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);
1970 char *ch = SvPV( *tmp, len );
1972 SV *sv=sv_newmortal();
1973 PerlIO_printf( Perl_debug_log,
1974 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1975 (int)depth * 2 + 2, "",
1977 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1978 PL_colors[0], PL_colors[1],
1979 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1980 PERL_PV_ESCAPE_FIRSTCHAR
1985 OP( convert ) = nodetype;
1986 str=STRING(convert);
1989 STR_LEN(convert) += len;
1995 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2001 regnode *n = convert+NODE_SZ_STR(convert);
2002 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2003 trie->startstate = state;
2004 trie->minlen -= (state - 1);
2005 trie->maxlen -= (state - 1);
2007 regnode *fix = convert;
2008 U32 word = trie->wordcount;
2010 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2011 while( ++fix < n ) {
2012 Set_Node_Offset_Length(fix, 0, 0);
2015 SV ** const tmp = av_fetch( trie_words, word, 0 );
2017 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2018 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2020 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2027 NEXT_OFF(convert) = (U16)(tail - convert);
2028 DEBUG_r(optimize= n);
2034 if ( trie->maxlen ) {
2035 NEXT_OFF( convert ) = (U16)(tail - convert);
2036 ARG_SET( convert, data_slot );
2037 /* Store the offset to the first unabsorbed branch in
2038 jump[0], which is otherwise unused by the jump logic.
2039 We use this when dumping a trie and during optimisation. */
2041 trie->jump[0] = (U16)(nextbranch - convert);
2044 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2045 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2047 OP( convert ) = TRIEC;
2048 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2049 PerlMemShared_free(trie->bitmap);
2052 OP( convert ) = TRIE;
2054 /* store the type in the flags */
2055 convert->flags = nodetype;
2059 + regarglen[ OP( convert ) ];
2061 /* XXX We really should free up the resource in trie now,
2062 as we won't use them - (which resources?) dmq */
2064 /* needed for dumping*/
2065 DEBUG_r(if (optimize) {
2066 regnode *opt = convert;
2068 while ( ++opt < optimize) {
2069 Set_Node_Offset_Length(opt,0,0);
2072 Try to clean up some of the debris left after the
2075 while( optimize < jumper ) {
2076 mjd_nodelen += Node_Length((optimize));
2077 OP( optimize ) = OPTIMIZED;
2078 Set_Node_Offset_Length(optimize,0,0);
2081 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2083 } /* end node insert */
2084 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2086 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2087 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2089 SvREFCNT_dec(revcharmap);
2093 : trie->startstate>1
2099 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2101 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2103 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2104 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2107 We find the fail state for each state in the trie, this state is the longest proper
2108 suffix of the current states 'word' that is also a proper prefix of another word in our
2109 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2110 the DFA not to have to restart after its tried and failed a word at a given point, it
2111 simply continues as though it had been matching the other word in the first place.
2113 'abcdgu'=~/abcdefg|cdgu/
2114 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2115 fail, which would bring use to the state representing 'd' in the second word where we would
2116 try 'g' and succeed, prodceding to match 'cdgu'.
2118 /* add a fail transition */
2119 const U32 trie_offset = ARG(source);
2120 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2122 const U32 ucharcount = trie->uniquecharcount;
2123 const U32 numstates = trie->statecount;
2124 const U32 ubound = trie->lasttrans + ucharcount;
2128 U32 base = trie->states[ 1 ].trans.base;
2131 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2132 GET_RE_DEBUG_FLAGS_DECL;
2134 PERL_UNUSED_ARG(depth);
2138 ARG_SET( stclass, data_slot );
2139 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2140 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2141 aho->trie=trie_offset;
2142 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2143 Copy( trie->states, aho->states, numstates, reg_trie_state );
2144 Newxz( q, numstates, U32);
2145 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2148 /* initialize fail[0..1] to be 1 so that we always have
2149 a valid final fail state */
2150 fail[ 0 ] = fail[ 1 ] = 1;
2152 for ( charid = 0; charid < ucharcount ; charid++ ) {
2153 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2155 q[ q_write ] = newstate;
2156 /* set to point at the root */
2157 fail[ q[ q_write++ ] ]=1;
2160 while ( q_read < q_write) {
2161 const U32 cur = q[ q_read++ % numstates ];
2162 base = trie->states[ cur ].trans.base;
2164 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2165 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2167 U32 fail_state = cur;
2170 fail_state = fail[ fail_state ];
2171 fail_base = aho->states[ fail_state ].trans.base;
2172 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2174 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2175 fail[ ch_state ] = fail_state;
2176 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2178 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2180 q[ q_write++ % numstates] = ch_state;
2184 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2185 when we fail in state 1, this allows us to use the
2186 charclass scan to find a valid start char. This is based on the principle
2187 that theres a good chance the string being searched contains lots of stuff
2188 that cant be a start char.
2190 fail[ 0 ] = fail[ 1 ] = 0;
2191 DEBUG_TRIE_COMPILE_r({
2192 PerlIO_printf(Perl_debug_log,
2193 "%*sStclass Failtable (%"UVuf" states): 0",
2194 (int)(depth * 2), "", (UV)numstates
2196 for( q_read=1; q_read<numstates; q_read++ ) {
2197 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2199 PerlIO_printf(Perl_debug_log, "\n");
2202 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2207 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2208 * These need to be revisited when a newer toolchain becomes available.
2210 #if defined(__sparc64__) && defined(__GNUC__)
2211 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2212 # undef SPARC64_GCC_WORKAROUND
2213 # define SPARC64_GCC_WORKAROUND 1
2217 #define DEBUG_PEEP(str,scan,depth) \
2218 DEBUG_OPTIMISE_r({if (scan){ \
2219 SV * const mysv=sv_newmortal(); \
2220 regnode *Next = regnext(scan); \
2221 regprop(RExC_rx, mysv, scan); \
2222 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2223 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2224 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2231 #define JOIN_EXACT(scan,min,flags) \
2232 if (PL_regkind[OP(scan)] == EXACT) \
2233 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2236 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2237 /* Merge several consecutive EXACTish nodes into one. */
2238 regnode *n = regnext(scan);
2240 regnode *next = scan + NODE_SZ_STR(scan);
2244 regnode *stop = scan;
2245 GET_RE_DEBUG_FLAGS_DECL;
2247 PERL_UNUSED_ARG(depth);
2249 #ifndef EXPERIMENTAL_INPLACESCAN
2250 PERL_UNUSED_ARG(flags);
2251 PERL_UNUSED_ARG(val);
2253 DEBUG_PEEP("join",scan,depth);
2255 /* Skip NOTHING, merge EXACT*. */
2257 ( PL_regkind[OP(n)] == NOTHING ||
2258 (stringok && (OP(n) == OP(scan))))
2260 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2262 if (OP(n) == TAIL || n > next)
2264 if (PL_regkind[OP(n)] == NOTHING) {
2265 DEBUG_PEEP("skip:",n,depth);
2266 NEXT_OFF(scan) += NEXT_OFF(n);
2267 next = n + NODE_STEP_REGNODE;
2274 else if (stringok) {
2275 const unsigned int oldl = STR_LEN(scan);
2276 regnode * const nnext = regnext(n);
2278 DEBUG_PEEP("merg",n,depth);
2281 if (oldl + STR_LEN(n) > U8_MAX)
2283 NEXT_OFF(scan) += NEXT_OFF(n);
2284 STR_LEN(scan) += STR_LEN(n);
2285 next = n + NODE_SZ_STR(n);
2286 /* Now we can overwrite *n : */
2287 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2295 #ifdef EXPERIMENTAL_INPLACESCAN
2296 if (flags && !NEXT_OFF(n)) {
2297 DEBUG_PEEP("atch", val, depth);
2298 if (reg_off_by_arg[OP(n)]) {
2299 ARG_SET(n, val - n);
2302 NEXT_OFF(n) = val - n;
2309 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2311 Two problematic code points in Unicode casefolding of EXACT nodes:
2313 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2314 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2320 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2321 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2323 This means that in case-insensitive matching (or "loose matching",
2324 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2325 length of the above casefolded versions) can match a target string
2326 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2327 This would rather mess up the minimum length computation.
2329 What we'll do is to look for the tail four bytes, and then peek
2330 at the preceding two bytes to see whether we need to decrease
2331 the minimum length by four (six minus two).
2333 Thanks to the design of UTF-8, there cannot be false matches:
2334 A sequence of valid UTF-8 bytes cannot be a subsequence of
2335 another valid sequence of UTF-8 bytes.
2338 char * const s0 = STRING(scan), *s, *t;
2339 char * const s1 = s0 + STR_LEN(scan) - 1;
2340 char * const s2 = s1 - 4;
2341 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2342 const char t0[] = "\xaf\x49\xaf\x42";
2344 const char t0[] = "\xcc\x88\xcc\x81";
2346 const char * const t1 = t0 + 3;
2349 s < s2 && (t = ninstr(s, s1, t0, t1));
2352 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2353 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2355 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2356 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2364 n = scan + NODE_SZ_STR(scan);
2366 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2373 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2377 /* REx optimizer. Converts nodes into quickier variants "in place".
2378 Finds fixed substrings. */
2380 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2381 to the position after last scanned or to NULL. */
2383 #define INIT_AND_WITHP \
2384 assert(!and_withp); \
2385 Newx(and_withp,1,struct regnode_charclass_class); \
2386 SAVEFREEPV(and_withp)
2388 /* this is a chain of data about sub patterns we are processing that
2389 need to be handled seperately/specially in study_chunk. Its so
2390 we can simulate recursion without losing state. */
2392 typedef struct scan_frame {
2393 regnode *last; /* last node to process in this frame */
2394 regnode *next; /* next node to process when last is reached */
2395 struct scan_frame *prev; /*previous frame*/
2396 I32 stop; /* what stopparen do we use */
2400 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2402 #define CASE_SYNST_FNC(nAmE) \
2404 if (flags & SCF_DO_STCLASS_AND) { \
2405 for (value = 0; value < 256; value++) \
2406 if (!is_ ## nAmE ## _cp(value)) \
2407 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2410 for (value = 0; value < 256; value++) \
2411 if (is_ ## nAmE ## _cp(value)) \
2412 ANYOF_BITMAP_SET(data->start_class, value); \
2416 if (flags & SCF_DO_STCLASS_AND) { \
2417 for (value = 0; value < 256; value++) \
2418 if (is_ ## nAmE ## _cp(value)) \
2419 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2422 for (value = 0; value < 256; value++) \
2423 if (!is_ ## nAmE ## _cp(value)) \
2424 ANYOF_BITMAP_SET(data->start_class, value); \
2431 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2432 I32 *minlenp, I32 *deltap,
2437 struct regnode_charclass_class *and_withp,
2438 U32 flags, U32 depth)
2439 /* scanp: Start here (read-write). */
2440 /* deltap: Write maxlen-minlen here. */
2441 /* last: Stop before this one. */
2442 /* data: string data about the pattern */
2443 /* stopparen: treat close N as END */
2444 /* recursed: which subroutines have we recursed into */
2445 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2448 I32 min = 0, pars = 0, code;
2449 regnode *scan = *scanp, *next;
2451 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2452 int is_inf_internal = 0; /* The studied chunk is infinite */
2453 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2454 scan_data_t data_fake;
2455 SV *re_trie_maxbuff = NULL;
2456 regnode *first_non_open = scan;
2457 I32 stopmin = I32_MAX;
2458 scan_frame *frame = NULL;
2460 GET_RE_DEBUG_FLAGS_DECL;
2463 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2467 while (first_non_open && OP(first_non_open) == OPEN)
2468 first_non_open=regnext(first_non_open);
2473 while ( scan && OP(scan) != END && scan < last ){
2474 /* Peephole optimizer: */
2475 DEBUG_STUDYDATA("Peep:", data,depth);
2476 DEBUG_PEEP("Peep",scan,depth);
2477 JOIN_EXACT(scan,&min,0);
2479 /* Follow the next-chain of the current node and optimize
2480 away all the NOTHINGs from it. */
2481 if (OP(scan) != CURLYX) {
2482 const int max = (reg_off_by_arg[OP(scan)]
2484 /* I32 may be smaller than U16 on CRAYs! */
2485 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2486 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2490 /* Skip NOTHING and LONGJMP. */
2491 while ((n = regnext(n))
2492 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2493 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2494 && off + noff < max)
2496 if (reg_off_by_arg[OP(scan)])
2499 NEXT_OFF(scan) = off;
2504 /* The principal pseudo-switch. Cannot be a switch, since we
2505 look into several different things. */
2506 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2507 || OP(scan) == IFTHEN) {
2508 next = regnext(scan);
2510 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2512 if (OP(next) == code || code == IFTHEN) {
2513 /* NOTE - There is similar code to this block below for handling
2514 TRIE nodes on a re-study. If you change stuff here check there
2516 I32 max1 = 0, min1 = I32_MAX, num = 0;
2517 struct regnode_charclass_class accum;
2518 regnode * const startbranch=scan;
2520 if (flags & SCF_DO_SUBSTR)
2521 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2522 if (flags & SCF_DO_STCLASS)
2523 cl_init_zero(pRExC_state, &accum);
2525 while (OP(scan) == code) {
2526 I32 deltanext, minnext, f = 0, fake;
2527 struct regnode_charclass_class this_class;
2530 data_fake.flags = 0;
2532 data_fake.whilem_c = data->whilem_c;
2533 data_fake.last_closep = data->last_closep;
2536 data_fake.last_closep = &fake;
2538 data_fake.pos_delta = delta;
2539 next = regnext(scan);
2540 scan = NEXTOPER(scan);
2542 scan = NEXTOPER(scan);
2543 if (flags & SCF_DO_STCLASS) {
2544 cl_init(pRExC_state, &this_class);
2545 data_fake.start_class = &this_class;
2546 f = SCF_DO_STCLASS_AND;
2548 if (flags & SCF_WHILEM_VISITED_POS)
2549 f |= SCF_WHILEM_VISITED_POS;
2551 /* we suppose the run is continuous, last=next...*/
2552 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2554 stopparen, recursed, NULL, f,depth+1);
2557 if (max1 < minnext + deltanext)
2558 max1 = minnext + deltanext;
2559 if (deltanext == I32_MAX)
2560 is_inf = is_inf_internal = 1;
2562 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2564 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2565 if ( stopmin > minnext)
2566 stopmin = min + min1;
2567 flags &= ~SCF_DO_SUBSTR;
2569 data->flags |= SCF_SEEN_ACCEPT;
2572 if (data_fake.flags & SF_HAS_EVAL)
2573 data->flags |= SF_HAS_EVAL;
2574 data->whilem_c = data_fake.whilem_c;
2576 if (flags & SCF_DO_STCLASS)
2577 cl_or(pRExC_state, &accum, &this_class);
2579 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2581 if (flags & SCF_DO_SUBSTR) {
2582 data->pos_min += min1;
2583 data->pos_delta += max1 - min1;
2584 if (max1 != min1 || is_inf)
2585 data->longest = &(data->longest_float);
2588 delta += max1 - min1;
2589 if (flags & SCF_DO_STCLASS_OR) {
2590 cl_or(pRExC_state, data->start_class, &accum);
2592 cl_and(data->start_class, and_withp);
2593 flags &= ~SCF_DO_STCLASS;
2596 else if (flags & SCF_DO_STCLASS_AND) {
2598 cl_and(data->start_class, &accum);
2599 flags &= ~SCF_DO_STCLASS;
2602 /* Switch to OR mode: cache the old value of
2603 * data->start_class */
2605 StructCopy(data->start_class, and_withp,
2606 struct regnode_charclass_class);
2607 flags &= ~SCF_DO_STCLASS_AND;
2608 StructCopy(&accum, data->start_class,
2609 struct regnode_charclass_class);
2610 flags |= SCF_DO_STCLASS_OR;
2611 data->start_class->flags |= ANYOF_EOS;
2615 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2618 Assuming this was/is a branch we are dealing with: 'scan' now
2619 points at the item that follows the branch sequence, whatever
2620 it is. We now start at the beginning of the sequence and look
2627 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2629 If we can find such a subseqence we need to turn the first
2630 element into a trie and then add the subsequent branch exact
2631 strings to the trie.
2635 1. patterns where the whole set of branch can be converted.
2637 2. patterns where only a subset can be converted.
2639 In case 1 we can replace the whole set with a single regop
2640 for the trie. In case 2 we need to keep the start and end
2643 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2644 becomes BRANCH TRIE; BRANCH X;
2646 There is an additional case, that being where there is a
2647 common prefix, which gets split out into an EXACT like node
2648 preceding the TRIE node.
2650 If x(1..n)==tail then we can do a simple trie, if not we make
2651 a "jump" trie, such that when we match the appropriate word
2652 we "jump" to the appopriate tail node. Essentailly we turn
2653 a nested if into a case structure of sorts.
2658 if (!re_trie_maxbuff) {
2659 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2660 if (!SvIOK(re_trie_maxbuff))
2661 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2663 if ( SvIV(re_trie_maxbuff)>=0 ) {
2665 regnode *first = (regnode *)NULL;
2666 regnode *last = (regnode *)NULL;
2667 regnode *tail = scan;
2672 SV * const mysv = sv_newmortal(); /* for dumping */
2674 /* var tail is used because there may be a TAIL
2675 regop in the way. Ie, the exacts will point to the
2676 thing following the TAIL, but the last branch will
2677 point at the TAIL. So we advance tail. If we
2678 have nested (?:) we may have to move through several
2682 while ( OP( tail ) == TAIL ) {
2683 /* this is the TAIL generated by (?:) */
2684 tail = regnext( tail );
2689 regprop(RExC_rx, mysv, tail );
2690 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2691 (int)depth * 2 + 2, "",
2692 "Looking for TRIE'able sequences. Tail node is: ",
2693 SvPV_nolen_const( mysv )
2699 step through the branches, cur represents each
2700 branch, noper is the first thing to be matched
2701 as part of that branch and noper_next is the
2702 regnext() of that node. if noper is an EXACT
2703 and noper_next is the same as scan (our current
2704 position in the regex) then the EXACT branch is
2705 a possible optimization target. Once we have
2706 two or more consequetive such branches we can
2707 create a trie of the EXACT's contents and stich
2708 it in place. If the sequence represents all of
2709 the branches we eliminate the whole thing and
2710 replace it with a single TRIE. If it is a
2711 subsequence then we need to stitch it in. This
2712 means the first branch has to remain, and needs
2713 to be repointed at the item on the branch chain
2714 following the last branch optimized. This could
2715 be either a BRANCH, in which case the
2716 subsequence is internal, or it could be the
2717 item following the branch sequence in which
2718 case the subsequence is at the end.
2722 /* dont use tail as the end marker for this traverse */
2723 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2724 regnode * const noper = NEXTOPER( cur );
2725 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2726 regnode * const noper_next = regnext( noper );
2730 regprop(RExC_rx, mysv, cur);
2731 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2732 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2734 regprop(RExC_rx, mysv, noper);
2735 PerlIO_printf( Perl_debug_log, " -> %s",
2736 SvPV_nolen_const(mysv));
2739 regprop(RExC_rx, mysv, noper_next );
2740 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2741 SvPV_nolen_const(mysv));
2743 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2744 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2746 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2747 : PL_regkind[ OP( noper ) ] == EXACT )
2748 || OP(noper) == NOTHING )
2750 && noper_next == tail
2755 if ( !first || optype == NOTHING ) {
2756 if (!first) first = cur;
2757 optype = OP( noper );
2763 make_trie( pRExC_state,
2764 startbranch, first, cur, tail, count,
2767 if ( PL_regkind[ OP( noper ) ] == EXACT
2769 && noper_next == tail
2774 optype = OP( noper );
2784 regprop(RExC_rx, mysv, cur);
2785 PerlIO_printf( Perl_debug_log,
2786 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2787 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2791 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2792 #ifdef TRIE_STUDY_OPT
2793 if ( ((made == MADE_EXACT_TRIE &&
2794 startbranch == first)
2795 || ( first_non_open == first )) &&
2797 flags |= SCF_TRIE_RESTUDY;
2798 if ( startbranch == first
2801 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2811 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2812 scan = NEXTOPER(NEXTOPER(scan));
2813 } else /* single branch is optimized. */
2814 scan = NEXTOPER(scan);
2816 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2817 scan_frame *newframe = NULL;
2822 if (OP(scan) != SUSPEND) {
2823 /* set the pointer */
2824 if (OP(scan) == GOSUB) {
2826 RExC_recurse[ARG2L(scan)] = scan;
2827 start = RExC_open_parens[paren-1];
2828 end = RExC_close_parens[paren-1];
2831 start = RExC_rxi->program + 1;
2835 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2836 SAVEFREEPV(recursed);
2838 if (!PAREN_TEST(recursed,paren+1)) {
2839 PAREN_SET(recursed,paren+1);
2840 Newx(newframe,1,scan_frame);
2842 if (flags & SCF_DO_SUBSTR) {
2843 SCAN_COMMIT(pRExC_state,data,minlenp);
2844 data->longest = &(data->longest_float);
2846 is_inf = is_inf_internal = 1;
2847 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2848 cl_anything(pRExC_state, data->start_class);
2849 flags &= ~SCF_DO_STCLASS;
2852 Newx(newframe,1,scan_frame);
2855 end = regnext(scan);
2860 SAVEFREEPV(newframe);
2861 newframe->next = regnext(scan);
2862 newframe->last = last;
2863 newframe->stop = stopparen;
2864 newframe->prev = frame;
2874 else if (OP(scan) == EXACT) {
2875 I32 l = STR_LEN(scan);
2878 const U8 * const s = (U8*)STRING(scan);
2879 l = utf8_length(s, s + l);
2880 uc = utf8_to_uvchr(s, NULL);
2882 uc = *((U8*)STRING(scan));
2885 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2886 /* The code below prefers earlier match for fixed
2887 offset, later match for variable offset. */
2888 if (data->last_end == -1) { /* Update the start info. */
2889 data->last_start_min = data->pos_min;
2890 data->last_start_max = is_inf
2891 ? I32_MAX : data->pos_min + data->pos_delta;
2893 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2895 SvUTF8_on(data->last_found);
2897 SV * const sv = data->last_found;
2898 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2899 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2900 if (mg && mg->mg_len >= 0)
2901 mg->mg_len += utf8_length((U8*)STRING(scan),
2902 (U8*)STRING(scan)+STR_LEN(scan));
2904 data->last_end = data->pos_min + l;
2905 data->pos_min += l; /* As in the first entry. */
2906 data->flags &= ~SF_BEFORE_EOL;
2908 if (flags & SCF_DO_STCLASS_AND) {
2909 /* Check whether it is compatible with what we know already! */
2913 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2914 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2915 && (!(data->start_class->flags & ANYOF_FOLD)
2916 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2919 ANYOF_CLASS_ZERO(data->start_class);
2920 ANYOF_BITMAP_ZERO(data->start_class);
2922 ANYOF_BITMAP_SET(data->start_class, uc);
2923 data->start_class->flags &= ~ANYOF_EOS;
2925 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2927 else if (flags & SCF_DO_STCLASS_OR) {
2928 /* false positive possible if the class is case-folded */
2930 ANYOF_BITMAP_SET(data->start_class, uc);
2932 data->start_class->flags |= ANYOF_UNICODE_ALL;
2933 data->start_class->flags &= ~ANYOF_EOS;
2934 cl_and(data->start_class, and_withp);
2936 flags &= ~SCF_DO_STCLASS;
2938 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2939 I32 l = STR_LEN(scan);
2940 UV uc = *((U8*)STRING(scan));
2942 /* Search for fixed substrings supports EXACT only. */
2943 if (flags & SCF_DO_SUBSTR) {
2945 SCAN_COMMIT(pRExC_state, data, minlenp);
2948 const U8 * const s = (U8 *)STRING(scan);
2949 l = utf8_length(s, s + l);
2950 uc = utf8_to_uvchr(s, NULL);
2953 if (flags & SCF_DO_SUBSTR)
2955 if (flags & SCF_DO_STCLASS_AND) {
2956 /* Check whether it is compatible with what we know already! */
2960 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2961 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2962 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2964 ANYOF_CLASS_ZERO(data->start_class);
2965 ANYOF_BITMAP_ZERO(data->start_class);
2967 ANYOF_BITMAP_SET(data->start_class, uc);
2968 data->start_class->flags &= ~ANYOF_EOS;
2969 data->start_class->flags |= ANYOF_FOLD;
2970 if (OP(scan) == EXACTFL)
2971 data->start_class->flags |= ANYOF_LOCALE;
2974 else if (flags & SCF_DO_STCLASS_OR) {
2975 if (data->start_class->flags & ANYOF_FOLD) {
2976 /* false positive possible if the class is case-folded.
2977 Assume that the locale settings are the same... */
2979 ANYOF_BITMAP_SET(data->start_class, uc);
2980 data->start_class->flags &= ~ANYOF_EOS;
2982 cl_and(data->start_class, and_withp);
2984 flags &= ~SCF_DO_STCLASS;
2986 else if (strchr((const char*)PL_varies,OP(scan))) {
2987 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2988 I32 f = flags, pos_before = 0;
2989 regnode * const oscan = scan;
2990 struct regnode_charclass_class this_class;
2991 struct regnode_charclass_class *oclass = NULL;
2992 I32 next_is_eval = 0;
2994 switch (PL_regkind[OP(scan)]) {
2995 case WHILEM: /* End of (?:...)* . */
2996 scan = NEXTOPER(scan);
2999 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3000 next = NEXTOPER(scan);
3001 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3003 maxcount = REG_INFTY;
3004 next = regnext(scan);
3005 scan = NEXTOPER(scan);
3009 if (flags & SCF_DO_SUBSTR)
3014 if (flags & SCF_DO_STCLASS) {
3016 maxcount = REG_INFTY;
3017 next = regnext(scan);
3018 scan = NEXTOPER(scan);
3021 is_inf = is_inf_internal = 1;
3022 scan = regnext(scan);
3023 if (flags & SCF_DO_SUBSTR) {
3024 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3025 data->longest = &(data->longest_float);
3027 goto optimize_curly_tail;
3029 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3030 && (scan->flags == stopparen))
3035 mincount = ARG1(scan);
3036 maxcount = ARG2(scan);
3038 next = regnext(scan);
3039 if (OP(scan) == CURLYX) {
3040 I32 lp = (data ? *(data->last_closep) : 0);
3041 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3043 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3044 next_is_eval = (OP(scan) == EVAL);
3046 if (flags & SCF_DO_SUBSTR) {
3047 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3048 pos_before = data->pos_min;
3052 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3054 data->flags |= SF_IS_INF;
3056 if (flags & SCF_DO_STCLASS) {
3057 cl_init(pRExC_state, &this_class);
3058 oclass = data->start_class;
3059 data->start_class = &this_class;
3060 f |= SCF_DO_STCLASS_AND;
3061 f &= ~SCF_DO_STCLASS_OR;
3063 /* These are the cases when once a subexpression
3064 fails at a particular position, it cannot succeed
3065 even after backtracking at the enclosing scope.
3067 XXXX what if minimal match and we are at the
3068 initial run of {n,m}? */
3069 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3070 f &= ~SCF_WHILEM_VISITED_POS;
3072 /* This will finish on WHILEM, setting scan, or on NULL: */
3073 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3074 last, data, stopparen, recursed, NULL,
3076 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3078 if (flags & SCF_DO_STCLASS)
3079 data->start_class = oclass;
3080 if (mincount == 0 || minnext == 0) {
3081 if (flags & SCF_DO_STCLASS_OR) {
3082 cl_or(pRExC_state, data->start_class, &this_class);
3084 else if (flags & SCF_DO_STCLASS_AND) {
3085 /* Switch to OR mode: cache the old value of
3086 * data->start_class */
3088 StructCopy(data->start_class, and_withp,
3089 struct regnode_charclass_class);
3090 flags &= ~SCF_DO_STCLASS_AND;
3091 StructCopy(&this_class, data->start_class,
3092 struct regnode_charclass_class);
3093 flags |= SCF_DO_STCLASS_OR;
3094 data->start_class->flags |= ANYOF_EOS;
3096 } else { /* Non-zero len */
3097 if (flags & SCF_DO_STCLASS_OR) {
3098 cl_or(pRExC_state, data->start_class, &this_class);
3099 cl_and(data->start_class, and_withp);
3101 else if (flags & SCF_DO_STCLASS_AND)
3102 cl_and(data->start_class, &this_class);
3103 flags &= ~SCF_DO_STCLASS;
3105 if (!scan) /* It was not CURLYX, but CURLY. */
3107 if ( /* ? quantifier ok, except for (?{ ... }) */
3108 (next_is_eval || !(mincount == 0 && maxcount == 1))
3109 && (minnext == 0) && (deltanext == 0)
3110 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3111 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3112 && ckWARN(WARN_REGEXP))
3115 "Quantifier unexpected on zero-length expression");
3118 min += minnext * mincount;
3119 is_inf_internal |= ((maxcount == REG_INFTY
3120 && (minnext + deltanext) > 0)
3121 || deltanext == I32_MAX);
3122 is_inf |= is_inf_internal;
3123 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3125 /* Try powerful optimization CURLYX => CURLYN. */
3126 if ( OP(oscan) == CURLYX && data
3127 && data->flags & SF_IN_PAR
3128 && !(data->flags & SF_HAS_EVAL)
3129 && !deltanext && minnext == 1 ) {
3130 /* Try to optimize to CURLYN. */
3131 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3132 regnode * const nxt1 = nxt;
3139 if (!strchr((const char*)PL_simple,OP(nxt))
3140 && !(PL_regkind[OP(nxt)] == EXACT
3141 && STR_LEN(nxt) == 1))
3147 if (OP(nxt) != CLOSE)
3149 if (RExC_open_parens) {
3150 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3151 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3153 /* Now we know that nxt2 is the only contents: */
3154 oscan->flags = (U8)ARG(nxt);
3156 OP(nxt1) = NOTHING; /* was OPEN. */
3159 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3160 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3161 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3162 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3163 OP(nxt + 1) = OPTIMIZED; /* was count. */
3164 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3169 /* Try optimization CURLYX => CURLYM. */
3170 if ( OP(oscan) == CURLYX && data
3171 && !(data->flags & SF_HAS_PAR)
3172 && !(data->flags & SF_HAS_EVAL)
3173 && !deltanext /* atom is fixed width */
3174 && minnext != 0 /* CURLYM can't handle zero width */
3176 /* XXXX How to optimize if data == 0? */
3177 /* Optimize to a simpler form. */
3178 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3182 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3183 && (OP(nxt2) != WHILEM))
3185 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3186 /* Need to optimize away parenths. */
3187 if (data->flags & SF_IN_PAR) {
3188 /* Set the parenth number. */
3189 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3191 if (OP(nxt) != CLOSE)
3192 FAIL("Panic opt close");
3193 oscan->flags = (U8)ARG(nxt);
3194 if (RExC_open_parens) {
3195 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3196 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3198 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3199 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3202 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3203 OP(nxt + 1) = OPTIMIZED; /* was count. */
3204 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3205 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3208 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3209 regnode *nnxt = regnext(nxt1);
3212 if (reg_off_by_arg[OP(nxt1)])
3213 ARG_SET(nxt1, nxt2 - nxt1);
3214 else if (nxt2 - nxt1 < U16_MAX)
3215 NEXT_OFF(nxt1) = nxt2 - nxt1;
3217 OP(nxt) = NOTHING; /* Cannot beautify */
3222 /* Optimize again: */
3223 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3224 NULL, stopparen, recursed, NULL, 0,depth+1);
3229 else if ((OP(oscan) == CURLYX)
3230 && (flags & SCF_WHILEM_VISITED_POS)
3231 /* See the comment on a similar expression above.
3232 However, this time it not a subexpression
3233 we care about, but the expression itself. */
3234 && (maxcount == REG_INFTY)
3235 && data && ++data->whilem_c < 16) {
3236 /* This stays as CURLYX, we can put the count/of pair. */
3237 /* Find WHILEM (as in regexec.c) */
3238 regnode *nxt = oscan + NEXT_OFF(oscan);
3240 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3242 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3243 | (RExC_whilem_seen << 4)); /* On WHILEM */
3245 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3247 if (flags & SCF_DO_SUBSTR) {
3248 SV *last_str = NULL;
3249 int counted = mincount != 0;
3251 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3252 #if defined(SPARC64_GCC_WORKAROUND)
3255 const char *s = NULL;
3258 if (pos_before >= data->last_start_min)
3261 b = data->last_start_min;
3264 s = SvPV_const(data->last_found, l);
3265 old = b - data->last_start_min;
3268 I32 b = pos_before >= data->last_start_min
3269 ? pos_before : data->last_start_min;
3271 const char * const s = SvPV_const(data->last_found, l);
3272 I32 old = b - data->last_start_min;
3276 old = utf8_hop((U8*)s, old) - (U8*)s;
3279 /* Get the added string: */
3280 last_str = newSVpvn(s + old, l);
3282 SvUTF8_on(last_str);
3283 if (deltanext == 0 && pos_before == b) {
3284 /* What was added is a constant string */
3286 SvGROW(last_str, (mincount * l) + 1);
3287 repeatcpy(SvPVX(last_str) + l,
3288 SvPVX_const(last_str), l, mincount - 1);
3289 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3290 /* Add additional parts. */
3291 SvCUR_set(data->last_found,
3292 SvCUR(data->last_found) - l);
3293 sv_catsv(data->last_found, last_str);
3295 SV * sv = data->last_found;
3297 SvUTF8(sv) && SvMAGICAL(sv) ?
3298 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3299 if (mg && mg->mg_len >= 0)
3300 mg->mg_len += CHR_SVLEN(last_str);
3302 data->last_end += l * (mincount - 1);
3305 /* start offset must point into the last copy */
3306 data->last_start_min += minnext * (mincount - 1);
3307 data->last_start_max += is_inf ? I32_MAX
3308 : (maxcount - 1) * (minnext + data->pos_delta);
3311 /* It is counted once already... */
3312 data->pos_min += minnext * (mincount - counted);
3313 data->pos_delta += - counted * deltanext +
3314 (minnext + deltanext) * maxcount - minnext * mincount;
3315 if (mincount != maxcount) {
3316 /* Cannot extend fixed substrings found inside
3318 SCAN_COMMIT(pRExC_state,data,minlenp);
3319 if (mincount && last_str) {
3320 SV * const sv = data->last_found;
3321 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3322 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3326 sv_setsv(sv, last_str);
3327 data->last_end = data->pos_min;
3328 data->last_start_min =
3329 data->pos_min - CHR_SVLEN(last_str);
3330 data->last_start_max = is_inf
3332 : data->pos_min + data->pos_delta
3333 - CHR_SVLEN(last_str);
3335 data->longest = &(data->longest_float);
3337 SvREFCNT_dec(last_str);
3339 if (data && (fl & SF_HAS_EVAL))
3340 data->flags |= SF_HAS_EVAL;
3341 optimize_curly_tail:
3342 if (OP(oscan) != CURLYX) {
3343 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3345 NEXT_OFF(oscan) += NEXT_OFF(next);
3348 default: /* REF and CLUMP only? */
3349 if (flags & SCF_DO_SUBSTR) {
3350 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3351 data->longest = &(data->longest_float);
3353 is_inf = is_inf_internal = 1;
3354 if (flags & SCF_DO_STCLASS_OR)
3355 cl_anything(pRExC_state, data->start_class);
3356 flags &= ~SCF_DO_STCLASS;
3360 else if (OP(scan) == LNBREAK) {
3361 if (flags & SCF_DO_STCLASS) {
3363 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3364 if (flags & SCF_DO_STCLASS_AND) {
3365 for (value = 0; value < 256; value++)
3366 if (!is_VERTWS_cp(value))
3367 ANYOF_BITMAP_CLEAR(data->start_class, value);
3370 for (value = 0; value < 256; value++)
3371 if (is_VERTWS_cp(value))
3372 ANYOF_BITMAP_SET(data->start_class, value);
3374 if (flags & SCF_DO_STCLASS_OR)
3375 cl_and(data->start_class, and_withp);
3376 flags &= ~SCF_DO_STCLASS;
3380 if (flags & SCF_DO_SUBSTR) {
3381 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3383 data->pos_delta += 1;
3384 data->longest = &(data->longest_float);
3388 else if (OP(scan) == FOLDCHAR) {
3389 int d = ARG(scan)==0xDF ? 1 : 2;
3390 flags &= ~SCF_DO_STCLASS;
3393 if (flags & SCF_DO_SUBSTR) {
3394 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3396 data->pos_delta += d;
3397 data->longest = &(data->longest_float);
3400 else if (strchr((const char*)PL_simple,OP(scan))) {
3403 if (flags & SCF_DO_SUBSTR) {
3404 SCAN_COMMIT(pRExC_state,data,minlenp);
3408 if (flags & SCF_DO_STCLASS) {
3409 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3411 /* Some of the logic below assumes that switching
3412 locale on will only add false positives. */
3413 switch (PL_regkind[OP(scan)]) {
3417 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3418 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3419 cl_anything(pRExC_state, data->start_class);
3422 if (OP(scan) == SANY)
3424 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3425 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3426 || (data->start_class->flags & ANYOF_CLASS));
3427 cl_anything(pRExC_state, data->start_class);
3429 if (flags & SCF_DO_STCLASS_AND || !value)
3430 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3433 if (flags & SCF_DO_STCLASS_AND)
3434 cl_and(data->start_class,
3435 (struct regnode_charclass_class*)scan);
3437 cl_or(pRExC_state, data->start_class,
3438 (struct regnode_charclass_class*)scan);
3441 if (flags & SCF_DO_STCLASS_AND) {
3442 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3443 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3444 for (value = 0; value < 256; value++)
3445 if (!isALNUM(value))
3446 ANYOF_BITMAP_CLEAR(data->start_class, value);
3450 if (data->start_class->flags & ANYOF_LOCALE)
3451 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3453 for (value = 0; value < 256; value++)
3455 ANYOF_BITMAP_SET(data->start_class, value);
3460 if (flags & SCF_DO_STCLASS_AND) {
3461 if (data->start_class->flags & ANYOF_LOCALE)
3462 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3465 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3466 data->start_class->flags |= ANYOF_LOCALE;
3470 if (flags & SCF_DO_STCLASS_AND) {
3471 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3472 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3473 for (value = 0; value < 256; value++)
3475 ANYOF_BITMAP_CLEAR(data->start_class, value);
3479 if (data->start_class->flags & ANYOF_LOCALE)
3480 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3482 for (value = 0; value < 256; value++)
3483 if (!isALNUM(value))
3484 ANYOF_BITMAP_SET(data->start_class, value);
3489 if (flags & SCF_DO_STCLASS_AND) {
3490 if (data->start_class->flags & ANYOF_LOCALE)
3491 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3494 data->start_class->flags |= ANYOF_LOCALE;
3495 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3499 if (flags & SCF_DO_STCLASS_AND) {
3500 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3501 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3502 for (value = 0; value < 256; value++)
3503 if (!isSPACE(value))
3504 ANYOF_BITMAP_CLEAR(data->start_class, value);
3508 if (data->start_class->flags & ANYOF_LOCALE)
3509 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3511 for (value = 0; value < 256; value++)
3513 ANYOF_BITMAP_SET(data->start_class, value);
3518 if (flags & SCF_DO_STCLASS_AND) {
3519 if (data->start_class->flags & ANYOF_LOCALE)
3520 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3523 data->start_class->flags |= ANYOF_LOCALE;
3524 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3528 if (flags & SCF_DO_STCLASS_AND) {
3529 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3530 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3531 for (value = 0; value < 256; value++)
3533 ANYOF_BITMAP_CLEAR(data->start_class, value);
3537 if (data->start_class->flags & ANYOF_LOCALE)
3538 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3540 for (value = 0; value < 256; value++)
3541 if (!isSPACE(value))
3542 ANYOF_BITMAP_SET(data->start_class, value);
3547 if (flags & SCF_DO_STCLASS_AND) {
3548 if (data->start_class->flags & ANYOF_LOCALE) {
3549 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3550 for (value = 0; value < 256; value++)
3551 if (!isSPACE(value))
3552 ANYOF_BITMAP_CLEAR(data->start_class, value);
3556 data->start_class->flags |= ANYOF_LOCALE;
3557 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3561 if (flags & SCF_DO_STCLASS_AND) {
3562 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3563 for (value = 0; value < 256; value++)
3564 if (!isDIGIT(value))
3565 ANYOF_BITMAP_CLEAR(data->start_class, value);
3568 if (data->start_class->flags & ANYOF_LOCALE)
3569 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3571 for (value = 0; value < 256; value++)
3573 ANYOF_BITMAP_SET(data->start_class, value);
3578 if (flags & SCF_DO_STCLASS_AND) {
3579 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3580 for (value = 0; value < 256; value++)
3582 ANYOF_BITMAP_CLEAR(data->start_class, value);
3585 if (data->start_class->flags & ANYOF_LOCALE)
3586 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3588 for (value = 0; value < 256; value++)
3589 if (!isDIGIT(value))
3590 ANYOF_BITMAP_SET(data->start_class, value);
3594 CASE_SYNST_FNC(VERTWS);
3595 CASE_SYNST_FNC(HORIZWS);
3598 if (flags & SCF_DO_STCLASS_OR)
3599 cl_and(data->start_class, and_withp);
3600 flags &= ~SCF_DO_STCLASS;
3603 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3604 data->flags |= (OP(scan) == MEOL
3608 else if ( PL_regkind[OP(scan)] == BRANCHJ
3609 /* Lookbehind, or need to calculate parens/evals/stclass: */
3610 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3611 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3612 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3613 || OP(scan) == UNLESSM )
3615 /* Negative Lookahead/lookbehind
3616 In this case we can't do fixed string optimisation.
3619 I32 deltanext, minnext, fake = 0;
3621 struct regnode_charclass_class intrnl;
3624 data_fake.flags = 0;
3626 data_fake.whilem_c = data->whilem_c;
3627 data_fake.last_closep = data->last_closep;
3630 data_fake.last_closep = &fake;
3631 data_fake.pos_delta = delta;
3632 if ( flags & SCF_DO_STCLASS && !scan->flags
3633 && OP(scan) == IFMATCH ) { /* Lookahead */
3634 cl_init(pRExC_state, &intrnl);
3635 data_fake.start_class = &intrnl;
3636 f |= SCF_DO_STCLASS_AND;
3638 if (flags & SCF_WHILEM_VISITED_POS)
3639 f |= SCF_WHILEM_VISITED_POS;
3640 next = regnext(scan);
3641 nscan = NEXTOPER(NEXTOPER(scan));
3642 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3643 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3646 FAIL("Variable length lookbehind not implemented");
3648 else if (minnext > (I32)U8_MAX) {
3649 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3651 scan->flags = (U8)minnext;
3654 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3656 if (data_fake.flags & SF_HAS_EVAL)
3657 data->flags |= SF_HAS_EVAL;
3658 data->whilem_c = data_fake.whilem_c;
3660 if (f & SCF_DO_STCLASS_AND) {
3661 const int was = (data->start_class->flags & ANYOF_EOS);
3663 cl_and(data->start_class, &intrnl);
3665 data->start_class->flags |= ANYOF_EOS;
3668 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3670 /* Positive Lookahead/lookbehind
3671 In this case we can do fixed string optimisation,
3672 but we must be careful about it. Note in the case of
3673 lookbehind the positions will be offset by the minimum
3674 length of the pattern, something we won't know about
3675 until after the recurse.
3677 I32 deltanext, fake = 0;
3679 struct regnode_charclass_class intrnl;
3681 /* We use SAVEFREEPV so that when the full compile
3682 is finished perl will clean up the allocated
3683 minlens when its all done. This was we don't
3684 have to worry about freeing them when we know
3685 they wont be used, which would be a pain.
3688 Newx( minnextp, 1, I32 );
3689 SAVEFREEPV(minnextp);
3692 StructCopy(data, &data_fake, scan_data_t);
3693 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3696 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3697 data_fake.last_found=newSVsv(data->last_found);
3701 data_fake.last_closep = &fake;
3702 data_fake.flags = 0;
3703 data_fake.pos_delta = delta;
3705 data_fake.flags |= SF_IS_INF;
3706 if ( flags & SCF_DO_STCLASS && !scan->flags
3707 && OP(scan) == IFMATCH ) { /* Lookahead */
3708 cl_init(pRExC_state, &intrnl);
3709 data_fake.start_class = &intrnl;
3710 f |= SCF_DO_STCLASS_AND;
3712 if (flags & SCF_WHILEM_VISITED_POS)
3713 f |= SCF_WHILEM_VISITED_POS;
3714 next = regnext(scan);
3715 nscan = NEXTOPER(NEXTOPER(scan));
3717 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3718 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3721 FAIL("Variable length lookbehind not implemented");
3723 else if (*minnextp > (I32)U8_MAX) {
3724 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3726 scan->flags = (U8)*minnextp;
3731 if (f & SCF_DO_STCLASS_AND) {
3732 const int was = (data->start_class->flags & ANYOF_EOS);
3734 cl_and(data->start_class, &intrnl);
3736 data->start_class->flags |= ANYOF_EOS;
3739 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3741 if (data_fake.flags & SF_HAS_EVAL)
3742 data->flags |= SF_HAS_EVAL;
3743 data->whilem_c = data_fake.whilem_c;
3744 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3745 if (RExC_rx->minlen<*minnextp)
3746 RExC_rx->minlen=*minnextp;
3747 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3748 SvREFCNT_dec(data_fake.last_found);
3750 if ( data_fake.minlen_fixed != minlenp )
3752 data->offset_fixed= data_fake.offset_fixed;
3753 data->minlen_fixed= data_fake.minlen_fixed;
3754 data->lookbehind_fixed+= scan->flags;
3756 if ( data_fake.minlen_float != minlenp )
3758 data->minlen_float= data_fake.minlen_float;
3759 data->offset_float_min=data_fake.offset_float_min;
3760 data->offset_float_max=data_fake.offset_float_max;
3761 data->lookbehind_float+= scan->flags;
3770 else if (OP(scan) == OPEN) {
3771 if (stopparen != (I32)ARG(scan))
3774 else if (OP(scan) == CLOSE) {
3775 if (stopparen == (I32)ARG(scan)) {
3778 if ((I32)ARG(scan) == is_par) {
3779 next = regnext(scan);
3781 if ( next && (OP(next) != WHILEM) && next < last)
3782 is_par = 0; /* Disable optimization */
3785 *(data->last_closep) = ARG(scan);
3787 else if (OP(scan) == EVAL) {
3789 data->flags |= SF_HAS_EVAL;
3791 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3792 if (flags & SCF_DO_SUBSTR) {
3793 SCAN_COMMIT(pRExC_state,data,minlenp);
3794 flags &= ~SCF_DO_SUBSTR;
3796 if (data && OP(scan)==ACCEPT) {
3797 data->flags |= SCF_SEEN_ACCEPT;
3802 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3804 if (flags & SCF_DO_SUBSTR) {
3805 SCAN_COMMIT(pRExC_state,data,minlenp);
3806 data->longest = &(data->longest_float);
3808 is_inf = is_inf_internal = 1;
3809 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3810 cl_anything(pRExC_state, data->start_class);
3811 flags &= ~SCF_DO_STCLASS;
3813 else if (OP(scan) == GPOS) {
3814 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3815 !(delta || is_inf || (data && data->pos_delta)))
3817 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3818 RExC_rx->extflags |= RXf_ANCH_GPOS;
3819 if (RExC_rx->gofs < (U32)min)
3820 RExC_rx->gofs = min;
3822 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3826 #ifdef TRIE_STUDY_OPT
3827 #ifdef FULL_TRIE_STUDY
3828 else if (PL_regkind[OP(scan)] == TRIE) {
3829 /* NOTE - There is similar code to this block above for handling
3830 BRANCH nodes on the initial study. If you change stuff here
3832 regnode *trie_node= scan;
3833 regnode *tail= regnext(scan);
3834 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3835 I32 max1 = 0, min1 = I32_MAX;
3836 struct regnode_charclass_class accum;
3838 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3839 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3840 if (flags & SCF_DO_STCLASS)
3841 cl_init_zero(pRExC_state, &accum);
3847 const regnode *nextbranch= NULL;
3850 for ( word=1 ; word <= trie->wordcount ; word++)
3852 I32 deltanext=0, minnext=0, f = 0, fake;
3853 struct regnode_charclass_class this_class;
3855 data_fake.flags = 0;
3857 data_fake.whilem_c = data->whilem_c;
3858 data_fake.last_closep = data->last_closep;
3861 data_fake.last_closep = &fake;
3862 data_fake.pos_delta = delta;
3863 if (flags & SCF_DO_STCLASS) {
3864 cl_init(pRExC_state, &this_class);
3865 data_fake.start_class = &this_class;
3866 f = SCF_DO_STCLASS_AND;
3868 if (flags & SCF_WHILEM_VISITED_POS)
3869 f |= SCF_WHILEM_VISITED_POS;
3871 if (trie->jump[word]) {
3873 nextbranch = trie_node + trie->jump[0];
3874 scan= trie_node + trie->jump[word];
3875 /* We go from the jump point to the branch that follows
3876 it. Note this means we need the vestigal unused branches
3877 even though they arent otherwise used.
3879 minnext = study_chunk(pRExC_state, &scan, minlenp,
3880 &deltanext, (regnode *)nextbranch, &data_fake,
3881 stopparen, recursed, NULL, f,depth+1);
3883 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3884 nextbranch= regnext((regnode*)nextbranch);
3886 if (min1 > (I32)(minnext + trie->minlen))
3887 min1 = minnext + trie->minlen;
3888 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3889 max1 = minnext + deltanext + trie->maxlen;
3890 if (deltanext == I32_MAX)
3891 is_inf = is_inf_internal = 1;
3893 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3895 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3896 if ( stopmin > min + min1)
3897 stopmin = min + min1;
3898 flags &= ~SCF_DO_SUBSTR;
3900 data->flags |= SCF_SEEN_ACCEPT;
3903 if (data_fake.flags & SF_HAS_EVAL)
3904 data->flags |= SF_HAS_EVAL;
3905 data->whilem_c = data_fake.whilem_c;
3907 if (flags & SCF_DO_STCLASS)
3908 cl_or(pRExC_state, &accum, &this_class);
3911 if (flags & SCF_DO_SUBSTR) {
3912 data->pos_min += min1;
3913 data->pos_delta += max1 - min1;
3914 if (max1 != min1 || is_inf)
3915 data->longest = &(data->longest_float);
3918 delta += max1 - min1;
3919 if (flags & SCF_DO_STCLASS_OR) {
3920 cl_or(pRExC_state, data->start_class, &accum);
3922 cl_and(data->start_class, and_withp);
3923 flags &= ~SCF_DO_STCLASS;
3926 else if (flags & SCF_DO_STCLASS_AND) {
3928 cl_and(data->start_class, &accum);
3929 flags &= ~SCF_DO_STCLASS;
3932 /* Switch to OR mode: cache the old value of
3933 * data->start_class */
3935 StructCopy(data->start_class, and_withp,
3936 struct regnode_charclass_class);
3937 flags &= ~SCF_DO_STCLASS_AND;
3938 StructCopy(&accum, data->start_class,
3939 struct regnode_charclass_class);
3940 flags |= SCF_DO_STCLASS_OR;
3941 data->start_class->flags |= ANYOF_EOS;
3948 else if (PL_regkind[OP(scan)] == TRIE) {
3949 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3952 min += trie->minlen;
3953 delta += (trie->maxlen - trie->minlen);
3954 flags &= ~SCF_DO_STCLASS; /* xxx */
3955 if (flags & SCF_DO_SUBSTR) {
3956 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3957 data->pos_min += trie->minlen;
3958 data->pos_delta += (trie->maxlen - trie->minlen);
3959 if (trie->maxlen != trie->minlen)
3960 data->longest = &(data->longest_float);
3962 if (trie->jump) /* no more substrings -- for now /grr*/
3963 flags &= ~SCF_DO_SUBSTR;
3965 #endif /* old or new */
3966 #endif /* TRIE_STUDY_OPT */
3968 /* Else: zero-length, ignore. */
3969 scan = regnext(scan);
3974 stopparen = frame->stop;
3975 frame = frame->prev;
3976 goto fake_study_recurse;
3981 DEBUG_STUDYDATA("pre-fin:",data,depth);
3984 *deltap = is_inf_internal ? I32_MAX : delta;
3985 if (flags & SCF_DO_SUBSTR && is_inf)
3986 data->pos_delta = I32_MAX - data->pos_min;
3987 if (is_par > (I32)U8_MAX)
3989 if (is_par && pars==1 && data) {
3990 data->flags |= SF_IN_PAR;
3991 data->flags &= ~SF_HAS_PAR;
3993 else if (pars && data) {
3994 data->flags |= SF_HAS_PAR;
3995 data->flags &= ~SF_IN_PAR;
3997 if (flags & SCF_DO_STCLASS_OR)
3998 cl_and(data->start_class, and_withp);
3999 if (flags & SCF_TRIE_RESTUDY)
4000 data->flags |= SCF_TRIE_RESTUDY;
4002 DEBUG_STUDYDATA("post-fin:",data,depth);
4004 return min < stopmin ? min : stopmin;
4008 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4010 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4012 Renewc(RExC_rxi->data,
4013 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4014 char, struct reg_data);
4016 Renew(RExC_rxi->data->what, count + n, U8);
4018 Newx(RExC_rxi->data->what, n, U8);
4019 RExC_rxi->data->count = count + n;
4020 Copy(s, RExC_rxi->data->what + count, n, U8);
4024 /*XXX: todo make this not included in a non debugging perl */
4025 #ifndef PERL_IN_XSUB_RE
4027 Perl_reginitcolors(pTHX)
4030 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4032 char *t = savepv(s);
4036 t = strchr(t, '\t');
4042 PL_colors[i] = t = (char *)"";
4047 PL_colors[i++] = (char *)"";
4054 #ifdef TRIE_STUDY_OPT
4055 #define CHECK_RESTUDY_GOTO \
4057 (data.flags & SCF_TRIE_RESTUDY) \
4061 #define CHECK_RESTUDY_GOTO
4065 - pregcomp - compile a regular expression into internal code
4067 * We can't allocate space until we know how big the compiled form will be,
4068 * but we can't compile it (and thus know how big it is) until we've got a
4069 * place to put the code. So we cheat: we compile it twice, once with code
4070 * generation turned off and size counting turned on, and once "for real".
4071 * This also means that we don't allocate space until we are sure that the
4072 * thing really will compile successfully, and we never have to move the
4073 * code and thus invalidate pointers into it. (Note that it has to be in
4074 * one piece because free() must be able to free it all.) [NB: not true in perl]
4076 * Beware that the optimization-preparation code in here knows about some
4077 * of the structure of the compiled regexp. [I'll say.]
4082 #ifndef PERL_IN_XSUB_RE
4083 #define RE_ENGINE_PTR &PL_core_reg_engine
4085 extern const struct regexp_engine my_reg_engine;
4086 #define RE_ENGINE_PTR &my_reg_engine
4089 #ifndef PERL_IN_XSUB_RE
4091 Perl_pregcomp(pTHX_ const SV * const pattern, const U32 flags)
4094 HV * const table = GvHV(PL_hintgv);
4095 /* Dispatch a request to compile a regexp to correct
4098 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4099 GET_RE_DEBUG_FLAGS_DECL;
4100 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4101 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4103 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4106 return CALLREGCOMP_ENG(eng, pattern, flags);
4109 return Perl_re_compile(aTHX_ pattern, flags);
4114 Perl_re_compile(pTHX_ const SV * const pattern, const U32 pm_flags)
4118 register regexp_internal *ri;
4120 char* exp = SvPV((SV*)pattern, plen);
4121 char* xend = exp + plen;
4128 RExC_state_t RExC_state;
4129 RExC_state_t * const pRExC_state = &RExC_state;
4130 #ifdef TRIE_STUDY_OPT
4132 RExC_state_t copyRExC_state;
4134 GET_RE_DEBUG_FLAGS_DECL;
4135 DEBUG_r(if (!PL_colorset) reginitcolors());
4137 RExC_utf8 = RExC_orig_utf8 = pm_flags & RXf_UTF8;
4140 SV *dsv= sv_newmortal();
4141 RE_PV_QUOTED_DECL(s, RExC_utf8,
4142 dsv, exp, plen, 60);
4143 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4144 PL_colors[4],PL_colors[5],s);
4149 RExC_flags = pm_flags;
4153 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4154 RExC_seen_evals = 0;
4157 /* First pass: determine size, legality. */
4165 RExC_emit = &PL_regdummy;
4166 RExC_whilem_seen = 0;
4167 RExC_charnames = NULL;
4168 RExC_open_parens = NULL;
4169 RExC_close_parens = NULL;
4171 RExC_paren_names = NULL;
4173 RExC_paren_name_list = NULL;
4175 RExC_recurse = NULL;
4176 RExC_recurse_count = 0;
4178 #if 0 /* REGC() is (currently) a NOP at the first pass.
4179 * Clever compilers notice this and complain. --jhi */
4180 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4182 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4183 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4184 RExC_precomp = NULL;
4187 if (RExC_utf8 && !RExC_orig_utf8) {
4188 /* It's possible to write a regexp in ascii that represents Unicode
4189 codepoints outside of the byte range, such as via \x{100}. If we
4190 detect such a sequence we have to convert the entire pattern to utf8
4191 and then recompile, as our sizing calculation will have been based
4192 on 1 byte == 1 character, but we will need to use utf8 to encode
4193 at least some part of the pattern, and therefore must convert the whole
4195 XXX: somehow figure out how to make this less expensive...
4198 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4199 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4200 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4202 RExC_orig_utf8 = RExC_utf8;
4204 goto redo_first_pass;
4207 PerlIO_printf(Perl_debug_log,
4208 "Required size %"IVdf" nodes\n"
4209 "Starting second pass (creation)\n",
4212 RExC_lastparse=NULL;
4214 /* Small enough for pointer-storage convention?
4215 If extralen==0, this means that we will not need long jumps. */
4216 if (RExC_size >= 0x10000L && RExC_extralen)
4217 RExC_size += RExC_extralen;
4220 if (RExC_whilem_seen > 15)
4221 RExC_whilem_seen = 15;
4223 /* Allocate space and zero-initialize. Note, the two step process
4224 of zeroing when in debug mode, thus anything assigned has to
4225 happen after that */
4226 Newxz(r, 1, regexp);
4227 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4228 char, regexp_internal);
4229 if ( r == NULL || ri == NULL )
4230 FAIL("Regexp out of space");
4232 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4233 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4235 /* bulk initialize base fields with 0. */
4236 Zero(ri, sizeof(regexp_internal), char);
4239 /* non-zero initialization begins here */
4241 r->engine= RE_ENGINE_PTR;
4244 r->extflags = pm_flags;
4246 bool has_k = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4247 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4248 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4249 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) >> 12);
4250 const char *fptr = STD_PAT_MODS; /*"msix"*/
4252 r->wraplen = r->prelen + has_minus + has_k + has_runon
4253 + (sizeof(STD_PAT_MODS) - 1)
4254 + (sizeof("(?:)") - 1);
4256 Newx(r->wrapped, r->wraplen + 1, char );
4260 *p++ = KEEPCOPY_PAT_MOD; /*'k'*/
4262 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4263 char *colon = r + 1;
4266 while((ch = *fptr++)) {
4280 Copy(RExC_precomp, p, r->prelen, char);
4290 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4292 if (RExC_seen & REG_SEEN_RECURSE) {
4293 Newxz(RExC_open_parens, RExC_npar,regnode *);
4294 SAVEFREEPV(RExC_open_parens);
4295 Newxz(RExC_close_parens,RExC_npar,regnode *);
4296 SAVEFREEPV(RExC_close_parens);
4299 /* Useful during FAIL. */
4300 #ifdef RE_TRACK_PATTERN_OFFSETS
4301 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4302 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4303 "%s %"UVuf" bytes for offset annotations.\n",
4304 ri->u.offsets ? "Got" : "Couldn't get",
4305 (UV)((2*RExC_size+1) * sizeof(U32))));
4307 SetProgLen(ri,RExC_size);
4311 /* Second pass: emit code. */
4312 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4317 RExC_emit_start = ri->program;
4318 RExC_emit = ri->program;
4319 RExC_emit_bound = ri->program + RExC_size + 1;
4321 /* Store the count of eval-groups for security checks: */
4322 RExC_rx->seen_evals = RExC_seen_evals;
4323 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4324 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4328 /* XXXX To minimize changes to RE engine we always allocate
4329 3-units-long substrs field. */
4330 Newx(r->substrs, 1, struct reg_substr_data);
4331 if (RExC_recurse_count) {
4332 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4333 SAVEFREEPV(RExC_recurse);
4337 r->minlen = minlen = sawplus = sawopen = 0;
4338 Zero(r->substrs, 1, struct reg_substr_data);
4340 #ifdef TRIE_STUDY_OPT
4343 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4345 RExC_state = copyRExC_state;
4346 if (seen & REG_TOP_LEVEL_BRANCHES)
4347 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4349 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4350 if (data.last_found) {
4351 SvREFCNT_dec(data.longest_fixed);
4352 SvREFCNT_dec(data.longest_float);
4353 SvREFCNT_dec(data.last_found);
4355 StructCopy(&zero_scan_data, &data, scan_data_t);
4357 StructCopy(&zero_scan_data, &data, scan_data_t);
4358 copyRExC_state = RExC_state;
4361 StructCopy(&zero_scan_data, &data, scan_data_t);
4364 /* Dig out information for optimizations. */
4365 r->extflags = pm_flags; /* Again? */
4366 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4369 r->extflags |= RXf_UTF8; /* Unicode in it? */
4370 ri->regstclass = NULL;
4371 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4372 r->intflags |= PREGf_NAUGHTY;
4373 scan = ri->program + 1; /* First BRANCH. */
4375 /* testing for BRANCH here tells us whether there is "must appear"
4376 data in the pattern. If there is then we can use it for optimisations */
4377 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4379 STRLEN longest_float_length, longest_fixed_length;
4380 struct regnode_charclass_class ch_class; /* pointed to by data */
4382 I32 last_close = 0; /* pointed to by data */
4383 regnode *first= scan;
4384 regnode *first_next= regnext(first);
4386 /* Skip introductions and multiplicators >= 1. */
4387 while ((OP(first) == OPEN && (sawopen = 1)) ||
4388 /* An OR of *one* alternative - should not happen now. */
4389 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4390 /* for now we can't handle lookbehind IFMATCH*/
4391 (OP(first) == IFMATCH && !first->flags) ||
4392 (OP(first) == PLUS) ||
4393 (OP(first) == MINMOD) ||
4394 /* An {n,m} with n>0 */
4395 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4396 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4399 if (OP(first) == PLUS)
4402 first += regarglen[OP(first)];
4403 if (OP(first) == IFMATCH) {
4404 first = NEXTOPER(first);
4405 first += EXTRA_STEP_2ARGS;
4406 } else /* XXX possible optimisation for /(?=)/ */
4407 first = NEXTOPER(first);
4408 first_next= regnext(first);
4411 /* Starting-point info. */
4413 DEBUG_PEEP("first:",first,0);
4414 /* Ignore EXACT as we deal with it later. */
4415 if (PL_regkind[OP(first)] == EXACT) {
4416 if (OP(first) == EXACT)
4417 NOOP; /* Empty, get anchored substr later. */
4418 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4419 ri->regstclass = first;
4422 else if (PL_regkind[OP(first)] == TRIE &&
4423 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4426 /* this can happen only on restudy */
4427 if ( OP(first) == TRIE ) {
4428 struct regnode_1 *trieop = (struct regnode_1 *)
4429 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4430 StructCopy(first,trieop,struct regnode_1);
4431 trie_op=(regnode *)trieop;
4433 struct regnode_charclass *trieop = (struct regnode_charclass *)
4434 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4435 StructCopy(first,trieop,struct regnode_charclass);
4436 trie_op=(regnode *)trieop;
4439 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4440 ri->regstclass = trie_op;
4443 else if (strchr((const char*)PL_simple,OP(first)))
4444 ri->regstclass = first;
4445 else if (PL_regkind[OP(first)] == BOUND ||
4446 PL_regkind[OP(first)] == NBOUND)
4447 ri->regstclass = first;
4448 else if (PL_regkind[OP(first)] == BOL) {
4449 r->extflags |= (OP(first) == MBOL
4451 : (OP(first) == SBOL
4454 first = NEXTOPER(first);
4457 else if (OP(first) == GPOS) {
4458 r->extflags |= RXf_ANCH_GPOS;
4459 first = NEXTOPER(first);
4462 else if ((!sawopen || !RExC_sawback) &&
4463 (OP(first) == STAR &&
4464 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4465 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4467 /* turn .* into ^.* with an implied $*=1 */
4469 (OP(NEXTOPER(first)) == REG_ANY)
4472 r->extflags |= type;
4473 r->intflags |= PREGf_IMPLICIT;
4474 first = NEXTOPER(first);
4477 if (sawplus && (!sawopen || !RExC_sawback)
4478 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4479 /* x+ must match at the 1st pos of run of x's */
4480 r->intflags |= PREGf_SKIP;
4482 /* Scan is after the zeroth branch, first is atomic matcher. */
4483 #ifdef TRIE_STUDY_OPT
4486 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4487 (IV)(first - scan + 1))
4491 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4492 (IV)(first - scan + 1))
4498 * If there's something expensive in the r.e., find the
4499 * longest literal string that must appear and make it the
4500 * regmust. Resolve ties in favor of later strings, since
4501 * the regstart check works with the beginning of the r.e.
4502 * and avoiding duplication strengthens checking. Not a
4503 * strong reason, but sufficient in the absence of others.
4504 * [Now we resolve ties in favor of the earlier string if
4505 * it happens that c_offset_min has been invalidated, since the
4506 * earlier string may buy us something the later one won't.]
4509 data.longest_fixed = newSVpvs("");
4510 data.longest_float = newSVpvs("");
4511 data.last_found = newSVpvs("");
4512 data.longest = &(data.longest_fixed);
4514 if (!ri->regstclass) {
4515 cl_init(pRExC_state, &ch_class);
4516 data.start_class = &ch_class;
4517 stclass_flag = SCF_DO_STCLASS_AND;
4518 } else /* XXXX Check for BOUND? */
4520 data.last_closep = &last_close;
4522 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4523 &data, -1, NULL, NULL,
4524 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4530 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4531 && data.last_start_min == 0 && data.last_end > 0
4532 && !RExC_seen_zerolen
4533 && !(RExC_seen & REG_SEEN_VERBARG)
4534 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4535 r->extflags |= RXf_CHECK_ALL;
4536 scan_commit(pRExC_state, &data,&minlen,0);
4537 SvREFCNT_dec(data.last_found);
4539 /* Note that code very similar to this but for anchored string
4540 follows immediately below, changes may need to be made to both.
4543 longest_float_length = CHR_SVLEN(data.longest_float);
4544 if (longest_float_length
4545 || (data.flags & SF_FL_BEFORE_EOL
4546 && (!(data.flags & SF_FL_BEFORE_MEOL)
4547 || (RExC_flags & RXf_PMf_MULTILINE))))
4551 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4552 && data.offset_fixed == data.offset_float_min
4553 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4554 goto remove_float; /* As in (a)+. */
4556 /* copy the information about the longest float from the reg_scan_data
4557 over to the program. */
4558 if (SvUTF8(data.longest_float)) {
4559 r->float_utf8 = data.longest_float;
4560 r->float_substr = NULL;
4562 r->float_substr = data.longest_float;
4563 r->float_utf8 = NULL;
4565 /* float_end_shift is how many chars that must be matched that
4566 follow this item. We calculate it ahead of time as once the
4567 lookbehind offset is added in we lose the ability to correctly
4569 ml = data.minlen_float ? *(data.minlen_float)
4570 : (I32)longest_float_length;
4571 r->float_end_shift = ml - data.offset_float_min
4572 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4573 + data.lookbehind_float;
4574 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4575 r->float_max_offset = data.offset_float_max;
4576 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4577 r->float_max_offset -= data.lookbehind_float;
4579 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4580 && (!(data.flags & SF_FL_BEFORE_MEOL)
4581 || (RExC_flags & RXf_PMf_MULTILINE)));
4582 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4586 r->float_substr = r->float_utf8 = NULL;
4587 SvREFCNT_dec(data.longest_float);
4588 longest_float_length = 0;
4591 /* Note that code very similar to this but for floating string
4592 is immediately above, changes may need to be made to both.
4595 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4596 if (longest_fixed_length
4597 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4598 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4599 || (RExC_flags & RXf_PMf_MULTILINE))))
4603 /* copy the information about the longest fixed
4604 from the reg_scan_data over to the program. */
4605 if (SvUTF8(data.longest_fixed)) {
4606 r->anchored_utf8 = data.longest_fixed;
4607 r->anchored_substr = NULL;
4609 r->anchored_substr = data.longest_fixed;
4610 r->anchored_utf8 = NULL;
4612 /* fixed_end_shift is how many chars that must be matched that
4613 follow this item. We calculate it ahead of time as once the
4614 lookbehind offset is added in we lose the ability to correctly
4616 ml = data.minlen_fixed ? *(data.minlen_fixed)
4617 : (I32)longest_fixed_length;
4618 r->anchored_end_shift = ml - data.offset_fixed
4619 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4620 + data.lookbehind_fixed;
4621 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4623 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4624 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4625 || (RExC_flags & RXf_PMf_MULTILINE)));
4626 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4629 r->anchored_substr = r->anchored_utf8 = NULL;
4630 SvREFCNT_dec(data.longest_fixed);
4631 longest_fixed_length = 0;
4634 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4635 ri->regstclass = NULL;
4636 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4638 && !(data.start_class->flags & ANYOF_EOS)
4639 && !cl_is_anything(data.start_class))
4641 const U32 n = add_data(pRExC_state, 1, "f");
4643 Newx(RExC_rxi->data->data[n], 1,
4644 struct regnode_charclass_class);
4645 StructCopy(data.start_class,
4646 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4647 struct regnode_charclass_class);
4648 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4649 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4650 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4651 regprop(r, sv, (regnode*)data.start_class);
4652 PerlIO_printf(Perl_debug_log,
4653 "synthetic stclass \"%s\".\n",
4654 SvPVX_const(sv));});
4657 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4658 if (longest_fixed_length > longest_float_length) {
4659 r->check_end_shift = r->anchored_end_shift;
4660 r->check_substr = r->anchored_substr;
4661 r->check_utf8 = r->anchored_utf8;
4662 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4663 if (r->extflags & RXf_ANCH_SINGLE)
4664 r->extflags |= RXf_NOSCAN;
4667 r->check_end_shift = r->float_end_shift;
4668 r->check_substr = r->float_substr;
4669 r->check_utf8 = r->float_utf8;
4670 r->check_offset_min = r->float_min_offset;
4671 r->check_offset_max = r->float_max_offset;
4673 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4674 This should be changed ASAP! */
4675 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4676 r->extflags |= RXf_USE_INTUIT;
4677 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4678 r->extflags |= RXf_INTUIT_TAIL;
4680 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4681 if ( (STRLEN)minlen < longest_float_length )
4682 minlen= longest_float_length;
4683 if ( (STRLEN)minlen < longest_fixed_length )
4684 minlen= longest_fixed_length;
4688 /* Several toplevels. Best we can is to set minlen. */
4690 struct regnode_charclass_class ch_class;
4693 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4695 scan = ri->program + 1;
4696 cl_init(pRExC_state, &ch_class);
4697 data.start_class = &ch_class;
4698 data.last_closep = &last_close;
4701 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4702 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4706 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4707 = r->float_substr = r->float_utf8 = NULL;
4708 if (!(data.start_class->flags & ANYOF_EOS)
4709 && !cl_is_anything(data.start_class))
4711 const U32 n = add_data(pRExC_state, 1, "f");
4713 Newx(RExC_rxi->data->data[n], 1,
4714 struct regnode_charclass_class);
4715 StructCopy(data.start_class,
4716 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4717 struct regnode_charclass_class);
4718 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4719 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4720 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4721 regprop(r, sv, (regnode*)data.start_class);
4722 PerlIO_printf(Perl_debug_log,
4723 "synthetic stclass \"%s\".\n",
4724 SvPVX_const(sv));});
4728 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4729 the "real" pattern. */
4731 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4732 (IV)minlen, (IV)r->minlen);
4734 r->minlenret = minlen;
4735 if (r->minlen < minlen)
4738 if (RExC_seen & REG_SEEN_GPOS)
4739 r->extflags |= RXf_GPOS_SEEN;
4740 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4741 r->extflags |= RXf_LOOKBEHIND_SEEN;
4742 if (RExC_seen & REG_SEEN_EVAL)
4743 r->extflags |= RXf_EVAL_SEEN;
4744 if (RExC_seen & REG_SEEN_CANY)
4745 r->extflags |= RXf_CANY_SEEN;
4746 if (RExC_seen & REG_SEEN_VERBARG)
4747 r->intflags |= PREGf_VERBARG_SEEN;
4748 if (RExC_seen & REG_SEEN_CUTGROUP)
4749 r->intflags |= PREGf_CUTGROUP_SEEN;
4750 if (RExC_paren_names)
4751 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4753 r->paren_names = NULL;
4755 #ifdef STUPID_PATTERN_CHECKS
4756 if (r->extflags & RXf_SPLIT && r->prelen == 1 && r->precomp[0] == ' ')
4757 /* XXX: this should happen BEFORE we compile */
4758 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4759 else if (r->prelen == 3 && memEQ("\\s+", r->precomp, 3))
4760 r->extflags |= RXf_WHITE;
4761 else if (r->prelen == 1 && r->precomp[0] == '^')
4762 r->extflags |= RXf_START_ONLY;
4764 if (r->extflags & RXf_SPLIT && r->prelen == 1 && r->precomp[0] == ' ')
4765 /* XXX: this should happen BEFORE we compile */
4766 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4768 regnode *first = ri->program + 1;
4769 char fop = OP(first);
4770 char nop = OP(NEXTOPER(first));
4772 if (PL_regkind[fop] == BOL && nop == END)
4773 r->extflags |= RXf_START_ONLY;
4774 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4775 r->extflags |= RXf_WHITE;
4779 if (RExC_paren_names) {
4780 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4781 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4784 ri->name_list_idx = 0;
4786 if (RExC_recurse_count) {
4787 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4788 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4789 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4792 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4793 /* assume we don't need to swap parens around before we match */
4796 PerlIO_printf(Perl_debug_log,"Final program:\n");
4799 #ifdef RE_TRACK_PATTERN_OFFSETS
4800 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4801 const U32 len = ri->u.offsets[0];
4803 GET_RE_DEBUG_FLAGS_DECL;
4804 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4805 for (i = 1; i <= len; i++) {
4806 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4807 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4808 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4810 PerlIO_printf(Perl_debug_log, "\n");
4816 #undef RE_ENGINE_PTR
4820 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4823 PERL_UNUSED_ARG(value);
4825 if (flags & RXf_HASH_FETCH) {
4826 return reg_named_buff_fetch(rx, key, flags);
4827 } else if (flags & (RXf_HASH_STORE | RXf_HASH_DELETE | RXf_HASH_CLEAR)) {
4828 Perl_croak(aTHX_ PL_no_modify);
4830 } else if (flags & RXf_HASH_EXISTS) {
4831 return reg_named_buff_exists(rx, key, flags)
4834 } else if (flags & RXf_HASH_REGNAMES) {
4835 return reg_named_buff_all(rx, flags);
4836 } else if (flags & (RXf_HASH_SCALAR | RXf_HASH_REGNAMES_COUNT)) {
4837 return reg_named_buff_scalar(rx, flags);
4839 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
4845 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
4848 PERL_UNUSED_ARG(lastkey);
4850 if (flags & RXf_HASH_FIRSTKEY)
4851 return reg_named_buff_firstkey(rx, flags);
4852 else if (flags & RXf_HASH_NEXTKEY)
4853 return reg_named_buff_nextkey(rx, flags);
4855 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
4861 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const rx, SV * const namesv, const U32 flags)
4863 AV *retarray = NULL;
4865 if (flags & RXf_HASH_ALL)
4868 if (rx && rx->paren_names) {
4869 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4872 SV* sv_dat=HeVAL(he_str);
4873 I32 *nums=(I32*)SvPVX(sv_dat);
4874 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4875 if ((I32)(rx->nparens) >= nums[i]
4876 && rx->offs[nums[i]].start != -1
4877 && rx->offs[nums[i]].end != -1)
4880 CALLREG_NUMBUF_FETCH(rx,nums[i],ret);
4884 ret = newSVsv(&PL_sv_undef);
4887 SvREFCNT_inc_simple_void(ret);
4888 av_push(retarray, ret);
4892 return newRV((SV*)retarray);
4899 Perl_reg_named_buff_exists(pTHX_ REGEXP * const rx, SV * const key,
4902 if (rx && rx->paren_names) {
4903 if (flags & RXf_HASH_ALL) {
4904 return hv_exists_ent(rx->paren_names, key, 0);
4906 SV *sv = CALLREG_NAMED_BUFF_FETCH(rx, key, flags);
4920 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const rx, const U32 flags)
4922 (void)hv_iterinit(rx->paren_names);
4924 return CALLREG_NAMED_BUFF_NEXTKEY(rx, NULL, flags & ~RXf_HASH_FIRSTKEY);
4928 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const rx, const U32 flags)
4930 if (rx && rx->paren_names) {
4931 HV *hv = rx->paren_names;
4933 while ( (temphe = hv_iternext_flags(hv,0)) ) {
4936 SV* sv_dat = HeVAL(temphe);
4937 I32 *nums = (I32*)SvPVX(sv_dat);
4938 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
4939 if ((I32)(rx->lastcloseparen) >= nums[i] &&
4940 rx->offs[nums[i]].start != -1 &&
4941 rx->offs[nums[i]].end != -1)
4947 if (parno || flags & RXf_HASH_ALL) {
4949 char *pv = HePV(temphe, len);
4950 return newSVpvn(pv,len);
4958 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const rx, const U32 flags)
4964 if (rx && rx->paren_names) {
4965 if (flags & (RXf_HASH_ALL | RXf_HASH_REGNAMES_COUNT)) {
4966 return newSViv(HvTOTALKEYS(rx->paren_names));
4967 } else if (flags & RXf_HASH_ONE) {
4968 ret = CALLREG_NAMED_BUFF_ALL(rx, (flags | RXf_HASH_REGNAMES));
4969 av = (AV*)SvRV(ret);
4970 length = av_len(av);
4971 return newSViv(length + 1);
4973 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
4977 return &PL_sv_undef;
4981 Perl_reg_named_buff_all(pTHX_ REGEXP * const rx, const U32 flags)
4985 if (rx && rx->paren_names) {
4986 HV *hv= rx->paren_names;
4988 (void)hv_iterinit(hv);
4989 while ( (temphe = hv_iternext_flags(hv,0)) ) {
4992 SV* sv_dat = HeVAL(temphe);
4993 I32 *nums = (I32*)SvPVX(sv_dat);
4994 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
4995 if ((I32)(rx->lastcloseparen) >= nums[i] &&
4996 rx->offs[nums[i]].start != -1 &&
4997 rx->offs[nums[i]].end != -1)
5003 if (parno || flags & RXf_HASH_ALL) {
5005 char *pv = HePV(temphe, len);
5006 av_push(av, newSVpvn(pv,len));
5011 return newRV((SV*)av);
5015 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const rx, const I32 paren, SV * const sv)
5022 sv_setsv(sv,&PL_sv_undef);
5026 if (paren == RXf_PREMATCH && rx->offs[0].start != -1) {
5028 i = rx->offs[0].start;
5032 if (paren == RXf_POSTMATCH && rx->offs[0].end != -1) {
5034 s = rx->subbeg + rx->offs[0].end;
5035 i = rx->sublen - rx->offs[0].end;
5038 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5039 (s1 = rx->offs[paren].start) != -1 &&
5040 (t1 = rx->offs[paren].end) != -1)
5044 s = rx->subbeg + s1;
5046 sv_setsv(sv,&PL_sv_undef);
5049 assert(rx->sublen >= (s - rx->subbeg) + i );
5051 const int oldtainted = PL_tainted;
5053 sv_setpvn(sv, s, i);
5054 PL_tainted = oldtainted;
5055 if ( (rx->extflags & RXf_CANY_SEEN)
5056 ? (RX_MATCH_UTF8(rx)
5057 && (!i || is_utf8_string((U8*)s, i)))
5058 : (RX_MATCH_UTF8(rx)) )
5065 if (RX_MATCH_TAINTED(rx)) {
5066 if (SvTYPE(sv) >= SVt_PVMG) {
5067 MAGIC* const mg = SvMAGIC(sv);
5070 SvMAGIC_set(sv, mg->mg_moremagic);
5072 if ((mgt = SvMAGIC(sv))) {
5073 mg->mg_moremagic = mgt;
5074 SvMAGIC_set(sv, mg);
5084 sv_setsv(sv,&PL_sv_undef);
5090 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5091 SV const * const value)
5093 PERL_UNUSED_ARG(rx);
5094 PERL_UNUSED_ARG(paren);
5095 PERL_UNUSED_ARG(value);
5098 Perl_croak(aTHX_ PL_no_modify);
5102 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const rx, const SV * const sv,
5108 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5110 /* $` / ${^PREMATCH} */
5112 if (rx->offs[0].start != -1) {
5113 i = rx->offs[0].start;
5121 /* $' / ${^POSTMATCH} */
5123 if (rx->offs[0].end != -1) {
5124 i = rx->sublen - rx->offs[0].end;
5126 s1 = rx->offs[0].end;
5132 /* $& / ${^MATCH}, $1, $2, ... */
5134 if (paren <= (I32)rx->nparens &&
5135 (s1 = rx->offs[paren].start) != -1 &&
5136 (t1 = rx->offs[paren].end) != -1)
5141 if (ckWARN(WARN_UNINITIALIZED))
5142 report_uninit((SV*)sv);
5147 if (i > 0 && RX_MATCH_UTF8(rx)) {
5148 const char * const s = rx->subbeg + s1;
5153 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5160 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5162 PERL_UNUSED_ARG(rx);
5163 return newSVpvs("Regexp");
5166 /* Scans the name of a named buffer from the pattern.
5167 * If flags is REG_RSN_RETURN_NULL returns null.
5168 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5169 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5170 * to the parsed name as looked up in the RExC_paren_names hash.
5171 * If there is an error throws a vFAIL().. type exception.
5174 #define REG_RSN_RETURN_NULL 0
5175 #define REG_RSN_RETURN_NAME 1
5176 #define REG_RSN_RETURN_DATA 2
5179 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
5180 char *name_start = RExC_parse;
5182 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5183 /* skip IDFIRST by using do...while */
5186 RExC_parse += UTF8SKIP(RExC_parse);
5187 } while (isALNUM_utf8((U8*)RExC_parse));
5191 } while (isALNUM(*RExC_parse));
5195 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
5196 (int)(RExC_parse - name_start)));
5199 if ( flags == REG_RSN_RETURN_NAME)
5201 else if (flags==REG_RSN_RETURN_DATA) {
5204 if ( ! sv_name ) /* should not happen*/
5205 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5206 if (RExC_paren_names)
5207 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5209 sv_dat = HeVAL(he_str);
5211 vFAIL("Reference to nonexistent named group");
5215 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5222 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5223 int rem=(int)(RExC_end - RExC_parse); \
5232 if (RExC_lastparse!=RExC_parse) \
5233 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5236 iscut ? "..." : "<" \
5239 PerlIO_printf(Perl_debug_log,"%16s",""); \
5242 num = RExC_size + 1; \
5244 num=REG_NODE_NUM(RExC_emit); \
5245 if (RExC_lastnum!=num) \
5246 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5248 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5249 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5250 (int)((depth*2)), "", \
5254 RExC_lastparse=RExC_parse; \
5259 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5260 DEBUG_PARSE_MSG((funcname)); \
5261 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5263 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5264 DEBUG_PARSE_MSG((funcname)); \
5265 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5268 - reg - regular expression, i.e. main body or parenthesized thing
5270 * Caller must absorb opening parenthesis.
5272 * Combining parenthesis handling with the base level of regular expression
5273 * is a trifle forced, but the need to tie the tails of the branches to what
5274 * follows makes it hard to avoid.
5276 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5278 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5280 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5284 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5285 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5288 register regnode *ret; /* Will be the head of the group. */
5289 register regnode *br;
5290 register regnode *lastbr;
5291 register regnode *ender = NULL;
5292 register I32 parno = 0;
5294 const I32 oregflags = RExC_flags;
5295 bool have_branch = 0;
5297 I32 freeze_paren = 0;
5298 I32 after_freeze = 0;
5300 /* for (?g), (?gc), and (?o) warnings; warning
5301 about (?c) will warn about (?g) -- japhy */
5303 #define WASTED_O 0x01
5304 #define WASTED_G 0x02
5305 #define WASTED_C 0x04
5306 #define WASTED_GC (0x02|0x04)
5307 I32 wastedflags = 0x00;
5309 char * parse_start = RExC_parse; /* MJD */
5310 char * const oregcomp_parse = RExC_parse;
5312 GET_RE_DEBUG_FLAGS_DECL;
5313 DEBUG_PARSE("reg ");
5315 *flagp = 0; /* Tentatively. */
5318 /* Make an OPEN node, if parenthesized. */
5320 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5321 char *start_verb = RExC_parse;
5322 STRLEN verb_len = 0;
5323 char *start_arg = NULL;
5324 unsigned char op = 0;
5326 int internal_argval = 0; /* internal_argval is only useful if !argok */
5327 while ( *RExC_parse && *RExC_parse != ')' ) {
5328 if ( *RExC_parse == ':' ) {
5329 start_arg = RExC_parse + 1;
5335 verb_len = RExC_parse - start_verb;
5338 while ( *RExC_parse && *RExC_parse != ')' )
5340 if ( *RExC_parse != ')' )
5341 vFAIL("Unterminated verb pattern argument");
5342 if ( RExC_parse == start_arg )
5345 if ( *RExC_parse != ')' )
5346 vFAIL("Unterminated verb pattern");
5349 switch ( *start_verb ) {
5350 case 'A': /* (*ACCEPT) */
5351 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5353 internal_argval = RExC_nestroot;
5356 case 'C': /* (*COMMIT) */
5357 if ( memEQs(start_verb,verb_len,"COMMIT") )
5360 case 'F': /* (*FAIL) */
5361 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5366 case ':': /* (*:NAME) */
5367 case 'M': /* (*MARK:NAME) */
5368 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5373 case 'P': /* (*PRUNE) */
5374 if ( memEQs(start_verb,verb_len,"PRUNE") )
5377 case 'S': /* (*SKIP) */
5378 if ( memEQs(start_verb,verb_len,"SKIP") )
5381 case 'T': /* (*THEN) */
5382 /* [19:06] <TimToady> :: is then */
5383 if ( memEQs(start_verb,verb_len,"THEN") ) {
5385 RExC_seen |= REG_SEEN_CUTGROUP;
5391 vFAIL3("Unknown verb pattern '%.*s'",
5392 verb_len, start_verb);
5395 if ( start_arg && internal_argval ) {
5396 vFAIL3("Verb pattern '%.*s' may not have an argument",
5397 verb_len, start_verb);
5398 } else if ( argok < 0 && !start_arg ) {
5399 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5400 verb_len, start_verb);
5402 ret = reganode(pRExC_state, op, internal_argval);
5403 if ( ! internal_argval && ! SIZE_ONLY ) {
5405 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5406 ARG(ret) = add_data( pRExC_state, 1, "S" );
5407 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5414 if (!internal_argval)
5415 RExC_seen |= REG_SEEN_VERBARG;
5416 } else if ( start_arg ) {
5417 vFAIL3("Verb pattern '%.*s' may not have an argument",
5418 verb_len, start_verb);
5420 ret = reg_node(pRExC_state, op);
5422 nextchar(pRExC_state);
5425 if (*RExC_parse == '?') { /* (?...) */
5426 bool is_logical = 0;
5427 const char * const seqstart = RExC_parse;
5430 paren = *RExC_parse++;
5431 ret = NULL; /* For look-ahead/behind. */
5434 case 'P': /* (?P...) variants for those used to PCRE/Python */
5435 paren = *RExC_parse++;
5436 if ( paren == '<') /* (?P<...>) named capture */
5438 else if (paren == '>') { /* (?P>name) named recursion */
5439 goto named_recursion;
5441 else if (paren == '=') { /* (?P=...) named backref */
5442 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5443 you change this make sure you change that */
5444 char* name_start = RExC_parse;
5446 SV *sv_dat = reg_scan_name(pRExC_state,
5447 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5448 if (RExC_parse == name_start || *RExC_parse != ')')
5449 vFAIL2("Sequence %.3s... not terminated",parse_start);
5452 num = add_data( pRExC_state, 1, "S" );
5453 RExC_rxi->data->data[num]=(void*)sv_dat;
5454 SvREFCNT_inc_simple_void(sv_dat);
5457 ret = reganode(pRExC_state,
5458 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5462 Set_Node_Offset(ret, parse_start+1);
5463 Set_Node_Cur_Length(ret); /* MJD */
5465 nextchar(pRExC_state);
5469 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5471 case '<': /* (?<...) */
5472 if (*RExC_parse == '!')
5474 else if (*RExC_parse != '=')
5480 case '\'': /* (?'...') */
5481 name_start= RExC_parse;
5482 svname = reg_scan_name(pRExC_state,
5483 SIZE_ONLY ? /* reverse test from the others */
5484 REG_RSN_RETURN_NAME :
5485 REG_RSN_RETURN_NULL);
5486 if (RExC_parse == name_start) {
5488 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5491 if (*RExC_parse != paren)
5492 vFAIL2("Sequence (?%c... not terminated",
5493 paren=='>' ? '<' : paren);
5497 if (!svname) /* shouldnt happen */
5499 "panic: reg_scan_name returned NULL");
5500 if (!RExC_paren_names) {
5501 RExC_paren_names= newHV();
5502 sv_2mortal((SV*)RExC_paren_names);
5504 RExC_paren_name_list= newAV();
5505 sv_2mortal((SV*)RExC_paren_name_list);
5508 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5510 sv_dat = HeVAL(he_str);
5512 /* croak baby croak */
5514 "panic: paren_name hash element allocation failed");
5515 } else if ( SvPOK(sv_dat) ) {
5516 /* (?|...) can mean we have dupes so scan to check
5517 its already been stored. Maybe a flag indicating
5518 we are inside such a construct would be useful,
5519 but the arrays are likely to be quite small, so
5520 for now we punt -- dmq */
5521 IV count = SvIV(sv_dat);
5522 I32 *pv = (I32*)SvPVX(sv_dat);
5524 for ( i = 0 ; i < count ; i++ ) {
5525 if ( pv[i] == RExC_npar ) {
5531 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5532 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5533 pv[count] = RExC_npar;
5537 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5538 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5543 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5544 SvREFCNT_dec(svname);
5547 /*sv_dump(sv_dat);*/
5549 nextchar(pRExC_state);
5551 goto capturing_parens;
5553 RExC_seen |= REG_SEEN_LOOKBEHIND;
5555 case '=': /* (?=...) */
5556 case '!': /* (?!...) */
5557 RExC_seen_zerolen++;
5558 if (*RExC_parse == ')') {
5559 ret=reg_node(pRExC_state, OPFAIL);
5560 nextchar(pRExC_state);
5564 case '|': /* (?|...) */
5565 /* branch reset, behave like a (?:...) except that
5566 buffers in alternations share the same numbers */
5568 after_freeze = freeze_paren = RExC_npar;
5570 case ':': /* (?:...) */
5571 case '>': /* (?>...) */
5573 case '$': /* (?$...) */
5574 case '@': /* (?@...) */
5575 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5577 case '#': /* (?#...) */
5578 while (*RExC_parse && *RExC_parse != ')')
5580 if (*RExC_parse != ')')
5581 FAIL("Sequence (?#... not terminated");
5582 nextchar(pRExC_state);
5585 case '0' : /* (?0) */
5586 case 'R' : /* (?R) */
5587 if (*RExC_parse != ')')
5588 FAIL("Sequence (?R) not terminated");
5589 ret = reg_node(pRExC_state, GOSTART);
5590 *flagp |= POSTPONED;
5591 nextchar(pRExC_state);
5594 { /* named and numeric backreferences */
5596 case '&': /* (?&NAME) */
5597 parse_start = RExC_parse - 1;
5600 SV *sv_dat = reg_scan_name(pRExC_state,
5601 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5602 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5604 goto gen_recurse_regop;
5607 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5609 vFAIL("Illegal pattern");
5611 goto parse_recursion;
5613 case '-': /* (?-1) */
5614 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5615 RExC_parse--; /* rewind to let it be handled later */
5619 case '1': case '2': case '3': case '4': /* (?1) */
5620 case '5': case '6': case '7': case '8': case '9':
5623 num = atoi(RExC_parse);
5624 parse_start = RExC_parse - 1; /* MJD */
5625 if (*RExC_parse == '-')
5627 while (isDIGIT(*RExC_parse))
5629 if (*RExC_parse!=')')
5630 vFAIL("Expecting close bracket");
5633 if ( paren == '-' ) {
5635 Diagram of capture buffer numbering.
5636 Top line is the normal capture buffer numbers
5637 Botton line is the negative indexing as from
5641 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5645 num = RExC_npar + num;
5648 vFAIL("Reference to nonexistent group");
5650 } else if ( paren == '+' ) {
5651 num = RExC_npar + num - 1;
5654 ret = reganode(pRExC_state, GOSUB, num);
5656 if (num > (I32)RExC_rx->nparens) {
5658 vFAIL("Reference to nonexistent group");
5660 ARG2L_SET( ret, RExC_recurse_count++);
5662 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5663 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5667 RExC_seen |= REG_SEEN_RECURSE;
5668 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5669 Set_Node_Offset(ret, parse_start); /* MJD */
5671 *flagp |= POSTPONED;
5672 nextchar(pRExC_state);
5674 } /* named and numeric backreferences */
5677 case '?': /* (??...) */
5679 if (*RExC_parse != '{') {
5681 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5684 *flagp |= POSTPONED;
5685 paren = *RExC_parse++;
5687 case '{': /* (?{...}) */
5692 char *s = RExC_parse;
5694 RExC_seen_zerolen++;
5695 RExC_seen |= REG_SEEN_EVAL;
5696 while (count && (c = *RExC_parse)) {
5707 if (*RExC_parse != ')') {
5709 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5713 OP_4tree *sop, *rop;
5714 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5717 Perl_save_re_context(aTHX);
5718 rop = sv_compile_2op(sv, &sop, "re", &pad);
5719 sop->op_private |= OPpREFCOUNTED;
5720 /* re_dup will OpREFCNT_inc */
5721 OpREFCNT_set(sop, 1);
5724 n = add_data(pRExC_state, 3, "nop");
5725 RExC_rxi->data->data[n] = (void*)rop;
5726 RExC_rxi->data->data[n+1] = (void*)sop;
5727 RExC_rxi->data->data[n+2] = (void*)pad;
5730 else { /* First pass */
5731 if (PL_reginterp_cnt < ++RExC_seen_evals
5733 /* No compiled RE interpolated, has runtime
5734 components ===> unsafe. */
5735 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5736 if (PL_tainting && PL_tainted)
5737 FAIL("Eval-group in insecure regular expression");
5738 #if PERL_VERSION > 8
5739 if (IN_PERL_COMPILETIME)
5744 nextchar(pRExC_state);
5746 ret = reg_node(pRExC_state, LOGICAL);
5749 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5750 /* deal with the length of this later - MJD */
5753 ret = reganode(pRExC_state, EVAL, n);
5754 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5755 Set_Node_Offset(ret, parse_start);
5758 case '(': /* (?(?{...})...) and (?(?=...)...) */
5761 if (RExC_parse[0] == '?') { /* (?(?...)) */
5762 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5763 || RExC_parse[1] == '<'
5764 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5767 ret = reg_node(pRExC_state, LOGICAL);
5770 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5774 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5775 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5777 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5778 char *name_start= RExC_parse++;
5780 SV *sv_dat=reg_scan_name(pRExC_state,
5781 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5782 if (RExC_parse == name_start || *RExC_parse != ch)
5783 vFAIL2("Sequence (?(%c... not terminated",
5784 (ch == '>' ? '<' : ch));
5787 num = add_data( pRExC_state, 1, "S" );
5788 RExC_rxi->data->data[num]=(void*)sv_dat;
5789 SvREFCNT_inc_simple_void(sv_dat);
5791 ret = reganode(pRExC_state,NGROUPP,num);
5792 goto insert_if_check_paren;
5794 else if (RExC_parse[0] == 'D' &&
5795 RExC_parse[1] == 'E' &&
5796 RExC_parse[2] == 'F' &&
5797 RExC_parse[3] == 'I' &&
5798 RExC_parse[4] == 'N' &&
5799 RExC_parse[5] == 'E')
5801 ret = reganode(pRExC_state,DEFINEP,0);
5804 goto insert_if_check_paren;
5806 else if (RExC_parse[0] == 'R') {
5809 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5810 parno = atoi(RExC_parse++);
5811 while (isDIGIT(*RExC_parse))
5813 } else if (RExC_parse[0] == '&') {
5816 sv_dat = reg_scan_name(pRExC_state,
5817 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5818 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5820 ret = reganode(pRExC_state,INSUBP,parno);
5821 goto insert_if_check_paren;
5823 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5826 parno = atoi(RExC_parse++);
5828 while (isDIGIT(*RExC_parse))
5830 ret = reganode(pRExC_state, GROUPP, parno);
5832 insert_if_check_paren:
5833 if ((c = *nextchar(pRExC_state)) != ')')
5834 vFAIL("Switch condition not recognized");
5836 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5837 br = regbranch(pRExC_state, &flags, 1,depth+1);
5839 br = reganode(pRExC_state, LONGJMP, 0);
5841 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5842 c = *nextchar(pRExC_state);
5847 vFAIL("(?(DEFINE)....) does not allow branches");
5848 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5849 regbranch(pRExC_state, &flags, 1,depth+1);
5850 REGTAIL(pRExC_state, ret, lastbr);
5853 c = *nextchar(pRExC_state);
5858 vFAIL("Switch (?(condition)... contains too many branches");
5859 ender = reg_node(pRExC_state, TAIL);
5860 REGTAIL(pRExC_state, br, ender);
5862 REGTAIL(pRExC_state, lastbr, ender);
5863 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5866 REGTAIL(pRExC_state, ret, ender);
5867 RExC_size++; /* XXX WHY do we need this?!!
5868 For large programs it seems to be required
5869 but I can't figure out why. -- dmq*/
5873 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5877 RExC_parse--; /* for vFAIL to print correctly */
5878 vFAIL("Sequence (? incomplete");
5882 parse_flags: /* (?i) */
5884 U32 posflags = 0, negflags = 0;
5885 U32 *flagsp = &posflags;
5887 while (*RExC_parse) {
5888 /* && strchr("iogcmsx", *RExC_parse) */
5889 /* (?g), (?gc) and (?o) are useless here
5890 and must be globally applied -- japhy */
5891 switch (*RExC_parse) {
5892 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5895 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5896 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5897 if (! (wastedflags & wflagbit) ) {
5898 wastedflags |= wflagbit;
5901 "Useless (%s%c) - %suse /%c modifier",
5902 flagsp == &negflags ? "?-" : "?",
5904 flagsp == &negflags ? "don't " : "",
5912 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5913 if (! (wastedflags & WASTED_C) ) {
5914 wastedflags |= WASTED_GC;
5917 "Useless (%sc) - %suse /gc modifier",
5918 flagsp == &negflags ? "?-" : "?",
5919 flagsp == &negflags ? "don't " : ""
5925 if (flagsp == &negflags) {
5926 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5927 vWARN(RExC_parse + 1,"Useless use of (?-k)");
5929 *flagsp |= RXf_PMf_KEEPCOPY;
5933 if (flagsp == &negflags) {
5935 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5939 wastedflags = 0; /* reset so (?g-c) warns twice */
5945 RExC_flags |= posflags;
5946 RExC_flags &= ~negflags;
5947 nextchar(pRExC_state);
5958 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5963 }} /* one for the default block, one for the switch */
5970 ret = reganode(pRExC_state, OPEN, parno);
5973 RExC_nestroot = parno;
5974 if (RExC_seen & REG_SEEN_RECURSE
5975 && !RExC_open_parens[parno-1])
5977 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5978 "Setting open paren #%"IVdf" to %d\n",
5979 (IV)parno, REG_NODE_NUM(ret)));
5980 RExC_open_parens[parno-1]= ret;
5983 Set_Node_Length(ret, 1); /* MJD */
5984 Set_Node_Offset(ret, RExC_parse); /* MJD */
5992 /* Pick up the branches, linking them together. */
5993 parse_start = RExC_parse; /* MJD */
5994 br = regbranch(pRExC_state, &flags, 1,depth+1);
5995 /* branch_len = (paren != 0); */
5999 if (*RExC_parse == '|') {
6000 if (!SIZE_ONLY && RExC_extralen) {
6001 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6004 reginsert(pRExC_state, BRANCH, br, depth+1);
6005 Set_Node_Length(br, paren != 0);
6006 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6010 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6012 else if (paren == ':') {
6013 *flagp |= flags&SIMPLE;
6015 if (is_open) { /* Starts with OPEN. */
6016 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6018 else if (paren != '?') /* Not Conditional */
6020 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6022 while (*RExC_parse == '|') {
6023 if (!SIZE_ONLY && RExC_extralen) {
6024 ender = reganode(pRExC_state, LONGJMP,0);
6025 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6028 RExC_extralen += 2; /* Account for LONGJMP. */
6029 nextchar(pRExC_state);
6031 if (RExC_npar > after_freeze)
6032 after_freeze = RExC_npar;
6033 RExC_npar = freeze_paren;
6035 br = regbranch(pRExC_state, &flags, 0, depth+1);
6039 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6041 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6044 if (have_branch || paren != ':') {
6045 /* Make a closing node, and hook it on the end. */
6048 ender = reg_node(pRExC_state, TAIL);
6051 ender = reganode(pRExC_state, CLOSE, parno);
6052 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6053 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6054 "Setting close paren #%"IVdf" to %d\n",
6055 (IV)parno, REG_NODE_NUM(ender)));
6056 RExC_close_parens[parno-1]= ender;
6057 if (RExC_nestroot == parno)
6060 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6061 Set_Node_Length(ender,1); /* MJD */
6067 *flagp &= ~HASWIDTH;
6070 ender = reg_node(pRExC_state, SUCCEED);
6073 ender = reg_node(pRExC_state, END);
6075 assert(!RExC_opend); /* there can only be one! */
6080 REGTAIL(pRExC_state, lastbr, ender);
6082 if (have_branch && !SIZE_ONLY) {
6084 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6086 /* Hook the tails of the branches to the closing node. */
6087 for (br = ret; br; br = regnext(br)) {
6088 const U8 op = PL_regkind[OP(br)];
6090 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6092 else if (op == BRANCHJ) {
6093 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6101 static const char parens[] = "=!<,>";
6103 if (paren && (p = strchr(parens, paren))) {
6104 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6105 int flag = (p - parens) > 1;
6108 node = SUSPEND, flag = 0;
6109 reginsert(pRExC_state, node,ret, depth+1);
6110 Set_Node_Cur_Length(ret);
6111 Set_Node_Offset(ret, parse_start + 1);
6113 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6117 /* Check for proper termination. */
6119 RExC_flags = oregflags;
6120 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6121 RExC_parse = oregcomp_parse;
6122 vFAIL("Unmatched (");
6125 else if (!paren && RExC_parse < RExC_end) {
6126 if (*RExC_parse == ')') {
6128 vFAIL("Unmatched )");
6131 FAIL("Junk on end of regexp"); /* "Can't happen". */
6135 RExC_npar = after_freeze;
6140 - regbranch - one alternative of an | operator
6142 * Implements the concatenation operator.
6145 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6148 register regnode *ret;
6149 register regnode *chain = NULL;
6150 register regnode *latest;
6151 I32 flags = 0, c = 0;
6152 GET_RE_DEBUG_FLAGS_DECL;
6153 DEBUG_PARSE("brnc");
6158 if (!SIZE_ONLY && RExC_extralen)
6159 ret = reganode(pRExC_state, BRANCHJ,0);
6161 ret = reg_node(pRExC_state, BRANCH);
6162 Set_Node_Length(ret, 1);
6166 if (!first && SIZE_ONLY)
6167 RExC_extralen += 1; /* BRANCHJ */
6169 *flagp = WORST; /* Tentatively. */
6172 nextchar(pRExC_state);
6173 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6175 latest = regpiece(pRExC_state, &flags,depth+1);
6176 if (latest == NULL) {
6177 if (flags & TRYAGAIN)
6181 else if (ret == NULL)
6183 *flagp |= flags&(HASWIDTH|POSTPONED);
6184 if (chain == NULL) /* First piece. */
6185 *flagp |= flags&SPSTART;
6188 REGTAIL(pRExC_state, chain, latest);
6193 if (chain == NULL) { /* Loop ran zero times. */
6194 chain = reg_node(pRExC_state, NOTHING);
6199 *flagp |= flags&SIMPLE;
6206 - regpiece - something followed by possible [*+?]
6208 * Note that the branching code sequences used for ? and the general cases
6209 * of * and + are somewhat optimized: they use the same NOTHING node as
6210 * both the endmarker for their branch list and the body of the last branch.
6211 * It might seem that this node could be dispensed with entirely, but the
6212 * endmarker role is not redundant.
6215 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6218 register regnode *ret;
6220 register char *next;
6222 const char * const origparse = RExC_parse;
6224 I32 max = REG_INFTY;
6226 const char *maxpos = NULL;
6227 GET_RE_DEBUG_FLAGS_DECL;
6228 DEBUG_PARSE("piec");
6230 ret = regatom(pRExC_state, &flags,depth+1);
6232 if (flags & TRYAGAIN)
6239 if (op == '{' && regcurly(RExC_parse)) {
6241 parse_start = RExC_parse; /* MJD */
6242 next = RExC_parse + 1;
6243 while (isDIGIT(*next) || *next == ',') {
6252 if (*next == '}') { /* got one */
6256 min = atoi(RExC_parse);
6260 maxpos = RExC_parse;
6262 if (!max && *maxpos != '0')
6263 max = REG_INFTY; /* meaning "infinity" */
6264 else if (max >= REG_INFTY)
6265 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6267 nextchar(pRExC_state);
6270 if ((flags&SIMPLE)) {
6271 RExC_naughty += 2 + RExC_naughty / 2;
6272 reginsert(pRExC_state, CURLY, ret, depth+1);
6273 Set_Node_Offset(ret, parse_start+1); /* MJD */
6274 Set_Node_Cur_Length(ret);
6277 regnode * const w = reg_node(pRExC_state, WHILEM);
6280 REGTAIL(pRExC_state, ret, w);
6281 if (!SIZE_ONLY && RExC_extralen) {
6282 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6283 reginsert(pRExC_state, NOTHING,ret, depth+1);
6284 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6286 reginsert(pRExC_state, CURLYX,ret, depth+1);
6288 Set_Node_Offset(ret, parse_start+1);
6289 Set_Node_Length(ret,
6290 op == '{' ? (RExC_parse - parse_start) : 1);
6292 if (!SIZE_ONLY && RExC_extralen)
6293 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6294 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6296 RExC_whilem_seen++, RExC_extralen += 3;
6297 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6305 if (max && max < min)
6306 vFAIL("Can't do {n,m} with n > m");
6308 ARG1_SET(ret, (U16)min);
6309 ARG2_SET(ret, (U16)max);
6321 #if 0 /* Now runtime fix should be reliable. */
6323 /* if this is reinstated, don't forget to put this back into perldiag:
6325 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6327 (F) The part of the regexp subject to either the * or + quantifier
6328 could match an empty string. The {#} shows in the regular
6329 expression about where the problem was discovered.
6333 if (!(flags&HASWIDTH) && op != '?')
6334 vFAIL("Regexp *+ operand could be empty");
6337 parse_start = RExC_parse;
6338 nextchar(pRExC_state);
6340 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6342 if (op == '*' && (flags&SIMPLE)) {
6343 reginsert(pRExC_state, STAR, ret, depth+1);
6347 else if (op == '*') {
6351 else if (op == '+' && (flags&SIMPLE)) {
6352 reginsert(pRExC_state, PLUS, ret, depth+1);
6356 else if (op == '+') {
6360 else if (op == '?') {
6365 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6367 "%.*s matches null string many times",
6368 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6372 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6373 nextchar(pRExC_state);
6374 reginsert(pRExC_state, MINMOD, ret, depth+1);
6375 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6377 #ifndef REG_ALLOW_MINMOD_SUSPEND
6380 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6382 nextchar(pRExC_state);
6383 ender = reg_node(pRExC_state, SUCCEED);
6384 REGTAIL(pRExC_state, ret, ender);
6385 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6387 ender = reg_node(pRExC_state, TAIL);
6388 REGTAIL(pRExC_state, ret, ender);
6392 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6394 vFAIL("Nested quantifiers");
6401 /* reg_namedseq(pRExC_state,UVp)
6403 This is expected to be called by a parser routine that has
6404 recognized'\N' and needs to handle the rest. RExC_parse is
6405 expected to point at the first char following the N at the time
6408 If valuep is non-null then it is assumed that we are parsing inside
6409 of a charclass definition and the first codepoint in the resolved
6410 string is returned via *valuep and the routine will return NULL.
6411 In this mode if a multichar string is returned from the charnames
6412 handler a warning will be issued, and only the first char in the
6413 sequence will be examined. If the string returned is zero length
6414 then the value of *valuep is undefined and NON-NULL will
6415 be returned to indicate failure. (This will NOT be a valid pointer
6418 If value is null then it is assumed that we are parsing normal text
6419 and inserts a new EXACT node into the program containing the resolved
6420 string and returns a pointer to the new node. If the string is
6421 zerolength a NOTHING node is emitted.
6423 On success RExC_parse is set to the char following the endbrace.
6424 Parsing failures will generate a fatal errorvia vFAIL(...)
6426 NOTE: We cache all results from the charnames handler locally in
6427 the RExC_charnames hash (created on first use) to prevent a charnames
6428 handler from playing silly-buggers and returning a short string and
6429 then a long string for a given pattern. Since the regexp program
6430 size is calculated during an initial parse this would result
6431 in a buffer overrun so we cache to prevent the charname result from
6432 changing during the course of the parse.
6436 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6438 char * name; /* start of the content of the name */
6439 char * endbrace; /* endbrace following the name */
6442 STRLEN len; /* this has various purposes throughout the code */
6443 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6444 regnode *ret = NULL;
6446 if (*RExC_parse != '{') {
6447 vFAIL("Missing braces on \\N{}");
6449 name = RExC_parse+1;
6450 endbrace = strchr(RExC_parse, '}');
6453 vFAIL("Missing right brace on \\N{}");
6455 RExC_parse = endbrace + 1;
6458 /* RExC_parse points at the beginning brace,
6459 endbrace points at the last */
6460 if ( name[0]=='U' && name[1]=='+' ) {
6461 /* its a "Unicode hex" notation {U+89AB} */
6462 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6463 | PERL_SCAN_DISALLOW_PREFIX
6464 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6466 len = (STRLEN)(endbrace - name - 2);
6467 cp = grok_hex(name + 2, &len, &fl, NULL);
6468 if ( len != (STRLEN)(endbrace - name - 2) ) {
6477 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6479 /* fetch the charnames handler for this scope */
6480 HV * const table = GvHV(PL_hintgv);
6482 hv_fetchs(table, "charnames", FALSE) :
6484 SV *cv= cvp ? *cvp : NULL;
6487 /* create an SV with the name as argument */
6488 sv_name = newSVpvn(name, endbrace - name);
6490 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6491 vFAIL2("Constant(\\N{%s}) unknown: "
6492 "(possibly a missing \"use charnames ...\")",
6495 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6496 vFAIL2("Constant(\\N{%s}): "
6497 "$^H{charnames} is not defined",SvPVX(sv_name));
6502 if (!RExC_charnames) {
6503 /* make sure our cache is allocated */
6504 RExC_charnames = newHV();
6505 sv_2mortal((SV*)RExC_charnames);
6507 /* see if we have looked this one up before */
6508 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6510 sv_str = HeVAL(he_str);
6523 count= call_sv(cv, G_SCALAR);
6525 if (count == 1) { /* XXXX is this right? dmq */
6527 SvREFCNT_inc_simple_void(sv_str);
6535 if ( !sv_str || !SvOK(sv_str) ) {
6536 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6537 "did not return a defined value",SvPVX(sv_name));
6539 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6544 char *p = SvPV(sv_str, len);
6547 if ( SvUTF8(sv_str) ) {
6548 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6552 We have to turn on utf8 for high bit chars otherwise
6553 we get failures with
6555 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6556 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6558 This is different from what \x{} would do with the same
6559 codepoint, where the condition is > 0xFF.
6566 /* warn if we havent used the whole string? */
6568 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6570 "Ignoring excess chars from \\N{%s} in character class",
6574 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6576 "Ignoring zero length \\N{%s} in character class",
6581 SvREFCNT_dec(sv_name);
6583 SvREFCNT_dec(sv_str);
6584 return len ? NULL : (regnode *)&len;
6585 } else if(SvCUR(sv_str)) {
6591 char * parse_start = name-3; /* needed for the offsets */
6593 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6595 ret = reg_node(pRExC_state,
6596 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6599 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6600 sv_utf8_upgrade(sv_str);
6601 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6605 p = SvPV(sv_str, len);
6607 /* len is the length written, charlen is the size the char read */
6608 for ( len = 0; p < pend; p += charlen ) {
6610 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6612 STRLEN foldlen,numlen;
6613 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6614 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6615 /* Emit all the Unicode characters. */
6617 for (foldbuf = tmpbuf;
6621 uvc = utf8_to_uvchr(foldbuf, &numlen);
6623 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6626 /* In EBCDIC the numlen
6627 * and unilen can differ. */
6629 if (numlen >= foldlen)
6633 break; /* "Can't happen." */
6636 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6648 RExC_size += STR_SZ(len);
6651 RExC_emit += STR_SZ(len);
6653 Set_Node_Cur_Length(ret); /* MJD */
6655 nextchar(pRExC_state);
6657 ret = reg_node(pRExC_state,NOTHING);
6660 SvREFCNT_dec(sv_str);
6663 SvREFCNT_dec(sv_name);
6673 * It returns the code point in utf8 for the value in *encp.
6674 * value: a code value in the source encoding
6675 * encp: a pointer to an Encode object
6677 * If the result from Encode is not a single character,
6678 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6681 S_reg_recode(pTHX_ const char value, SV **encp)
6684 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6685 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6686 const STRLEN newlen = SvCUR(sv);
6687 UV uv = UNICODE_REPLACEMENT;
6691 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6694 if (!newlen || numlen != newlen) {
6695 uv = UNICODE_REPLACEMENT;
6703 - regatom - the lowest level
6705 Try to identify anything special at the start of the pattern. If there
6706 is, then handle it as required. This may involve generating a single regop,
6707 such as for an assertion; or it may involve recursing, such as to
6708 handle a () structure.
6710 If the string doesn't start with something special then we gobble up
6711 as much literal text as we can.
6713 Once we have been able to handle whatever type of thing started the
6714 sequence, we return.
6716 Note: we have to be careful with escapes, as they can be both literal
6717 and special, and in the case of \10 and friends can either, depending
6718 on context. Specifically there are two seperate switches for handling
6719 escape sequences, with the one for handling literal escapes requiring
6720 a dummy entry for all of the special escapes that are actually handled
6725 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6728 register regnode *ret = NULL;
6730 char *parse_start = RExC_parse;
6731 GET_RE_DEBUG_FLAGS_DECL;
6732 DEBUG_PARSE("atom");
6733 *flagp = WORST; /* Tentatively. */
6737 switch ((U8)*RExC_parse) {
6739 RExC_seen_zerolen++;
6740 nextchar(pRExC_state);
6741 if (RExC_flags & RXf_PMf_MULTILINE)
6742 ret = reg_node(pRExC_state, MBOL);
6743 else if (RExC_flags & RXf_PMf_SINGLELINE)
6744 ret = reg_node(pRExC_state, SBOL);
6746 ret = reg_node(pRExC_state, BOL);
6747 Set_Node_Length(ret, 1); /* MJD */
6750 nextchar(pRExC_state);
6752 RExC_seen_zerolen++;
6753 if (RExC_flags & RXf_PMf_MULTILINE)
6754 ret = reg_node(pRExC_state, MEOL);
6755 else if (RExC_flags & RXf_PMf_SINGLELINE)
6756 ret = reg_node(pRExC_state, SEOL);
6758 ret = reg_node(pRExC_state, EOL);
6759 Set_Node_Length(ret, 1); /* MJD */
6762 nextchar(pRExC_state);
6763 if (RExC_flags & RXf_PMf_SINGLELINE)
6764 ret = reg_node(pRExC_state, SANY);
6766 ret = reg_node(pRExC_state, REG_ANY);
6767 *flagp |= HASWIDTH|SIMPLE;
6769 Set_Node_Length(ret, 1); /* MJD */
6773 char * const oregcomp_parse = ++RExC_parse;
6774 ret = regclass(pRExC_state,depth+1);
6775 if (*RExC_parse != ']') {
6776 RExC_parse = oregcomp_parse;
6777 vFAIL("Unmatched [");
6779 nextchar(pRExC_state);
6780 *flagp |= HASWIDTH|SIMPLE;
6781 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6785 nextchar(pRExC_state);
6786 ret = reg(pRExC_state, 1, &flags,depth+1);
6788 if (flags & TRYAGAIN) {
6789 if (RExC_parse == RExC_end) {
6790 /* Make parent create an empty node if needed. */
6798 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6802 if (flags & TRYAGAIN) {
6806 vFAIL("Internal urp");
6807 /* Supposed to be caught earlier. */
6810 if (!regcurly(RExC_parse)) {
6819 vFAIL("Quantifier follows nothing");
6826 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
6827 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
6828 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
6829 ret = reganode(pRExC_state, FOLDCHAR, cp);
6830 Set_Node_Length(ret, 1); /* MJD */
6831 nextchar(pRExC_state); /* kill whitespace under /x */
6839 This switch handles escape sequences that resolve to some kind
6840 of special regop and not to literal text. Escape sequnces that
6841 resolve to literal text are handled below in the switch marked
6844 Every entry in this switch *must* have a corresponding entry
6845 in the literal escape switch. However, the opposite is not
6846 required, as the default for this switch is to jump to the
6847 literal text handling code.
6849 switch (*++RExC_parse) {
6850 /* Special Escapes */
6852 RExC_seen_zerolen++;
6853 ret = reg_node(pRExC_state, SBOL);
6855 goto finish_meta_pat;
6857 ret = reg_node(pRExC_state, GPOS);
6858 RExC_seen |= REG_SEEN_GPOS;
6860 goto finish_meta_pat;
6862 RExC_seen_zerolen++;
6863 ret = reg_node(pRExC_state, KEEPS);
6865 goto finish_meta_pat;
6867 ret = reg_node(pRExC_state, SEOL);
6869 RExC_seen_zerolen++; /* Do not optimize RE away */
6870 goto finish_meta_pat;
6872 ret = reg_node(pRExC_state, EOS);
6874 RExC_seen_zerolen++; /* Do not optimize RE away */
6875 goto finish_meta_pat;
6877 ret = reg_node(pRExC_state, CANY);
6878 RExC_seen |= REG_SEEN_CANY;
6879 *flagp |= HASWIDTH|SIMPLE;
6880 goto finish_meta_pat;
6882 ret = reg_node(pRExC_state, CLUMP);
6884 goto finish_meta_pat;
6886 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6887 *flagp |= HASWIDTH|SIMPLE;
6888 goto finish_meta_pat;
6890 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6891 *flagp |= HASWIDTH|SIMPLE;
6892 goto finish_meta_pat;
6894 RExC_seen_zerolen++;
6895 RExC_seen |= REG_SEEN_LOOKBEHIND;
6896 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6898 goto finish_meta_pat;
6900 RExC_seen_zerolen++;
6901 RExC_seen |= REG_SEEN_LOOKBEHIND;
6902 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6904 goto finish_meta_pat;
6906 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6907 *flagp |= HASWIDTH|SIMPLE;
6908 goto finish_meta_pat;
6910 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6911 *flagp |= HASWIDTH|SIMPLE;
6912 goto finish_meta_pat;
6914 ret = reg_node(pRExC_state, DIGIT);
6915 *flagp |= HASWIDTH|SIMPLE;
6916 goto finish_meta_pat;
6918 ret = reg_node(pRExC_state, NDIGIT);
6919 *flagp |= HASWIDTH|SIMPLE;
6920 goto finish_meta_pat;
6922 ret = reg_node(pRExC_state, LNBREAK);
6923 *flagp |= HASWIDTH|SIMPLE;
6924 goto finish_meta_pat;
6926 ret = reg_node(pRExC_state, HORIZWS);
6927 *flagp |= HASWIDTH|SIMPLE;
6928 goto finish_meta_pat;
6930 ret = reg_node(pRExC_state, NHORIZWS);
6931 *flagp |= HASWIDTH|SIMPLE;
6932 goto finish_meta_pat;
6934 ret = reg_node(pRExC_state, VERTWS);
6935 *flagp |= HASWIDTH|SIMPLE;
6936 goto finish_meta_pat;
6938 ret = reg_node(pRExC_state, NVERTWS);
6939 *flagp |= HASWIDTH|SIMPLE;
6941 nextchar(pRExC_state);
6942 Set_Node_Length(ret, 2); /* MJD */
6947 char* const oldregxend = RExC_end;
6949 char* parse_start = RExC_parse - 2;
6952 if (RExC_parse[1] == '{') {
6953 /* a lovely hack--pretend we saw [\pX] instead */
6954 RExC_end = strchr(RExC_parse, '}');
6956 const U8 c = (U8)*RExC_parse;
6958 RExC_end = oldregxend;
6959 vFAIL2("Missing right brace on \\%c{}", c);
6964 RExC_end = RExC_parse + 2;
6965 if (RExC_end > oldregxend)
6966 RExC_end = oldregxend;
6970 ret = regclass(pRExC_state,depth+1);
6972 RExC_end = oldregxend;
6975 Set_Node_Offset(ret, parse_start + 2);
6976 Set_Node_Cur_Length(ret);
6977 nextchar(pRExC_state);
6978 *flagp |= HASWIDTH|SIMPLE;
6982 /* Handle \N{NAME} here and not below because it can be
6983 multicharacter. join_exact() will join them up later on.
6984 Also this makes sure that things like /\N{BLAH}+/ and
6985 \N{BLAH} being multi char Just Happen. dmq*/
6987 ret= reg_namedseq(pRExC_state, NULL);
6989 case 'k': /* Handle \k<NAME> and \k'NAME' */
6992 char ch= RExC_parse[1];
6993 if (ch != '<' && ch != '\'' && ch != '{') {
6995 vFAIL2("Sequence %.2s... not terminated",parse_start);
6997 /* this pretty much dupes the code for (?P=...) in reg(), if
6998 you change this make sure you change that */
6999 char* name_start = (RExC_parse += 2);
7001 SV *sv_dat = reg_scan_name(pRExC_state,
7002 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7003 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7004 if (RExC_parse == name_start || *RExC_parse != ch)
7005 vFAIL2("Sequence %.3s... not terminated",parse_start);
7008 num = add_data( pRExC_state, 1, "S" );
7009 RExC_rxi->data->data[num]=(void*)sv_dat;
7010 SvREFCNT_inc_simple_void(sv_dat);
7014 ret = reganode(pRExC_state,
7015 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7019 /* override incorrect value set in reganode MJD */
7020 Set_Node_Offset(ret, parse_start+1);
7021 Set_Node_Cur_Length(ret); /* MJD */
7022 nextchar(pRExC_state);
7028 case '1': case '2': case '3': case '4':
7029 case '5': case '6': case '7': case '8': case '9':
7032 bool isg = *RExC_parse == 'g';
7037 if (*RExC_parse == '{') {
7041 if (*RExC_parse == '-') {
7045 if (hasbrace && !isDIGIT(*RExC_parse)) {
7046 if (isrel) RExC_parse--;
7048 goto parse_named_seq;
7050 num = atoi(RExC_parse);
7052 num = RExC_npar - num;
7054 vFAIL("Reference to nonexistent or unclosed group");
7056 if (!isg && num > 9 && num >= RExC_npar)
7059 char * const parse_start = RExC_parse - 1; /* MJD */
7060 while (isDIGIT(*RExC_parse))
7062 if (parse_start == RExC_parse - 1)
7063 vFAIL("Unterminated \\g... pattern");
7065 if (*RExC_parse != '}')
7066 vFAIL("Unterminated \\g{...} pattern");
7070 if (num > (I32)RExC_rx->nparens)
7071 vFAIL("Reference to nonexistent group");
7074 ret = reganode(pRExC_state,
7075 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7079 /* override incorrect value set in reganode MJD */
7080 Set_Node_Offset(ret, parse_start+1);
7081 Set_Node_Cur_Length(ret); /* MJD */
7083 nextchar(pRExC_state);
7088 if (RExC_parse >= RExC_end)
7089 FAIL("Trailing \\");
7092 /* Do not generate "unrecognized" warnings here, we fall
7093 back into the quick-grab loop below */
7100 if (RExC_flags & RXf_PMf_EXTENDED) {
7101 if ( reg_skipcomment( pRExC_state ) )
7108 register STRLEN len;
7113 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7115 parse_start = RExC_parse - 1;
7121 ret = reg_node(pRExC_state,
7122 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7124 for (len = 0, p = RExC_parse - 1;
7125 len < 127 && p < RExC_end;
7128 char * const oldp = p;
7130 if (RExC_flags & RXf_PMf_EXTENDED)
7131 p = regwhite( pRExC_state, p );
7136 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7137 goto normal_default;
7147 /* Literal Escapes Switch
7149 This switch is meant to handle escape sequences that
7150 resolve to a literal character.
7152 Every escape sequence that represents something
7153 else, like an assertion or a char class, is handled
7154 in the switch marked 'Special Escapes' above in this
7155 routine, but also has an entry here as anything that
7156 isn't explicitly mentioned here will be treated as
7157 an unescaped equivalent literal.
7161 /* These are all the special escapes. */
7162 case 'A': /* Start assertion */
7163 case 'b': case 'B': /* Word-boundary assertion*/
7164 case 'C': /* Single char !DANGEROUS! */
7165 case 'd': case 'D': /* digit class */
7166 case 'g': case 'G': /* generic-backref, pos assertion */
7167 case 'h': case 'H': /* HORIZWS */
7168 case 'k': case 'K': /* named backref, keep marker */
7169 case 'N': /* named char sequence */
7170 case 'p': case 'P': /* Unicode property */
7171 case 'R': /* LNBREAK */
7172 case 's': case 'S': /* space class */
7173 case 'v': case 'V': /* VERTWS */
7174 case 'w': case 'W': /* word class */
7175 case 'X': /* eXtended Unicode "combining character sequence" */
7176 case 'z': case 'Z': /* End of line/string assertion */
7180 /* Anything after here is an escape that resolves to a
7181 literal. (Except digits, which may or may not)
7200 ender = ASCII_TO_NATIVE('\033');
7204 ender = ASCII_TO_NATIVE('\007');
7209 char* const e = strchr(p, '}');
7213 vFAIL("Missing right brace on \\x{}");
7216 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7217 | PERL_SCAN_DISALLOW_PREFIX;
7218 STRLEN numlen = e - p - 1;
7219 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7226 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7228 ender = grok_hex(p, &numlen, &flags, NULL);
7231 if (PL_encoding && ender < 0x100)
7232 goto recode_encoding;
7236 ender = UCHARAT(p++);
7237 ender = toCTRL(ender);
7239 case '0': case '1': case '2': case '3':case '4':
7240 case '5': case '6': case '7': case '8':case '9':
7242 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7245 ender = grok_oct(p, &numlen, &flags, NULL);
7252 if (PL_encoding && ender < 0x100)
7253 goto recode_encoding;
7257 SV* enc = PL_encoding;
7258 ender = reg_recode((const char)(U8)ender, &enc);
7259 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7260 vWARN(p, "Invalid escape in the specified encoding");
7266 FAIL("Trailing \\");
7269 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
7270 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7271 goto normal_default;
7276 if (UTF8_IS_START(*p) && UTF) {
7278 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7279 &numlen, UTF8_ALLOW_DEFAULT);
7286 if ( RExC_flags & RXf_PMf_EXTENDED)
7287 p = regwhite( pRExC_state, p );
7289 /* Prime the casefolded buffer. */
7290 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7292 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7297 /* Emit all the Unicode characters. */
7299 for (foldbuf = tmpbuf;
7301 foldlen -= numlen) {
7302 ender = utf8_to_uvchr(foldbuf, &numlen);
7304 const STRLEN unilen = reguni(pRExC_state, ender, s);
7307 /* In EBCDIC the numlen
7308 * and unilen can differ. */
7310 if (numlen >= foldlen)
7314 break; /* "Can't happen." */
7318 const STRLEN unilen = reguni(pRExC_state, ender, s);
7327 REGC((char)ender, s++);
7333 /* Emit all the Unicode characters. */
7335 for (foldbuf = tmpbuf;
7337 foldlen -= numlen) {
7338 ender = utf8_to_uvchr(foldbuf, &numlen);
7340 const STRLEN unilen = reguni(pRExC_state, ender, s);
7343 /* In EBCDIC the numlen
7344 * and unilen can differ. */
7346 if (numlen >= foldlen)
7354 const STRLEN unilen = reguni(pRExC_state, ender, s);
7363 REGC((char)ender, s++);
7367 Set_Node_Cur_Length(ret); /* MJD */
7368 nextchar(pRExC_state);
7370 /* len is STRLEN which is unsigned, need to copy to signed */
7373 vFAIL("Internal disaster");
7377 if (len == 1 && UNI_IS_INVARIANT(ender))
7381 RExC_size += STR_SZ(len);
7384 RExC_emit += STR_SZ(len);
7394 S_regwhite( RExC_state_t *pRExC_state, char *p )
7396 const char *e = RExC_end;
7400 else if (*p == '#') {
7409 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7417 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7418 Character classes ([:foo:]) can also be negated ([:^foo:]).
7419 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7420 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7421 but trigger failures because they are currently unimplemented. */
7423 #define POSIXCC_DONE(c) ((c) == ':')
7424 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7425 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7428 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7431 I32 namedclass = OOB_NAMEDCLASS;
7433 if (value == '[' && RExC_parse + 1 < RExC_end &&
7434 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7435 POSIXCC(UCHARAT(RExC_parse))) {
7436 const char c = UCHARAT(RExC_parse);
7437 char* const s = RExC_parse++;
7439 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7441 if (RExC_parse == RExC_end)
7442 /* Grandfather lone [:, [=, [. */
7445 const char* const t = RExC_parse++; /* skip over the c */
7448 if (UCHARAT(RExC_parse) == ']') {
7449 const char *posixcc = s + 1;
7450 RExC_parse++; /* skip over the ending ] */
7453 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7454 const I32 skip = t - posixcc;
7456 /* Initially switch on the length of the name. */
7459 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7460 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7463 /* Names all of length 5. */
7464 /* alnum alpha ascii blank cntrl digit graph lower
7465 print punct space upper */
7466 /* Offset 4 gives the best switch position. */
7467 switch (posixcc[4]) {
7469 if (memEQ(posixcc, "alph", 4)) /* alpha */
7470 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7473 if (memEQ(posixcc, "spac", 4)) /* space */
7474 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7477 if (memEQ(posixcc, "grap", 4)) /* graph */
7478 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7481 if (memEQ(posixcc, "asci", 4)) /* ascii */
7482 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7485 if (memEQ(posixcc, "blan", 4)) /* blank */
7486 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7489 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7490 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7493 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7494 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7497 if (memEQ(posixcc, "lowe", 4)) /* lower */
7498 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7499 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7500 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7503 if (memEQ(posixcc, "digi", 4)) /* digit */
7504 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7505 else if (memEQ(posixcc, "prin", 4)) /* print */
7506 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7507 else if (memEQ(posixcc, "punc", 4)) /* punct */
7508 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7513 if (memEQ(posixcc, "xdigit", 6))
7514 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7518 if (namedclass == OOB_NAMEDCLASS)
7519 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7521 assert (posixcc[skip] == ':');
7522 assert (posixcc[skip+1] == ']');
7523 } else if (!SIZE_ONLY) {
7524 /* [[=foo=]] and [[.foo.]] are still future. */
7526 /* adjust RExC_parse so the warning shows after
7528 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7530 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7533 /* Maternal grandfather:
7534 * "[:" ending in ":" but not in ":]" */
7544 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7547 if (POSIXCC(UCHARAT(RExC_parse))) {
7548 const char *s = RExC_parse;
7549 const char c = *s++;
7553 if (*s && c == *s && s[1] == ']') {
7554 if (ckWARN(WARN_REGEXP))
7556 "POSIX syntax [%c %c] belongs inside character classes",
7559 /* [[=foo=]] and [[.foo.]] are still future. */
7560 if (POSIXCC_NOTYET(c)) {
7561 /* adjust RExC_parse so the error shows after
7563 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7565 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7572 #define _C_C_T_(NAME,TEST,WORD) \
7575 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7577 for (value = 0; value < 256; value++) \
7579 ANYOF_BITMAP_SET(ret, value); \
7584 case ANYOF_N##NAME: \
7586 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7588 for (value = 0; value < 256; value++) \
7590 ANYOF_BITMAP_SET(ret, value); \
7596 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7598 for (value = 0; value < 256; value++) \
7600 ANYOF_BITMAP_SET(ret, value); \
7604 case ANYOF_N##NAME: \
7605 for (value = 0; value < 256; value++) \
7607 ANYOF_BITMAP_SET(ret, value); \
7613 parse a class specification and produce either an ANYOF node that
7614 matches the pattern or if the pattern matches a single char only and
7615 that char is < 256 and we are case insensitive then we produce an
7620 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7623 register UV nextvalue;
7624 register IV prevvalue = OOB_UNICODE;
7625 register IV range = 0;
7626 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7627 register regnode *ret;
7630 char *rangebegin = NULL;
7631 bool need_class = 0;
7634 bool optimize_invert = TRUE;
7635 AV* unicode_alternate = NULL;
7637 UV literal_endpoint = 0;
7639 UV stored = 0; /* number of chars stored in the class */
7641 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7642 case we need to change the emitted regop to an EXACT. */
7643 const char * orig_parse = RExC_parse;
7644 GET_RE_DEBUG_FLAGS_DECL;
7646 PERL_UNUSED_ARG(depth);
7649 DEBUG_PARSE("clas");
7651 /* Assume we are going to generate an ANYOF node. */
7652 ret = reganode(pRExC_state, ANYOF, 0);
7655 ANYOF_FLAGS(ret) = 0;
7657 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7661 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7665 RExC_size += ANYOF_SKIP;
7666 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7669 RExC_emit += ANYOF_SKIP;
7671 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7673 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7674 ANYOF_BITMAP_ZERO(ret);
7675 listsv = newSVpvs("# comment\n");
7678 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7680 if (!SIZE_ONLY && POSIXCC(nextvalue))
7681 checkposixcc(pRExC_state);
7683 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7684 if (UCHARAT(RExC_parse) == ']')
7688 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7692 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7695 rangebegin = RExC_parse;
7697 value = utf8n_to_uvchr((U8*)RExC_parse,
7698 RExC_end - RExC_parse,
7699 &numlen, UTF8_ALLOW_DEFAULT);
7700 RExC_parse += numlen;
7703 value = UCHARAT(RExC_parse++);
7705 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7706 if (value == '[' && POSIXCC(nextvalue))
7707 namedclass = regpposixcc(pRExC_state, value);
7708 else if (value == '\\') {
7710 value = utf8n_to_uvchr((U8*)RExC_parse,
7711 RExC_end - RExC_parse,
7712 &numlen, UTF8_ALLOW_DEFAULT);
7713 RExC_parse += numlen;
7716 value = UCHARAT(RExC_parse++);
7717 /* Some compilers cannot handle switching on 64-bit integer
7718 * values, therefore value cannot be an UV. Yes, this will
7719 * be a problem later if we want switch on Unicode.
7720 * A similar issue a little bit later when switching on
7721 * namedclass. --jhi */
7722 switch ((I32)value) {
7723 case 'w': namedclass = ANYOF_ALNUM; break;
7724 case 'W': namedclass = ANYOF_NALNUM; break;
7725 case 's': namedclass = ANYOF_SPACE; break;
7726 case 'S': namedclass = ANYOF_NSPACE; break;
7727 case 'd': namedclass = ANYOF_DIGIT; break;
7728 case 'D': namedclass = ANYOF_NDIGIT; break;
7729 case 'v': namedclass = ANYOF_VERTWS; break;
7730 case 'V': namedclass = ANYOF_NVERTWS; break;
7731 case 'h': namedclass = ANYOF_HORIZWS; break;
7732 case 'H': namedclass = ANYOF_NHORIZWS; break;
7733 case 'N': /* Handle \N{NAME} in class */
7735 /* We only pay attention to the first char of
7736 multichar strings being returned. I kinda wonder
7737 if this makes sense as it does change the behaviour
7738 from earlier versions, OTOH that behaviour was broken
7740 UV v; /* value is register so we cant & it /grrr */
7741 if (reg_namedseq(pRExC_state, &v)) {
7751 if (RExC_parse >= RExC_end)
7752 vFAIL2("Empty \\%c{}", (U8)value);
7753 if (*RExC_parse == '{') {
7754 const U8 c = (U8)value;
7755 e = strchr(RExC_parse++, '}');
7757 vFAIL2("Missing right brace on \\%c{}", c);
7758 while (isSPACE(UCHARAT(RExC_parse)))
7760 if (e == RExC_parse)
7761 vFAIL2("Empty \\%c{}", c);
7763 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7771 if (UCHARAT(RExC_parse) == '^') {
7774 value = value == 'p' ? 'P' : 'p'; /* toggle */
7775 while (isSPACE(UCHARAT(RExC_parse))) {
7780 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7781 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7784 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7785 namedclass = ANYOF_MAX; /* no official name, but it's named */
7788 case 'n': value = '\n'; break;
7789 case 'r': value = '\r'; break;
7790 case 't': value = '\t'; break;
7791 case 'f': value = '\f'; break;
7792 case 'b': value = '\b'; break;
7793 case 'e': value = ASCII_TO_NATIVE('\033');break;
7794 case 'a': value = ASCII_TO_NATIVE('\007');break;
7796 if (*RExC_parse == '{') {
7797 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7798 | PERL_SCAN_DISALLOW_PREFIX;
7799 char * const e = strchr(RExC_parse++, '}');
7801 vFAIL("Missing right brace on \\x{}");
7803 numlen = e - RExC_parse;
7804 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7808 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7810 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7811 RExC_parse += numlen;
7813 if (PL_encoding && value < 0x100)
7814 goto recode_encoding;
7817 value = UCHARAT(RExC_parse++);
7818 value = toCTRL(value);
7820 case '0': case '1': case '2': case '3': case '4':
7821 case '5': case '6': case '7': case '8': case '9':
7825 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7826 RExC_parse += numlen;
7827 if (PL_encoding && value < 0x100)
7828 goto recode_encoding;
7833 SV* enc = PL_encoding;
7834 value = reg_recode((const char)(U8)value, &enc);
7835 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7837 "Invalid escape in the specified encoding");
7841 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7843 "Unrecognized escape \\%c in character class passed through",
7847 } /* end of \blah */
7853 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7855 if (!SIZE_ONLY && !need_class)
7856 ANYOF_CLASS_ZERO(ret);
7860 /* a bad range like a-\d, a-[:digit:] ? */
7863 if (ckWARN(WARN_REGEXP)) {
7865 RExC_parse >= rangebegin ?
7866 RExC_parse - rangebegin : 0;
7868 "False [] range \"%*.*s\"",
7871 if (prevvalue < 256) {
7872 ANYOF_BITMAP_SET(ret, prevvalue);
7873 ANYOF_BITMAP_SET(ret, '-');
7876 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7877 Perl_sv_catpvf(aTHX_ listsv,
7878 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7882 range = 0; /* this was not a true range */
7888 const char *what = NULL;
7891 if (namedclass > OOB_NAMEDCLASS)
7892 optimize_invert = FALSE;
7893 /* Possible truncation here but in some 64-bit environments
7894 * the compiler gets heartburn about switch on 64-bit values.
7895 * A similar issue a little earlier when switching on value.
7897 switch ((I32)namedclass) {
7898 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7899 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7900 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7901 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7902 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7903 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7904 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7905 case _C_C_T_(PRINT, isPRINT(value), "Print");
7906 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7907 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7908 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7909 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7910 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7911 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
7912 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
7915 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7918 for (value = 0; value < 128; value++)
7919 ANYOF_BITMAP_SET(ret, value);
7921 for (value = 0; value < 256; value++) {
7923 ANYOF_BITMAP_SET(ret, value);
7932 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7935 for (value = 128; value < 256; value++)
7936 ANYOF_BITMAP_SET(ret, value);
7938 for (value = 0; value < 256; value++) {
7939 if (!isASCII(value))
7940 ANYOF_BITMAP_SET(ret, value);
7949 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7951 /* consecutive digits assumed */
7952 for (value = '0'; value <= '9'; value++)
7953 ANYOF_BITMAP_SET(ret, value);
7960 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7962 /* consecutive digits assumed */
7963 for (value = 0; value < '0'; value++)
7964 ANYOF_BITMAP_SET(ret, value);
7965 for (value = '9' + 1; value < 256; value++)
7966 ANYOF_BITMAP_SET(ret, value);
7972 /* this is to handle \p and \P */
7975 vFAIL("Invalid [::] class");
7979 /* Strings such as "+utf8::isWord\n" */
7980 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7983 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7986 } /* end of namedclass \blah */
7989 if (prevvalue > (IV)value) /* b-a */ {
7990 const int w = RExC_parse - rangebegin;
7991 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7992 range = 0; /* not a valid range */
7996 prevvalue = value; /* save the beginning of the range */
7997 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7998 RExC_parse[1] != ']') {
8001 /* a bad range like \w-, [:word:]- ? */
8002 if (namedclass > OOB_NAMEDCLASS) {
8003 if (ckWARN(WARN_REGEXP)) {
8005 RExC_parse >= rangebegin ?
8006 RExC_parse - rangebegin : 0;
8008 "False [] range \"%*.*s\"",
8012 ANYOF_BITMAP_SET(ret, '-');
8014 range = 1; /* yeah, it's a range! */
8015 continue; /* but do it the next time */
8019 /* now is the next time */
8020 /*stored += (value - prevvalue + 1);*/
8022 if (prevvalue < 256) {
8023 const IV ceilvalue = value < 256 ? value : 255;
8026 /* In EBCDIC [\x89-\x91] should include
8027 * the \x8e but [i-j] should not. */
8028 if (literal_endpoint == 2 &&
8029 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8030 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8032 if (isLOWER(prevvalue)) {
8033 for (i = prevvalue; i <= ceilvalue; i++)
8035 ANYOF_BITMAP_SET(ret, i);
8037 for (i = prevvalue; i <= ceilvalue; i++)
8039 ANYOF_BITMAP_SET(ret, i);
8044 for (i = prevvalue; i <= ceilvalue; i++) {
8045 if (!ANYOF_BITMAP_TEST(ret,i)) {
8047 ANYOF_BITMAP_SET(ret, i);
8051 if (value > 255 || UTF) {
8052 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8053 const UV natvalue = NATIVE_TO_UNI(value);
8054 stored+=2; /* can't optimize this class */
8055 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8056 if (prevnatvalue < natvalue) { /* what about > ? */
8057 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8058 prevnatvalue, natvalue);
8060 else if (prevnatvalue == natvalue) {
8061 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8063 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8065 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8067 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8068 if (RExC_precomp[0] == ':' &&
8069 RExC_precomp[1] == '[' &&
8070 (f == 0xDF || f == 0x92)) {
8071 f = NATIVE_TO_UNI(f);
8074 /* If folding and foldable and a single
8075 * character, insert also the folded version
8076 * to the charclass. */
8078 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8079 if ((RExC_precomp[0] == ':' &&
8080 RExC_precomp[1] == '[' &&
8082 (value == 0xFB05 || value == 0xFB06))) ?
8083 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8084 foldlen == (STRLEN)UNISKIP(f) )
8086 if (foldlen == (STRLEN)UNISKIP(f))
8088 Perl_sv_catpvf(aTHX_ listsv,
8091 /* Any multicharacter foldings
8092 * require the following transform:
8093 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8094 * where E folds into "pq" and F folds
8095 * into "rst", all other characters
8096 * fold to single characters. We save
8097 * away these multicharacter foldings,
8098 * to be later saved as part of the
8099 * additional "s" data. */
8102 if (!unicode_alternate)
8103 unicode_alternate = newAV();
8104 sv = newSVpvn((char*)foldbuf, foldlen);
8106 av_push(unicode_alternate, sv);
8110 /* If folding and the value is one of the Greek
8111 * sigmas insert a few more sigmas to make the
8112 * folding rules of the sigmas to work right.
8113 * Note that not all the possible combinations
8114 * are handled here: some of them are handled
8115 * by the standard folding rules, and some of
8116 * them (literal or EXACTF cases) are handled
8117 * during runtime in regexec.c:S_find_byclass(). */
8118 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8119 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8120 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8121 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8122 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8124 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8125 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8126 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8131 literal_endpoint = 0;
8135 range = 0; /* this range (if it was one) is done now */
8139 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8141 RExC_size += ANYOF_CLASS_ADD_SKIP;
8143 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8149 /****** !SIZE_ONLY AFTER HERE *********/
8151 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8152 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8154 /* optimize single char class to an EXACT node
8155 but *only* when its not a UTF/high char */
8156 const char * cur_parse= RExC_parse;
8157 RExC_emit = (regnode *)orig_emit;
8158 RExC_parse = (char *)orig_parse;
8159 ret = reg_node(pRExC_state,
8160 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8161 RExC_parse = (char *)cur_parse;
8162 *STRING(ret)= (char)value;
8164 RExC_emit += STR_SZ(1);
8167 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8168 if ( /* If the only flag is folding (plus possibly inversion). */
8169 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8171 for (value = 0; value < 256; ++value) {
8172 if (ANYOF_BITMAP_TEST(ret, value)) {
8173 UV fold = PL_fold[value];
8176 ANYOF_BITMAP_SET(ret, fold);
8179 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8182 /* optimize inverted simple patterns (e.g. [^a-z]) */
8183 if (optimize_invert &&
8184 /* If the only flag is inversion. */
8185 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8186 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8187 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8188 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8191 AV * const av = newAV();
8193 /* The 0th element stores the character class description
8194 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8195 * to initialize the appropriate swash (which gets stored in
8196 * the 1st element), and also useful for dumping the regnode.
8197 * The 2nd element stores the multicharacter foldings,
8198 * used later (regexec.c:S_reginclass()). */
8199 av_store(av, 0, listsv);
8200 av_store(av, 1, NULL);
8201 av_store(av, 2, (SV*)unicode_alternate);
8202 rv = newRV_noinc((SV*)av);
8203 n = add_data(pRExC_state, 1, "s");
8204 RExC_rxi->data->data[n] = (void*)rv;
8212 /* reg_skipcomment()
8214 Absorbs an /x style # comments from the input stream.
8215 Returns true if there is more text remaining in the stream.
8216 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8217 terminates the pattern without including a newline.
8219 Note its the callers responsibility to ensure that we are
8225 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8228 while (RExC_parse < RExC_end)
8229 if (*RExC_parse++ == '\n') {
8234 /* we ran off the end of the pattern without ending
8235 the comment, so we have to add an \n when wrapping */
8236 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8244 Advance that parse position, and optionall absorbs
8245 "whitespace" from the inputstream.
8247 Without /x "whitespace" means (?#...) style comments only,
8248 with /x this means (?#...) and # comments and whitespace proper.
8250 Returns the RExC_parse point from BEFORE the scan occurs.
8252 This is the /x friendly way of saying RExC_parse++.
8256 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8258 char* const retval = RExC_parse++;
8261 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8262 RExC_parse[2] == '#') {
8263 while (*RExC_parse != ')') {
8264 if (RExC_parse == RExC_end)
8265 FAIL("Sequence (?#... not terminated");
8271 if (RExC_flags & RXf_PMf_EXTENDED) {
8272 if (isSPACE(*RExC_parse)) {
8276 else if (*RExC_parse == '#') {
8277 if ( reg_skipcomment( pRExC_state ) )
8286 - reg_node - emit a node
8288 STATIC regnode * /* Location. */
8289 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8292 register regnode *ptr;
8293 regnode * const ret = RExC_emit;
8294 GET_RE_DEBUG_FLAGS_DECL;
8297 SIZE_ALIGN(RExC_size);
8301 if (RExC_emit >= RExC_emit_bound)
8302 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8304 NODE_ALIGN_FILL(ret);
8306 FILL_ADVANCE_NODE(ptr, op);
8307 #ifdef RE_TRACK_PATTERN_OFFSETS
8308 if (RExC_offsets) { /* MJD */
8309 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8310 "reg_node", __LINE__,
8312 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8313 ? "Overwriting end of array!\n" : "OK",
8314 (UV)(RExC_emit - RExC_emit_start),
8315 (UV)(RExC_parse - RExC_start),
8316 (UV)RExC_offsets[0]));
8317 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8325 - reganode - emit a node with an argument
8327 STATIC regnode * /* Location. */
8328 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8331 register regnode *ptr;
8332 regnode * const ret = RExC_emit;
8333 GET_RE_DEBUG_FLAGS_DECL;
8336 SIZE_ALIGN(RExC_size);
8341 assert(2==regarglen[op]+1);
8343 Anything larger than this has to allocate the extra amount.
8344 If we changed this to be:
8346 RExC_size += (1 + regarglen[op]);
8348 then it wouldn't matter. Its not clear what side effect
8349 might come from that so its not done so far.
8354 if (RExC_emit >= RExC_emit_bound)
8355 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8357 NODE_ALIGN_FILL(ret);
8359 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8360 #ifdef RE_TRACK_PATTERN_OFFSETS
8361 if (RExC_offsets) { /* MJD */
8362 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8366 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8367 "Overwriting end of array!\n" : "OK",
8368 (UV)(RExC_emit - RExC_emit_start),
8369 (UV)(RExC_parse - RExC_start),
8370 (UV)RExC_offsets[0]));
8371 Set_Cur_Node_Offset;
8379 - reguni - emit (if appropriate) a Unicode character
8382 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8385 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8389 - reginsert - insert an operator in front of already-emitted operand
8391 * Means relocating the operand.
8394 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8397 register regnode *src;
8398 register regnode *dst;
8399 register regnode *place;
8400 const int offset = regarglen[(U8)op];
8401 const int size = NODE_STEP_REGNODE + offset;
8402 GET_RE_DEBUG_FLAGS_DECL;
8403 PERL_UNUSED_ARG(depth);
8404 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8405 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8414 if (RExC_open_parens) {
8416 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8417 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8418 if ( RExC_open_parens[paren] >= opnd ) {
8419 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8420 RExC_open_parens[paren] += size;
8422 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8424 if ( RExC_close_parens[paren] >= opnd ) {
8425 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8426 RExC_close_parens[paren] += size;
8428 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8433 while (src > opnd) {
8434 StructCopy(--src, --dst, regnode);
8435 #ifdef RE_TRACK_PATTERN_OFFSETS
8436 if (RExC_offsets) { /* MJD 20010112 */
8437 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8441 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8442 ? "Overwriting end of array!\n" : "OK",
8443 (UV)(src - RExC_emit_start),
8444 (UV)(dst - RExC_emit_start),
8445 (UV)RExC_offsets[0]));
8446 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8447 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8453 place = opnd; /* Op node, where operand used to be. */
8454 #ifdef RE_TRACK_PATTERN_OFFSETS
8455 if (RExC_offsets) { /* MJD */
8456 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8460 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8461 ? "Overwriting end of array!\n" : "OK",
8462 (UV)(place - RExC_emit_start),
8463 (UV)(RExC_parse - RExC_start),
8464 (UV)RExC_offsets[0]));
8465 Set_Node_Offset(place, RExC_parse);
8466 Set_Node_Length(place, 1);
8469 src = NEXTOPER(place);
8470 FILL_ADVANCE_NODE(place, op);
8471 Zero(src, offset, regnode);
8475 - regtail - set the next-pointer at the end of a node chain of p to val.
8476 - SEE ALSO: regtail_study
8478 /* TODO: All three parms should be const */
8480 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8483 register regnode *scan;
8484 GET_RE_DEBUG_FLAGS_DECL;
8486 PERL_UNUSED_ARG(depth);
8492 /* Find last node. */
8495 regnode * const temp = regnext(scan);
8497 SV * const mysv=sv_newmortal();
8498 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8499 regprop(RExC_rx, mysv, scan);
8500 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8501 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8502 (temp == NULL ? "->" : ""),
8503 (temp == NULL ? PL_reg_name[OP(val)] : "")
8511 if (reg_off_by_arg[OP(scan)]) {
8512 ARG_SET(scan, val - scan);
8515 NEXT_OFF(scan) = val - scan;
8521 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8522 - Look for optimizable sequences at the same time.
8523 - currently only looks for EXACT chains.
8525 This is expermental code. The idea is to use this routine to perform
8526 in place optimizations on branches and groups as they are constructed,
8527 with the long term intention of removing optimization from study_chunk so
8528 that it is purely analytical.
8530 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8531 to control which is which.
8534 /* TODO: All four parms should be const */
8537 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8540 register regnode *scan;
8542 #ifdef EXPERIMENTAL_INPLACESCAN
8546 GET_RE_DEBUG_FLAGS_DECL;
8552 /* Find last node. */
8556 regnode * const temp = regnext(scan);
8557 #ifdef EXPERIMENTAL_INPLACESCAN
8558 if (PL_regkind[OP(scan)] == EXACT)
8559 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8567 if( exact == PSEUDO )
8569 else if ( exact != OP(scan) )
8578 SV * const mysv=sv_newmortal();
8579 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8580 regprop(RExC_rx, mysv, scan);
8581 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8582 SvPV_nolen_const(mysv),
8584 PL_reg_name[exact]);
8591 SV * const mysv_val=sv_newmortal();
8592 DEBUG_PARSE_MSG("");
8593 regprop(RExC_rx, mysv_val, val);
8594 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8595 SvPV_nolen_const(mysv_val),
8596 (IV)REG_NODE_NUM(val),
8600 if (reg_off_by_arg[OP(scan)]) {
8601 ARG_SET(scan, val - scan);
8604 NEXT_OFF(scan) = val - scan;
8612 - regcurly - a little FSA that accepts {\d+,?\d*}
8615 S_regcurly(register const char *s)
8634 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8637 Perl_regdump(pTHX_ const regexp *r)
8641 SV * const sv = sv_newmortal();
8642 SV *dsv= sv_newmortal();
8645 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8647 /* Header fields of interest. */
8648 if (r->anchored_substr) {
8649 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8650 RE_SV_DUMPLEN(r->anchored_substr), 30);
8651 PerlIO_printf(Perl_debug_log,
8652 "anchored %s%s at %"IVdf" ",
8653 s, RE_SV_TAIL(r->anchored_substr),
8654 (IV)r->anchored_offset);
8655 } else if (r->anchored_utf8) {
8656 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8657 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8658 PerlIO_printf(Perl_debug_log,
8659 "anchored utf8 %s%s at %"IVdf" ",
8660 s, RE_SV_TAIL(r->anchored_utf8),
8661 (IV)r->anchored_offset);
8663 if (r->float_substr) {
8664 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8665 RE_SV_DUMPLEN(r->float_substr), 30);
8666 PerlIO_printf(Perl_debug_log,
8667 "floating %s%s at %"IVdf"..%"UVuf" ",
8668 s, RE_SV_TAIL(r->float_substr),
8669 (IV)r->float_min_offset, (UV)r->float_max_offset);
8670 } else if (r->float_utf8) {
8671 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8672 RE_SV_DUMPLEN(r->float_utf8), 30);
8673 PerlIO_printf(Perl_debug_log,
8674 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8675 s, RE_SV_TAIL(r->float_utf8),
8676 (IV)r->float_min_offset, (UV)r->float_max_offset);
8678 if (r->check_substr || r->check_utf8)
8679 PerlIO_printf(Perl_debug_log,
8681 (r->check_substr == r->float_substr
8682 && r->check_utf8 == r->float_utf8
8683 ? "(checking floating" : "(checking anchored"));
8684 if (r->extflags & RXf_NOSCAN)
8685 PerlIO_printf(Perl_debug_log, " noscan");
8686 if (r->extflags & RXf_CHECK_ALL)
8687 PerlIO_printf(Perl_debug_log, " isall");
8688 if (r->check_substr || r->check_utf8)
8689 PerlIO_printf(Perl_debug_log, ") ");
8691 if (ri->regstclass) {
8692 regprop(r, sv, ri->regstclass);
8693 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8695 if (r->extflags & RXf_ANCH) {
8696 PerlIO_printf(Perl_debug_log, "anchored");
8697 if (r->extflags & RXf_ANCH_BOL)
8698 PerlIO_printf(Perl_debug_log, "(BOL)");
8699 if (r->extflags & RXf_ANCH_MBOL)
8700 PerlIO_printf(Perl_debug_log, "(MBOL)");
8701 if (r->extflags & RXf_ANCH_SBOL)
8702 PerlIO_printf(Perl_debug_log, "(SBOL)");
8703 if (r->extflags & RXf_ANCH_GPOS)
8704 PerlIO_printf(Perl_debug_log, "(GPOS)");
8705 PerlIO_putc(Perl_debug_log, ' ');
8707 if (r->extflags & RXf_GPOS_SEEN)
8708 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8709 if (r->intflags & PREGf_SKIP)
8710 PerlIO_printf(Perl_debug_log, "plus ");
8711 if (r->intflags & PREGf_IMPLICIT)
8712 PerlIO_printf(Perl_debug_log, "implicit ");
8713 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8714 if (r->extflags & RXf_EVAL_SEEN)
8715 PerlIO_printf(Perl_debug_log, "with eval ");
8716 PerlIO_printf(Perl_debug_log, "\n");
8718 PERL_UNUSED_CONTEXT;
8720 #endif /* DEBUGGING */
8724 - regprop - printable representation of opcode
8727 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8732 RXi_GET_DECL(prog,progi);
8733 GET_RE_DEBUG_FLAGS_DECL;
8736 sv_setpvn(sv, "", 0);
8738 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8739 /* It would be nice to FAIL() here, but this may be called from
8740 regexec.c, and it would be hard to supply pRExC_state. */
8741 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8742 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
8744 k = PL_regkind[OP(o)];
8747 SV * const dsv = sv_2mortal(newSVpvs(""));
8748 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8749 * is a crude hack but it may be the best for now since
8750 * we have no flag "this EXACTish node was UTF-8"
8752 const char * const s =
8753 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8754 PL_colors[0], PL_colors[1],
8755 PERL_PV_ESCAPE_UNI_DETECT |
8756 PERL_PV_PRETTY_ELIPSES |
8759 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8760 } else if (k == TRIE) {
8761 /* print the details of the trie in dumpuntil instead, as
8762 * progi->data isn't available here */
8763 const char op = OP(o);
8764 const U32 n = ARG(o);
8765 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8766 (reg_ac_data *)progi->data->data[n] :
8768 const reg_trie_data * const trie
8769 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8771 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
8772 DEBUG_TRIE_COMPILE_r(
8773 Perl_sv_catpvf(aTHX_ sv,
8774 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8775 (UV)trie->startstate,
8776 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8777 (UV)trie->wordcount,
8780 (UV)TRIE_CHARCOUNT(trie),
8781 (UV)trie->uniquecharcount
8784 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8786 int rangestart = -1;
8787 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8788 Perl_sv_catpvf(aTHX_ sv, "[");
8789 for (i = 0; i <= 256; i++) {
8790 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8791 if (rangestart == -1)
8793 } else if (rangestart != -1) {
8794 if (i <= rangestart + 3)
8795 for (; rangestart < i; rangestart++)
8796 put_byte(sv, rangestart);
8798 put_byte(sv, rangestart);
8800 put_byte(sv, i - 1);
8805 Perl_sv_catpvf(aTHX_ sv, "]");
8808 } else if (k == CURLY) {
8809 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8810 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8811 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8813 else if (k == WHILEM && o->flags) /* Ordinal/of */
8814 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8815 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8816 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8817 if ( prog->paren_names ) {
8818 if ( k != REF || OP(o) < NREF) {
8819 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8820 SV **name= av_fetch(list, ARG(o), 0 );
8822 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8825 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8826 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8827 I32 *nums=(I32*)SvPVX(sv_dat);
8828 SV **name= av_fetch(list, nums[0], 0 );
8831 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8832 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8833 (n ? "," : ""), (IV)nums[n]);
8835 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8839 } else if (k == GOSUB)
8840 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8841 else if (k == VERB) {
8843 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8844 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8845 } else if (k == LOGICAL)
8846 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8847 else if (k == FOLDCHAR)
8848 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]",ARG(o) );
8849 else if (k == ANYOF) {
8850 int i, rangestart = -1;
8851 const U8 flags = ANYOF_FLAGS(o);
8853 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8854 static const char * const anyofs[] = {
8887 if (flags & ANYOF_LOCALE)
8888 sv_catpvs(sv, "{loc}");
8889 if (flags & ANYOF_FOLD)
8890 sv_catpvs(sv, "{i}");
8891 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8892 if (flags & ANYOF_INVERT)
8894 for (i = 0; i <= 256; i++) {
8895 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8896 if (rangestart == -1)
8898 } else if (rangestart != -1) {
8899 if (i <= rangestart + 3)
8900 for (; rangestart < i; rangestart++)
8901 put_byte(sv, rangestart);
8903 put_byte(sv, rangestart);
8905 put_byte(sv, i - 1);
8911 if (o->flags & ANYOF_CLASS)
8912 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8913 if (ANYOF_CLASS_TEST(o,i))
8914 sv_catpv(sv, anyofs[i]);
8916 if (flags & ANYOF_UNICODE)
8917 sv_catpvs(sv, "{unicode}");
8918 else if (flags & ANYOF_UNICODE_ALL)
8919 sv_catpvs(sv, "{unicode_all}");
8923 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8927 U8 s[UTF8_MAXBYTES_CASE+1];
8929 for (i = 0; i <= 256; i++) { /* just the first 256 */
8930 uvchr_to_utf8(s, i);
8932 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8933 if (rangestart == -1)
8935 } else if (rangestart != -1) {
8936 if (i <= rangestart + 3)
8937 for (; rangestart < i; rangestart++) {
8938 const U8 * const e = uvchr_to_utf8(s,rangestart);
8940 for(p = s; p < e; p++)
8944 const U8 *e = uvchr_to_utf8(s,rangestart);
8946 for (p = s; p < e; p++)
8949 e = uvchr_to_utf8(s, i-1);
8950 for (p = s; p < e; p++)
8957 sv_catpvs(sv, "..."); /* et cetera */
8961 char *s = savesvpv(lv);
8962 char * const origs = s;
8964 while (*s && *s != '\n')
8968 const char * const t = ++s;
8986 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8988 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8989 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8991 PERL_UNUSED_CONTEXT;
8992 PERL_UNUSED_ARG(sv);
8994 PERL_UNUSED_ARG(prog);
8995 #endif /* DEBUGGING */
8999 Perl_re_intuit_string(pTHX_ REGEXP * const prog)
9000 { /* Assume that RE_INTUIT is set */
9002 GET_RE_DEBUG_FLAGS_DECL;
9003 PERL_UNUSED_CONTEXT;
9007 const char * const s = SvPV_nolen_const(prog->check_substr
9008 ? prog->check_substr : prog->check_utf8);
9010 if (!PL_colorset) reginitcolors();
9011 PerlIO_printf(Perl_debug_log,
9012 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9014 prog->check_substr ? "" : "utf8 ",
9015 PL_colors[5],PL_colors[0],
9018 (strlen(s) > 60 ? "..." : ""));
9021 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9027 handles refcounting and freeing the perl core regexp structure. When
9028 it is necessary to actually free the structure the first thing it
9029 does is call the 'free' method of the regexp_engine associated to to
9030 the regexp, allowing the handling of the void *pprivate; member
9031 first. (This routine is not overridable by extensions, which is why
9032 the extensions free is called first.)
9034 See regdupe and regdupe_internal if you change anything here.
9036 #ifndef PERL_IN_XSUB_RE
9038 Perl_pregfree(pTHX_ struct regexp *r)
9041 GET_RE_DEBUG_FLAGS_DECL;
9043 if (!r || (--r->refcnt > 0))
9046 ReREFCNT_dec(r->mother_re);
9048 CALLREGFREE_PVT(r); /* free the private data */
9050 SvREFCNT_dec(r->paren_names);
9051 Safefree(r->wrapped);
9054 if (r->anchored_substr)
9055 SvREFCNT_dec(r->anchored_substr);
9056 if (r->anchored_utf8)
9057 SvREFCNT_dec(r->anchored_utf8);
9058 if (r->float_substr)
9059 SvREFCNT_dec(r->float_substr);
9061 SvREFCNT_dec(r->float_utf8);
9062 Safefree(r->substrs);
9064 RX_MATCH_COPY_FREE(r);
9065 #ifdef PERL_OLD_COPY_ON_WRITE
9067 SvREFCNT_dec(r->saved_copy);
9076 This is a hacky workaround to the structural issue of match results
9077 being stored in the regexp structure which is in turn stored in
9078 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9079 could be PL_curpm in multiple contexts, and could require multiple
9080 result sets being associated with the pattern simultaneously, such
9081 as when doing a recursive match with (??{$qr})
9083 The solution is to make a lightweight copy of the regexp structure
9084 when a qr// is returned from the code executed by (??{$qr}) this
9085 lightweight copy doesnt actually own any of its data except for
9086 the starp/end and the actual regexp structure itself.
9092 Perl_reg_temp_copy (pTHX_ struct regexp *r) {
9094 register const I32 npar = r->nparens+1;
9095 (void)ReREFCNT_inc(r);
9096 Newx(ret, 1, regexp);
9097 StructCopy(r, ret, regexp);
9098 Newx(ret->offs, npar, regexp_paren_pair);
9099 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9102 Newx(ret->substrs, 1, struct reg_substr_data);
9103 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9105 SvREFCNT_inc_void(ret->anchored_substr);
9106 SvREFCNT_inc_void(ret->anchored_utf8);
9107 SvREFCNT_inc_void(ret->float_substr);
9108 SvREFCNT_inc_void(ret->float_utf8);
9110 /* check_substr and check_utf8, if non-NULL, point to either their
9111 anchored or float namesakes, and don't hold a second reference. */
9113 RX_MATCH_COPIED_off(ret);
9114 #ifdef PERL_OLD_COPY_ON_WRITE
9115 ret->saved_copy = NULL;
9124 /* regfree_internal()
9126 Free the private data in a regexp. This is overloadable by
9127 extensions. Perl takes care of the regexp structure in pregfree(),
9128 this covers the *pprivate pointer which technically perldoesnt
9129 know about, however of course we have to handle the
9130 regexp_internal structure when no extension is in use.
9132 Note this is called before freeing anything in the regexp
9137 Perl_regfree_internal(pTHX_ REGEXP * const r)
9141 GET_RE_DEBUG_FLAGS_DECL;
9147 SV *dsv= sv_newmortal();
9148 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
9149 dsv, r->precomp, r->prelen, 60);
9150 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9151 PL_colors[4],PL_colors[5],s);
9154 #ifdef RE_TRACK_PATTERN_OFFSETS
9156 Safefree(ri->u.offsets); /* 20010421 MJD */
9159 int n = ri->data->count;
9160 PAD* new_comppad = NULL;
9165 /* If you add a ->what type here, update the comment in regcomp.h */
9166 switch (ri->data->what[n]) {
9170 SvREFCNT_dec((SV*)ri->data->data[n]);
9173 Safefree(ri->data->data[n]);
9176 new_comppad = (AV*)ri->data->data[n];
9179 if (new_comppad == NULL)
9180 Perl_croak(aTHX_ "panic: pregfree comppad");
9181 PAD_SAVE_LOCAL(old_comppad,
9182 /* Watch out for global destruction's random ordering. */
9183 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9186 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9189 op_free((OP_4tree*)ri->data->data[n]);
9191 PAD_RESTORE_LOCAL(old_comppad);
9192 SvREFCNT_dec((SV*)new_comppad);
9198 { /* Aho Corasick add-on structure for a trie node.
9199 Used in stclass optimization only */
9201 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9203 refcount = --aho->refcount;
9206 PerlMemShared_free(aho->states);
9207 PerlMemShared_free(aho->fail);
9208 /* do this last!!!! */
9209 PerlMemShared_free(ri->data->data[n]);
9210 PerlMemShared_free(ri->regstclass);
9216 /* trie structure. */
9218 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9220 refcount = --trie->refcount;
9223 PerlMemShared_free(trie->charmap);
9224 PerlMemShared_free(trie->states);
9225 PerlMemShared_free(trie->trans);
9227 PerlMemShared_free(trie->bitmap);
9229 PerlMemShared_free(trie->wordlen);
9231 PerlMemShared_free(trie->jump);
9233 PerlMemShared_free(trie->nextword);
9234 /* do this last!!!! */
9235 PerlMemShared_free(ri->data->data[n]);
9240 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9243 Safefree(ri->data->what);
9250 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
9251 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9252 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9253 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9256 re_dup - duplicate a regexp.
9258 This routine is expected to clone a given regexp structure. It is not
9259 compiler under USE_ITHREADS.
9261 After all of the core data stored in struct regexp is duplicated
9262 the regexp_engine.dupe method is used to copy any private data
9263 stored in the *pprivate pointer. This allows extensions to handle
9264 any duplication it needs to do.
9266 See pregfree() and regfree_internal() if you change anything here.
9268 #if defined(USE_ITHREADS)
9269 #ifndef PERL_IN_XSUB_RE
9271 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
9278 return (REGEXP *)NULL;
9280 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
9284 npar = r->nparens+1;
9285 Newx(ret, 1, regexp);
9286 StructCopy(r, ret, regexp);
9287 Newx(ret->offs, npar, regexp_paren_pair);
9288 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9290 /* no need to copy these */
9291 Newx(ret->swap, npar, regexp_paren_pair);
9295 /* Do it this way to avoid reading from *r after the StructCopy().
9296 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9297 cache, it doesn't matter. */
9298 const bool anchored = r->check_substr == r->anchored_substr;
9299 Newx(ret->substrs, 1, struct reg_substr_data);
9300 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9302 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9303 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9304 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9305 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9307 /* check_substr and check_utf8, if non-NULL, point to either their
9308 anchored or float namesakes, and don't hold a second reference. */
9310 if (ret->check_substr) {
9312 assert(r->check_utf8 == r->anchored_utf8);
9313 ret->check_substr = ret->anchored_substr;
9314 ret->check_utf8 = ret->anchored_utf8;
9316 assert(r->check_substr == r->float_substr);
9317 assert(r->check_utf8 == r->float_utf8);
9318 ret->check_substr = ret->float_substr;
9319 ret->check_utf8 = ret->float_utf8;
9324 ret->wrapped = SAVEPVN(ret->wrapped, ret->wraplen+1);
9325 ret->precomp = ret->wrapped + (ret->precomp - ret->wrapped);
9326 ret->paren_names = hv_dup_inc(ret->paren_names, param);
9329 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
9331 if (RX_MATCH_COPIED(ret))
9332 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9335 #ifdef PERL_OLD_COPY_ON_WRITE
9336 ret->saved_copy = NULL;
9339 ret->mother_re = NULL;
9341 ret->seen_evals = 0;
9343 ptr_table_store(PL_ptr_table, r, ret);
9346 #endif /* PERL_IN_XSUB_RE */
9351 This is the internal complement to regdupe() which is used to copy
9352 the structure pointed to by the *pprivate pointer in the regexp.
9353 This is the core version of the extension overridable cloning hook.
9354 The regexp structure being duplicated will be copied by perl prior
9355 to this and will be provided as the regexp *r argument, however
9356 with the /old/ structures pprivate pointer value. Thus this routine
9357 may override any copying normally done by perl.
9359 It returns a pointer to the new regexp_internal structure.
9363 Perl_regdupe_internal(pTHX_ REGEXP * const r, CLONE_PARAMS *param)
9366 regexp_internal *reti;
9370 npar = r->nparens+1;
9373 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
9374 Copy(ri->program, reti->program, len+1, regnode);
9377 reti->regstclass = NULL;
9381 const int count = ri->data->count;
9384 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9385 char, struct reg_data);
9386 Newx(d->what, count, U8);
9389 for (i = 0; i < count; i++) {
9390 d->what[i] = ri->data->what[i];
9391 switch (d->what[i]) {
9392 /* legal options are one of: sSfpontTu
9393 see also regcomp.h and pregfree() */
9396 case 'p': /* actually an AV, but the dup function is identical. */
9397 case 'u': /* actually an HV, but the dup function is identical. */
9398 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
9401 /* This is cheating. */
9402 Newx(d->data[i], 1, struct regnode_charclass_class);
9403 StructCopy(ri->data->data[i], d->data[i],
9404 struct regnode_charclass_class);
9405 reti->regstclass = (regnode*)d->data[i];
9408 /* Compiled op trees are readonly and in shared memory,
9409 and can thus be shared without duplication. */
9411 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9415 /* Trie stclasses are readonly and can thus be shared
9416 * without duplication. We free the stclass in pregfree
9417 * when the corresponding reg_ac_data struct is freed.
9419 reti->regstclass= ri->regstclass;
9423 ((reg_trie_data*)ri->data->data[i])->refcount++;
9427 d->data[i] = ri->data->data[i];
9430 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9439 reti->name_list_idx = ri->name_list_idx;
9441 #ifdef RE_TRACK_PATTERN_OFFSETS
9442 if (ri->u.offsets) {
9443 Newx(reti->u.offsets, 2*len+1, U32);
9444 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9447 SetProgLen(reti,len);
9453 #endif /* USE_ITHREADS */
9458 converts a regexp embedded in a MAGIC struct to its stringified form,
9459 caching the converted form in the struct and returns the cached
9462 If lp is nonnull then it is used to return the length of the
9465 If flags is nonnull and the returned string contains UTF8 then
9466 (*flags & 1) will be true.
9468 If haseval is nonnull then it is used to return whether the pattern
9471 Normally called via macro:
9473 CALLREG_STRINGIFY(mg,&len,&utf8);
9477 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9479 See sv_2pv_flags() in sv.c for an example of internal usage.
9482 #ifndef PERL_IN_XSUB_RE
9485 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9487 const regexp * const re = (regexp *)mg->mg_obj;
9489 *haseval = re->seen_evals;
9491 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
9498 - regnext - dig the "next" pointer out of a node
9501 Perl_regnext(pTHX_ register regnode *p)
9504 register I32 offset;
9509 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9518 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9521 STRLEN l1 = strlen(pat1);
9522 STRLEN l2 = strlen(pat2);
9525 const char *message;
9531 Copy(pat1, buf, l1 , char);
9532 Copy(pat2, buf + l1, l2 , char);
9533 buf[l1 + l2] = '\n';
9534 buf[l1 + l2 + 1] = '\0';
9536 /* ANSI variant takes additional second argument */
9537 va_start(args, pat2);
9541 msv = vmess(buf, &args);
9543 message = SvPV_const(msv,l1);
9546 Copy(message, buf, l1 , char);
9547 buf[l1-1] = '\0'; /* Overwrite \n */
9548 Perl_croak(aTHX_ "%s", buf);
9551 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9553 #ifndef PERL_IN_XSUB_RE
9555 Perl_save_re_context(pTHX)
9559 struct re_save_state *state;
9561 SAVEVPTR(PL_curcop);
9562 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9564 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9565 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9566 SSPUSHINT(SAVEt_RE_STATE);
9568 Copy(&PL_reg_state, state, 1, struct re_save_state);
9570 PL_reg_start_tmp = 0;
9571 PL_reg_start_tmpl = 0;
9572 PL_reg_oldsaved = NULL;
9573 PL_reg_oldsavedlen = 0;
9575 PL_reg_leftiter = 0;
9576 PL_reg_poscache = NULL;
9577 PL_reg_poscache_size = 0;
9578 #ifdef PERL_OLD_COPY_ON_WRITE
9582 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9584 const REGEXP * const rx = PM_GETRE(PL_curpm);
9587 for (i = 1; i <= rx->nparens; i++) {
9588 char digits[TYPE_CHARS(long)];
9589 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9590 GV *const *const gvp
9591 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9594 GV * const gv = *gvp;
9595 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9605 clear_re(pTHX_ void *r)
9608 ReREFCNT_dec((regexp *)r);
9614 S_put_byte(pTHX_ SV *sv, int c)
9616 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9617 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9618 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9619 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9621 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9625 #define CLEAR_OPTSTART \
9626 if (optstart) STMT_START { \
9627 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9631 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9633 STATIC const regnode *
9634 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9635 const regnode *last, const regnode *plast,
9636 SV* sv, I32 indent, U32 depth)
9639 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9640 register const regnode *next;
9641 const regnode *optstart= NULL;
9644 GET_RE_DEBUG_FLAGS_DECL;
9646 #ifdef DEBUG_DUMPUNTIL
9647 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9648 last ? last-start : 0,plast ? plast-start : 0);
9651 if (plast && plast < last)
9654 while (PL_regkind[op] != END && (!last || node < last)) {
9655 /* While that wasn't END last time... */
9658 if (op == CLOSE || op == WHILEM)
9660 next = regnext((regnode *)node);
9663 if (OP(node) == OPTIMIZED) {
9664 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9671 regprop(r, sv, node);
9672 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9673 (int)(2*indent + 1), "", SvPVX_const(sv));
9675 if (OP(node) != OPTIMIZED) {
9676 if (next == NULL) /* Next ptr. */
9677 PerlIO_printf(Perl_debug_log, " (0)");
9678 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9679 PerlIO_printf(Perl_debug_log, " (FAIL)");
9681 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9682 (void)PerlIO_putc(Perl_debug_log, '\n');
9686 if (PL_regkind[(U8)op] == BRANCHJ) {
9689 register const regnode *nnode = (OP(next) == LONGJMP
9690 ? regnext((regnode *)next)
9692 if (last && nnode > last)
9694 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9697 else if (PL_regkind[(U8)op] == BRANCH) {
9699 DUMPUNTIL(NEXTOPER(node), next);
9701 else if ( PL_regkind[(U8)op] == TRIE ) {
9702 const regnode *this_trie = node;
9703 const char op = OP(node);
9704 const U32 n = ARG(node);
9705 const reg_ac_data * const ac = op>=AHOCORASICK ?
9706 (reg_ac_data *)ri->data->data[n] :
9708 const reg_trie_data * const trie =
9709 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9711 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9713 const regnode *nextbranch= NULL;
9715 sv_setpvn(sv, "", 0);
9716 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9717 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9719 PerlIO_printf(Perl_debug_log, "%*s%s ",
9720 (int)(2*(indent+3)), "",
9721 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9722 PL_colors[0], PL_colors[1],
9723 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9724 PERL_PV_PRETTY_ELIPSES |
9730 U16 dist= trie->jump[word_idx+1];
9731 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9732 (UV)((dist ? this_trie + dist : next) - start));
9735 nextbranch= this_trie + trie->jump[0];
9736 DUMPUNTIL(this_trie + dist, nextbranch);
9738 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9739 nextbranch= regnext((regnode *)nextbranch);
9741 PerlIO_printf(Perl_debug_log, "\n");
9744 if (last && next > last)
9749 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9750 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9751 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9753 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9755 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9757 else if ( op == PLUS || op == STAR) {
9758 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9760 else if (op == ANYOF) {
9761 /* arglen 1 + class block */
9762 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9763 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9764 node = NEXTOPER(node);
9766 else if (PL_regkind[(U8)op] == EXACT) {
9767 /* Literal string, where present. */
9768 node += NODE_SZ_STR(node) - 1;
9769 node = NEXTOPER(node);
9772 node = NEXTOPER(node);
9773 node += regarglen[(U8)op];
9775 if (op == CURLYX || op == OPEN)
9779 #ifdef DEBUG_DUMPUNTIL
9780 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9785 #endif /* DEBUGGING */
9789 * c-indentation-style: bsd
9791 * indent-tabs-mode: t
9794 * ex: set ts=8 sts=4 sw=4 noet: