2193 PerlIO_printf(Perl_debug_log,
2194 "%*sStclass Failtable (%"UVuf" states): 0",
2195 (int)(depth * 2), "", (UV)numstates
2197 for( q_read=1; q_read<numstates; q_read++ ) {
2198 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2200 PerlIO_printf(Perl_debug_log, "\n");
2203 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2208 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2209 * These need to be revisited when a newer toolchain becomes available.
2211 #if defined(__sparc64__) && defined(__GNUC__)
2212 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2213 # undef SPARC64_GCC_WORKAROUND
2214 # define SPARC64_GCC_WORKAROUND 1
2218 #define DEBUG_PEEP(str,scan,depth) \
2219 DEBUG_OPTIMISE_r({if (scan){ \
2220 SV * const mysv=sv_newmortal(); \
2221 regnode *Next = regnext(scan); \
2222 regprop(RExC_rx, mysv, scan); \
2223 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2224 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2225 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2232 #define JOIN_EXACT(scan,min,flags) \
2233 if (PL_regkind[OP(scan)] == EXACT) \
2234 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2237 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2238 /* Merge several consecutive EXACTish nodes into one. */
2239 regnode *n = regnext(scan);
2241 regnode *next = scan + NODE_SZ_STR(scan);
2245 regnode *stop = scan;
2246 GET_RE_DEBUG_FLAGS_DECL;
2248 PERL_UNUSED_ARG(depth);
2250 #ifndef EXPERIMENTAL_INPLACESCAN
2251 PERL_UNUSED_ARG(flags);
2252 PERL_UNUSED_ARG(val);
2254 DEBUG_PEEP("join",scan,depth);
2256 /* Skip NOTHING, merge EXACT*. */
2258 ( PL_regkind[OP(n)] == NOTHING ||
2259 (stringok && (OP(n) == OP(scan))))
2261 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2263 if (OP(n) == TAIL || n > next)
2265 if (PL_regkind[OP(n)] == NOTHING) {
2266 DEBUG_PEEP("skip:",n,depth);
2267 NEXT_OFF(scan) += NEXT_OFF(n);
2268 next = n + NODE_STEP_REGNODE;
2275 else if (stringok) {
2276 const unsigned int oldl = STR_LEN(scan);
2277 regnode * const nnext = regnext(n);
2279 DEBUG_PEEP("merg",n,depth);
2282 if (oldl + STR_LEN(n) > U8_MAX)
2284 NEXT_OFF(scan) += NEXT_OFF(n);
2285 STR_LEN(scan) += STR_LEN(n);
2286 next = n + NODE_SZ_STR(n);
2287 /* Now we can overwrite *n : */
2288 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2296 #ifdef EXPERIMENTAL_INPLACESCAN
2297 if (flags && !NEXT_OFF(n)) {
2298 DEBUG_PEEP("atch", val, depth);
2299 if (reg_off_by_arg[OP(n)]) {
2300 ARG_SET(n, val - n);
2303 NEXT_OFF(n) = val - n;
2310 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2312 Two problematic code points in Unicode casefolding of EXACT nodes:
2314 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2315 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2321 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2322 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2324 This means that in case-insensitive matching (or "loose matching",
2325 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2326 length of the above casefolded versions) can match a target string
2327 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2328 This would rather mess up the minimum length computation.
2330 What we'll do is to look for the tail four bytes, and then peek
2331 at the preceding two bytes to see whether we need to decrease
2332 the minimum length by four (six minus two).
2334 Thanks to the design of UTF-8, there cannot be false matches:
2335 A sequence of valid UTF-8 bytes cannot be a subsequence of
2336 another valid sequence of UTF-8 bytes.
2339 char * const s0 = STRING(scan), *s, *t;
2340 char * const s1 = s0 + STR_LEN(scan) - 1;
2341 char * const s2 = s1 - 4;
2342 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2343 const char t0[] = "\xaf\x49\xaf\x42";
2345 const char t0[] = "\xcc\x88\xcc\x81";
2347 const char * const t1 = t0 + 3;
2350 s < s2 && (t = ninstr(s, s1, t0, t1));
2353 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2354 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2356 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2357 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2365 n = scan + NODE_SZ_STR(scan);
2367 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2374 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2378 /* REx optimizer. Converts nodes into quickier variants "in place".
2379 Finds fixed substrings. */
2381 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2382 to the position after last scanned or to NULL. */
2384 #define INIT_AND_WITHP \
2385 assert(!and_withp); \
2386 Newx(and_withp,1,struct regnode_charclass_class); \
2387 SAVEFREEPV(and_withp)
2389 /* this is a chain of data about sub patterns we are processing that
2390 need to be handled seperately/specially in study_chunk. Its so
2391 we can simulate recursion without losing state. */
2393 typedef struct scan_frame {
2394 regnode *last; /* last node to process in this frame */
2395 regnode *next; /* next node to process when last is reached */
2396 struct scan_frame *prev; /*previous frame*/
2397 I32 stop; /* what stopparen do we use */
2401 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2404 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2405 I32 *minlenp, I32 *deltap,
2410 struct regnode_charclass_class *and_withp,
2411 U32 flags, U32 depth)
2412 /* scanp: Start here (read-write). */
2413 /* deltap: Write maxlen-minlen here. */
2414 /* last: Stop before this one. */
2415 /* data: string data about the pattern */
2416 /* stopparen: treat close N as END */
2417 /* recursed: which subroutines have we recursed into */
2418 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2421 I32 min = 0, pars = 0, code;
2422 regnode *scan = *scanp, *next;
2424 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2425 int is_inf_internal = 0; /* The studied chunk is infinite */
2426 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2427 scan_data_t data_fake;
2428 SV *re_trie_maxbuff = NULL;
2429 regnode *first_non_open = scan;
2430 I32 stopmin = I32_MAX;
2431 scan_frame *frame = NULL;
2433 GET_RE_DEBUG_FLAGS_DECL;
2436 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2440 while (first_non_open && OP(first_non_open) == OPEN)
2441 first_non_open=regnext(first_non_open);
2446 while ( scan && OP(scan) != END && scan < last ){
2447 /* Peephole optimizer: */
2448 DEBUG_STUDYDATA("Peep:", data,depth);
2449 DEBUG_PEEP("Peep",scan,depth);
2450 JOIN_EXACT(scan,&min,0);
2452 /* Follow the next-chain of the current node and optimize
2453 away all the NOTHINGs from it. */
2454 if (OP(scan) != CURLYX) {
2455 const int max = (reg_off_by_arg[OP(scan)]
2457 /* I32 may be smaller than U16 on CRAYs! */
2458 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2459 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2463 /* Skip NOTHING and LONGJMP. */
2464 while ((n = regnext(n))
2465 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2466 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2467 && off + noff < max)
2469 if (reg_off_by_arg[OP(scan)])
2472 NEXT_OFF(scan) = off;
2477 /* The principal pseudo-switch. Cannot be a switch, since we
2478 look into several different things. */
2479 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2480 || OP(scan) == IFTHEN) {
2481 next = regnext(scan);
2483 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2485 if (OP(next) == code || code == IFTHEN) {
2486 /* NOTE - There is similar code to this block below for handling
2487 TRIE nodes on a re-study. If you change stuff here check there
2489 I32 max1 = 0, min1 = I32_MAX, num = 0;
2490 struct regnode_charclass_class accum;
2491 regnode * const startbranch=scan;
2493 if (flags & SCF_DO_SUBSTR)
2494 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2495 if (flags & SCF_DO_STCLASS)
2496 cl_init_zero(pRExC_state, &accum);
2498 while (OP(scan) == code) {
2499 I32 deltanext, minnext, f = 0, fake;
2500 struct regnode_charclass_class this_class;
2503 data_fake.flags = 0;
2505 data_fake.whilem_c = data->whilem_c;
2506 data_fake.last_closep = data->last_closep;
2509 data_fake.last_closep = &fake;
2511 data_fake.pos_delta = delta;
2512 next = regnext(scan);
2513 scan = NEXTOPER(scan);
2515 scan = NEXTOPER(scan);
2516 if (flags & SCF_DO_STCLASS) {
2517 cl_init(pRExC_state, &this_class);
2518 data_fake.start_class = &this_class;
2519 f = SCF_DO_STCLASS_AND;
2521 if (flags & SCF_WHILEM_VISITED_POS)
2522 f |= SCF_WHILEM_VISITED_POS;
2524 /* we suppose the run is continuous, last=next...*/
2525 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2527 stopparen, recursed, NULL, f,depth+1);
2530 if (max1 < minnext + deltanext)
2531 max1 = minnext + deltanext;
2532 if (deltanext == I32_MAX)
2533 is_inf = is_inf_internal = 1;
2535 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2537 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2538 if ( stopmin > minnext)
2539 stopmin = min + min1;
2540 flags &= ~SCF_DO_SUBSTR;
2542 data->flags |= SCF_SEEN_ACCEPT;
2545 if (data_fake.flags & SF_HAS_EVAL)
2546 data->flags |= SF_HAS_EVAL;
2547 data->whilem_c = data_fake.whilem_c;
2549 if (flags & SCF_DO_STCLASS)
2550 cl_or(pRExC_state, &accum, &this_class);
2552 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2554 if (flags & SCF_DO_SUBSTR) {
2555 data->pos_min += min1;
2556 data->pos_delta += max1 - min1;
2557 if (max1 != min1 || is_inf)
2558 data->longest = &(data->longest_float);
2561 delta += max1 - min1;
2562 if (flags & SCF_DO_STCLASS_OR) {
2563 cl_or(pRExC_state, data->start_class, &accum);
2565 cl_and(data->start_class, and_withp);
2566 flags &= ~SCF_DO_STCLASS;
2569 else if (flags & SCF_DO_STCLASS_AND) {
2571 cl_and(data->start_class, &accum);
2572 flags &= ~SCF_DO_STCLASS;
2575 /* Switch to OR mode: cache the old value of
2576 * data->start_class */
2578 StructCopy(data->start_class, and_withp,
2579 struct regnode_charclass_class);
2580 flags &= ~SCF_DO_STCLASS_AND;
2581 StructCopy(&accum, data->start_class,
2582 struct regnode_charclass_class);
2583 flags |= SCF_DO_STCLASS_OR;
2584 data->start_class->flags |= ANYOF_EOS;
2588 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2591 Assuming this was/is a branch we are dealing with: 'scan' now
2592 points at the item that follows the branch sequence, whatever
2593 it is. We now start at the beginning of the sequence and look
2600 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2602 If we can find such a subseqence we need to turn the first
2603 element into a trie and then add the subsequent branch exact
2604 strings to the trie.
2608 1. patterns where the whole set of branch can be converted.
2610 2. patterns where only a subset can be converted.
2612 In case 1 we can replace the whole set with a single regop
2613 for the trie. In case 2 we need to keep the start and end
2616 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2617 becomes BRANCH TRIE; BRANCH X;
2619 There is an additional case, that being where there is a
2620 common prefix, which gets split out into an EXACT like node
2621 preceding the TRIE node.
2623 If x(1..n)==tail then we can do a simple trie, if not we make
2624 a "jump" trie, such that when we match the appropriate word
2625 we "jump" to the appopriate tail node. Essentailly we turn
2626 a nested if into a case structure of sorts.
2631 if (!re_trie_maxbuff) {
2632 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2633 if (!SvIOK(re_trie_maxbuff))
2634 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2636 if ( SvIV(re_trie_maxbuff)>=0 ) {
2638 regnode *first = (regnode *)NULL;
2639 regnode *last = (regnode *)NULL;
2640 regnode *tail = scan;
2645 SV * const mysv = sv_newmortal(); /* for dumping */
2647 /* var tail is used because there may be a TAIL
2648 regop in the way. Ie, the exacts will point to the
2649 thing following the TAIL, but the last branch will
2650 point at the TAIL. So we advance tail. If we
2651 have nested (?:) we may have to move through several
2655 while ( OP( tail ) == TAIL ) {
2656 /* this is the TAIL generated by (?:) */
2657 tail = regnext( tail );
2662 regprop(RExC_rx, mysv, tail );
2663 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2664 (int)depth * 2 + 2, "",
2665 "Looking for TRIE'able sequences. Tail node is: ",
2666 SvPV_nolen_const( mysv )
2672 step through the branches, cur represents each
2673 branch, noper is the first thing to be matched
2674 as part of that branch and noper_next is the
2675 regnext() of that node. if noper is an EXACT
2676 and noper_next is the same as scan (our current
2677 position in the regex) then the EXACT branch is
2678 a possible optimization target. Once we have
2679 two or more consequetive such branches we can
2680 create a trie of the EXACT's contents and stich
2681 it in place. If the sequence represents all of
2682 the branches we eliminate the whole thing and
2683 replace it with a single TRIE. If it is a
2684 subsequence then we need to stitch it in. This
2685 means the first branch has to remain, and needs
2686 to be repointed at the item on the branch chain
2687 following the last branch optimized. This could
2688 be either a BRANCH, in which case the
2689 subsequence is internal, or it could be the
2690 item following the branch sequence in which
2691 case the subsequence is at the end.
2695 /* dont use tail as the end marker for this traverse */
2696 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2697 regnode * const noper = NEXTOPER( cur );
2698 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2699 regnode * const noper_next = regnext( noper );
2703 regprop(RExC_rx, mysv, cur);
2704 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2705 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2707 regprop(RExC_rx, mysv, noper);
2708 PerlIO_printf( Perl_debug_log, " -> %s",
2709 SvPV_nolen_const(mysv));
2712 regprop(RExC_rx, mysv, noper_next );
2713 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2714 SvPV_nolen_const(mysv));
2716 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2717 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2719 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2720 : PL_regkind[ OP( noper ) ] == EXACT )
2721 || OP(noper) == NOTHING )
2723 && noper_next == tail
2728 if ( !first || optype == NOTHING ) {
2729 if (!first) first = cur;
2730 optype = OP( noper );
2736 make_trie( pRExC_state,
2737 startbranch, first, cur, tail, count,
2740 if ( PL_regkind[ OP( noper ) ] == EXACT
2742 && noper_next == tail
2747 optype = OP( noper );
2757 regprop(RExC_rx, mysv, cur);
2758 PerlIO_printf( Perl_debug_log,
2759 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2760 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2764 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2765 #ifdef TRIE_STUDY_OPT
2766 if ( ((made == MADE_EXACT_TRIE &&
2767 startbranch == first)
2768 || ( first_non_open == first )) &&
2770 flags |= SCF_TRIE_RESTUDY;
2771 if ( startbranch == first
2774 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2784 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2785 scan = NEXTOPER(NEXTOPER(scan));
2786 } else /* single branch is optimized. */
2787 scan = NEXTOPER(scan);
2789 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2790 scan_frame *newframe = NULL;
2795 if (OP(scan) != SUSPEND) {
2796 /* set the pointer */
2797 if (OP(scan) == GOSUB) {
2799 RExC_recurse[ARG2L(scan)] = scan;
2800 start = RExC_open_parens[paren-1];
2801 end = RExC_close_parens[paren-1];
2804 start = RExC_rxi->program + 1;
2808 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2809 SAVEFREEPV(recursed);
2811 if (!PAREN_TEST(recursed,paren+1)) {
2812 PAREN_SET(recursed,paren+1);
2813 Newx(newframe,1,scan_frame);
2815 if (flags & SCF_DO_SUBSTR) {
2816 SCAN_COMMIT(pRExC_state,data,minlenp);
2817 data->longest = &(data->longest_float);
2819 is_inf = is_inf_internal = 1;
2820 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2821 cl_anything(pRExC_state, data->start_class);
2822 flags &= ~SCF_DO_STCLASS;
2825 Newx(newframe,1,scan_frame);
2828 end = regnext(scan);
2833 SAVEFREEPV(newframe);
2834 newframe->next = regnext(scan);
2835 newframe->last = last;
2836 newframe->stop = stopparen;
2837 newframe->prev = frame;
2847 else if (OP(scan) == EXACT) {
2848 I32 l = STR_LEN(scan);
2851 const U8 * const s = (U8*)STRING(scan);
2852 l = utf8_length(s, s + l);
2853 uc = utf8_to_uvchr(s, NULL);
2855 uc = *((U8*)STRING(scan));
2858 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2859 /* The code below prefers earlier match for fixed
2860 offset, later match for variable offset. */
2861 if (data->last_end == -1) { /* Update the start info. */
2862 data->last_start_min = data->pos_min;
2863 data->last_start_max = is_inf
2864 ? I32_MAX : data->pos_min + data->pos_delta;
2866 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2868 SvUTF8_on(data->last_found);
2870 SV * const sv = data->last_found;
2871 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2872 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2873 if (mg && mg->mg_len >= 0)
2874 mg->mg_len += utf8_length((U8*)STRING(scan),
2875 (U8*)STRING(scan)+STR_LEN(scan));
2877 data->last_end = data->pos_min + l;
2878 data->pos_min += l; /* As in the first entry. */
2879 data->flags &= ~SF_BEFORE_EOL;
2881 if (flags & SCF_DO_STCLASS_AND) {
2882 /* Check whether it is compatible with what we know already! */
2886 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2887 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2888 && (!(data->start_class->flags & ANYOF_FOLD)
2889 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2892 ANYOF_CLASS_ZERO(data->start_class);
2893 ANYOF_BITMAP_ZERO(data->start_class);
2895 ANYOF_BITMAP_SET(data->start_class, uc);
2896 data->start_class->flags &= ~ANYOF_EOS;
2898 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2900 else if (flags & SCF_DO_STCLASS_OR) {
2901 /* false positive possible if the class is case-folded */
2903 ANYOF_BITMAP_SET(data->start_class, uc);
2905 data->start_class->flags |= ANYOF_UNICODE_ALL;
2906 data->start_class->flags &= ~ANYOF_EOS;
2907 cl_and(data->start_class, and_withp);
2909 flags &= ~SCF_DO_STCLASS;
2911 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2912 I32 l = STR_LEN(scan);
2913 UV uc = *((U8*)STRING(scan));
2915 /* Search for fixed substrings supports EXACT only. */
2916 if (flags & SCF_DO_SUBSTR) {
2918 SCAN_COMMIT(pRExC_state, data, minlenp);
2921 const U8 * const s = (U8 *)STRING(scan);
2922 l = utf8_length(s, s + l);
2923 uc = utf8_to_uvchr(s, NULL);
2926 if (flags & SCF_DO_SUBSTR)
2928 if (flags & SCF_DO_STCLASS_AND) {
2929 /* Check whether it is compatible with what we know already! */
2933 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2934 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2935 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2937 ANYOF_CLASS_ZERO(data->start_class);
2938 ANYOF_BITMAP_ZERO(data->start_class);
2940 ANYOF_BITMAP_SET(data->start_class, uc);
2941 data->start_class->flags &= ~ANYOF_EOS;
2942 data->start_class->flags |= ANYOF_FOLD;
2943 if (OP(scan) == EXACTFL)
2944 data->start_class->flags |= ANYOF_LOCALE;
2947 else if (flags & SCF_DO_STCLASS_OR) {
2948 if (data->start_class->flags & ANYOF_FOLD) {
2949 /* false positive possible if the class is case-folded.
2950 Assume that the locale settings are the same... */
2952 ANYOF_BITMAP_SET(data->start_class, uc);
2953 data->start_class->flags &= ~ANYOF_EOS;
2955 cl_and(data->start_class, and_withp);
2957 flags &= ~SCF_DO_STCLASS;
2959 else if (strchr((const char*)PL_varies,OP(scan))) {
2960 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2961 I32 f = flags, pos_before = 0;
2962 regnode * const oscan = scan;
2963 struct regnode_charclass_class this_class;
2964 struct regnode_charclass_class *oclass = NULL;
2965 I32 next_is_eval = 0;
2967 switch (PL_regkind[OP(scan)]) {
2968 case WHILEM: /* End of (?:...)* . */
2969 scan = NEXTOPER(scan);
2972 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2973 next = NEXTOPER(scan);
2974 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2976 maxcount = REG_INFTY;
2977 next = regnext(scan);
2978 scan = NEXTOPER(scan);
2982 if (flags & SCF_DO_SUBSTR)
2987 if (flags & SCF_DO_STCLASS) {
2989 maxcount = REG_INFTY;
2990 next = regnext(scan);
2991 scan = NEXTOPER(scan);
2994 is_inf = is_inf_internal = 1;
2995 scan = regnext(scan);
2996 if (flags & SCF_DO_SUBSTR) {
2997 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2998 data->longest = &(data->longest_float);
3000 goto optimize_curly_tail;
3002 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3003 && (scan->flags == stopparen))
3008 mincount = ARG1(scan);
3009 maxcount = ARG2(scan);
3011 next = regnext(scan);
3012 if (OP(scan) == CURLYX) {
3013 I32 lp = (data ? *(data->last_closep) : 0);
3014 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3016 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3017 next_is_eval = (OP(scan) == EVAL);
3019 if (flags & SCF_DO_SUBSTR) {
3020 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3021 pos_before = data->pos_min;
3025 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3027 data->flags |= SF_IS_INF;
3029 if (flags & SCF_DO_STCLASS) {
3030 cl_init(pRExC_state, &this_class);
3031 oclass = data->start_class;
3032 data->start_class = &this_class;
3033 f |= SCF_DO_STCLASS_AND;
3034 f &= ~SCF_DO_STCLASS_OR;
3036 /* These are the cases when once a subexpression
3037 fails at a particular position, it cannot succeed
3038 even after backtracking at the enclosing scope.
3040 XXXX what if minimal match and we are at the
3041 initial run of {n,m}? */
3042 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3043 f &= ~SCF_WHILEM_VISITED_POS;
3045 /* This will finish on WHILEM, setting scan, or on NULL: */
3046 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3047 last, data, stopparen, recursed, NULL,
3049 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3051 if (flags & SCF_DO_STCLASS)
3052 data->start_class = oclass;
3053 if (mincount == 0 || minnext == 0) {
3054 if (flags & SCF_DO_STCLASS_OR) {
3055 cl_or(pRExC_state, data->start_class, &this_class);
3057 else if (flags & SCF_DO_STCLASS_AND) {
3058 /* Switch to OR mode: cache the old value of
3059 * data->start_class */
3061 StructCopy(data->start_class, and_withp,
3062 struct regnode_charclass_class);
3063 flags &= ~SCF_DO_STCLASS_AND;
3064 StructCopy(&this_class, data->start_class,
3065 struct regnode_charclass_class);
3066 flags |= SCF_DO_STCLASS_OR;
3067 data->start_class->flags |= ANYOF_EOS;
3069 } else { /* Non-zero len */
3070 if (flags & SCF_DO_STCLASS_OR) {
3071 cl_or(pRExC_state, data->start_class, &this_class);
3072 cl_and(data->start_class, and_withp);
3074 else if (flags & SCF_DO_STCLASS_AND)
3075 cl_and(data->start_class, &this_class);
3076 flags &= ~SCF_DO_STCLASS;
3078 if (!scan) /* It was not CURLYX, but CURLY. */
3080 if ( /* ? quantifier ok, except for (?{ ... }) */
3081 (next_is_eval || !(mincount == 0 && maxcount == 1))
3082 && (minnext == 0) && (deltanext == 0)
3083 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3084 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3085 && ckWARN(WARN_REGEXP))
3088 "Quantifier unexpected on zero-length expression");
3091 min += minnext * mincount;
3092 is_inf_internal |= ((maxcount == REG_INFTY
3093 && (minnext + deltanext) > 0)
3094 || deltanext == I32_MAX);
3095 is_inf |= is_inf_internal;
3096 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3098 /* Try powerful optimization CURLYX => CURLYN. */
3099 if ( OP(oscan) == CURLYX && data
3100 && data->flags & SF_IN_PAR
3101 && !(data->flags & SF_HAS_EVAL)
3102 && !deltanext && minnext == 1 ) {
3103 /* Try to optimize to CURLYN. */
3104 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3105 regnode * const nxt1 = nxt;
3112 if (!strchr((const char*)PL_simple,OP(nxt))
3113 && !(PL_regkind[OP(nxt)] == EXACT
3114 && STR_LEN(nxt) == 1))
3120 if (OP(nxt) != CLOSE)
3122 if (RExC_open_parens) {
3123 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3124 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3126 /* Now we know that nxt2 is the only contents: */
3127 oscan->flags = (U8)ARG(nxt);
3129 OP(nxt1) = NOTHING; /* was OPEN. */
3132 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3133 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3134 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3135 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3136 OP(nxt + 1) = OPTIMIZED; /* was count. */
3137 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3142 /* Try optimization CURLYX => CURLYM. */
3143 if ( OP(oscan) == CURLYX && data
3144 && !(data->flags & SF_HAS_PAR)
3145 && !(data->flags & SF_HAS_EVAL)
3146 && !deltanext /* atom is fixed width */
3147 && minnext != 0 /* CURLYM can't handle zero width */
3149 /* XXXX How to optimize if data == 0? */
3150 /* Optimize to a simpler form. */
3151 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3155 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3156 && (OP(nxt2) != WHILEM))
3158 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3159 /* Need to optimize away parenths. */
3160 if (data->flags & SF_IN_PAR) {
3161 /* Set the parenth number. */
3162 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3164 if (OP(nxt) != CLOSE)
3165 FAIL("Panic opt close");
3166 oscan->flags = (U8)ARG(nxt);
3167 if (RExC_open_parens) {
3168 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3169 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3171 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3172 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3175 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3176 OP(nxt + 1) = OPTIMIZED; /* was count. */
3177 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3178 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3181 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3182 regnode *nnxt = regnext(nxt1);
3185 if (reg_off_by_arg[OP(nxt1)])
3186 ARG_SET(nxt1, nxt2 - nxt1);
3187 else if (nxt2 - nxt1 < U16_MAX)
3188 NEXT_OFF(nxt1) = nxt2 - nxt1;
3190 OP(nxt) = NOTHING; /* Cannot beautify */
3195 /* Optimize again: */
3196 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3197 NULL, stopparen, recursed, NULL, 0,depth+1);
3202 else if ((OP(oscan) == CURLYX)
3203 && (flags & SCF_WHILEM_VISITED_POS)
3204 /* See the comment on a similar expression above.
3205 However, this time it not a subexpression
3206 we care about, but the expression itself. */
3207 && (maxcount == REG_INFTY)
3208 && data && ++data->whilem_c < 16) {
3209 /* This stays as CURLYX, we can put the count/of pair. */
3210 /* Find WHILEM (as in regexec.c) */
3211 regnode *nxt = oscan + NEXT_OFF(oscan);
3213 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3215 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3216 | (RExC_whilem_seen << 4)); /* On WHILEM */
3218 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3220 if (flags & SCF_DO_SUBSTR) {
3221 SV *last_str = NULL;
3222 int counted = mincount != 0;
3224 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3225 #if defined(SPARC64_GCC_WORKAROUND)
3228 const char *s = NULL;
3231 if (pos_before >= data->last_start_min)
3234 b = data->last_start_min;
3237 s = SvPV_const(data->last_found, l);
3238 old = b - data->last_start_min;
3241 I32 b = pos_before >= data->last_start_min
3242 ? pos_before : data->last_start_min;
3244 const char * const s = SvPV_const(data->last_found, l);
3245 I32 old = b - data->last_start_min;
3249 old = utf8_hop((U8*)s, old) - (U8*)s;
3252 /* Get the added string: */
3253 last_str = newSVpvn(s + old, l);
3255 SvUTF8_on(last_str);
3256 if (deltanext == 0 && pos_before == b) {
3257 /* What was added is a constant string */
3259 SvGROW(last_str, (mincount * l) + 1);
3260 repeatcpy(SvPVX(last_str) + l,
3261 SvPVX_const(last_str), l, mincount - 1);
3262 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3263 /* Add additional parts. */
3264 SvCUR_set(data->last_found,
3265 SvCUR(data->last_found) - l);
3266 sv_catsv(data->last_found, last_str);
3268 SV * sv = data->last_found;
3270 SvUTF8(sv) && SvMAGICAL(sv) ?
3271 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3272 if (mg && mg->mg_len >= 0)
3273 mg->mg_len += CHR_SVLEN(last_str);
3275 data->last_end += l * (mincount - 1);
3278 /* start offset must point into the last copy */
3279 data->last_start_min += minnext * (mincount - 1);
3280 data->last_start_max += is_inf ? I32_MAX
3281 : (maxcount - 1) * (minnext + data->pos_delta);
3284 /* It is counted once already... */
3285 data->pos_min += minnext * (mincount - counted);
3286 data->pos_delta += - counted * deltanext +
3287 (minnext + deltanext) * maxcount - minnext * mincount;
3288 if (mincount != maxcount) {
3289 /* Cannot extend fixed substrings found inside
3291 SCAN_COMMIT(pRExC_state,data,minlenp);
3292 if (mincount && last_str) {
3293 SV * const sv = data->last_found;
3294 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3295 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3299 sv_setsv(sv, last_str);
3300 data->last_end = data->pos_min;
3301 data->last_start_min =
3302 data->pos_min - CHR_SVLEN(last_str);
3303 data->last_start_max = is_inf
3305 : data->pos_min + data->pos_delta
3306 - CHR_SVLEN(last_str);
3308 data->longest = &(data->longest_float);
3310 SvREFCNT_dec(last_str);
3312 if (data && (fl & SF_HAS_EVAL))
3313 data->flags |= SF_HAS_EVAL;
3314 optimize_curly_tail:
3315 if (OP(oscan) != CURLYX) {
3316 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3318 NEXT_OFF(oscan) += NEXT_OFF(next);
3321 default: /* REF and CLUMP only? */
3322 if (flags & SCF_DO_SUBSTR) {
3323 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3324 data->longest = &(data->longest_float);
3326 is_inf = is_inf_internal = 1;
3327 if (flags & SCF_DO_STCLASS_OR)
3328 cl_anything(pRExC_state, data->start_class);
3329 flags &= ~SCF_DO_STCLASS;
3333 else if (strchr((const char*)PL_simple,OP(scan))) {
3336 if (flags & SCF_DO_SUBSTR) {
3337 SCAN_COMMIT(pRExC_state,data,minlenp);
3341 if (flags & SCF_DO_STCLASS) {
3342 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3344 /* Some of the logic below assumes that switching
3345 locale on will only add false positives. */
3346 switch (PL_regkind[OP(scan)]) {
3350 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3351 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3352 cl_anything(pRExC_state, data->start_class);
3355 if (OP(scan) == SANY)
3357 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3358 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3359 || (data->start_class->flags & ANYOF_CLASS));
3360 cl_anything(pRExC_state, data->start_class);
3362 if (flags & SCF_DO_STCLASS_AND || !value)
3363 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3366 if (flags & SCF_DO_STCLASS_AND)
3367 cl_and(data->start_class,
3368 (struct regnode_charclass_class*)scan);
3370 cl_or(pRExC_state, data->start_class,
3371 (struct regnode_charclass_class*)scan);
3374 if (flags & SCF_DO_STCLASS_AND) {
3375 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3376 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3377 for (value = 0; value < 256; value++)
3378 if (!isALNUM(value))
3379 ANYOF_BITMAP_CLEAR(data->start_class, value);
3383 if (data->start_class->flags & ANYOF_LOCALE)
3384 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3386 for (value = 0; value < 256; value++)
3388 ANYOF_BITMAP_SET(data->start_class, value);
3393 if (flags & SCF_DO_STCLASS_AND) {
3394 if (data->start_class->flags & ANYOF_LOCALE)
3395 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3398 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3399 data->start_class->flags |= ANYOF_LOCALE;
3403 if (flags & SCF_DO_STCLASS_AND) {
3404 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3405 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3406 for (value = 0; value < 256; value++)
3408 ANYOF_BITMAP_CLEAR(data->start_class, value);
3412 if (data->start_class->flags & ANYOF_LOCALE)
3413 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3415 for (value = 0; value < 256; value++)
3416 if (!isALNUM(value))
3417 ANYOF_BITMAP_SET(data->start_class, value);
3422 if (flags & SCF_DO_STCLASS_AND) {
3423 if (data->start_class->flags & ANYOF_LOCALE)
3424 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3427 data->start_class->flags |= ANYOF_LOCALE;
3428 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3432 if (flags & SCF_DO_STCLASS_AND) {
3433 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3434 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3435 for (value = 0; value < 256; value++)
3436 if (!isSPACE(value))
3437 ANYOF_BITMAP_CLEAR(data->start_class, value);
3441 if (data->start_class->flags & ANYOF_LOCALE)
3442 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3444 for (value = 0; value < 256; value++)
3446 ANYOF_BITMAP_SET(data->start_class, value);
3451 if (flags & SCF_DO_STCLASS_AND) {
3452 if (data->start_class->flags & ANYOF_LOCALE)
3453 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3456 data->start_class->flags |= ANYOF_LOCALE;
3457 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3461 if (flags & SCF_DO_STCLASS_AND) {
3462 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3463 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3464 for (value = 0; value < 256; value++)
3466 ANYOF_BITMAP_CLEAR(data->start_class, value);
3470 if (data->start_class->flags & ANYOF_LOCALE)
3471 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3473 for (value = 0; value < 256; value++)
3474 if (!isSPACE(value))
3475 ANYOF_BITMAP_SET(data->start_class, value);
3480 if (flags & SCF_DO_STCLASS_AND) {
3481 if (data->start_class->flags & ANYOF_LOCALE) {
3482 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3483 for (value = 0; value < 256; value++)
3484 if (!isSPACE(value))
3485 ANYOF_BITMAP_CLEAR(data->start_class, value);
3489 data->start_class->flags |= ANYOF_LOCALE;
3490 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3494 if (flags & SCF_DO_STCLASS_AND) {
3495 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3496 for (value = 0; value < 256; value++)
3497 if (!isDIGIT(value))
3498 ANYOF_BITMAP_CLEAR(data->start_class, value);
3501 if (data->start_class->flags & ANYOF_LOCALE)
3502 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3504 for (value = 0; value < 256; value++)
3506 ANYOF_BITMAP_SET(data->start_class, value);
3511 if (flags & SCF_DO_STCLASS_AND) {
3512 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3513 for (value = 0; value < 256; value++)
3515 ANYOF_BITMAP_CLEAR(data->start_class, value);
3518 if (data->start_class->flags & ANYOF_LOCALE)
3519 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3521 for (value = 0; value < 256; value++)
3522 if (!isDIGIT(value))
3523 ANYOF_BITMAP_SET(data->start_class, value);
3528 if (flags & SCF_DO_STCLASS_OR)
3529 cl_and(data->start_class, and_withp);
3530 flags &= ~SCF_DO_STCLASS;
3533 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3534 data->flags |= (OP(scan) == MEOL
3538 else if ( PL_regkind[OP(scan)] == BRANCHJ
3539 /* Lookbehind, or need to calculate parens/evals/stclass: */
3540 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3541 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3542 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3543 || OP(scan) == UNLESSM )
3545 /* Negative Lookahead/lookbehind
3546 In this case we can't do fixed string optimisation.
3549 I32 deltanext, minnext, fake = 0;
3551 struct regnode_charclass_class intrnl;
3554 data_fake.flags = 0;
3556 data_fake.whilem_c = data->whilem_c;
3557 data_fake.last_closep = data->last_closep;
3560 data_fake.last_closep = &fake;
3561 data_fake.pos_delta = delta;
3562 if ( flags & SCF_DO_STCLASS && !scan->flags
3563 && OP(scan) == IFMATCH ) { /* Lookahead */
3564 cl_init(pRExC_state, &intrnl);
3565 data_fake.start_class = &intrnl;
3566 f |= SCF_DO_STCLASS_AND;
3568 if (flags & SCF_WHILEM_VISITED_POS)
3569 f |= SCF_WHILEM_VISITED_POS;
3570 next = regnext(scan);
3571 nscan = NEXTOPER(NEXTOPER(scan));
3572 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3573 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3576 FAIL("Variable length lookbehind not implemented");
3578 else if (minnext > (I32)U8_MAX) {
3579 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3581 scan->flags = (U8)minnext;
3584 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3586 if (data_fake.flags & SF_HAS_EVAL)
3587 data->flags |= SF_HAS_EVAL;
3588 data->whilem_c = data_fake.whilem_c;
3590 if (f & SCF_DO_STCLASS_AND) {
3591 const int was = (data->start_class->flags & ANYOF_EOS);
3593 cl_and(data->start_class, &intrnl);
3595 data->start_class->flags |= ANYOF_EOS;
3598 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3600 /* Positive Lookahead/lookbehind
3601 In this case we can do fixed string optimisation,
3602 but we must be careful about it. Note in the case of
3603 lookbehind the positions will be offset by the minimum
3604 length of the pattern, something we won't know about
3605 until after the recurse.
3607 I32 deltanext, fake = 0;
3609 struct regnode_charclass_class intrnl;
3611 /* We use SAVEFREEPV so that when the full compile
3612 is finished perl will clean up the allocated
3613 minlens when its all done. This was we don't
3614 have to worry about freeing them when we know
3615 they wont be used, which would be a pain.
3618 Newx( minnextp, 1, I32 );
3619 SAVEFREEPV(minnextp);
3622 StructCopy(data, &data_fake, scan_data_t);
3623 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3626 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3627 data_fake.last_found=newSVsv(data->last_found);
3631 data_fake.last_closep = &fake;
3632 data_fake.flags = 0;
3633 data_fake.pos_delta = delta;
3635 data_fake.flags |= SF_IS_INF;
3636 if ( flags & SCF_DO_STCLASS && !scan->flags
3637 && OP(scan) == IFMATCH ) { /* Lookahead */
3638 cl_init(pRExC_state, &intrnl);
3639 data_fake.start_class = &intrnl;
3640 f |= SCF_DO_STCLASS_AND;
3642 if (flags & SCF_WHILEM_VISITED_POS)
3643 f |= SCF_WHILEM_VISITED_POS;
3644 next = regnext(scan);
3645 nscan = NEXTOPER(NEXTOPER(scan));
3647 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3648 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3651 FAIL("Variable length lookbehind not implemented");
3653 else if (*minnextp > (I32)U8_MAX) {
3654 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3656 scan->flags = (U8)*minnextp;
3661 if (f & SCF_DO_STCLASS_AND) {
3662 const int was = (data->start_class->flags & ANYOF_EOS);
3664 cl_and(data->start_class, &intrnl);
3666 data->start_class->flags |= ANYOF_EOS;
3669 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3671 if (data_fake.flags & SF_HAS_EVAL)
3672 data->flags |= SF_HAS_EVAL;
3673 data->whilem_c = data_fake.whilem_c;
3674 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3675 if (RExC_rx->minlen<*minnextp)
3676 RExC_rx->minlen=*minnextp;
3677 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3678 SvREFCNT_dec(data_fake.last_found);
3680 if ( data_fake.minlen_fixed != minlenp )
3682 data->offset_fixed= data_fake.offset_fixed;
3683 data->minlen_fixed= data_fake.minlen_fixed;
3684 data->lookbehind_fixed+= scan->flags;
3686 if ( data_fake.minlen_float != minlenp )
3688 data->minlen_float= data_fake.minlen_float;
3689 data->offset_float_min=data_fake.offset_float_min;
3690 data->offset_float_max=data_fake.offset_float_max;
3691 data->lookbehind_float+= scan->flags;
3700 else if (OP(scan) == OPEN) {
3701 if (stopparen != (I32)ARG(scan))
3704 else if (OP(scan) == CLOSE) {
3705 if (stopparen == (I32)ARG(scan)) {
3708 if ((I32)ARG(scan) == is_par) {
3709 next = regnext(scan);
3711 if ( next && (OP(next) != WHILEM) && next < last)
3712 is_par = 0; /* Disable optimization */
3715 *(data->last_closep) = ARG(scan);
3717 else if (OP(scan) == EVAL) {
3719 data->flags |= SF_HAS_EVAL;
3721 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3722 if (flags & SCF_DO_SUBSTR) {
3723 SCAN_COMMIT(pRExC_state,data,minlenp);
3724 flags &= ~SCF_DO_SUBSTR;
3726 if (data && OP(scan)==ACCEPT) {
3727 data->flags |= SCF_SEEN_ACCEPT;
3732 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3734 if (flags & SCF_DO_SUBSTR) {
3735 SCAN_COMMIT(pRExC_state,data,minlenp);
3736 data->longest = &(data->longest_float);
3738 is_inf = is_inf_internal = 1;
3739 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3740 cl_anything(pRExC_state, data->start_class);
3741 flags &= ~SCF_DO_STCLASS;
3743 else if (OP(scan) == GPOS) {
3744 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3745 !(delta || is_inf || (data && data->pos_delta)))
3747 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3748 RExC_rx->extflags |= RXf_ANCH_GPOS;
3749 if (RExC_rx->gofs < (U32)min)
3750 RExC_rx->gofs = min;
3752 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3756 #ifdef TRIE_STUDY_OPT
3757 #ifdef FULL_TRIE_STUDY
3758 else if (PL_regkind[OP(scan)] == TRIE) {
3759 /* NOTE - There is similar code to this block above for handling
3760 BRANCH nodes on the initial study. If you change stuff here
3762 regnode *trie_node= scan;
3763 regnode *tail= regnext(scan);
3764 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3765 I32 max1 = 0, min1 = I32_MAX;
3766 struct regnode_charclass_class accum;
3768 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3769 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3770 if (flags & SCF_DO_STCLASS)
3771 cl_init_zero(pRExC_state, &accum);
3777 const regnode *nextbranch= NULL;
3780 for ( word=1 ; word <= trie->wordcount ; word++)
3782 I32 deltanext=0, minnext=0, f = 0, fake;
3783 struct regnode_charclass_class this_class;
3785 data_fake.flags = 0;
3787 data_fake.whilem_c = data->whilem_c;
3788 data_fake.last_closep = data->last_closep;
3791 data_fake.last_closep = &fake;
3792 data_fake.pos_delta = delta;
3793 if (flags & SCF_DO_STCLASS) {
3794 cl_init(pRExC_state, &this_class);
3795 data_fake.start_class = &this_class;
3796 f = SCF_DO_STCLASS_AND;
3798 if (flags & SCF_WHILEM_VISITED_POS)
3799 f |= SCF_WHILEM_VISITED_POS;
3801 if (trie->jump[word]) {
3803 nextbranch = trie_node + trie->jump[0];
3804 scan= trie_node + trie->jump[word];
3805 /* We go from the jump point to the branch that follows
3806 it. Note this means we need the vestigal unused branches
3807 even though they arent otherwise used.
3809 minnext = study_chunk(pRExC_state, &scan, minlenp,
3810 &deltanext, (regnode *)nextbranch, &data_fake,
3811 stopparen, recursed, NULL, f,depth+1);
3813 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3814 nextbranch= regnext((regnode*)nextbranch);
3816 if (min1 > (I32)(minnext + trie->minlen))
3817 min1 = minnext + trie->minlen;
3818 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3819 max1 = minnext + deltanext + trie->maxlen;
3820 if (deltanext == I32_MAX)
3821 is_inf = is_inf_internal = 1;
3823 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3825 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3826 if ( stopmin > min + min1)
3827 stopmin = min + min1;
3828 flags &= ~SCF_DO_SUBSTR;
3830 data->flags |= SCF_SEEN_ACCEPT;
3833 if (data_fake.flags & SF_HAS_EVAL)
3834 data->flags |= SF_HAS_EVAL;
3835 data->whilem_c = data_fake.whilem_c;
3837 if (flags & SCF_DO_STCLASS)
3838 cl_or(pRExC_state, &accum, &this_class);
3841 if (flags & SCF_DO_SUBSTR) {
3842 data->pos_min += min1;
3843 data->pos_delta += max1 - min1;
3844 if (max1 != min1 || is_inf)
3845 data->longest = &(data->longest_float);
3848 delta += max1 - min1;
3849 if (flags & SCF_DO_STCLASS_OR) {
3850 cl_or(pRExC_state, data->start_class, &accum);
3852 cl_and(data->start_class, and_withp);
3853 flags &= ~SCF_DO_STCLASS;
3856 else if (flags & SCF_DO_STCLASS_AND) {
3858 cl_and(data->start_class, &accum);
3859 flags &= ~SCF_DO_STCLASS;
3862 /* Switch to OR mode: cache the old value of
3863 * data->start_class */
3865 StructCopy(data->start_class, and_withp,
3866 struct regnode_charclass_class);
3867 flags &= ~SCF_DO_STCLASS_AND;
3868 StructCopy(&accum, data->start_class,
3869 struct regnode_charclass_class);
3870 flags |= SCF_DO_STCLASS_OR;
3871 data->start_class->flags |= ANYOF_EOS;
3878 else if (PL_regkind[OP(scan)] == TRIE) {
3879 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3882 min += trie->minlen;
3883 delta += (trie->maxlen - trie->minlen);
3884 flags &= ~SCF_DO_STCLASS; /* xxx */
3885 if (flags & SCF_DO_SUBSTR) {
3886 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3887 data->pos_min += trie->minlen;
3888 data->pos_delta += (trie->maxlen - trie->minlen);
3889 if (trie->maxlen != trie->minlen)
3890 data->longest = &(data->longest_float);
3892 if (trie->jump) /* no more substrings -- for now /grr*/
3893 flags &= ~SCF_DO_SUBSTR;
3895 #endif /* old or new */
3896 #endif /* TRIE_STUDY_OPT */
3897 /* Else: zero-length, ignore. */
3898 scan = regnext(scan);
3903 stopparen = frame->stop;
3904 frame = frame->prev;
3905 goto fake_study_recurse;
3910 DEBUG_STUDYDATA("pre-fin:",data,depth);
3913 *deltap = is_inf_internal ? I32_MAX : delta;
3914 if (flags & SCF_DO_SUBSTR && is_inf)
3915 data->pos_delta = I32_MAX - data->pos_min;
3916 if (is_par > (I32)U8_MAX)
3918 if (is_par && pars==1 && data) {
3919 data->flags |= SF_IN_PAR;
3920 data->flags &= ~SF_HAS_PAR;
3922 else if (pars && data) {
3923 data->flags |= SF_HAS_PAR;
3924 data->flags &= ~SF_IN_PAR;
3926 if (flags & SCF_DO_STCLASS_OR)
3927 cl_and(data->start_class, and_withp);
3928 if (flags & SCF_TRIE_RESTUDY)
3929 data->flags |= SCF_TRIE_RESTUDY;
3931 DEBUG_STUDYDATA("post-fin:",data,depth);
3933 return min < stopmin ? min : stopmin;
3937 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3939 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3941 Renewc(RExC_rxi->data,
3942 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3943 char, struct reg_data);
3945 Renew(RExC_rxi->data->what, count + n, U8);
3947 Newx(RExC_rxi->data->what, n, U8);
3948 RExC_rxi->data->count = count + n;
3949 Copy(s, RExC_rxi->data->what + count, n, U8);
3953 /*XXX: todo make this not included in a non debugging perl */
3954 #ifndef PERL_IN_XSUB_RE
3956 Perl_reginitcolors(pTHX)
3959 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3961 char *t = savepv(s);
3965 t = strchr(t, '\t');
3971 PL_colors[i] = t = (char *)"";
3976 PL_colors[i++] = (char *)"";
3983 #ifdef TRIE_STUDY_OPT
3984 #define CHECK_RESTUDY_GOTO \
3986 (data.flags & SCF_TRIE_RESTUDY) \
3990 #define CHECK_RESTUDY_GOTO
3994 - pregcomp - compile a regular expression into internal code
3996 * We can't allocate space until we know how big the compiled form will be,
3997 * but we can't compile it (and thus know how big it is) until we've got a
3998 * place to put the code. So we cheat: we compile it twice, once with code
3999 * generation turned off and size counting turned on, and once "for real".
4000 * This also means that we don't allocate space until we are sure that the
4001 * thing really will compile successfully, and we never have to move the
4002 * code and thus invalidate pointers into it. (Note that it has to be in
4003 * one piece because free() must be able to free it all.) [NB: not true in perl]
4005 * Beware that the optimization-preparation code in here knows about some
4006 * of the structure of the compiled regexp. [I'll say.]
4011 #ifndef PERL_IN_XSUB_RE
4012 #define RE_ENGINE_PTR &PL_core_reg_engine
4014 extern const struct regexp_engine my_reg_engine;
4015 #define RE_ENGINE_PTR &my_reg_engine
4018 #ifndef PERL_IN_XSUB_RE
4020 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
4023 HV * const table = GvHV(PL_hintgv);
4024 /* Dispatch a request to compile a regexp to correct
4027 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4028 GET_RE_DEBUG_FLAGS_DECL;
4029 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4030 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4032 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4035 return CALLREGCOMP_ENG(eng, exp, xend, pm);
4038 return Perl_re_compile(aTHX_ exp, xend, pm);
4043 Perl_re_compile(pTHX_ char *exp, char *xend, PMOP *pm)
4047 register regexp_internal *ri;
4055 RExC_state_t RExC_state;
4056 RExC_state_t * const pRExC_state = &RExC_state;
4057 #ifdef TRIE_STUDY_OPT
4059 RExC_state_t copyRExC_state;
4061 GET_RE_DEBUG_FLAGS_DECL;
4062 DEBUG_r(if (!PL_colorset) reginitcolors());
4065 FAIL("NULL regexp argument");
4067 RExC_utf8 = RExC_orig_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4070 SV *dsv= sv_newmortal();
4071 RE_PV_QUOTED_DECL(s, RExC_utf8,
4072 dsv, exp, (xend - exp), 60);
4073 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4074 PL_colors[4],PL_colors[5],s);
4079 RExC_flags = pm->op_pmflags;
4083 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4084 RExC_seen_evals = 0;
4087 /* First pass: determine size, legality. */
4095 RExC_emit = &PL_regdummy;
4096 RExC_whilem_seen = 0;
4097 RExC_charnames = NULL;
4098 RExC_open_parens = NULL;
4099 RExC_close_parens = NULL;
4101 RExC_paren_names = NULL;
4103 RExC_paren_name_list = NULL;
4105 RExC_recurse = NULL;
4106 RExC_recurse_count = 0;
4108 #if 0 /* REGC() is (currently) a NOP at the first pass.
4109 * Clever compilers notice this and complain. --jhi */
4110 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4112 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4113 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4114 RExC_precomp = NULL;
4117 if (RExC_utf8 && !RExC_orig_utf8) {
4118 /* It's possible to write a regexp in ascii that represents unicode
4119 codepoints outside of the byte range, such as via \x{100}. If we
4120 detect such a sequence we have to convert the entire pattern to utf8
4121 and then recompile, as our sizing calculation will have been based
4122 on 1 byte == 1 character, but we will need to use utf8 to encode
4123 at least some part of the pattern, and therefore must convert the whole
4125 XXX: somehow figure out how to make this less expensive...
4127 STRLEN len = xend-exp;
4128 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4129 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4130 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4132 RExC_orig_utf8 = RExC_utf8;
4134 goto redo_first_pass;
4137 PerlIO_printf(Perl_debug_log,
4138 "Required size %"IVdf" nodes\n"
4139 "Starting second pass (creation)\n",
4142 RExC_lastparse=NULL;
4144 /* Small enough for pointer-storage convention?
4145 If extralen==0, this means that we will not need long jumps. */
4146 if (RExC_size >= 0x10000L && RExC_extralen)
4147 RExC_size += RExC_extralen;
4150 if (RExC_whilem_seen > 15)
4151 RExC_whilem_seen = 15;
4153 /* Allocate space and zero-initialize. Note, the two step process
4154 of zeroing when in debug mode, thus anything assigned has to
4155 happen after that */
4156 Newxz(r, 1, regexp);
4157 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4158 char, regexp_internal);
4159 if ( r == NULL || ri == NULL )
4160 FAIL("Regexp out of space");
4162 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4163 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4165 /* bulk initialize base fields with 0. */
4166 Zero(ri, sizeof(regexp_internal), char);
4169 /* non-zero initialization begins here */
4171 r->engine= RE_ENGINE_PTR;
4173 r->prelen = xend - exp;
4174 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4176 bool has_k = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4177 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4178 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4179 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) >> 12);
4180 const char *fptr = STD_PAT_MODS; /*"msix"*/
4182 r->wraplen = r->prelen + has_minus + has_k + has_runon
4183 + (sizeof(STD_PAT_MODS) - 1)
4184 + (sizeof("(?:)") - 1);
4186 Newx(r->wrapped, r->wraplen, char );
4190 *p++ = KEEPCOPY_PAT_MOD; /*'k'*/
4192 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4193 char *colon = r + 1;
4196 while((ch = *fptr++)) {
4210 Copy(RExC_precomp, p, r->prelen, char);
4219 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4221 if (RExC_seen & REG_SEEN_RECURSE) {
4222 Newxz(RExC_open_parens, RExC_npar,regnode *);
4223 SAVEFREEPV(RExC_open_parens);
4224 Newxz(RExC_close_parens,RExC_npar,regnode *);
4225 SAVEFREEPV(RExC_close_parens);
4228 /* Useful during FAIL. */
4229 #ifdef RE_TRACK_PATTERN_OFFSETS
4230 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4231 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4232 "%s %"UVuf" bytes for offset annotations.\n",
4233 ri->u.offsets ? "Got" : "Couldn't get",
4234 (UV)((2*RExC_size+1) * sizeof(U32))));
4236 SetProgLen(ri,RExC_size);
4240 /* Second pass: emit code. */
4241 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4246 RExC_emit_start = ri->program;
4247 RExC_emit = ri->program;
4248 RExC_emit_bound = ri->program + RExC_size + 1;
4250 /* Store the count of eval-groups for security checks: */
4251 RExC_rx->seen_evals = RExC_seen_evals;
4252 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4253 if (reg(pRExC_state, 0, &flags,1) == NULL)
4256 /* XXXX To minimize changes to RE engine we always allocate
4257 3-units-long substrs field. */
4258 Newx(r->substrs, 1, struct reg_substr_data);
4259 if (RExC_recurse_count) {
4260 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4261 SAVEFREEPV(RExC_recurse);
4265 r->minlen = minlen = sawplus = sawopen = 0;
4266 Zero(r->substrs, 1, struct reg_substr_data);
4268 #ifdef TRIE_STUDY_OPT
4271 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4273 RExC_state = copyRExC_state;
4274 if (seen & REG_TOP_LEVEL_BRANCHES)
4275 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4277 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4278 if (data.last_found) {
4279 SvREFCNT_dec(data.longest_fixed);
4280 SvREFCNT_dec(data.longest_float);
4281 SvREFCNT_dec(data.last_found);
4283 StructCopy(&zero_scan_data, &data, scan_data_t);
4285 StructCopy(&zero_scan_data, &data, scan_data_t);
4286 copyRExC_state = RExC_state;
4289 StructCopy(&zero_scan_data, &data, scan_data_t);
4292 /* Dig out information for optimizations. */
4293 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4294 pm->op_pmflags = RExC_flags;
4296 r->extflags |= RXf_UTF8; /* Unicode in it? */
4297 ri->regstclass = NULL;
4298 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4299 r->intflags |= PREGf_NAUGHTY;
4300 scan = ri->program + 1; /* First BRANCH. */
4302 /* testing for BRANCH here tells us whether there is "must appear"
4303 data in the pattern. If there is then we can use it for optimisations */
4304 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4306 STRLEN longest_float_length, longest_fixed_length;
4307 struct regnode_charclass_class ch_class; /* pointed to by data */
4309 I32 last_close = 0; /* pointed to by data */
4312 /* Skip introductions and multiplicators >= 1. */
4313 while ((OP(first) == OPEN && (sawopen = 1)) ||
4314 /* An OR of *one* alternative - should not happen now. */
4315 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4316 /* for now we can't handle lookbehind IFMATCH*/
4317 (OP(first) == IFMATCH && !first->flags) ||
4318 (OP(first) == PLUS) ||
4319 (OP(first) == MINMOD) ||
4320 /* An {n,m} with n>0 */
4321 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4324 if (OP(first) == PLUS)
4327 first += regarglen[OP(first)];
4328 if (OP(first) == IFMATCH) {
4329 first = NEXTOPER(first);
4330 first += EXTRA_STEP_2ARGS;
4331 } else /* XXX possible optimisation for /(?=)/ */
4332 first = NEXTOPER(first);
4335 /* Starting-point info. */
4337 DEBUG_PEEP("first:",first,0);
4338 /* Ignore EXACT as we deal with it later. */
4339 if (PL_regkind[OP(first)] == EXACT) {
4340 if (OP(first) == EXACT)
4341 NOOP; /* Empty, get anchored substr later. */
4342 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4343 ri->regstclass = first;
4346 else if (PL_regkind[OP(first)] == TRIE &&
4347 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4350 /* this can happen only on restudy */
4351 if ( OP(first) == TRIE ) {
4352 struct regnode_1 *trieop = (struct regnode_1 *)
4353 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4354 StructCopy(first,trieop,struct regnode_1);
4355 trie_op=(regnode *)trieop;
4357 struct regnode_charclass *trieop = (struct regnode_charclass *)
4358 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4359 StructCopy(first,trieop,struct regnode_charclass);
4360 trie_op=(regnode *)trieop;
4363 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4364 ri->regstclass = trie_op;
4367 else if (strchr((const char*)PL_simple,OP(first)))
4368 ri->regstclass = first;
4369 else if (PL_regkind[OP(first)] == BOUND ||
4370 PL_regkind[OP(first)] == NBOUND)
4371 ri->regstclass = first;
4372 else if (PL_regkind[OP(first)] == BOL) {
4373 r->extflags |= (OP(first) == MBOL
4375 : (OP(first) == SBOL
4378 first = NEXTOPER(first);
4381 else if (OP(first) == GPOS) {
4382 r->extflags |= RXf_ANCH_GPOS;
4383 first = NEXTOPER(first);
4386 else if ((!sawopen || !RExC_sawback) &&
4387 (OP(first) == STAR &&
4388 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4389 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4391 /* turn .* into ^.* with an implied $*=1 */
4393 (OP(NEXTOPER(first)) == REG_ANY)
4396 r->extflags |= type;
4397 r->intflags |= PREGf_IMPLICIT;
4398 first = NEXTOPER(first);
4401 if (sawplus && (!sawopen || !RExC_sawback)
4402 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4403 /* x+ must match at the 1st pos of run of x's */
4404 r->intflags |= PREGf_SKIP;
4406 /* Scan is after the zeroth branch, first is atomic matcher. */
4407 #ifdef TRIE_STUDY_OPT
4410 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4411 (IV)(first - scan + 1))
4415 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4416 (IV)(first - scan + 1))
4422 * If there's something expensive in the r.e., find the
4423 * longest literal string that must appear and make it the
4424 * regmust. Resolve ties in favor of later strings, since
4425 * the regstart check works with the beginning of the r.e.
4426 * and avoiding duplication strengthens checking. Not a
4427 * strong reason, but sufficient in the absence of others.
4428 * [Now we resolve ties in favor of the earlier string if
4429 * it happens that c_offset_min has been invalidated, since the
4430 * earlier string may buy us something the later one won't.]
4433 data.longest_fixed = newSVpvs("");
4434 data.longest_float = newSVpvs("");
4435 data.last_found = newSVpvs("");
4436 data.longest = &(data.longest_fixed);
4438 if (!ri->regstclass) {
4439 cl_init(pRExC_state, &ch_class);
4440 data.start_class = &ch_class;
4441 stclass_flag = SCF_DO_STCLASS_AND;
4442 } else /* XXXX Check for BOUND? */
4444 data.last_closep = &last_close;
4446 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4447 &data, -1, NULL, NULL,
4448 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4454 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4455 && data.last_start_min == 0 && data.last_end > 0
4456 && !RExC_seen_zerolen
4457 && !(RExC_seen & REG_SEEN_VERBARG)
4458 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4459 r->extflags |= RXf_CHECK_ALL;
4460 scan_commit(pRExC_state, &data,&minlen,0);
4461 SvREFCNT_dec(data.last_found);
4463 /* Note that code very similar to this but for anchored string
4464 follows immediately below, changes may need to be made to both.
4467 longest_float_length = CHR_SVLEN(data.longest_float);
4468 if (longest_float_length
4469 || (data.flags & SF_FL_BEFORE_EOL
4470 && (!(data.flags & SF_FL_BEFORE_MEOL)
4471 || (RExC_flags & RXf_PMf_MULTILINE))))
4475 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4476 && data.offset_fixed == data.offset_float_min
4477 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4478 goto remove_float; /* As in (a)+. */
4480 /* copy the information about the longest float from the reg_scan_data
4481 over to the program. */
4482 if (SvUTF8(data.longest_float)) {
4483 r->float_utf8 = data.longest_float;
4484 r->float_substr = NULL;
4486 r->float_substr = data.longest_float;
4487 r->float_utf8 = NULL;
4489 /* float_end_shift is how many chars that must be matched that
4490 follow this item. We calculate it ahead of time as once the
4491 lookbehind offset is added in we lose the ability to correctly
4493 ml = data.minlen_float ? *(data.minlen_float)
4494 : (I32)longest_float_length;
4495 r->float_end_shift = ml - data.offset_float_min
4496 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4497 + data.lookbehind_float;
4498 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4499 r->float_max_offset = data.offset_float_max;
4500 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4501 r->float_max_offset -= data.lookbehind_float;
4503 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4504 && (!(data.flags & SF_FL_BEFORE_MEOL)
4505 || (RExC_flags & RXf_PMf_MULTILINE)));
4506 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4510 r->float_substr = r->float_utf8 = NULL;
4511 SvREFCNT_dec(data.longest_float);
4512 longest_float_length = 0;
4515 /* Note that code very similar to this but for floating string
4516 is immediately above, changes may need to be made to both.
4519 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4520 if (longest_fixed_length
4521 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4522 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4523 || (RExC_flags & RXf_PMf_MULTILINE))))
4527 /* copy the information about the longest fixed
4528 from the reg_scan_data over to the program. */
4529 if (SvUTF8(data.longest_fixed)) {
4530 r->anchored_utf8 = data.longest_fixed;
4531 r->anchored_substr = NULL;
4533 r->anchored_substr = data.longest_fixed;
4534 r->anchored_utf8 = NULL;
4536 /* fixed_end_shift is how many chars that must be matched that
4537 follow this item. We calculate it ahead of time as once the
4538 lookbehind offset is added in we lose the ability to correctly
4540 ml = data.minlen_fixed ? *(data.minlen_fixed)
4541 : (I32)longest_fixed_length;
4542 r->anchored_end_shift = ml - data.offset_fixed
4543 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4544 + data.lookbehind_fixed;
4545 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4547 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4548 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4549 || (RExC_flags & RXf_PMf_MULTILINE)));
4550 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4553 r->anchored_substr = r->anchored_utf8 = NULL;
4554 SvREFCNT_dec(data.longest_fixed);
4555 longest_fixed_length = 0;
4558 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4559 ri->regstclass = NULL;
4560 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4562 && !(data.start_class->flags & ANYOF_EOS)
4563 && !cl_is_anything(data.start_class))
4565 const U32 n = add_data(pRExC_state, 1, "f");
4567 Newx(RExC_rxi->data->data[n], 1,
4568 struct regnode_charclass_class);
4569 StructCopy(data.start_class,
4570 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4571 struct regnode_charclass_class);
4572 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4573 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4574 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4575 regprop(r, sv, (regnode*)data.start_class);
4576 PerlIO_printf(Perl_debug_log,
4577 "synthetic stclass \"%s\".\n",
4578 SvPVX_const(sv));});
4581 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4582 if (longest_fixed_length > longest_float_length) {
4583 r->check_end_shift = r->anchored_end_shift;
4584 r->check_substr = r->anchored_substr;
4585 r->check_utf8 = r->anchored_utf8;
4586 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4587 if (r->extflags & RXf_ANCH_SINGLE)
4588 r->extflags |= RXf_NOSCAN;
4591 r->check_end_shift = r->float_end_shift;
4592 r->check_substr = r->float_substr;
4593 r->check_utf8 = r->float_utf8;
4594 r->check_offset_min = r->float_min_offset;
4595 r->check_offset_max = r->float_max_offset;
4597 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4598 This should be changed ASAP! */
4599 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4600 r->extflags |= RXf_USE_INTUIT;
4601 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4602 r->extflags |= RXf_INTUIT_TAIL;
4604 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4605 if ( (STRLEN)minlen < longest_float_length )
4606 minlen= longest_float_length;
4607 if ( (STRLEN)minlen < longest_fixed_length )
4608 minlen= longest_fixed_length;
4612 /* Several toplevels. Best we can is to set minlen. */
4614 struct regnode_charclass_class ch_class;
4617 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4619 scan = ri->program + 1;
4620 cl_init(pRExC_state, &ch_class);
4621 data.start_class = &ch_class;
4622 data.last_closep = &last_close;
4625 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4626 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4630 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4631 = r->float_substr = r->float_utf8 = NULL;
4632 if (!(data.start_class->flags & ANYOF_EOS)
4633 && !cl_is_anything(data.start_class))
4635 const U32 n = add_data(pRExC_state, 1, "f");
4637 Newx(RExC_rxi->data->data[n], 1,
4638 struct regnode_charclass_class);
4639 StructCopy(data.start_class,
4640 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4641 struct regnode_charclass_class);
4642 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4643 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4644 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4645 regprop(r, sv, (regnode*)data.start_class);
4646 PerlIO_printf(Perl_debug_log,
4647 "synthetic stclass \"%s\".\n",
4648 SvPVX_const(sv));});
4652 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4653 the "real" pattern. */
4655 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4656 (IV)minlen, (IV)r->minlen);
4658 r->minlenret = minlen;
4659 if (r->minlen < minlen)
4662 if (RExC_seen & REG_SEEN_GPOS)
4663 r->extflags |= RXf_GPOS_SEEN;
4664 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4665 r->extflags |= RXf_LOOKBEHIND_SEEN;
4666 if (RExC_seen & REG_SEEN_EVAL)
4667 r->extflags |= RXf_EVAL_SEEN;
4668 if (RExC_seen & REG_SEEN_CANY)
4669 r->extflags |= RXf_CANY_SEEN;
4670 if (RExC_seen & REG_SEEN_VERBARG)
4671 r->intflags |= PREGf_VERBARG_SEEN;
4672 if (RExC_seen & REG_SEEN_CUTGROUP)
4673 r->intflags |= PREGf_CUTGROUP_SEEN;
4674 if (RExC_paren_names)
4675 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4677 r->paren_names = NULL;
4678 if (r->prelen == 3 && strEQ("\\s+", r->precomp))
4679 r->extflags |= RXf_WHITE;
4680 else if (r->prelen == 1 && r->precomp[0] == '^')
4681 r->extflags |= RXf_START_ONLY;
4684 if (RExC_paren_names) {
4685 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4686 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4689 ri->name_list_idx = 0;
4691 if (RExC_recurse_count) {
4692 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4693 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4694 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4697 Newxz(r->startp, RExC_npar, I32);
4698 Newxz(r->endp, RExC_npar, I32);
4699 /* assume we don't need to swap parens around before we match */
4702 PerlIO_printf(Perl_debug_log,"Final program:\n");
4705 #ifdef RE_TRACK_PATTERN_OFFSETS
4706 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4707 const U32 len = ri->u.offsets[0];
4709 GET_RE_DEBUG_FLAGS_DECL;
4710 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4711 for (i = 1; i <= len; i++) {
4712 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4713 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4714 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4716 PerlIO_printf(Perl_debug_log, "\n");
4722 #undef RE_ENGINE_PTR
4726 Perl_reg_named_buff_get(pTHX_ const REGEXP * const rx, SV* namesv, U32 flags)
4728 AV *retarray = NULL;
4733 if (rx && rx->paren_names) {
4734 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4737 SV* sv_dat=HeVAL(he_str);
4738 I32 *nums=(I32*)SvPVX(sv_dat);
4739 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4740 if ((I32)(rx->nparens) >= nums[i]
4741 && rx->startp[nums[i]] != -1
4742 && rx->endp[nums[i]] != -1)
4744 ret = CALLREG_NUMBUF(rx,nums[i],NULL);
4748 ret = newSVsv(&PL_sv_undef);
4752 av_push(retarray, ret);
4756 return (SV*)retarray;
4763 Perl_reg_numbered_buff_get(pTHX_ const REGEXP * const rx, I32 paren, SV* usesv)
4768 SV *sv = usesv ? usesv : newSVpvs("");
4771 sv_setsv(sv,&PL_sv_undef);
4775 if (paren == -2 && rx->startp[0] != -1) {
4781 if (paren == -1 && rx->endp[0] != -1) {
4783 s = rx->subbeg + rx->endp[0];
4784 i = rx->sublen - rx->endp[0];
4787 if ( 0 <= paren && paren <= (I32)rx->nparens &&
4788 (s1 = rx->startp[paren]) != -1 &&
4789 (t1 = rx->endp[paren]) != -1)
4793 s = rx->subbeg + s1;
4795 sv_setsv(sv,&PL_sv_undef);
4798 assert(rx->sublen >= (s - rx->subbeg) + i );
4800 const int oldtainted = PL_tainted;
4802 sv_setpvn(sv, s, i);
4803 PL_tainted = oldtainted;
4804 if ( (rx->extflags & RXf_CANY_SEEN)
4805 ? (RX_MATCH_UTF8(rx)
4806 && (!i || is_utf8_string((U8*)s, i)))
4807 : (RX_MATCH_UTF8(rx)) )
4814 if (RX_MATCH_TAINTED(rx)) {
4815 if (SvTYPE(sv) >= SVt_PVMG) {
4816 MAGIC* const mg = SvMAGIC(sv);
4819 SvMAGIC_set(sv, mg->mg_moremagic);
4821 if ((mgt = SvMAGIC(sv))) {
4822 mg->mg_moremagic = mgt;
4823 SvMAGIC_set(sv, mg);
4833 sv_setsv(sv,&PL_sv_undef);
4839 /* Scans the name of a named buffer from the pattern.
4840 * If flags is REG_RSN_RETURN_NULL returns null.
4841 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4842 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4843 * to the parsed name as looked up in the RExC_paren_names hash.
4844 * If there is an error throws a vFAIL().. type exception.
4847 #define REG_RSN_RETURN_NULL 0
4848 #define REG_RSN_RETURN_NAME 1
4849 #define REG_RSN_RETURN_DATA 2
4852 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4853 char *name_start = RExC_parse;
4855 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
4856 /* skip IDFIRST by using do...while */
4859 RExC_parse += UTF8SKIP(RExC_parse);
4860 } while (isALNUM_utf8((U8*)RExC_parse));
4864 } while (isALNUM(*RExC_parse));
4868 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4869 (int)(RExC_parse - name_start)));
4872 if ( flags == REG_RSN_RETURN_NAME)
4874 else if (flags==REG_RSN_RETURN_DATA) {
4877 if ( ! sv_name ) /* should not happen*/
4878 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4879 if (RExC_paren_names)
4880 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4882 sv_dat = HeVAL(he_str);
4884 vFAIL("Reference to nonexistent named group");
4888 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4895 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4896 int rem=(int)(RExC_end - RExC_parse); \
4905 if (RExC_lastparse!=RExC_parse) \
4906 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4909 iscut ? "..." : "<" \
4912 PerlIO_printf(Perl_debug_log,"%16s",""); \
4915 num = RExC_size + 1; \
4917 num=REG_NODE_NUM(RExC_emit); \
4918 if (RExC_lastnum!=num) \
4919 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4921 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4922 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4923 (int)((depth*2)), "", \
4927 RExC_lastparse=RExC_parse; \
4932 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4933 DEBUG_PARSE_MSG((funcname)); \
4934 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4936 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4937 DEBUG_PARSE_MSG((funcname)); \
4938 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4941 - reg - regular expression, i.e. main body or parenthesized thing
4943 * Caller must absorb opening parenthesis.
4945 * Combining parenthesis handling with the base level of regular expression
4946 * is a trifle forced, but the need to tie the tails of the branches to what
4947 * follows makes it hard to avoid.
4949 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4951 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4953 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4956 /* this idea is borrowed from STR_WITH_LEN in handy.h */
4957 #define CHECK_WORD(s,v,l) \
4958 (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
4961 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4962 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4965 register regnode *ret; /* Will be the head of the group. */
4966 register regnode *br;
4967 register regnode *lastbr;
4968 register regnode *ender = NULL;
4969 register I32 parno = 0;
4971 const I32 oregflags = RExC_flags;
4972 bool have_branch = 0;
4974 I32 freeze_paren = 0;
4975 I32 after_freeze = 0;
4977 /* for (?g), (?gc), and (?o) warnings; warning
4978 about (?c) will warn about (?g) -- japhy */
4980 #define WASTED_O 0x01
4981 #define WASTED_G 0x02
4982 #define WASTED_C 0x04
4983 #define WASTED_GC (0x02|0x04)
4984 I32 wastedflags = 0x00;
4986 char * parse_start = RExC_parse; /* MJD */
4987 char * const oregcomp_parse = RExC_parse;
4989 GET_RE_DEBUG_FLAGS_DECL;
4990 DEBUG_PARSE("reg ");
4992 *flagp = 0; /* Tentatively. */
4995 /* Make an OPEN node, if parenthesized. */
4997 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4998 char *start_verb = RExC_parse;
4999 STRLEN verb_len = 0;
5000 char *start_arg = NULL;
5001 unsigned char op = 0;
5003 int internal_argval = 0; /* internal_argval is only useful if !argok */
5004 while ( *RExC_parse && *RExC_parse != ')' ) {
5005 if ( *RExC_parse == ':' ) {
5006 start_arg = RExC_parse + 1;
5012 verb_len = RExC_parse - start_verb;
5015 while ( *RExC_parse && *RExC_parse != ')' )
5017 if ( *RExC_parse != ')' )
5018 vFAIL("Unterminated verb pattern argument");
5019 if ( RExC_parse == start_arg )
5022 if ( *RExC_parse != ')' )
5023 vFAIL("Unterminated verb pattern");
5026 switch ( *start_verb ) {
5027 case 'A': /* (*ACCEPT) */
5028 if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
5030 internal_argval = RExC_nestroot;
5033 case 'C': /* (*COMMIT) */
5034 if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
5037 case 'F': /* (*FAIL) */
5038 if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
5043 case ':': /* (*:NAME) */
5044 case 'M': /* (*MARK:NAME) */
5045 if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
5050 case 'P': /* (*PRUNE) */
5051 if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
5054 case 'S': /* (*SKIP) */
5055 if ( CHECK_WORD("SKIP",start_verb,verb_len) )
5058 case 'T': /* (*THEN) */
5059 /* [19:06] <TimToady> :: is then */
5060 if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
5062 RExC_seen |= REG_SEEN_CUTGROUP;
5068 vFAIL3("Unknown verb pattern '%.*s'",
5069 verb_len, start_verb);
5072 if ( start_arg && internal_argval ) {
5073 vFAIL3("Verb pattern '%.*s' may not have an argument",
5074 verb_len, start_verb);
5075 } else if ( argok < 0 && !start_arg ) {
5076 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5077 verb_len, start_verb);
5079 ret = reganode(pRExC_state, op, internal_argval);
5080 if ( ! internal_argval && ! SIZE_ONLY ) {
5082 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5083 ARG(ret) = add_data( pRExC_state, 1, "S" );
5084 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5091 if (!internal_argval)
5092 RExC_seen |= REG_SEEN_VERBARG;
5093 } else if ( start_arg ) {
5094 vFAIL3("Verb pattern '%.*s' may not have an argument",
5095 verb_len, start_verb);
5097 ret = reg_node(pRExC_state, op);
5099 nextchar(pRExC_state);
5102 if (*RExC_parse == '?') { /* (?...) */
5103 bool is_logical = 0;
5104 const char * const seqstart = RExC_parse;
5107 paren = *RExC_parse++;
5108 ret = NULL; /* For look-ahead/behind. */
5111 case 'P': /* (?P...) variants for those used to PCRE/Python */
5112 paren = *RExC_parse++;
5113 if ( paren == '<') /* (?P<...>) named capture */
5115 else if (paren == '>') { /* (?P>name) named recursion */
5116 goto named_recursion;
5118 else if (paren == '=') { /* (?P=...) named backref */
5119 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5120 you change this make sure you change that */
5121 char* name_start = RExC_parse;
5123 SV *sv_dat = reg_scan_name(pRExC_state,
5124 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5125 if (RExC_parse == name_start || *RExC_parse != ')')
5126 vFAIL2("Sequence %.3s... not terminated",parse_start);
5129 num = add_data( pRExC_state, 1, "S" );
5130 RExC_rxi->data->data[num]=(void*)sv_dat;
5131 SvREFCNT_inc(sv_dat);
5134 ret = reganode(pRExC_state,
5135 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5139 Set_Node_Offset(ret, parse_start+1);
5140 Set_Node_Cur_Length(ret); /* MJD */
5142 nextchar(pRExC_state);
5146 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5148 case '<': /* (?<...) */
5149 if (*RExC_parse == '!')
5151 else if (*RExC_parse != '=')
5157 case '\'': /* (?'...') */
5158 name_start= RExC_parse;
5159 svname = reg_scan_name(pRExC_state,
5160 SIZE_ONLY ? /* reverse test from the others */
5161 REG_RSN_RETURN_NAME :
5162 REG_RSN_RETURN_NULL);
5163 if (RExC_parse == name_start) {
5165 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5168 if (*RExC_parse != paren)
5169 vFAIL2("Sequence (?%c... not terminated",
5170 paren=='>' ? '<' : paren);
5174 if (!svname) /* shouldnt happen */
5176 "panic: reg_scan_name returned NULL");
5177 if (!RExC_paren_names) {
5178 RExC_paren_names= newHV();
5179 sv_2mortal((SV*)RExC_paren_names);
5181 RExC_paren_name_list= newAV();
5182 sv_2mortal((SV*)RExC_paren_name_list);
5185 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5187 sv_dat = HeVAL(he_str);
5189 /* croak baby croak */
5191 "panic: paren_name hash element allocation failed");
5192 } else if ( SvPOK(sv_dat) ) {
5193 /* (?|...) can mean we have dupes so scan to check
5194 its already been stored. Maybe a flag indicating
5195 we are inside such a construct would be useful,
5196 but the arrays are likely to be quite small, so
5197 for now we punt -- dmq */
5198 IV count = SvIV(sv_dat);
5199 I32 *pv = (I32*)SvPVX(sv_dat);
5201 for ( i = 0 ; i < count ; i++ ) {
5202 if ( pv[i] == RExC_npar ) {
5208 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5209 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5210 pv[count] = RExC_npar;
5214 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5215 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5220 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5221 SvREFCNT_dec(svname);
5224 /*sv_dump(sv_dat);*/
5226 nextchar(pRExC_state);
5228 goto capturing_parens;
5230 RExC_seen |= REG_SEEN_LOOKBEHIND;
5232 case '=': /* (?=...) */
5233 case '!': /* (?!...) */
5234 RExC_seen_zerolen++;
5235 if (*RExC_parse == ')') {
5236 ret=reg_node(pRExC_state, OPFAIL);
5237 nextchar(pRExC_state);
5241 case '|': /* (?|...) */
5242 /* branch reset, behave like a (?:...) except that
5243 buffers in alternations share the same numbers */
5245 after_freeze = freeze_paren = RExC_npar;
5247 case ':': /* (?:...) */
5248 case '>': /* (?>...) */
5250 case '$': /* (?$...) */
5251 case '@': /* (?@...) */
5252 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5254 case '#': /* (?#...) */
5255 while (*RExC_parse && *RExC_parse != ')')
5257 if (*RExC_parse != ')')
5258 FAIL("Sequence (?#... not terminated");
5259 nextchar(pRExC_state);
5262 case '0' : /* (?0) */
5263 case 'R' : /* (?R) */
5264 if (*RExC_parse != ')')
5265 FAIL("Sequence (?R) not terminated");
5266 ret = reg_node(pRExC_state, GOSTART);
5267 *flagp |= POSTPONED;
5268 nextchar(pRExC_state);
5271 { /* named and numeric backreferences */
5273 case '&': /* (?&NAME) */
5274 parse_start = RExC_parse - 1;
5277 SV *sv_dat = reg_scan_name(pRExC_state,
5278 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5279 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5281 goto gen_recurse_regop;
5284 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5286 vFAIL("Illegal pattern");
5288 goto parse_recursion;
5290 case '-': /* (?-1) */
5291 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5292 RExC_parse--; /* rewind to let it be handled later */
5296 case '1': case '2': case '3': case '4': /* (?1) */
5297 case '5': case '6': case '7': case '8': case '9':
5300 num = atoi(RExC_parse);
5301 parse_start = RExC_parse - 1; /* MJD */
5302 if (*RExC_parse == '-')
5304 while (isDIGIT(*RExC_parse))
5306 if (*RExC_parse!=')')
5307 vFAIL("Expecting close bracket");
5310 if ( paren == '-' ) {
5312 Diagram of capture buffer numbering.
5313 Top line is the normal capture buffer numbers
5314 Botton line is the negative indexing as from
5318 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5322 num = RExC_npar + num;
5325 vFAIL("Reference to nonexistent group");
5327 } else if ( paren == '+' ) {
5328 num = RExC_npar + num - 1;
5331 ret = reganode(pRExC_state, GOSUB, num);
5333 if (num > (I32)RExC_rx->nparens) {
5335 vFAIL("Reference to nonexistent group");
5337 ARG2L_SET( ret, RExC_recurse_count++);
5339 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5340 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5344 RExC_seen |= REG_SEEN_RECURSE;
5345 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5346 Set_Node_Offset(ret, parse_start); /* MJD */
5348 *flagp |= POSTPONED;
5349 nextchar(pRExC_state);
5351 } /* named and numeric backreferences */
5354 case '?': /* (??...) */
5356 if (*RExC_parse != '{') {
5358 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5361 *flagp |= POSTPONED;
5362 paren = *RExC_parse++;
5364 case '{': /* (?{...}) */
5369 char *s = RExC_parse;
5371 RExC_seen_zerolen++;
5372 RExC_seen |= REG_SEEN_EVAL;
5373 while (count && (c = *RExC_parse)) {
5384 if (*RExC_parse != ')') {
5386 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5390 OP_4tree *sop, *rop;
5391 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5394 Perl_save_re_context(aTHX);
5395 rop = sv_compile_2op(sv, &sop, "re", &pad);
5396 sop->op_private |= OPpREFCOUNTED;
5397 /* re_dup will OpREFCNT_inc */
5398 OpREFCNT_set(sop, 1);
5401 n = add_data(pRExC_state, 3, "nop");
5402 RExC_rxi->data->data[n] = (void*)rop;
5403 RExC_rxi->data->data[n+1] = (void*)sop;
5404 RExC_rxi->data->data[n+2] = (void*)pad;
5407 else { /* First pass */
5408 if (PL_reginterp_cnt < ++RExC_seen_evals
5410 /* No compiled RE interpolated, has runtime
5411 components ===> unsafe. */
5412 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5413 if (PL_tainting && PL_tainted)
5414 FAIL("Eval-group in insecure regular expression");
5415 #if PERL_VERSION > 8
5416 if (IN_PERL_COMPILETIME)
5421 nextchar(pRExC_state);
5423 ret = reg_node(pRExC_state, LOGICAL);
5426 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5427 /* deal with the length of this later - MJD */
5430 ret = reganode(pRExC_state, EVAL, n);
5431 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5432 Set_Node_Offset(ret, parse_start);
5435 case '(': /* (?(?{...})...) and (?(?=...)...) */
5438 if (RExC_parse[0] == '?') { /* (?(?...)) */
5439 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5440 || RExC_parse[1] == '<'
5441 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5444 ret = reg_node(pRExC_state, LOGICAL);
5447 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5451 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5452 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5454 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5455 char *name_start= RExC_parse++;
5457 SV *sv_dat=reg_scan_name(pRExC_state,
5458 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5459 if (RExC_parse == name_start || *RExC_parse != ch)
5460 vFAIL2("Sequence (?(%c... not terminated",
5461 (ch == '>' ? '<' : ch));
5464 num = add_data( pRExC_state, 1, "S" );
5465 RExC_rxi->data->data[num]=(void*)sv_dat;
5466 SvREFCNT_inc(sv_dat);
5468 ret = reganode(pRExC_state,NGROUPP,num);
5469 goto insert_if_check_paren;
5471 else if (RExC_parse[0] == 'D' &&
5472 RExC_parse[1] == 'E' &&
5473 RExC_parse[2] == 'F' &&
5474 RExC_parse[3] == 'I' &&
5475 RExC_parse[4] == 'N' &&
5476 RExC_parse[5] == 'E')
5478 ret = reganode(pRExC_state,DEFINEP,0);
5481 goto insert_if_check_paren;
5483 else if (RExC_parse[0] == 'R') {
5486 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5487 parno = atoi(RExC_parse++);
5488 while (isDIGIT(*RExC_parse))
5490 } else if (RExC_parse[0] == '&') {
5493 sv_dat = reg_scan_name(pRExC_state,
5494 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5495 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5497 ret = reganode(pRExC_state,INSUBP,parno);
5498 goto insert_if_check_paren;
5500 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5503 parno = atoi(RExC_parse++);
5505 while (isDIGIT(*RExC_parse))
5507 ret = reganode(pRExC_state, GROUPP, parno);
5509 insert_if_check_paren:
5510 if ((c = *nextchar(pRExC_state)) != ')')
5511 vFAIL("Switch condition not recognized");
5513 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5514 br = regbranch(pRExC_state, &flags, 1,depth+1);
5516 br = reganode(pRExC_state, LONGJMP, 0);
5518 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5519 c = *nextchar(pRExC_state);
5524 vFAIL("(?(DEFINE)....) does not allow branches");
5525 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5526 regbranch(pRExC_state, &flags, 1,depth+1);
5527 REGTAIL(pRExC_state, ret, lastbr);
5530 c = *nextchar(pRExC_state);
5535 vFAIL("Switch (?(condition)... contains too many branches");
5536 ender = reg_node(pRExC_state, TAIL);
5537 REGTAIL(pRExC_state, br, ender);
5539 REGTAIL(pRExC_state, lastbr, ender);
5540 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5543 REGTAIL(pRExC_state, ret, ender);
5544 RExC_size++; /* XXX WHY do we need this?!!
5545 For large programs it seems to be required
5546 but I can't figure out why. -- dmq*/
5550 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5554 RExC_parse--; /* for vFAIL to print correctly */
5555 vFAIL("Sequence (? incomplete");
5559 parse_flags: /* (?i) */
5561 U32 posflags = 0, negflags = 0;
5562 U32 *flagsp = &posflags;
5564 while (*RExC_parse) {
5565 /* && strchr("iogcmsx", *RExC_parse) */
5566 /* (?g), (?gc) and (?o) are useless here
5567 and must be globally applied -- japhy */
5568 switch (*RExC_parse) {
5569 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5572 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5573 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5574 if (! (wastedflags & wflagbit) ) {
5575 wastedflags |= wflagbit;
5578 "Useless (%s%c) - %suse /%c modifier",
5579 flagsp == &negflags ? "?-" : "?",
5581 flagsp == &negflags ? "don't " : "",
5589 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5590 if (! (wastedflags & WASTED_C) ) {
5591 wastedflags |= WASTED_GC;
5594 "Useless (%sc) - %suse /gc modifier",
5595 flagsp == &negflags ? "?-" : "?",
5596 flagsp == &negflags ? "don't " : ""
5602 if (flagsp == &negflags) {
5603 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5604 vWARN(RExC_parse + 1,"Useless use of (?-k)");
5606 *flagsp |= RXf_PMf_KEEPCOPY;
5610 if (flagsp == &negflags) {
5612 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5616 wastedflags = 0; /* reset so (?g-c) warns twice */
5622 RExC_flags |= posflags;
5623 RExC_flags &= ~negflags;
5624 nextchar(pRExC_state);
5635 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5640 }} /* one for the default block, one for the switch */
5647 ret = reganode(pRExC_state, OPEN, parno);
5650 RExC_nestroot = parno;
5651 if (RExC_seen & REG_SEEN_RECURSE
5652 && !RExC_open_parens[parno-1])
5654 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5655 "Setting open paren #%"IVdf" to %d\n",
5656 (IV)parno, REG_NODE_NUM(ret)));
5657 RExC_open_parens[parno-1]= ret;
5660 Set_Node_Length(ret, 1); /* MJD */
5661 Set_Node_Offset(ret, RExC_parse); /* MJD */
5669 /* Pick up the branches, linking them together. */
5670 parse_start = RExC_parse; /* MJD */
5671 br = regbranch(pRExC_state, &flags, 1,depth+1);
5672 /* branch_len = (paren != 0); */
5676 if (*RExC_parse == '|') {
5677 if (!SIZE_ONLY && RExC_extralen) {
5678 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5681 reginsert(pRExC_state, BRANCH, br, depth+1);
5682 Set_Node_Length(br, paren != 0);
5683 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5687 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5689 else if (paren == ':') {
5690 *flagp |= flags&SIMPLE;
5692 if (is_open) { /* Starts with OPEN. */
5693 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5695 else if (paren != '?') /* Not Conditional */
5697 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5699 while (*RExC_parse == '|') {
5700 if (!SIZE_ONLY && RExC_extralen) {
5701 ender = reganode(pRExC_state, LONGJMP,0);
5702 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5705 RExC_extralen += 2; /* Account for LONGJMP. */
5706 nextchar(pRExC_state);
5708 if (RExC_npar > after_freeze)
5709 after_freeze = RExC_npar;
5710 RExC_npar = freeze_paren;
5712 br = regbranch(pRExC_state, &flags, 0, depth+1);
5716 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5718 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5721 if (have_branch || paren != ':') {
5722 /* Make a closing node, and hook it on the end. */
5725 ender = reg_node(pRExC_state, TAIL);
5728 ender = reganode(pRExC_state, CLOSE, parno);
5729 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5730 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5731 "Setting close paren #%"IVdf" to %d\n",
5732 (IV)parno, REG_NODE_NUM(ender)));
5733 RExC_close_parens[parno-1]= ender;
5734 if (RExC_nestroot == parno)
5737 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5738 Set_Node_Length(ender,1); /* MJD */
5744 *flagp &= ~HASWIDTH;
5747 ender = reg_node(pRExC_state, SUCCEED);
5750 ender = reg_node(pRExC_state, END);
5752 assert(!RExC_opend); /* there can only be one! */
5757 REGTAIL(pRExC_state, lastbr, ender);
5759 if (have_branch && !SIZE_ONLY) {
5761 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5763 /* Hook the tails of the branches to the closing node. */
5764 for (br = ret; br; br = regnext(br)) {
5765 const U8 op = PL_regkind[OP(br)];
5767 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5769 else if (op == BRANCHJ) {
5770 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5778 static const char parens[] = "=!<,>";
5780 if (paren && (p = strchr(parens, paren))) {
5781 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5782 int flag = (p - parens) > 1;
5785 node = SUSPEND, flag = 0;
5786 reginsert(pRExC_state, node,ret, depth+1);
5787 Set_Node_Cur_Length(ret);
5788 Set_Node_Offset(ret, parse_start + 1);
5790 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5794 /* Check for proper termination. */
5796 RExC_flags = oregflags;
5797 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5798 RExC_parse = oregcomp_parse;
5799 vFAIL("Unmatched (");
5802 else if (!paren && RExC_parse < RExC_end) {
5803 if (*RExC_parse == ')') {
5805 vFAIL("Unmatched )");
5808 FAIL("Junk on end of regexp"); /* "Can't happen". */
5812 RExC_npar = after_freeze;
5817 - regbranch - one alternative of an | operator
5819 * Implements the concatenation operator.
5822 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5825 register regnode *ret;
5826 register regnode *chain = NULL;
5827 register regnode *latest;
5828 I32 flags = 0, c = 0;
5829 GET_RE_DEBUG_FLAGS_DECL;
5830 DEBUG_PARSE("brnc");
5835 if (!SIZE_ONLY && RExC_extralen)
5836 ret = reganode(pRExC_state, BRANCHJ,0);
5838 ret = reg_node(pRExC_state, BRANCH);
5839 Set_Node_Length(ret, 1);
5843 if (!first && SIZE_ONLY)
5844 RExC_extralen += 1; /* BRANCHJ */
5846 *flagp = WORST; /* Tentatively. */
5849 nextchar(pRExC_state);
5850 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5852 latest = regpiece(pRExC_state, &flags,depth+1);
5853 if (latest == NULL) {
5854 if (flags & TRYAGAIN)
5858 else if (ret == NULL)
5860 *flagp |= flags&(HASWIDTH|POSTPONED);
5861 if (chain == NULL) /* First piece. */
5862 *flagp |= flags&SPSTART;
5865 REGTAIL(pRExC_state, chain, latest);
5870 if (chain == NULL) { /* Loop ran zero times. */
5871 chain = reg_node(pRExC_state, NOTHING);
5876 *flagp |= flags&SIMPLE;
5883 - regpiece - something followed by possible [*+?]
5885 * Note that the branching code sequences used for ? and the general cases
5886 * of * and + are somewhat optimized: they use the same NOTHING node as
5887 * both the endmarker for their branch list and the body of the last branch.
5888 * It might seem that this node could be dispensed with entirely, but the
5889 * endmarker role is not redundant.
5892 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5895 register regnode *ret;
5897 register char *next;
5899 const char * const origparse = RExC_parse;
5901 I32 max = REG_INFTY;
5903 const char *maxpos = NULL;
5904 GET_RE_DEBUG_FLAGS_DECL;
5905 DEBUG_PARSE("piec");
5907 ret = regatom(pRExC_state, &flags,depth+1);
5909 if (flags & TRYAGAIN)
5916 if (op == '{' && regcurly(RExC_parse)) {
5918 parse_start = RExC_parse; /* MJD */
5919 next = RExC_parse + 1;
5920 while (isDIGIT(*next) || *next == ',') {
5929 if (*next == '}') { /* got one */
5933 min = atoi(RExC_parse);
5937 maxpos = RExC_parse;
5939 if (!max && *maxpos != '0')
5940 max = REG_INFTY; /* meaning "infinity" */
5941 else if (max >= REG_INFTY)
5942 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5944 nextchar(pRExC_state);
5947 if ((flags&SIMPLE)) {
5948 RExC_naughty += 2 + RExC_naughty / 2;
5949 reginsert(pRExC_state, CURLY, ret, depth+1);
5950 Set_Node_Offset(ret, parse_start+1); /* MJD */
5951 Set_Node_Cur_Length(ret);
5954 regnode * const w = reg_node(pRExC_state, WHILEM);
5957 REGTAIL(pRExC_state, ret, w);
5958 if (!SIZE_ONLY && RExC_extralen) {
5959 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5960 reginsert(pRExC_state, NOTHING,ret, depth+1);
5961 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5963 reginsert(pRExC_state, CURLYX,ret, depth+1);
5965 Set_Node_Offset(ret, parse_start+1);
5966 Set_Node_Length(ret,
5967 op == '{' ? (RExC_parse - parse_start) : 1);
5969 if (!SIZE_ONLY && RExC_extralen)
5970 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5971 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5973 RExC_whilem_seen++, RExC_extralen += 3;
5974 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5982 if (max && max < min)
5983 vFAIL("Can't do {n,m} with n > m");
5985 ARG1_SET(ret, (U16)min);
5986 ARG2_SET(ret, (U16)max);
5998 #if 0 /* Now runtime fix should be reliable. */
6000 /* if this is reinstated, don't forget to put this back into perldiag:
6002 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6004 (F) The part of the regexp subject to either the * or + quantifier
6005 could match an empty string. The {#} shows in the regular
6006 expression about where the problem was discovered.
6010 if (!(flags&HASWIDTH) && op != '?')
6011 vFAIL("Regexp *+ operand could be empty");
6014 parse_start = RExC_parse;
6015 nextchar(pRExC_state);
6017 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6019 if (op == '*' && (flags&SIMPLE)) {
6020 reginsert(pRExC_state, STAR, ret, depth+1);
6024 else if (op == '*') {
6028 else if (op == '+' && (flags&SIMPLE)) {
6029 reginsert(pRExC_state, PLUS, ret, depth+1);
6033 else if (op == '+') {
6037 else if (op == '?') {
6042 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6044 "%.*s matches null string many times",
6045 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6049 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6050 nextchar(pRExC_state);
6051 reginsert(pRExC_state, MINMOD, ret, depth+1);
6052 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6054 #ifndef REG_ALLOW_MINMOD_SUSPEND
6057 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6059 nextchar(pRExC_state);
6060 ender = reg_node(pRExC_state, SUCCEED);
6061 REGTAIL(pRExC_state, ret, ender);
6062 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6064 ender = reg_node(pRExC_state, TAIL);
6065 REGTAIL(pRExC_state, ret, ender);
6069 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6071 vFAIL("Nested quantifiers");
6078 /* reg_namedseq(pRExC_state,UVp)
6080 This is expected to be called by a parser routine that has
6081 recognized'\N' and needs to handle the rest. RExC_parse is
6082 expected to point at the first char following the N at the time
6085 If valuep is non-null then it is assumed that we are parsing inside
6086 of a charclass definition and the first codepoint in the resolved
6087 string is returned via *valuep and the routine will return NULL.
6088 In this mode if a multichar string is returned from the charnames
6089 handler a warning will be issued, and only the first char in the
6090 sequence will be examined. If the string returned is zero length
6091 then the value of *valuep is undefined and NON-NULL will
6092 be returned to indicate failure. (This will NOT be a valid pointer
6095 If value is null then it is assumed that we are parsing normal text
6096 and inserts a new EXACT node into the program containing the resolved
6097 string and returns a pointer to the new node. If the string is
6098 zerolength a NOTHING node is emitted.
6100 On success RExC_parse is set to the char following the endbrace.
6101 Parsing failures will generate a fatal errorvia vFAIL(...)
6103 NOTE: We cache all results from the charnames handler locally in
6104 the RExC_charnames hash (created on first use) to prevent a charnames
6105 handler from playing silly-buggers and returning a short string and
6106 then a long string for a given pattern. Since the regexp program
6107 size is calculated during an initial parse this would result
6108 in a buffer overrun so we cache to prevent the charname result from
6109 changing during the course of the parse.
6113 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6115 char * name; /* start of the content of the name */
6116 char * endbrace; /* endbrace following the name */
6119 STRLEN len; /* this has various purposes throughout the code */
6120 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6121 regnode *ret = NULL;
6123 if (*RExC_parse != '{') {
6124 vFAIL("Missing braces on \\N{}");
6126 name = RExC_parse+1;
6127 endbrace = strchr(RExC_parse, '}');
6130 vFAIL("Missing right brace on \\N{}");
6132 RExC_parse = endbrace + 1;
6135 /* RExC_parse points at the beginning brace,
6136 endbrace points at the last */
6137 if ( name[0]=='U' && name[1]=='+' ) {
6138 /* its a "unicode hex" notation {U+89AB} */
6139 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6140 | PERL_SCAN_DISALLOW_PREFIX
6141 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6143 len = (STRLEN)(endbrace - name - 2);
6144 cp = grok_hex(name + 2, &len, &fl, NULL);
6145 if ( len != (STRLEN)(endbrace - name - 2) ) {
6154 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6156 /* fetch the charnames handler for this scope */
6157 HV * const table = GvHV(PL_hintgv);
6159 hv_fetchs(table, "charnames", FALSE) :
6161 SV *cv= cvp ? *cvp : NULL;
6164 /* create an SV with the name as argument */
6165 sv_name = newSVpvn(name, endbrace - name);
6167 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6168 vFAIL2("Constant(\\N{%s}) unknown: "
6169 "(possibly a missing \"use charnames ...\")",
6172 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6173 vFAIL2("Constant(\\N{%s}): "
6174 "$^H{charnames} is not defined",SvPVX(sv_name));
6179 if (!RExC_charnames) {
6180 /* make sure our cache is allocated */
6181 RExC_charnames = newHV();
6182 sv_2mortal((SV*)RExC_charnames);
6184 /* see if we have looked this one up before */
6185 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6187 sv_str = HeVAL(he_str);
6200 count= call_sv(cv, G_SCALAR);
6202 if (count == 1) { /* XXXX is this right? dmq */
6204 SvREFCNT_inc_simple_void(sv_str);
6212 if ( !sv_str || !SvOK(sv_str) ) {
6213 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6214 "did not return a defined value",SvPVX(sv_name));
6216 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6221 char *p = SvPV(sv_str, len);
6224 if ( SvUTF8(sv_str) ) {
6225 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6229 We have to turn on utf8 for high bit chars otherwise
6230 we get failures with
6232 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6233 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6235 This is different from what \x{} would do with the same
6236 codepoint, where the condition is > 0xFF.
6243 /* warn if we havent used the whole string? */
6245 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6247 "Ignoring excess chars from \\N{%s} in character class",
6251 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6253 "Ignoring zero length \\N{%s} in character class",
6258 SvREFCNT_dec(sv_name);
6260 SvREFCNT_dec(sv_str);
6261 return len ? NULL : (regnode *)&len;
6262 } else if(SvCUR(sv_str)) {
6268 char * parse_start = name-3; /* needed for the offsets */
6270 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6272 ret = reg_node(pRExC_state,
6273 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6276 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6277 sv_utf8_upgrade(sv_str);
6278 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6282 p = SvPV(sv_str, len);
6284 /* len is the length written, charlen is the size the char read */
6285 for ( len = 0; p < pend; p += charlen ) {
6287 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6289 STRLEN foldlen,numlen;
6290 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6291 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6292 /* Emit all the Unicode characters. */
6294 for (foldbuf = tmpbuf;
6298 uvc = utf8_to_uvchr(foldbuf, &numlen);
6300 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6303 /* In EBCDIC the numlen
6304 * and unilen can differ. */
6306 if (numlen >= foldlen)
6310 break; /* "Can't happen." */
6313 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6325 RExC_size += STR_SZ(len);
6328 RExC_emit += STR_SZ(len);
6330 Set_Node_Cur_Length(ret); /* MJD */
6332 nextchar(pRExC_state);
6334 ret = reg_node(pRExC_state,NOTHING);
6337 SvREFCNT_dec(sv_str);
6340 SvREFCNT_dec(sv_name);
6350 * It returns the code point in utf8 for the value in *encp.
6351 * value: a code value in the source encoding
6352 * encp: a pointer to an Encode object
6354 * If the result from Encode is not a single character,
6355 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6358 S_reg_recode(pTHX_ const char value, SV **encp)
6361 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6362 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6364 const STRLEN newlen = SvCUR(sv);
6365 UV uv = UNICODE_REPLACEMENT;
6369 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6372 if (!newlen || numlen != newlen) {
6373 uv = UNICODE_REPLACEMENT;
6382 - regatom - the lowest level
6384 Try to identify anything special at the start of the pattern. If there
6385 is, then handle it as required. This may involve generating a single regop,
6386 such as for an assertion; or it may involve recursing, such as to
6387 handle a () structure.
6389 If the string doesn't start with something special then we gobble up
6390 as much literal text as we can.
6392 Once we have been able to handle whatever type of thing started the
6393 sequence, we return.
6395 Note: we have to be careful with escapes, as they can be both literal
6396 and special, and in the case of \10 and friends can either, depending
6397 on context. Specifically there are two seperate switches for handling
6398 escape sequences, with the one for handling literal escapes requiring
6399 a dummy entry for all of the special escapes that are actually handled
6404 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6407 register regnode *ret = NULL;
6409 char *parse_start = RExC_parse;
6410 GET_RE_DEBUG_FLAGS_DECL;
6411 DEBUG_PARSE("atom");
6412 *flagp = WORST; /* Tentatively. */
6416 switch (*RExC_parse) {
6418 RExC_seen_zerolen++;
6419 nextchar(pRExC_state);
6420 if (RExC_flags & RXf_PMf_MULTILINE)
6421 ret = reg_node(pRExC_state, MBOL);
6422 else if (RExC_flags & RXf_PMf_SINGLELINE)
6423 ret = reg_node(pRExC_state, SBOL);
6425 ret = reg_node(pRExC_state, BOL);
6426 Set_Node_Length(ret, 1); /* MJD */
6429 nextchar(pRExC_state);
6431 RExC_seen_zerolen++;
6432 if (RExC_flags & RXf_PMf_MULTILINE)
6433 ret = reg_node(pRExC_state, MEOL);
6434 else if (RExC_flags & RXf_PMf_SINGLELINE)
6435 ret = reg_node(pRExC_state, SEOL);
6437 ret = reg_node(pRExC_state, EOL);
6438 Set_Node_Length(ret, 1); /* MJD */
6441 nextchar(pRExC_state);
6442 if (RExC_flags & RXf_PMf_SINGLELINE)
6443 ret = reg_node(pRExC_state, SANY);
6445 ret = reg_node(pRExC_state, REG_ANY);
6446 *flagp |= HASWIDTH|SIMPLE;
6448 Set_Node_Length(ret, 1); /* MJD */
6452 char * const oregcomp_parse = ++RExC_parse;
6453 ret = regclass(pRExC_state,depth+1);
6454 if (*RExC_parse != ']') {
6455 RExC_parse = oregcomp_parse;
6456 vFAIL("Unmatched [");
6458 nextchar(pRExC_state);
6459 *flagp |= HASWIDTH|SIMPLE;
6460 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6464 nextchar(pRExC_state);
6465 ret = reg(pRExC_state, 1, &flags,depth+1);
6467 if (flags & TRYAGAIN) {
6468 if (RExC_parse == RExC_end) {
6469 /* Make parent create an empty node if needed. */
6477 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6481 if (flags & TRYAGAIN) {
6485 vFAIL("Internal urp");
6486 /* Supposed to be caught earlier. */
6489 if (!regcurly(RExC_parse)) {
6498 vFAIL("Quantifier follows nothing");
6503 This switch handles escape sequences that resolve to some kind
6504 of special regop and not to literal text. Escape sequnces that
6505 resolve to literal text are handled below in the switch marked
6508 Every entry in this switch *must* have a corresponding entry
6509 in the literal escape switch. However, the opposite is not
6510 required, as the default for this switch is to jump to the
6511 literal text handling code.
6513 switch (*++RExC_parse) {
6514 /* Special Escapes */
6516 RExC_seen_zerolen++;
6517 ret = reg_node(pRExC_state, SBOL);
6519 goto finish_meta_pat;
6521 ret = reg_node(pRExC_state, GPOS);
6522 RExC_seen |= REG_SEEN_GPOS;
6524 goto finish_meta_pat;
6526 RExC_seen_zerolen++;
6527 ret = reg_node(pRExC_state, KEEPS);
6529 goto finish_meta_pat;
6531 ret = reg_node(pRExC_state, SEOL);
6533 RExC_seen_zerolen++; /* Do not optimize RE away */
6534 goto finish_meta_pat;
6536 ret = reg_node(pRExC_state, EOS);
6538 RExC_seen_zerolen++; /* Do not optimize RE away */
6539 goto finish_meta_pat;
6541 ret = reg_node(pRExC_state, CANY);
6542 RExC_seen |= REG_SEEN_CANY;
6543 *flagp |= HASWIDTH|SIMPLE;
6544 goto finish_meta_pat;
6546 ret = reg_node(pRExC_state, CLUMP);
6548 goto finish_meta_pat;
6550 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6551 *flagp |= HASWIDTH|SIMPLE;
6552 goto finish_meta_pat;
6554 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6555 *flagp |= HASWIDTH|SIMPLE;
6556 goto finish_meta_pat;
6558 RExC_seen_zerolen++;
6559 RExC_seen |= REG_SEEN_LOOKBEHIND;
6560 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6562 goto finish_meta_pat;
6564 RExC_seen_zerolen++;
6565 RExC_seen |= REG_SEEN_LOOKBEHIND;
6566 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6568 goto finish_meta_pat;
6570 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6571 *flagp |= HASWIDTH|SIMPLE;
6572 goto finish_meta_pat;
6574 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6575 *flagp |= HASWIDTH|SIMPLE;
6576 goto finish_meta_pat;
6578 ret = reg_node(pRExC_state, DIGIT);
6579 *flagp |= HASWIDTH|SIMPLE;
6580 goto finish_meta_pat;
6582 ret = reg_node(pRExC_state, NDIGIT);
6583 *flagp |= HASWIDTH|SIMPLE;
6584 goto finish_meta_pat;
6586 ret = reganode(pRExC_state, PRUNE, 0);
6589 goto finish_meta_pat;
6591 ret = reganode(pRExC_state, SKIP, 0);
6595 nextchar(pRExC_state);
6596 Set_Node_Length(ret, 2); /* MJD */
6601 char* const oldregxend = RExC_end;
6603 char* parse_start = RExC_parse - 2;
6606 if (RExC_parse[1] == '{') {
6607 /* a lovely hack--pretend we saw [\pX] instead */
6608 RExC_end = strchr(RExC_parse, '}');
6610 const U8 c = (U8)*RExC_parse;
6612 RExC_end = oldregxend;
6613 vFAIL2("Missing right brace on \\%c{}", c);
6618 RExC_end = RExC_parse + 2;
6619 if (RExC_end > oldregxend)
6620 RExC_end = oldregxend;
6624 ret = regclass(pRExC_state,depth+1);
6626 RExC_end = oldregxend;
6629 Set_Node_Offset(ret, parse_start + 2);
6630 Set_Node_Cur_Length(ret);
6631 nextchar(pRExC_state);
6632 *flagp |= HASWIDTH|SIMPLE;
6636 /* Handle \N{NAME} here and not below because it can be
6637 multicharacter. join_exact() will join them up later on.
6638 Also this makes sure that things like /\N{BLAH}+/ and
6639 \N{BLAH} being multi char Just Happen. dmq*/
6641 ret= reg_namedseq(pRExC_state, NULL);
6643 case 'k': /* Handle \k<NAME> and \k'NAME' */
6646 char ch= RExC_parse[1];
6647 if (ch != '<' && ch != '\'' && ch != '{') {
6649 vFAIL2("Sequence %.2s... not terminated",parse_start);
6651 /* this pretty much dupes the code for (?P=...) in reg(), if
6652 you change this make sure you change that */
6653 char* name_start = (RExC_parse += 2);
6655 SV *sv_dat = reg_scan_name(pRExC_state,
6656 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6657 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
6658 if (RExC_parse == name_start || *RExC_parse != ch)
6659 vFAIL2("Sequence %.3s... not terminated",parse_start);
6662 num = add_data( pRExC_state, 1, "S" );
6663 RExC_rxi->data->data[num]=(void*)sv_dat;
6664 SvREFCNT_inc(sv_dat);
6668 ret = reganode(pRExC_state,
6669 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6673 /* override incorrect value set in reganode MJD */
6674 Set_Node_Offset(ret, parse_start+1);
6675 Set_Node_Cur_Length(ret); /* MJD */
6676 nextchar(pRExC_state);
6682 case '1': case '2': case '3': case '4':
6683 case '5': case '6': case '7': case '8': case '9':
6686 bool isg = *RExC_parse == 'g';
6691 if (*RExC_parse == '{') {
6695 if (*RExC_parse == '-') {
6699 if (hasbrace && !isDIGIT(*RExC_parse)) {
6700 if (isrel) RExC_parse--;
6702 goto parse_named_seq;
6704 num = atoi(RExC_parse);
6706 num = RExC_npar - num;
6708 vFAIL("Reference to nonexistent or unclosed group");
6710 if (!isg && num > 9 && num >= RExC_npar)
6713 char * const parse_start = RExC_parse - 1; /* MJD */
6714 while (isDIGIT(*RExC_parse))
6716 if (parse_start == RExC_parse - 1)
6717 vFAIL("Unterminated \\g... pattern");
6719 if (*RExC_parse != '}')
6720 vFAIL("Unterminated \\g{...} pattern");
6724 if (num > (I32)RExC_rx->nparens)
6725 vFAIL("Reference to nonexistent group");
6728 ret = reganode(pRExC_state,
6729 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6733 /* override incorrect value set in reganode MJD */
6734 Set_Node_Offset(ret, parse_start+1);
6735 Set_Node_Cur_Length(ret); /* MJD */
6737 nextchar(pRExC_state);
6742 if (RExC_parse >= RExC_end)
6743 FAIL("Trailing \\");
6746 /* Do not generate "unrecognized" warnings here, we fall
6747 back into the quick-grab loop below */
6754 if (RExC_flags & RXf_PMf_EXTENDED) {
6755 if ( reg_skipcomment( pRExC_state ) )
6761 register STRLEN len;
6766 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6768 parse_start = RExC_parse - 1;
6774 ret = reg_node(pRExC_state,
6775 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6777 for (len = 0, p = RExC_parse - 1;
6778 len < 127 && p < RExC_end;
6781 char * const oldp = p;
6783 if (RExC_flags & RXf_PMf_EXTENDED)
6784 p = regwhite( pRExC_state, p );
6795 /* Literal Escapes Switch
6797 This switch is meant to handle escape sequences that
6798 resolve to a literal character.
6800 Every escape sequence that represents something
6801 else, like an assertion or a char class, is handled
6802 in the switch marked 'Special Escapes' above in this
6803 routine, but also has an entry here as anything that
6804 isn't explicitly mentioned here will be treated as
6805 an unescaped equivalent literal.
6809 /* These are all the special escapes. */
6810 case 'A': /* Start assertion */
6811 case 'b': case 'B': /* Word-boundary assertion*/
6812 case 'C': /* Single char !DANGEROUS! */
6813 case 'd': case 'D': /* digit class */
6814 case 'g': case 'G': /* generic-backref, pos assertion */
6815 case 'k': case 'K': /* named backref, keep marker */
6816 case 'N': /* named char sequence */
6817 case 'p': case 'P': /* unicode property */
6818 case 's': case 'S': /* space class */
6819 case 'v': case 'V': /* (*PRUNE) and (*SKIP) */
6820 case 'w': case 'W': /* word class */
6821 case 'X': /* eXtended Unicode "combining character sequence" */
6822 case 'z': case 'Z': /* End of line/string assertion */
6826 /* Anything after here is an escape that resolves to a
6827 literal. (Except digits, which may or may not)
6846 ender = ASCII_TO_NATIVE('\033');
6850 ender = ASCII_TO_NATIVE('\007');
6855 char* const e = strchr(p, '}');
6859 vFAIL("Missing right brace on \\x{}");
6862 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6863 | PERL_SCAN_DISALLOW_PREFIX;
6864 STRLEN numlen = e - p - 1;
6865 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6872 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6874 ender = grok_hex(p, &numlen, &flags, NULL);
6877 if (PL_encoding && ender < 0x100)
6878 goto recode_encoding;
6882 ender = UCHARAT(p++);
6883 ender = toCTRL(ender);
6885 case '0': case '1': case '2': case '3':case '4':
6886 case '5': case '6': case '7': case '8':case '9':
6888 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6891 ender = grok_oct(p, &numlen, &flags, NULL);
6898 if (PL_encoding && ender < 0x100)
6899 goto recode_encoding;
6903 SV* enc = PL_encoding;
6904 ender = reg_recode((const char)(U8)ender, &enc);
6905 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6906 vWARN(p, "Invalid escape in the specified encoding");
6912 FAIL("Trailing \\");
6915 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6916 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6917 goto normal_default;
6922 if (UTF8_IS_START(*p) && UTF) {
6924 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6925 &numlen, UTF8_ALLOW_DEFAULT);
6932 if ( RExC_flags & RXf_PMf_EXTENDED)
6933 p = regwhite( pRExC_state, p );
6935 /* Prime the casefolded buffer. */
6936 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6938 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
6943 /* Emit all the Unicode characters. */
6945 for (foldbuf = tmpbuf;
6947 foldlen -= numlen) {
6948 ender = utf8_to_uvchr(foldbuf, &numlen);
6950 const STRLEN unilen = reguni(pRExC_state, ender, s);
6953 /* In EBCDIC the numlen
6954 * and unilen can differ. */
6956 if (numlen >= foldlen)
6960 break; /* "Can't happen." */
6964 const STRLEN unilen = reguni(pRExC_state, ender, s);
6973 REGC((char)ender, s++);
6979 /* Emit all the Unicode characters. */
6981 for (foldbuf = tmpbuf;
6983 foldlen -= numlen) {
6984 ender = utf8_to_uvchr(foldbuf, &numlen);
6986 const STRLEN unilen = reguni(pRExC_state, ender, s);
6989 /* In EBCDIC the numlen
6990 * and unilen can differ. */
6992 if (numlen >= foldlen)
7000 const STRLEN unilen = reguni(pRExC_state, ender, s);
7009 REGC((char)ender, s++);
7013 Set_Node_Cur_Length(ret); /* MJD */
7014 nextchar(pRExC_state);
7016 /* len is STRLEN which is unsigned, need to copy to signed */
7019 vFAIL("Internal disaster");
7023 if (len == 1 && UNI_IS_INVARIANT(ender))
7027 RExC_size += STR_SZ(len);
7030 RExC_emit += STR_SZ(len);
7040 S_regwhite( RExC_state_t *pRExC_state, char *p )
7042 const char *e = RExC_end;
7046 else if (*p == '#') {
7055 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7063 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7064 Character classes ([:foo:]) can also be negated ([:^foo:]).
7065 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7066 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7067 but trigger failures because they are currently unimplemented. */
7069 #define POSIXCC_DONE(c) ((c) == ':')
7070 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7071 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7074 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7077 I32 namedclass = OOB_NAMEDCLASS;
7079 if (value == '[' && RExC_parse + 1 < RExC_end &&
7080 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7081 POSIXCC(UCHARAT(RExC_parse))) {
7082 const char c = UCHARAT(RExC_parse);
7083 char* const s = RExC_parse++;
7085 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7087 if (RExC_parse == RExC_end)
7088 /* Grandfather lone [:, [=, [. */
7091 const char* const t = RExC_parse++; /* skip over the c */
7094 if (UCHARAT(RExC_parse) == ']') {
7095 const char *posixcc = s + 1;
7096 RExC_parse++; /* skip over the ending ] */
7099 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7100 const I32 skip = t - posixcc;
7102 /* Initially switch on the length of the name. */
7105 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7106 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7109 /* Names all of length 5. */
7110 /* alnum alpha ascii blank cntrl digit graph lower
7111 print punct space upper */
7112 /* Offset 4 gives the best switch position. */
7113 switch (posixcc[4]) {
7115 if (memEQ(posixcc, "alph", 4)) /* alpha */
7116 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7119 if (memEQ(posixcc, "spac", 4)) /* space */
7120 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7123 if (memEQ(posixcc, "grap", 4)) /* graph */
7124 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7127 if (memEQ(posixcc, "asci", 4)) /* ascii */
7128 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7131 if (memEQ(posixcc, "blan", 4)) /* blank */
7132 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7135 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7136 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7139 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7140 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7143 if (memEQ(posixcc, "lowe", 4)) /* lower */
7144 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7145 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7146 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7149 if (memEQ(posixcc, "digi", 4)) /* digit */
7150 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7151 else if (memEQ(posixcc, "prin", 4)) /* print */
7152 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7153 else if (memEQ(posixcc, "punc", 4)) /* punct */
7154 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7159 if (memEQ(posixcc, "xdigit", 6))
7160 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7164 if (namedclass == OOB_NAMEDCLASS)
7165 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7167 assert (posixcc[skip] == ':');
7168 assert (posixcc[skip+1] == ']');
7169 } else if (!SIZE_ONLY) {
7170 /* [[=foo=]] and [[.foo.]] are still future. */
7172 /* adjust RExC_parse so the warning shows after
7174 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7176 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7179 /* Maternal grandfather:
7180 * "[:" ending in ":" but not in ":]" */
7190 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7193 if (POSIXCC(UCHARAT(RExC_parse))) {
7194 const char *s = RExC_parse;
7195 const char c = *s++;
7199 if (*s && c == *s && s[1] == ']') {
7200 if (ckWARN(WARN_REGEXP))
7202 "POSIX syntax [%c %c] belongs inside character classes",
7205 /* [[=foo=]] and [[.foo.]] are still future. */
7206 if (POSIXCC_NOTYET(c)) {
7207 /* adjust RExC_parse so the error shows after
7209 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7211 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7218 #define _C_C_T_(NAME,TEST,WORD) \
7221 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7223 for (value = 0; value < 256; value++) \
7225 ANYOF_BITMAP_SET(ret, value); \
7230 case ANYOF_N##NAME: \
7232 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7234 for (value = 0; value < 256; value++) \
7236 ANYOF_BITMAP_SET(ret, value); \
7244 parse a class specification and produce either an ANYOF node that
7245 matches the pattern or if the pattern matches a single char only and
7246 that char is < 256 and we are case insensitive then we produce an
7251 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7254 register UV value = 0;
7255 register UV nextvalue;
7256 register IV prevvalue = OOB_UNICODE;
7257 register IV range = 0;
7258 register regnode *ret;
7261 char *rangebegin = NULL;
7262 bool need_class = 0;
7265 bool optimize_invert = TRUE;
7266 AV* unicode_alternate = NULL;
7268 UV literal_endpoint = 0;
7270 UV stored = 0; /* number of chars stored in the class */
7272 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7273 case we need to change the emitted regop to an EXACT. */
7274 const char * orig_parse = RExC_parse;
7275 GET_RE_DEBUG_FLAGS_DECL;
7277 PERL_UNUSED_ARG(depth);
7280 DEBUG_PARSE("clas");
7282 /* Assume we are going to generate an ANYOF node. */
7283 ret = reganode(pRExC_state, ANYOF, 0);
7286 ANYOF_FLAGS(ret) = 0;
7288 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7292 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7296 RExC_size += ANYOF_SKIP;
7297 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7300 RExC_emit += ANYOF_SKIP;
7302 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7304 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7305 ANYOF_BITMAP_ZERO(ret);
7306 listsv = newSVpvs("# comment\n");
7309 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7311 if (!SIZE_ONLY && POSIXCC(nextvalue))
7312 checkposixcc(pRExC_state);
7314 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7315 if (UCHARAT(RExC_parse) == ']')
7319 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7323 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7326 rangebegin = RExC_parse;
7328 value = utf8n_to_uvchr((U8*)RExC_parse,
7329 RExC_end - RExC_parse,
7330 &numlen, UTF8_ALLOW_DEFAULT);
7331 RExC_parse += numlen;
7334 value = UCHARAT(RExC_parse++);
7336 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7337 if (value == '[' && POSIXCC(nextvalue))
7338 namedclass = regpposixcc(pRExC_state, value);
7339 else if (value == '\\') {
7341 value = utf8n_to_uvchr((U8*)RExC_parse,
7342 RExC_end - RExC_parse,
7343 &numlen, UTF8_ALLOW_DEFAULT);
7344 RExC_parse += numlen;
7347 value = UCHARAT(RExC_parse++);
7348 /* Some compilers cannot handle switching on 64-bit integer
7349 * values, therefore value cannot be an UV. Yes, this will
7350 * be a problem later if we want switch on Unicode.
7351 * A similar issue a little bit later when switching on
7352 * namedclass. --jhi */
7353 switch ((I32)value) {
7354 case 'w': namedclass = ANYOF_ALNUM; break;
7355 case 'W': namedclass = ANYOF_NALNUM; break;
7356 case 's': namedclass = ANYOF_SPACE; break;
7357 case 'S': namedclass = ANYOF_NSPACE; break;
7358 case 'd': namedclass = ANYOF_DIGIT; break;
7359 case 'D': namedclass = ANYOF_NDIGIT; break;
7360 case 'N': /* Handle \N{NAME} in class */
7362 /* We only pay attention to the first char of
7363 multichar strings being returned. I kinda wonder
7364 if this makes sense as it does change the behaviour
7365 from earlier versions, OTOH that behaviour was broken
7367 UV v; /* value is register so we cant & it /grrr */
7368 if (reg_namedseq(pRExC_state, &v)) {
7378 if (RExC_parse >= RExC_end)
7379 vFAIL2("Empty \\%c{}", (U8)value);
7380 if (*RExC_parse == '{') {
7381 const U8 c = (U8)value;
7382 e = strchr(RExC_parse++, '}');
7384 vFAIL2("Missing right brace on \\%c{}", c);
7385 while (isSPACE(UCHARAT(RExC_parse)))
7387 if (e == RExC_parse)
7388 vFAIL2("Empty \\%c{}", c);
7390 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7398 if (UCHARAT(RExC_parse) == '^') {
7401 value = value == 'p' ? 'P' : 'p'; /* toggle */
7402 while (isSPACE(UCHARAT(RExC_parse))) {
7407 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7408 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7411 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7412 namedclass = ANYOF_MAX; /* no official name, but it's named */
7415 case 'n': value = '\n'; break;
7416 case 'r': value = '\r'; break;
7417 case 't': value = '\t'; break;
7418 case 'f': value = '\f'; break;
7419 case 'b': value = '\b'; break;
7420 case 'e': value = ASCII_TO_NATIVE('\033');break;
7421 case 'a': value = ASCII_TO_NATIVE('\007');break;
7423 if (*RExC_parse == '{') {
7424 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7425 | PERL_SCAN_DISALLOW_PREFIX;
7426 char * const e = strchr(RExC_parse++, '}');
7428 vFAIL("Missing right brace on \\x{}");
7430 numlen = e - RExC_parse;
7431 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7435 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7437 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7438 RExC_parse += numlen;
7440 if (PL_encoding && value < 0x100)
7441 goto recode_encoding;
7444 value = UCHARAT(RExC_parse++);
7445 value = toCTRL(value);
7447 case '0': case '1': case '2': case '3': case '4':
7448 case '5': case '6': case '7': case '8': case '9':
7452 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7453 RExC_parse += numlen;
7454 if (PL_encoding && value < 0x100)
7455 goto recode_encoding;
7460 SV* enc = PL_encoding;
7461 value = reg_recode((const char)(U8)value, &enc);
7462 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7464 "Invalid escape in the specified encoding");
7468 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7470 "Unrecognized escape \\%c in character class passed through",
7474 } /* end of \blah */
7480 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7482 if (!SIZE_ONLY && !need_class)
7483 ANYOF_CLASS_ZERO(ret);
7487 /* a bad range like a-\d, a-[:digit:] ? */
7490 if (ckWARN(WARN_REGEXP)) {
7492 RExC_parse >= rangebegin ?
7493 RExC_parse - rangebegin : 0;
7495 "False [] range \"%*.*s\"",
7498 if (prevvalue < 256) {
7499 ANYOF_BITMAP_SET(ret, prevvalue);
7500 ANYOF_BITMAP_SET(ret, '-');
7503 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7504 Perl_sv_catpvf(aTHX_ listsv,
7505 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7509 range = 0; /* this was not a true range */
7515 const char *what = NULL;
7518 if (namedclass > OOB_NAMEDCLASS)
7519 optimize_invert = FALSE;
7520 /* Possible truncation here but in some 64-bit environments
7521 * the compiler gets heartburn about switch on 64-bit values.
7522 * A similar issue a little earlier when switching on value.
7524 switch ((I32)namedclass) {
7525 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7526 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7527 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7528 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7529 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7530 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7531 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7532 case _C_C_T_(PRINT, isPRINT(value), "Print");
7533 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7534 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7535 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7536 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7537 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7540 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7543 for (value = 0; value < 128; value++)
7544 ANYOF_BITMAP_SET(ret, value);
7546 for (value = 0; value < 256; value++) {
7548 ANYOF_BITMAP_SET(ret, value);
7557 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7560 for (value = 128; value < 256; value++)
7561 ANYOF_BITMAP_SET(ret, value);
7563 for (value = 0; value < 256; value++) {
7564 if (!isASCII(value))
7565 ANYOF_BITMAP_SET(ret, value);
7574 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7576 /* consecutive digits assumed */
7577 for (value = '0'; value <= '9'; value++)
7578 ANYOF_BITMAP_SET(ret, value);
7585 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7587 /* consecutive digits assumed */
7588 for (value = 0; value < '0'; value++)
7589 ANYOF_BITMAP_SET(ret, value);
7590 for (value = '9' + 1; value < 256; value++)
7591 ANYOF_BITMAP_SET(ret, value);
7597 /* this is to handle \p and \P */
7600 vFAIL("Invalid [::] class");
7604 /* Strings such as "+utf8::isWord\n" */
7605 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7608 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7611 } /* end of namedclass \blah */
7614 if (prevvalue > (IV)value) /* b-a */ {
7615 const int w = RExC_parse - rangebegin;
7616 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7617 range = 0; /* not a valid range */
7621 prevvalue = value; /* save the beginning of the range */
7622 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7623 RExC_parse[1] != ']') {
7626 /* a bad range like \w-, [:word:]- ? */
7627 if (namedclass > OOB_NAMEDCLASS) {
7628 if (ckWARN(WARN_REGEXP)) {
7630 RExC_parse >= rangebegin ?
7631 RExC_parse - rangebegin : 0;
7633 "False [] range \"%*.*s\"",
7637 ANYOF_BITMAP_SET(ret, '-');
7639 range = 1; /* yeah, it's a range! */
7640 continue; /* but do it the next time */
7644 /* now is the next time */
7645 /*stored += (value - prevvalue + 1);*/
7647 if (prevvalue < 256) {
7648 const IV ceilvalue = value < 256 ? value : 255;
7651 /* In EBCDIC [\x89-\x91] should include
7652 * the \x8e but [i-j] should not. */
7653 if (literal_endpoint == 2 &&
7654 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7655 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7657 if (isLOWER(prevvalue)) {
7658 for (i = prevvalue; i <= ceilvalue; i++)
7660 ANYOF_BITMAP_SET(ret, i);
7662 for (i = prevvalue; i <= ceilvalue; i++)
7664 ANYOF_BITMAP_SET(ret, i);
7669 for (i = prevvalue; i <= ceilvalue; i++) {
7670 if (!ANYOF_BITMAP_TEST(ret,i)) {
7672 ANYOF_BITMAP_SET(ret, i);
7676 if (value > 255 || UTF) {
7677 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7678 const UV natvalue = NATIVE_TO_UNI(value);
7679 stored+=2; /* can't optimize this class */
7680 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7681 if (prevnatvalue < natvalue) { /* what about > ? */
7682 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7683 prevnatvalue, natvalue);
7685 else if (prevnatvalue == natvalue) {
7686 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7688 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7690 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7692 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7693 if (RExC_precomp[0] == ':' &&
7694 RExC_precomp[1] == '[' &&
7695 (f == 0xDF || f == 0x92)) {
7696 f = NATIVE_TO_UNI(f);
7699 /* If folding and foldable and a single
7700 * character, insert also the folded version
7701 * to the charclass. */
7703 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7704 if ((RExC_precomp[0] == ':' &&
7705 RExC_precomp[1] == '[' &&
7707 (value == 0xFB05 || value == 0xFB06))) ?
7708 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7709 foldlen == (STRLEN)UNISKIP(f) )
7711 if (foldlen == (STRLEN)UNISKIP(f))
7713 Perl_sv_catpvf(aTHX_ listsv,
7716 /* Any multicharacter foldings
7717 * require the following transform:
7718 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7719 * where E folds into "pq" and F folds
7720 * into "rst", all other characters
7721 * fold to single characters. We save
7722 * away these multicharacter foldings,
7723 * to be later saved as part of the
7724 * additional "s" data. */
7727 if (!unicode_alternate)
7728 unicode_alternate = newAV();
7729 sv = newSVpvn((char*)foldbuf, foldlen);
7731 av_push(unicode_alternate, sv);
7735 /* If folding and the value is one of the Greek
7736 * sigmas insert a few more sigmas to make the
7737 * folding rules of the sigmas to work right.
7738 * Note that not all the possible combinations
7739 * are handled here: some of them are handled
7740 * by the standard folding rules, and some of
7741 * them (literal or EXACTF cases) are handled
7742 * during runtime in regexec.c:S_find_byclass(). */
7743 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7744 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7745 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7746 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7747 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7749 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7750 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7751 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7756 literal_endpoint = 0;
7760 range = 0; /* this range (if it was one) is done now */
7764 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7766 RExC_size += ANYOF_CLASS_ADD_SKIP;
7768 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7774 /****** !SIZE_ONLY AFTER HERE *********/
7776 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
7777 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7779 /* optimize single char class to an EXACT node
7780 but *only* when its not a UTF/high char */
7781 const char * cur_parse= RExC_parse;
7782 RExC_emit = (regnode *)orig_emit;
7783 RExC_parse = (char *)orig_parse;
7784 ret = reg_node(pRExC_state,
7785 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7786 RExC_parse = (char *)cur_parse;
7787 *STRING(ret)= (char)value;
7789 RExC_emit += STR_SZ(1);
7792 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7793 if ( /* If the only flag is folding (plus possibly inversion). */
7794 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7796 for (value = 0; value < 256; ++value) {
7797 if (ANYOF_BITMAP_TEST(ret, value)) {
7798 UV fold = PL_fold[value];
7801 ANYOF_BITMAP_SET(ret, fold);
7804 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7807 /* optimize inverted simple patterns (e.g. [^a-z]) */
7808 if (optimize_invert &&
7809 /* If the only flag is inversion. */
7810 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7811 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7812 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7813 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7816 AV * const av = newAV();
7818 /* The 0th element stores the character class description
7819 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7820 * to initialize the appropriate swash (which gets stored in
7821 * the 1st element), and also useful for dumping the regnode.
7822 * The 2nd element stores the multicharacter foldings,
7823 * used later (regexec.c:S_reginclass()). */
7824 av_store(av, 0, listsv);
7825 av_store(av, 1, NULL);
7826 av_store(av, 2, (SV*)unicode_alternate);
7827 rv = newRV_noinc((SV*)av);
7828 n = add_data(pRExC_state, 1, "s");
7829 RExC_rxi->data->data[n] = (void*)rv;
7837 /* reg_skipcomment()
7839 Absorbs an /x style # comments from the input stream.
7840 Returns true if there is more text remaining in the stream.
7841 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
7842 terminates the pattern without including a newline.
7844 Note its the callers responsibility to ensure that we are
7850 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
7853 while (RExC_parse < RExC_end)
7854 if (*RExC_parse++ == '\n') {
7859 /* we ran off the end of the pattern without ending
7860 the comment, so we have to add an \n when wrapping */
7861 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7869 Advance that parse position, and optionall absorbs
7870 "whitespace" from the inputstream.
7872 Without /x "whitespace" means (?#...) style comments only,
7873 with /x this means (?#...) and # comments and whitespace proper.
7875 Returns the RExC_parse point from BEFORE the scan occurs.
7877 This is the /x friendly way of saying RExC_parse++.
7881 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7883 char* const retval = RExC_parse++;
7886 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7887 RExC_parse[2] == '#') {
7888 while (*RExC_parse != ')') {
7889 if (RExC_parse == RExC_end)
7890 FAIL("Sequence (?#... not terminated");
7896 if (RExC_flags & RXf_PMf_EXTENDED) {
7897 if (isSPACE(*RExC_parse)) {
7901 else if (*RExC_parse == '#') {
7902 if ( reg_skipcomment( pRExC_state ) )
7911 - reg_node - emit a node
7913 STATIC regnode * /* Location. */
7914 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7917 register regnode *ptr;
7918 regnode * const ret = RExC_emit;
7919 GET_RE_DEBUG_FLAGS_DECL;
7922 SIZE_ALIGN(RExC_size);
7926 if (RExC_emit >= RExC_emit_bound)
7927 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
7929 NODE_ALIGN_FILL(ret);
7931 FILL_ADVANCE_NODE(ptr, op);
7932 #ifdef RE_TRACK_PATTERN_OFFSETS
7933 if (RExC_offsets) { /* MJD */
7934 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7935 "reg_node", __LINE__,
7937 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7938 ? "Overwriting end of array!\n" : "OK",
7939 (UV)(RExC_emit - RExC_emit_start),
7940 (UV)(RExC_parse - RExC_start),
7941 (UV)RExC_offsets[0]));
7942 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7950 - reganode - emit a node with an argument
7952 STATIC regnode * /* Location. */
7953 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7956 register regnode *ptr;
7957 regnode * const ret = RExC_emit;
7958 GET_RE_DEBUG_FLAGS_DECL;
7961 SIZE_ALIGN(RExC_size);
7966 assert(2==regarglen[op]+1);
7968 Anything larger than this has to allocate the extra amount.
7969 If we changed this to be:
7971 RExC_size += (1 + regarglen[op]);
7973 then it wouldn't matter. Its not clear what side effect
7974 might come from that so its not done so far.
7979 if (RExC_emit >= RExC_emit_bound)
7980 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
7982 NODE_ALIGN_FILL(ret);
7984 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7985 #ifdef RE_TRACK_PATTERN_OFFSETS
7986 if (RExC_offsets) { /* MJD */
7987 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7991 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7992 "Overwriting end of array!\n" : "OK",
7993 (UV)(RExC_emit - RExC_emit_start),
7994 (UV)(RExC_parse - RExC_start),
7995 (UV)RExC_offsets[0]));
7996 Set_Cur_Node_Offset;
8004 - reguni - emit (if appropriate) a Unicode character
8007 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8010 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8014 - reginsert - insert an operator in front of already-emitted operand
8016 * Means relocating the operand.
8019 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8022 register regnode *src;
8023 register regnode *dst;
8024 register regnode *place;
8025 const int offset = regarglen[(U8)op];
8026 const int size = NODE_STEP_REGNODE + offset;
8027 GET_RE_DEBUG_FLAGS_DECL;
8028 PERL_UNUSED_ARG(depth);
8029 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8030 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8039 if (RExC_open_parens) {
8041 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8042 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8043 if ( RExC_open_parens[paren] >= opnd ) {
8044 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8045 RExC_open_parens[paren] += size;
8047 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8049 if ( RExC_close_parens[paren] >= opnd ) {
8050 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8051 RExC_close_parens[paren] += size;
8053 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8058 while (src > opnd) {
8059 StructCopy(--src, --dst, regnode);
8060 #ifdef RE_TRACK_PATTERN_OFFSETS
8061 if (RExC_offsets) { /* MJD 20010112 */
8062 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8066 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8067 ? "Overwriting end of array!\n" : "OK",
8068 (UV)(src - RExC_emit_start),
8069 (UV)(dst - RExC_emit_start),
8070 (UV)RExC_offsets[0]));
8071 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8072 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8078 place = opnd; /* Op node, where operand used to be. */
8079 #ifdef RE_TRACK_PATTERN_OFFSETS
8080 if (RExC_offsets) { /* MJD */
8081 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8085 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8086 ? "Overwriting end of array!\n" : "OK",
8087 (UV)(place - RExC_emit_start),
8088 (UV)(RExC_parse - RExC_start),
8089 (UV)RExC_offsets[0]));
8090 Set_Node_Offset(place, RExC_parse);
8091 Set_Node_Length(place, 1);
8094 src = NEXTOPER(place);
8095 FILL_ADVANCE_NODE(place, op);
8096 Zero(src, offset, regnode);
8100 - regtail - set the next-pointer at the end of a node chain of p to val.
8101 - SEE ALSO: regtail_study
8103 /* TODO: All three parms should be const */
8105 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8108 register regnode *scan;
8109 GET_RE_DEBUG_FLAGS_DECL;
8111 PERL_UNUSED_ARG(depth);
8117 /* Find last node. */
8120 regnode * const temp = regnext(scan);
8122 SV * const mysv=sv_newmortal();
8123 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8124 regprop(RExC_rx, mysv, scan);
8125 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8126 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8127 (temp == NULL ? "->" : ""),
8128 (temp == NULL ? PL_reg_name[OP(val)] : "")
8136 if (reg_off_by_arg[OP(scan)]) {
8137 ARG_SET(scan, val - scan);
8140 NEXT_OFF(scan) = val - scan;
8146 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8147 - Look for optimizable sequences at the same time.
8148 - currently only looks for EXACT chains.
8150 This is expermental code. The idea is to use this routine to perform
8151 in place optimizations on branches and groups as they are constructed,
8152 with the long term intention of removing optimization from study_chunk so
8153 that it is purely analytical.
8155 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8156 to control which is which.
8159 /* TODO: All four parms should be const */
8162 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8165 register regnode *scan;
8167 #ifdef EXPERIMENTAL_INPLACESCAN
8171 GET_RE_DEBUG_FLAGS_DECL;
8177 /* Find last node. */
8181 regnode * const temp = regnext(scan);
8182 #ifdef EXPERIMENTAL_INPLACESCAN
8183 if (PL_regkind[OP(scan)] == EXACT)
8184 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8192 if( exact == PSEUDO )
8194 else if ( exact != OP(scan) )
8203 SV * const mysv=sv_newmortal();
8204 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8205 regprop(RExC_rx, mysv, scan);
8206 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8207 SvPV_nolen_const(mysv),
8209 PL_reg_name[exact]);
8216 SV * const mysv_val=sv_newmortal();
8217 DEBUG_PARSE_MSG("");
8218 regprop(RExC_rx, mysv_val, val);
8219 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8220 SvPV_nolen_const(mysv_val),
8221 (IV)REG_NODE_NUM(val),
8225 if (reg_off_by_arg[OP(scan)]) {
8226 ARG_SET(scan, val - scan);
8229 NEXT_OFF(scan) = val - scan;
8237 - regcurly - a little FSA that accepts {\d+,?\d*}
8240 S_regcurly(register const char *s)
8259 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8262 Perl_regdump(pTHX_ const regexp *r)
8266 SV * const sv = sv_newmortal();
8267 SV *dsv= sv_newmortal();
8270 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8272 /* Header fields of interest. */
8273 if (r->anchored_substr) {
8274 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8275 RE_SV_DUMPLEN(r->anchored_substr), 30);
8276 PerlIO_printf(Perl_debug_log,
8277 "anchored %s%s at %"IVdf" ",
8278 s, RE_SV_TAIL(r->anchored_substr),
8279 (IV)r->anchored_offset);
8280 } else if (r->anchored_utf8) {
8281 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8282 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8283 PerlIO_printf(Perl_debug_log,
8284 "anchored utf8 %s%s at %"IVdf" ",
8285 s, RE_SV_TAIL(r->anchored_utf8),
8286 (IV)r->anchored_offset);
8288 if (r->float_substr) {
8289 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8290 RE_SV_DUMPLEN(r->float_substr), 30);
8291 PerlIO_printf(Perl_debug_log,
8292 "floating %s%s at %"IVdf"..%"UVuf" ",
8293 s, RE_SV_TAIL(r->float_substr),
8294 (IV)r->float_min_offset, (UV)r->float_max_offset);
8295 } else if (r->float_utf8) {
8296 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8297 RE_SV_DUMPLEN(r->float_utf8), 30);
8298 PerlIO_printf(Perl_debug_log,
8299 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8300 s, RE_SV_TAIL(r->float_utf8),
8301 (IV)r->float_min_offset, (UV)r->float_max_offset);
8303 if (r->check_substr || r->check_utf8)
8304 PerlIO_printf(Perl_debug_log,
8306 (r->check_substr == r->float_substr
8307 && r->check_utf8 == r->float_utf8
8308 ? "(checking floating" : "(checking anchored"));
8309 if (r->extflags & RXf_NOSCAN)
8310 PerlIO_printf(Perl_debug_log, " noscan");
8311 if (r->extflags & RXf_CHECK_ALL)
8312 PerlIO_printf(Perl_debug_log, " isall");
8313 if (r->check_substr || r->check_utf8)
8314 PerlIO_printf(Perl_debug_log, ") ");
8316 if (ri->regstclass) {
8317 regprop(r, sv, ri->regstclass);
8318 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8320 if (r->extflags & RXf_ANCH) {
8321 PerlIO_printf(Perl_debug_log, "anchored");
8322 if (r->extflags & RXf_ANCH_BOL)
8323 PerlIO_printf(Perl_debug_log, "(BOL)");
8324 if (r->extflags & RXf_ANCH_MBOL)
8325 PerlIO_printf(Perl_debug_log, "(MBOL)");
8326 if (r->extflags & RXf_ANCH_SBOL)
8327 PerlIO_printf(Perl_debug_log, "(SBOL)");
8328 if (r->extflags & RXf_ANCH_GPOS)
8329 PerlIO_printf(Perl_debug_log, "(GPOS)");
8330 PerlIO_putc(Perl_debug_log, ' ');
8332 if (r->extflags & RXf_GPOS_SEEN)
8333 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8334 if (r->intflags & PREGf_SKIP)
8335 PerlIO_printf(Perl_debug_log, "plus ");
8336 if (r->intflags & PREGf_IMPLICIT)
8337 PerlIO_printf(Perl_debug_log, "implicit ");
8338 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8339 if (r->extflags & RXf_EVAL_SEEN)
8340 PerlIO_printf(Perl_debug_log, "with eval ");
8341 PerlIO_printf(Perl_debug_log, "\n");
8343 PERL_UNUSED_CONTEXT;
8345 #endif /* DEBUGGING */
8349 - regprop - printable representation of opcode
8352 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8357 RXi_GET_DECL(prog,progi);
8358 GET_RE_DEBUG_FLAGS_DECL;
8361 sv_setpvn(sv, "", 0);
8363 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8364 /* It would be nice to FAIL() here, but this may be called from
8365 regexec.c, and it would be hard to supply pRExC_state. */
8366 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8367 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
8369 k = PL_regkind[OP(o)];
8372 SV * const dsv = sv_2mortal(newSVpvs(""));
8373 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8374 * is a crude hack but it may be the best for now since
8375 * we have no flag "this EXACTish node was UTF-8"
8377 const char * const s =
8378 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8379 PL_colors[0], PL_colors[1],
8380 PERL_PV_ESCAPE_UNI_DETECT |
8381 PERL_PV_PRETTY_ELIPSES |
8384 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8385 } else if (k == TRIE) {
8386 /* print the details of the trie in dumpuntil instead, as
8387 * progi->data isn't available here */
8388 const char op = OP(o);
8389 const U32 n = ARG(o);
8390 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8391 (reg_ac_data *)progi->data->data[n] :
8393 const reg_trie_data * const trie
8394 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8396 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
8397 DEBUG_TRIE_COMPILE_r(
8398 Perl_sv_catpvf(aTHX_ sv,
8399 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8400 (UV)trie->startstate,
8401 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8402 (UV)trie->wordcount,
8405 (UV)TRIE_CHARCOUNT(trie),
8406 (UV)trie->uniquecharcount
8409 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8411 int rangestart = -1;
8412 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8413 Perl_sv_catpvf(aTHX_ sv, "[");
8414 for (i = 0; i <= 256; i++) {
8415 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8416 if (rangestart == -1)
8418 } else if (rangestart != -1) {
8419 if (i <= rangestart + 3)
8420 for (; rangestart < i; rangestart++)
8421 put_byte(sv, rangestart);
8423 put_byte(sv, rangestart);
8425 put_byte(sv, i - 1);
8430 Perl_sv_catpvf(aTHX_ sv, "]");
8433 } else if (k == CURLY) {
8434 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8435 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8436 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8438 else if (k == WHILEM && o->flags) /* Ordinal/of */
8439 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8440 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8441 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8442 if ( prog->paren_names ) {
8443 if ( k != REF || OP(o) < NREF) {
8444 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8445 SV **name= av_fetch(list, ARG(o), 0 );
8447 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8450 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8451 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8452 I32 *nums=(I32*)SvPVX(sv_dat);
8453 SV **name= av_fetch(list, nums[0], 0 );
8456 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8457 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8458 (n ? "," : ""), (IV)nums[n]);
8460 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8464 } else if (k == GOSUB)
8465 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8466 else if (k == VERB) {
8468 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8469 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8470 } else if (k == LOGICAL)
8471 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8472 else if (k == ANYOF) {
8473 int i, rangestart = -1;
8474 const U8 flags = ANYOF_FLAGS(o);
8476 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8477 static const char * const anyofs[] = {
8510 if (flags & ANYOF_LOCALE)
8511 sv_catpvs(sv, "{loc}");
8512 if (flags & ANYOF_FOLD)
8513 sv_catpvs(sv, "{i}");
8514 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8515 if (flags & ANYOF_INVERT)
8517 for (i = 0; i <= 256; i++) {
8518 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8519 if (rangestart == -1)
8521 } else if (rangestart != -1) {
8522 if (i <= rangestart + 3)
8523 for (; rangestart < i; rangestart++)
8524 put_byte(sv, rangestart);
8526 put_byte(sv, rangestart);
8528 put_byte(sv, i - 1);
8534 if (o->flags & ANYOF_CLASS)
8535 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8536 if (ANYOF_CLASS_TEST(o,i))
8537 sv_catpv(sv, anyofs[i]);
8539 if (flags & ANYOF_UNICODE)
8540 sv_catpvs(sv, "{unicode}");
8541 else if (flags & ANYOF_UNICODE_ALL)
8542 sv_catpvs(sv, "{unicode_all}");
8546 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8550 U8 s[UTF8_MAXBYTES_CASE+1];
8552 for (i = 0; i <= 256; i++) { /* just the first 256 */
8553 uvchr_to_utf8(s, i);
8555 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8556 if (rangestart == -1)
8558 } else if (rangestart != -1) {
8559 if (i <= rangestart + 3)
8560 for (; rangestart < i; rangestart++) {
8561 const U8 * const e = uvchr_to_utf8(s,rangestart);
8563 for(p = s; p < e; p++)
8567 const U8 *e = uvchr_to_utf8(s,rangestart);
8569 for (p = s; p < e; p++)
8572 e = uvchr_to_utf8(s, i-1);
8573 for (p = s; p < e; p++)
8580 sv_catpvs(sv, "..."); /* et cetera */
8584 char *s = savesvpv(lv);
8585 char * const origs = s;
8587 while (*s && *s != '\n')
8591 const char * const t = ++s;
8609 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8611 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8612 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8614 PERL_UNUSED_CONTEXT;
8615 PERL_UNUSED_ARG(sv);
8617 PERL_UNUSED_ARG(prog);
8618 #endif /* DEBUGGING */
8622 Perl_re_intuit_string(pTHX_ regexp *prog)
8623 { /* Assume that RE_INTUIT is set */
8625 GET_RE_DEBUG_FLAGS_DECL;
8626 PERL_UNUSED_CONTEXT;
8630 const char * const s = SvPV_nolen_const(prog->check_substr
8631 ? prog->check_substr : prog->check_utf8);
8633 if (!PL_colorset) reginitcolors();
8634 PerlIO_printf(Perl_debug_log,
8635 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8637 prog->check_substr ? "" : "utf8 ",
8638 PL_colors[5],PL_colors[0],
8641 (strlen(s) > 60 ? "..." : ""));
8644 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8650 handles refcounting and freeing the perl core regexp structure. When
8651 it is necessary to actually free the structure the first thing it
8652 does is call the 'free' method of the regexp_engine associated to to
8653 the regexp, allowing the handling of the void *pprivate; member
8654 first. (This routine is not overridable by extensions, which is why
8655 the extensions free is called first.)
8657 See regdupe and regdupe_internal if you change anything here.
8659 #ifndef PERL_IN_XSUB_RE
8661 Perl_pregfree(pTHX_ struct regexp *r)
8664 GET_RE_DEBUG_FLAGS_DECL;
8666 if (!r || (--r->refcnt > 0))
8669 CALLREGFREE_PVT(r); /* free the private data */
8670 RX_MATCH_COPY_FREE(r);
8671 #ifdef PERL_OLD_COPY_ON_WRITE
8673 SvREFCNT_dec(r->saved_copy);
8676 if (r->anchored_substr)
8677 SvREFCNT_dec(r->anchored_substr);
8678 if (r->anchored_utf8)
8679 SvREFCNT_dec(r->anchored_utf8);
8680 if (r->float_substr)
8681 SvREFCNT_dec(r->float_substr);
8683 SvREFCNT_dec(r->float_utf8);
8684 Safefree(r->substrs);
8687 SvREFCNT_dec(r->paren_names);
8688 Safefree(r->wrapped);
8689 Safefree(r->startp);
8695 /* regfree_internal()
8697 Free the private data in a regexp. This is overloadable by
8698 extensions. Perl takes care of the regexp structure in pregfree(),
8699 this covers the *pprivate pointer which technically perldoesnt
8700 know about, however of course we have to handle the
8701 regexp_internal structure when no extension is in use.
8703 Note this is called before freeing anything in the regexp
8708 Perl_regfree_internal(pTHX_ struct regexp *r)
8712 GET_RE_DEBUG_FLAGS_DECL;
8718 SV *dsv= sv_newmortal();
8719 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8720 dsv, r->precomp, r->prelen, 60);
8721 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8722 PL_colors[4],PL_colors[5],s);
8725 #ifdef RE_TRACK_PATTERN_OFFSETS
8727 Safefree(ri->u.offsets); /* 20010421 MJD */
8730 int n = ri->data->count;
8731 PAD* new_comppad = NULL;
8736 /* If you add a ->what type here, update the comment in regcomp.h */
8737 switch (ri->data->what[n]) {
8741 SvREFCNT_dec((SV*)ri->data->data[n]);
8744 Safefree(ri->data->data[n]);
8747 new_comppad = (AV*)ri->data->data[n];
8750 if (new_comppad == NULL)
8751 Perl_croak(aTHX_ "panic: pregfree comppad");
8752 PAD_SAVE_LOCAL(old_comppad,
8753 /* Watch out for global destruction's random ordering. */
8754 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8757 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8760 op_free((OP_4tree*)ri->data->data[n]);
8762 PAD_RESTORE_LOCAL(old_comppad);
8763 SvREFCNT_dec((SV*)new_comppad);
8769 { /* Aho Corasick add-on structure for a trie node.
8770 Used in stclass optimization only */
8772 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8774 refcount = --aho->refcount;
8777 PerlMemShared_free(aho->states);
8778 PerlMemShared_free(aho->fail);
8779 /* do this last!!!! */
8780 PerlMemShared_free(ri->data->data[n]);
8781 PerlMemShared_free(ri->regstclass);
8787 /* trie structure. */
8789 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8791 refcount = --trie->refcount;
8794 PerlMemShared_free(trie->charmap);
8795 PerlMemShared_free(trie->states);
8796 PerlMemShared_free(trie->trans);
8798 PerlMemShared_free(trie->bitmap);
8800 PerlMemShared_free(trie->wordlen);
8802 PerlMemShared_free(trie->jump);
8804 PerlMemShared_free(trie->nextword);
8805 /* do this last!!!! */
8806 PerlMemShared_free(ri->data->data[n]);
8811 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8814 Safefree(ri->data->what);
8818 Safefree(ri->swap->startp);
8819 Safefree(ri->swap->endp);
8825 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8826 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8827 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8828 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8831 regdupe - duplicate a regexp.
8833 This routine is called by sv.c's re_dup and is expected to clone a
8834 given regexp structure. It is a no-op when not under USE_ITHREADS.
8835 (Originally this *was* re_dup() for change history see sv.c)
8837 After all of the core data stored in struct regexp is duplicated
8838 the regexp_engine.dupe method is used to copy any private data
8839 stored in the *pprivate pointer. This allows extensions to handle
8840 any duplication it needs to do.
8842 See pregfree() and regfree_internal() if you change anything here.
8844 #if defined(USE_ITHREADS)
8845 #ifndef PERL_IN_XSUB_RE
8847 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
8852 struct reg_substr_datum *s;
8855 return (REGEXP *)NULL;
8857 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8861 npar = r->nparens+1;
8862 Newxz(ret, 1, regexp);
8863 Newx(ret->startp, npar, I32);
8864 Copy(r->startp, ret->startp, npar, I32);
8865 Newx(ret->endp, npar, I32);
8866 Copy(r->endp, ret->endp, npar, I32);
8869 Newx(ret->substrs, 1, struct reg_substr_data);
8870 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8871 s->min_offset = r->substrs->data[i].min_offset;
8872 s->max_offset = r->substrs->data[i].max_offset;
8873 s->end_shift = r->substrs->data[i].end_shift;
8874 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8875 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8878 ret->substrs = NULL;
8880 ret->wrapped = SAVEPVN(r->wrapped, r->wraplen);
8881 ret->precomp = ret->wrapped + (r->precomp - r->wrapped);
8882 ret->prelen = r->prelen;
8883 ret->wraplen = r->wraplen;
8885 ret->refcnt = r->refcnt;
8886 ret->minlen = r->minlen;
8887 ret->minlenret = r->minlenret;
8888 ret->nparens = r->nparens;
8889 ret->lastparen = r->lastparen;
8890 ret->lastcloseparen = r->lastcloseparen;
8891 ret->intflags = r->intflags;
8892 ret->extflags = r->extflags;
8894 ret->sublen = r->sublen;
8896 ret->engine = r->engine;
8898 ret->paren_names = hv_dup_inc(r->paren_names, param);
8900 if (RX_MATCH_COPIED(ret))
8901 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8904 #ifdef PERL_OLD_COPY_ON_WRITE
8905 ret->saved_copy = NULL;
8908 ret->pprivate = r->pprivate;
8910 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
8912 ptr_table_store(PL_ptr_table, r, ret);
8915 #endif /* PERL_IN_XSUB_RE */
8920 This is the internal complement to regdupe() which is used to copy
8921 the structure pointed to by the *pprivate pointer in the regexp.
8922 This is the core version of the extension overridable cloning hook.
8923 The regexp structure being duplicated will be copied by perl prior
8924 to this and will be provided as the regexp *r argument, however
8925 with the /old/ structures pprivate pointer value. Thus this routine
8926 may override any copying normally done by perl.
8928 It returns a pointer to the new regexp_internal structure.
8932 Perl_regdupe_internal(pTHX_ const regexp *r, CLONE_PARAMS *param)
8935 regexp_internal *reti;
8939 npar = r->nparens+1;
8942 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8943 Copy(ri->program, reti->program, len+1, regnode);
8946 Newx(reti->swap, 1, regexp_paren_ofs);
8947 /* no need to copy these */
8948 Newx(reti->swap->startp, npar, I32);
8949 Newx(reti->swap->endp, npar, I32);
8954 reti->regstclass = NULL;
8958 const int count = ri->data->count;
8961 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8962 char, struct reg_data);
8963 Newx(d->what, count, U8);
8966 for (i = 0; i < count; i++) {
8967 d->what[i] = ri->data->what[i];
8968 switch (d->what[i]) {
8969 /* legal options are one of: sSfpontTu
8970 see also regcomp.h and pregfree() */
8973 case 'p': /* actually an AV, but the dup function is identical. */
8974 case 'u': /* actually an HV, but the dup function is identical. */
8975 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
8978 /* This is cheating. */
8979 Newx(d->data[i], 1, struct regnode_charclass_class);
8980 StructCopy(ri->data->data[i], d->data[i],
8981 struct regnode_charclass_class);
8982 reti->regstclass = (regnode*)d->data[i];
8985 /* Compiled op trees are readonly and in shared memory,
8986 and can thus be shared without duplication. */
8988 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
8992 /* Trie stclasses are readonly and can thus be shared
8993 * without duplication. We free the stclass in pregfree
8994 * when the corresponding reg_ac_data struct is freed.
8996 reti->regstclass= ri->regstclass;
9000 ((reg_trie_data*)ri->data->data[i])->refcount++;
9004 d->data[i] = ri->data->data[i];
9007 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9016 reti->name_list_idx = ri->name_list_idx;
9018 #ifdef RE_TRACK_PATTERN_OFFSETS
9019 if (ri->u.offsets) {
9020 Newx(reti->u.offsets, 2*len+1, U32);
9021 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9024 SetProgLen(reti,len);
9030 #endif /* USE_ITHREADS */
9035 converts a regexp embedded in a MAGIC struct to its stringified form,
9036 caching the converted form in the struct and returns the cached
9039 If lp is nonnull then it is used to return the length of the
9042 If flags is nonnull and the returned string contains UTF8 then
9043 (*flags & 1) will be true.
9045 If haseval is nonnull then it is used to return whether the pattern
9048 Normally called via macro:
9050 CALLREG_STRINGIFY(mg,&len,&utf8);
9054 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9056 See sv_2pv_flags() in sv.c for an example of internal usage.
9059 #ifndef PERL_IN_XSUB_RE
9062 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9064 const regexp * const re = (regexp *)mg->mg_obj;
9066 *haseval = re->seen_evals;
9068 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
9075 - regnext - dig the "next" pointer out of a node
9078 Perl_regnext(pTHX_ register regnode *p)
9081 register I32 offset;
9086 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9095 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9098 STRLEN l1 = strlen(pat1);
9099 STRLEN l2 = strlen(pat2);
9102 const char *message;
9108 Copy(pat1, buf, l1 , char);
9109 Copy(pat2, buf + l1, l2 , char);
9110 buf[l1 + l2] = '\n';
9111 buf[l1 + l2 + 1] = '\0';
9113 /* ANSI variant takes additional second argument */
9114 va_start(args, pat2);
9118 msv = vmess(buf, &args);
9120 message = SvPV_const(msv,l1);
9123 Copy(message, buf, l1 , char);
9124 buf[l1-1] = '\0'; /* Overwrite \n */
9125 Perl_croak(aTHX_ "%s", buf);
9128 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9130 #ifndef PERL_IN_XSUB_RE
9132 Perl_save_re_context(pTHX)
9136 struct re_save_state *state;
9138 SAVEVPTR(PL_curcop);
9139 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9141 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9142 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9143 SSPUSHINT(SAVEt_RE_STATE);
9145 Copy(&PL_reg_state, state, 1, struct re_save_state);
9147 PL_reg_start_tmp = 0;
9148 PL_reg_start_tmpl = 0;
9149 PL_reg_oldsaved = NULL;
9150 PL_reg_oldsavedlen = 0;
9152 PL_reg_leftiter = 0;
9153 PL_reg_poscache = NULL;
9154 PL_reg_poscache_size = 0;
9155 #ifdef PERL_OLD_COPY_ON_WRITE
9159 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9161 const REGEXP * const rx = PM_GETRE(PL_curpm);
9164 for (i = 1; i <= rx->nparens; i++) {
9165 char digits[TYPE_CHARS(long)];
9166 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9167 GV *const *const gvp
9168 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9171 GV * const gv = *gvp;
9172 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9182 clear_re(pTHX_ void *r)
9185 ReREFCNT_dec((regexp *)r);
9191 S_put_byte(pTHX_ SV *sv, int c)
9193 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9194 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9195 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9196 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9198 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9202 #define CLEAR_OPTSTART \
9203 if (optstart) STMT_START { \
9204 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9208 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9210 STATIC const regnode *
9211 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9212 const regnode *last, const regnode *plast,
9213 SV* sv, I32 indent, U32 depth)
9216 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9217 register const regnode *next;
9218 const regnode *optstart= NULL;
9221 GET_RE_DEBUG_FLAGS_DECL;
9223 #ifdef DEBUG_DUMPUNTIL
9224 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9225 last ? last-start : 0,plast ? plast-start : 0);
9228 if (plast && plast < last)
9231 while (PL_regkind[op] != END && (!last || node < last)) {
9232 /* While that wasn't END last time... */
9235 if (op == CLOSE || op == WHILEM)
9237 next = regnext((regnode *)node);
9240 if (OP(node) == OPTIMIZED) {
9241 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9248 regprop(r, sv, node);
9249 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9250 (int)(2*indent + 1), "", SvPVX_const(sv));
9252 if (OP(node) != OPTIMIZED) {
9253 if (next == NULL) /* Next ptr. */
9254 PerlIO_printf(Perl_debug_log, " (0)");
9255 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9256 PerlIO_printf(Perl_debug_log, " (FAIL)");
9258 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9259 (void)PerlIO_putc(Perl_debug_log, '\n');
9263 if (PL_regkind[(U8)op] == BRANCHJ) {
9266 register const regnode *nnode = (OP(next) == LONGJMP
9267 ? regnext((regnode *)next)
9269 if (last && nnode > last)
9271 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9274 else if (PL_regkind[(U8)op] == BRANCH) {
9276 DUMPUNTIL(NEXTOPER(node), next);
9278 else if ( PL_regkind[(U8)op] == TRIE ) {
9279 const regnode *this_trie = node;
9280 const char op = OP(node);
9281 const U32 n = ARG(node);
9282 const reg_ac_data * const ac = op>=AHOCORASICK ?
9283 (reg_ac_data *)ri->data->data[n] :
9285 const reg_trie_data * const trie =
9286 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9288 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9290 const regnode *nextbranch= NULL;
9292 sv_setpvn(sv, "", 0);
9293 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9294 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9296 PerlIO_printf(Perl_debug_log, "%*s%s ",
9297 (int)(2*(indent+3)), "",
9298 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9299 PL_colors[0], PL_colors[1],
9300 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9301 PERL_PV_PRETTY_ELIPSES |
9307 U16 dist= trie->jump[word_idx+1];
9308 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9309 (UV)((dist ? this_trie + dist : next) - start));
9312 nextbranch= this_trie + trie->jump[0];
9313 DUMPUNTIL(this_trie + dist, nextbranch);
9315 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9316 nextbranch= regnext((regnode *)nextbranch);
9318 PerlIO_printf(Perl_debug_log, "\n");
9321 if (last && next > last)
9326 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9327 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9328 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9330 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9332 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9334 else if ( op == PLUS || op == STAR) {
9335 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9337 else if (op == ANYOF) {
9338 /* arglen 1 + class block */
9339 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9340 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9341 node = NEXTOPER(node);
9343 else if (PL_regkind[(U8)op] == EXACT) {
9344 /* Literal string, where present. */
9345 node += NODE_SZ_STR(node) - 1;
9346 node = NEXTOPER(node);
9349 node = NEXTOPER(node);
9350 node += regarglen[(U8)op];
9352 if (op == CURLYX || op == OPEN)
9356 #ifdef DEBUG_DUMPUNTIL
9357 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9362 #endif /* DEBUGGING */
9366 * c-indentation-style: bsd
9368 * indent-tabs-mode: t
9371 * ex: set ts=8 sts=4 sw=4 noet: