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_sv; /* The SV that is the regexp. */
106 regexp *rx; /* perl core regexp structure */
107 regexp_internal *rxi; /* internal data for regexp object pprivate field */
108 char *start; /* Start of input for compile */
109 char *end; /* End of input for compile */
110 char *parse; /* Input-scan pointer. */
111 I32 whilem_seen; /* number of WHILEM in this expr */
112 regnode *emit_start; /* Start of emitted-code area */
113 regnode *emit_bound; /* First regnode outside of the allocated space */
114 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
115 I32 naughty; /* How bad is this pattern? */
116 I32 sawback; /* Did we see \1, ...? */
118 I32 size; /* Code size. */
119 I32 npar; /* Capture buffer count, (OPEN). */
120 I32 cpar; /* Capture buffer count, (CLOSE). */
121 I32 nestroot; /* root parens we are in - used by accept */
125 regnode **open_parens; /* pointers to open parens */
126 regnode **close_parens; /* pointers to close parens */
127 regnode *opend; /* END node in program */
128 I32 utf8; /* whether the pattern is utf8 or not */
129 I32 orig_utf8; /* whether the pattern was originally in utf8 */
130 /* XXX use this for future optimisation of case
131 * where pattern must be upgraded to utf8. */
132 HV *charnames; /* cache of named sequences */
133 HV *paren_names; /* Paren names */
135 regnode **recurse; /* Recurse regops */
136 I32 recurse_count; /* Number of recurse regops */
138 char *starttry; /* -Dr: where regtry was called. */
139 #define RExC_starttry (pRExC_state->starttry)
142 const char *lastparse;
144 AV *paren_name_list; /* idx -> name */
145 #define RExC_lastparse (pRExC_state->lastparse)
146 #define RExC_lastnum (pRExC_state->lastnum)
147 #define RExC_paren_name_list (pRExC_state->paren_name_list)
151 #define RExC_flags (pRExC_state->flags)
152 #define RExC_precomp (pRExC_state->precomp)
153 #define RExC_rx_sv (pRExC_state->rx_sv)
154 #define RExC_rx (pRExC_state->rx)
155 #define RExC_rxi (pRExC_state->rxi)
156 #define RExC_start (pRExC_state->start)
157 #define RExC_end (pRExC_state->end)
158 #define RExC_parse (pRExC_state->parse)
159 #define RExC_whilem_seen (pRExC_state->whilem_seen)
160 #ifdef RE_TRACK_PATTERN_OFFSETS
161 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
163 #define RExC_emit (pRExC_state->emit)
164 #define RExC_emit_start (pRExC_state->emit_start)
165 #define RExC_emit_bound (pRExC_state->emit_bound)
166 #define RExC_naughty (pRExC_state->naughty)
167 #define RExC_sawback (pRExC_state->sawback)
168 #define RExC_seen (pRExC_state->seen)
169 #define RExC_size (pRExC_state->size)
170 #define RExC_npar (pRExC_state->npar)
171 #define RExC_nestroot (pRExC_state->nestroot)
172 #define RExC_extralen (pRExC_state->extralen)
173 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
174 #define RExC_seen_evals (pRExC_state->seen_evals)
175 #define RExC_utf8 (pRExC_state->utf8)
176 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
177 #define RExC_charnames (pRExC_state->charnames)
178 #define RExC_open_parens (pRExC_state->open_parens)
179 #define RExC_close_parens (pRExC_state->close_parens)
180 #define RExC_opend (pRExC_state->opend)
181 #define RExC_paren_names (pRExC_state->paren_names)
182 #define RExC_recurse (pRExC_state->recurse)
183 #define RExC_recurse_count (pRExC_state->recurse_count)
186 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
187 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
188 ((*s) == '{' && regcurly(s)))
191 #undef SPSTART /* dratted cpp namespace... */
194 * Flags to be passed up and down.
196 #define WORST 0 /* Worst case. */
197 #define HASWIDTH 0x01 /* Known to match non-null strings. */
198 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
199 #define SPSTART 0x04 /* Starts with * or +. */
200 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
201 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
203 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
205 /* whether trie related optimizations are enabled */
206 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
207 #define TRIE_STUDY_OPT
208 #define FULL_TRIE_STUDY
214 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
215 #define PBITVAL(paren) (1 << ((paren) & 7))
216 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
217 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
218 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
221 /* About scan_data_t.
223 During optimisation we recurse through the regexp program performing
224 various inplace (keyhole style) optimisations. In addition study_chunk
225 and scan_commit populate this data structure with information about
226 what strings MUST appear in the pattern. We look for the longest
227 string that must appear for at a fixed location, and we look for the
228 longest string that may appear at a floating location. So for instance
233 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
234 strings (because they follow a .* construct). study_chunk will identify
235 both FOO and BAR as being the longest fixed and floating strings respectively.
237 The strings can be composites, for instance
241 will result in a composite fixed substring 'foo'.
243 For each string some basic information is maintained:
245 - offset or min_offset
246 This is the position the string must appear at, or not before.
247 It also implicitly (when combined with minlenp) tells us how many
248 character must match before the string we are searching.
249 Likewise when combined with minlenp and the length of the string
250 tells us how many characters must appear after the string we have
254 Only used for floating strings. This is the rightmost point that
255 the string can appear at. Ifset to I32 max it indicates that the
256 string can occur infinitely far to the right.
259 A pointer to the minimum length of the pattern that the string
260 was found inside. This is important as in the case of positive
261 lookahead or positive lookbehind we can have multiple patterns
266 The minimum length of the pattern overall is 3, the minimum length
267 of the lookahead part is 3, but the minimum length of the part that
268 will actually match is 1. So 'FOO's minimum length is 3, but the
269 minimum length for the F is 1. This is important as the minimum length
270 is used to determine offsets in front of and behind the string being
271 looked for. Since strings can be composites this is the length of the
272 pattern at the time it was commited with a scan_commit. Note that
273 the length is calculated by study_chunk, so that the minimum lengths
274 are not known until the full pattern has been compiled, thus the
275 pointer to the value.
279 In the case of lookbehind the string being searched for can be
280 offset past the start point of the final matching string.
281 If this value was just blithely removed from the min_offset it would
282 invalidate some of the calculations for how many chars must match
283 before or after (as they are derived from min_offset and minlen and
284 the length of the string being searched for).
285 When the final pattern is compiled and the data is moved from the
286 scan_data_t structure into the regexp structure the information
287 about lookbehind is factored in, with the information that would
288 have been lost precalculated in the end_shift field for the
291 The fields pos_min and pos_delta are used to store the minimum offset
292 and the delta to the maximum offset at the current point in the pattern.
296 typedef struct scan_data_t {
297 /*I32 len_min; unused */
298 /*I32 len_delta; unused */
302 I32 last_end; /* min value, <0 unless valid. */
305 SV **longest; /* Either &l_fixed, or &l_float. */
306 SV *longest_fixed; /* longest fixed string found in pattern */
307 I32 offset_fixed; /* offset where it starts */
308 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
309 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
310 SV *longest_float; /* longest floating string found in pattern */
311 I32 offset_float_min; /* earliest point in string it can appear */
312 I32 offset_float_max; /* latest point in string it can appear */
313 I32 *minlen_float; /* pointer to the minlen relevent to the string */
314 I32 lookbehind_float; /* is the position of the string modified by LB */
318 struct regnode_charclass_class *start_class;
322 * Forward declarations for pregcomp()'s friends.
325 static const scan_data_t zero_scan_data =
326 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
328 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
329 #define SF_BEFORE_SEOL 0x0001
330 #define SF_BEFORE_MEOL 0x0002
331 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
332 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
335 # define SF_FIX_SHIFT_EOL (0+2)
336 # define SF_FL_SHIFT_EOL (0+4)
338 # define SF_FIX_SHIFT_EOL (+2)
339 # define SF_FL_SHIFT_EOL (+4)
342 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
343 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
345 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
346 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
347 #define SF_IS_INF 0x0040
348 #define SF_HAS_PAR 0x0080
349 #define SF_IN_PAR 0x0100
350 #define SF_HAS_EVAL 0x0200
351 #define SCF_DO_SUBSTR 0x0400
352 #define SCF_DO_STCLASS_AND 0x0800
353 #define SCF_DO_STCLASS_OR 0x1000
354 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
355 #define SCF_WHILEM_VISITED_POS 0x2000
357 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
358 #define SCF_SEEN_ACCEPT 0x8000
360 #define UTF (RExC_utf8 != 0)
361 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
362 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
364 #define OOB_UNICODE 12345678
365 #define OOB_NAMEDCLASS -1
367 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
368 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
371 /* length of regex to show in messages that don't mark a position within */
372 #define RegexLengthToShowInErrorMessages 127
375 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
376 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
377 * op/pragma/warn/regcomp.
379 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
380 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
382 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
385 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
386 * arg. Show regex, up to a maximum length. If it's too long, chop and add
389 #define _FAIL(code) STMT_START { \
390 const char *ellipses = ""; \
391 IV len = RExC_end - RExC_precomp; \
394 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
395 if (len > RegexLengthToShowInErrorMessages) { \
396 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
397 len = RegexLengthToShowInErrorMessages - 10; \
403 #define FAIL(msg) _FAIL( \
404 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
405 msg, (int)len, RExC_precomp, ellipses))
407 #define FAIL2(msg,arg) _FAIL( \
408 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
409 arg, (int)len, RExC_precomp, ellipses))
412 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
414 #define Simple_vFAIL(m) STMT_START { \
415 const IV offset = RExC_parse - RExC_precomp; \
416 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
417 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
421 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
423 #define vFAIL(m) STMT_START { \
425 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
430 * Like Simple_vFAIL(), but accepts two arguments.
432 #define Simple_vFAIL2(m,a1) STMT_START { \
433 const IV offset = RExC_parse - RExC_precomp; \
434 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
435 (int)offset, RExC_precomp, RExC_precomp + offset); \
439 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
441 #define vFAIL2(m,a1) STMT_START { \
443 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
444 Simple_vFAIL2(m, a1); \
449 * Like Simple_vFAIL(), but accepts three arguments.
451 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
452 const IV offset = RExC_parse - RExC_precomp; \
453 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
454 (int)offset, RExC_precomp, RExC_precomp + offset); \
458 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
460 #define vFAIL3(m,a1,a2) STMT_START { \
462 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
463 Simple_vFAIL3(m, a1, a2); \
467 * Like Simple_vFAIL(), but accepts four arguments.
469 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
470 const IV offset = RExC_parse - RExC_precomp; \
471 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
472 (int)offset, RExC_precomp, RExC_precomp + offset); \
475 #define vWARN(loc,m) STMT_START { \
476 const IV offset = loc - RExC_precomp; \
477 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
478 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
481 #define vWARNdep(loc,m) STMT_START { \
482 const IV offset = loc - RExC_precomp; \
483 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
484 "%s" REPORT_LOCATION, \
485 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
489 #define vWARN2(loc, m, a1) STMT_START { \
490 const IV offset = loc - RExC_precomp; \
491 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
492 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
495 #define vWARN3(loc, m, a1, a2) STMT_START { \
496 const IV offset = loc - RExC_precomp; \
497 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
498 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
501 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
502 const IV offset = loc - RExC_precomp; \
503 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
504 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
507 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
508 const IV offset = loc - RExC_precomp; \
509 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
510 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
514 /* Allow for side effects in s */
515 #define REGC(c,s) STMT_START { \
516 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
519 /* Macros for recording node offsets. 20001227 mjd@plover.com
520 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
521 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
522 * Element 0 holds the number n.
523 * Position is 1 indexed.
525 #ifndef RE_TRACK_PATTERN_OFFSETS
526 #define Set_Node_Offset_To_R(node,byte)
527 #define Set_Node_Offset(node,byte)
528 #define Set_Cur_Node_Offset
529 #define Set_Node_Length_To_R(node,len)
530 #define Set_Node_Length(node,len)
531 #define Set_Node_Cur_Length(node)
532 #define Node_Offset(n)
533 #define Node_Length(n)
534 #define Set_Node_Offset_Length(node,offset,len)
535 #define ProgLen(ri) ri->u.proglen
536 #define SetProgLen(ri,x) ri->u.proglen = x
538 #define ProgLen(ri) ri->u.offsets[0]
539 #define SetProgLen(ri,x) ri->u.offsets[0] = x
540 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
542 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
543 __LINE__, (int)(node), (int)(byte))); \
545 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
547 RExC_offsets[2*(node)-1] = (byte); \
552 #define Set_Node_Offset(node,byte) \
553 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
554 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
556 #define Set_Node_Length_To_R(node,len) STMT_START { \
558 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
559 __LINE__, (int)(node), (int)(len))); \
561 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
563 RExC_offsets[2*(node)] = (len); \
568 #define Set_Node_Length(node,len) \
569 Set_Node_Length_To_R((node)-RExC_emit_start, len)
570 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
571 #define Set_Node_Cur_Length(node) \
572 Set_Node_Length(node, RExC_parse - parse_start)
574 /* Get offsets and lengths */
575 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
576 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
578 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
579 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
580 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
584 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
585 #define EXPERIMENTAL_INPLACESCAN
586 #endif /*RE_TRACK_PATTERN_OFFSETS*/
588 #define DEBUG_STUDYDATA(str,data,depth) \
589 DEBUG_OPTIMISE_MORE_r(if(data){ \
590 PerlIO_printf(Perl_debug_log, \
591 "%*s" str "Pos:%"IVdf"/%"IVdf \
592 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
593 (int)(depth)*2, "", \
594 (IV)((data)->pos_min), \
595 (IV)((data)->pos_delta), \
596 (UV)((data)->flags), \
597 (IV)((data)->whilem_c), \
598 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
599 is_inf ? "INF " : "" \
601 if ((data)->last_found) \
602 PerlIO_printf(Perl_debug_log, \
603 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
604 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
605 SvPVX_const((data)->last_found), \
606 (IV)((data)->last_end), \
607 (IV)((data)->last_start_min), \
608 (IV)((data)->last_start_max), \
609 ((data)->longest && \
610 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
611 SvPVX_const((data)->longest_fixed), \
612 (IV)((data)->offset_fixed), \
613 ((data)->longest && \
614 (data)->longest==&((data)->longest_float)) ? "*" : "", \
615 SvPVX_const((data)->longest_float), \
616 (IV)((data)->offset_float_min), \
617 (IV)((data)->offset_float_max) \
619 PerlIO_printf(Perl_debug_log,"\n"); \
622 static void clear_re(pTHX_ void *r);
624 /* Mark that we cannot extend a found fixed substring at this point.
625 Update the longest found anchored substring and the longest found
626 floating substrings if needed. */
629 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
631 const STRLEN l = CHR_SVLEN(data->last_found);
632 const STRLEN old_l = CHR_SVLEN(*data->longest);
633 GET_RE_DEBUG_FLAGS_DECL;
635 PERL_ARGS_ASSERT_SCAN_COMMIT;
637 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
638 SvSetMagicSV(*data->longest, data->last_found);
639 if (*data->longest == data->longest_fixed) {
640 data->offset_fixed = l ? data->last_start_min : data->pos_min;
641 if (data->flags & SF_BEFORE_EOL)
643 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
645 data->flags &= ~SF_FIX_BEFORE_EOL;
646 data->minlen_fixed=minlenp;
647 data->lookbehind_fixed=0;
649 else { /* *data->longest == data->longest_float */
650 data->offset_float_min = l ? data->last_start_min : data->pos_min;
651 data->offset_float_max = (l
652 ? data->last_start_max
653 : data->pos_min + data->pos_delta);
654 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
655 data->offset_float_max = I32_MAX;
656 if (data->flags & SF_BEFORE_EOL)
658 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
660 data->flags &= ~SF_FL_BEFORE_EOL;
661 data->minlen_float=minlenp;
662 data->lookbehind_float=0;
665 SvCUR_set(data->last_found, 0);
667 SV * const sv = data->last_found;
668 if (SvUTF8(sv) && SvMAGICAL(sv)) {
669 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
675 data->flags &= ~SF_BEFORE_EOL;
676 DEBUG_STUDYDATA("commit: ",data,0);
679 /* Can match anything (initialization) */
681 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
683 PERL_ARGS_ASSERT_CL_ANYTHING;
685 ANYOF_CLASS_ZERO(cl);
686 ANYOF_BITMAP_SETALL(cl);
687 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
689 cl->flags |= ANYOF_LOCALE;
692 /* Can match anything (initialization) */
694 S_cl_is_anything(const struct regnode_charclass_class *cl)
698 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
700 for (value = 0; value <= ANYOF_MAX; value += 2)
701 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
703 if (!(cl->flags & ANYOF_UNICODE_ALL))
705 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
710 /* Can match anything (initialization) */
712 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
714 PERL_ARGS_ASSERT_CL_INIT;
716 Zero(cl, 1, struct regnode_charclass_class);
718 cl_anything(pRExC_state, cl);
722 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
724 PERL_ARGS_ASSERT_CL_INIT_ZERO;
726 Zero(cl, 1, struct regnode_charclass_class);
728 cl_anything(pRExC_state, cl);
730 cl->flags |= ANYOF_LOCALE;
733 /* 'And' a given class with another one. Can create false positives */
734 /* We assume that cl is not inverted */
736 S_cl_and(struct regnode_charclass_class *cl,
737 const struct regnode_charclass_class *and_with)
739 PERL_ARGS_ASSERT_CL_AND;
741 assert(and_with->type == ANYOF);
742 if (!(and_with->flags & ANYOF_CLASS)
743 && !(cl->flags & ANYOF_CLASS)
744 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
745 && !(and_with->flags & ANYOF_FOLD)
746 && !(cl->flags & ANYOF_FOLD)) {
749 if (and_with->flags & ANYOF_INVERT)
750 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
751 cl->bitmap[i] &= ~and_with->bitmap[i];
753 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
754 cl->bitmap[i] &= and_with->bitmap[i];
755 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
756 if (!(and_with->flags & ANYOF_EOS))
757 cl->flags &= ~ANYOF_EOS;
759 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
760 !(and_with->flags & ANYOF_INVERT)) {
761 cl->flags &= ~ANYOF_UNICODE_ALL;
762 cl->flags |= ANYOF_UNICODE;
763 ARG_SET(cl, ARG(and_with));
765 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
766 !(and_with->flags & ANYOF_INVERT))
767 cl->flags &= ~ANYOF_UNICODE_ALL;
768 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
769 !(and_with->flags & ANYOF_INVERT))
770 cl->flags &= ~ANYOF_UNICODE;
773 /* 'OR' a given class with another one. Can create false positives */
774 /* We assume that cl is not inverted */
776 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
778 PERL_ARGS_ASSERT_CL_OR;
780 if (or_with->flags & ANYOF_INVERT) {
782 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
783 * <= (B1 | !B2) | (CL1 | !CL2)
784 * which is wasteful if CL2 is small, but we ignore CL2:
785 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
786 * XXXX Can we handle case-fold? Unclear:
787 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
788 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
790 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
791 && !(or_with->flags & ANYOF_FOLD)
792 && !(cl->flags & ANYOF_FOLD) ) {
795 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
796 cl->bitmap[i] |= ~or_with->bitmap[i];
797 } /* XXXX: logic is complicated otherwise */
799 cl_anything(pRExC_state, cl);
802 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
803 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
804 && (!(or_with->flags & ANYOF_FOLD)
805 || (cl->flags & ANYOF_FOLD)) ) {
808 /* OR char bitmap and class bitmap separately */
809 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
810 cl->bitmap[i] |= or_with->bitmap[i];
811 if (or_with->flags & ANYOF_CLASS) {
812 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
813 cl->classflags[i] |= or_with->classflags[i];
814 cl->flags |= ANYOF_CLASS;
817 else { /* XXXX: logic is complicated, leave it along for a moment. */
818 cl_anything(pRExC_state, cl);
821 if (or_with->flags & ANYOF_EOS)
822 cl->flags |= ANYOF_EOS;
824 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
825 ARG(cl) != ARG(or_with)) {
826 cl->flags |= ANYOF_UNICODE_ALL;
827 cl->flags &= ~ANYOF_UNICODE;
829 if (or_with->flags & ANYOF_UNICODE_ALL) {
830 cl->flags |= ANYOF_UNICODE_ALL;
831 cl->flags &= ~ANYOF_UNICODE;
835 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
836 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
837 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
838 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
843 dump_trie(trie,widecharmap,revcharmap)
844 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
845 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
847 These routines dump out a trie in a somewhat readable format.
848 The _interim_ variants are used for debugging the interim
849 tables that are used to generate the final compressed
850 representation which is what dump_trie expects.
852 Part of the reason for their existance is to provide a form
853 of documentation as to how the different representations function.
858 Dumps the final compressed table form of the trie to Perl_debug_log.
859 Used for debugging make_trie().
863 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
864 AV *revcharmap, U32 depth)
867 SV *sv=sv_newmortal();
868 int colwidth= widecharmap ? 6 : 4;
869 GET_RE_DEBUG_FLAGS_DECL;
871 PERL_ARGS_ASSERT_DUMP_TRIE;
873 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
874 (int)depth * 2 + 2,"",
875 "Match","Base","Ofs" );
877 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
878 SV ** const tmp = av_fetch( revcharmap, state, 0);
880 PerlIO_printf( Perl_debug_log, "%*s",
882 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
883 PL_colors[0], PL_colors[1],
884 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
885 PERL_PV_ESCAPE_FIRSTCHAR
890 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
891 (int)depth * 2 + 2,"");
893 for( state = 0 ; state < trie->uniquecharcount ; state++ )
894 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
895 PerlIO_printf( Perl_debug_log, "\n");
897 for( state = 1 ; state < trie->statecount ; state++ ) {
898 const U32 base = trie->states[ state ].trans.base;
900 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
902 if ( trie->states[ state ].wordnum ) {
903 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
905 PerlIO_printf( Perl_debug_log, "%6s", "" );
908 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
913 while( ( base + ofs < trie->uniquecharcount ) ||
914 ( base + ofs - trie->uniquecharcount < trie->lasttrans
915 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
918 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
920 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
921 if ( ( base + ofs >= trie->uniquecharcount ) &&
922 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
923 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
925 PerlIO_printf( Perl_debug_log, "%*"UVXf,
927 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
929 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
933 PerlIO_printf( Perl_debug_log, "]");
936 PerlIO_printf( Perl_debug_log, "\n" );
940 Dumps a fully constructed but uncompressed trie in list form.
941 List tries normally only are used for construction when the number of
942 possible chars (trie->uniquecharcount) is very high.
943 Used for debugging make_trie().
946 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
947 HV *widecharmap, AV *revcharmap, U32 next_alloc,
951 SV *sv=sv_newmortal();
952 int colwidth= widecharmap ? 6 : 4;
953 GET_RE_DEBUG_FLAGS_DECL;
955 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
957 /* print out the table precompression. */
958 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
959 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
960 "------:-----+-----------------\n" );
962 for( state=1 ; state < next_alloc ; state ++ ) {
965 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
966 (int)depth * 2 + 2,"", (UV)state );
967 if ( ! trie->states[ state ].wordnum ) {
968 PerlIO_printf( Perl_debug_log, "%5s| ","");
970 PerlIO_printf( Perl_debug_log, "W%4x| ",
971 trie->states[ state ].wordnum
974 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
975 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
977 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
979 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
980 PL_colors[0], PL_colors[1],
981 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
982 PERL_PV_ESCAPE_FIRSTCHAR
984 TRIE_LIST_ITEM(state,charid).forid,
985 (UV)TRIE_LIST_ITEM(state,charid).newstate
988 PerlIO_printf(Perl_debug_log, "\n%*s| ",
989 (int)((depth * 2) + 14), "");
992 PerlIO_printf( Perl_debug_log, "\n");
997 Dumps a fully constructed but uncompressed trie in table form.
998 This is the normal DFA style state transition table, with a few
999 twists to facilitate compression later.
1000 Used for debugging make_trie().
1003 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1004 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1009 SV *sv=sv_newmortal();
1010 int colwidth= widecharmap ? 6 : 4;
1011 GET_RE_DEBUG_FLAGS_DECL;
1013 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1016 print out the table precompression so that we can do a visual check
1017 that they are identical.
1020 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1022 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1023 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1025 PerlIO_printf( Perl_debug_log, "%*s",
1027 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1028 PL_colors[0], PL_colors[1],
1029 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1030 PERL_PV_ESCAPE_FIRSTCHAR
1036 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1038 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1039 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1042 PerlIO_printf( Perl_debug_log, "\n" );
1044 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1046 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1047 (int)depth * 2 + 2,"",
1048 (UV)TRIE_NODENUM( state ) );
1050 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1051 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1053 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1055 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1057 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1058 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1060 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1061 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1068 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1069 startbranch: the first branch in the whole branch sequence
1070 first : start branch of sequence of branch-exact nodes.
1071 May be the same as startbranch
1072 last : Thing following the last branch.
1073 May be the same as tail.
1074 tail : item following the branch sequence
1075 count : words in the sequence
1076 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1077 depth : indent depth
1079 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1081 A trie is an N'ary tree where the branches are determined by digital
1082 decomposition of the key. IE, at the root node you look up the 1st character and
1083 follow that branch repeat until you find the end of the branches. Nodes can be
1084 marked as "accepting" meaning they represent a complete word. Eg:
1088 would convert into the following structure. Numbers represent states, letters
1089 following numbers represent valid transitions on the letter from that state, if
1090 the number is in square brackets it represents an accepting state, otherwise it
1091 will be in parenthesis.
1093 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1097 (1) +-i->(6)-+-s->[7]
1099 +-s->(3)-+-h->(4)-+-e->[5]
1101 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1103 This shows that when matching against the string 'hers' we will begin at state 1
1104 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1105 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1106 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1107 single traverse. We store a mapping from accepting to state to which word was
1108 matched, and then when we have multiple possibilities we try to complete the
1109 rest of the regex in the order in which they occured in the alternation.
1111 The only prior NFA like behaviour that would be changed by the TRIE support is
1112 the silent ignoring of duplicate alternations which are of the form:
1114 / (DUPE|DUPE) X? (?{ ... }) Y /x
1116 Thus EVAL blocks follwing a trie may be called a different number of times with
1117 and without the optimisation. With the optimisations dupes will be silently
1118 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1119 the following demonstrates:
1121 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1123 which prints out 'word' three times, but
1125 'words'=~/(word|word|word)(?{ print $1 })S/
1127 which doesnt print it out at all. This is due to other optimisations kicking in.
1129 Example of what happens on a structural level:
1131 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1133 1: CURLYM[1] {1,32767}(18)
1144 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1145 and should turn into:
1147 1: CURLYM[1] {1,32767}(18)
1149 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1157 Cases where tail != last would be like /(?foo|bar)baz/:
1167 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1168 and would end up looking like:
1171 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1178 d = uvuni_to_utf8_flags(d, uv, 0);
1180 is the recommended Unicode-aware way of saying
1185 #define TRIE_STORE_REVCHAR \
1188 SV *zlopp = newSV(2); \
1189 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1190 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1191 SvCUR_set(zlopp, kapow - flrbbbbb); \
1194 av_push(revcharmap, zlopp); \
1196 char ooooff = (char)uvc; \
1197 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1201 #define TRIE_READ_CHAR STMT_START { \
1205 if ( foldlen > 0 ) { \
1206 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1211 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1212 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1213 foldlen -= UNISKIP( uvc ); \
1214 scan = foldbuf + UNISKIP( uvc ); \
1217 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1227 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1228 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1229 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1230 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1232 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1233 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1234 TRIE_LIST_CUR( state )++; \
1237 #define TRIE_LIST_NEW(state) STMT_START { \
1238 Newxz( trie->states[ state ].trans.list, \
1239 4, reg_trie_trans_le ); \
1240 TRIE_LIST_CUR( state ) = 1; \
1241 TRIE_LIST_LEN( state ) = 4; \
1244 #define TRIE_HANDLE_WORD(state) STMT_START { \
1245 U16 dupe= trie->states[ state ].wordnum; \
1246 regnode * const noper_next = regnext( noper ); \
1248 if (trie->wordlen) \
1249 trie->wordlen[ curword ] = wordlen; \
1251 /* store the word for dumping */ \
1253 if (OP(noper) != NOTHING) \
1254 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1256 tmp = newSVpvn_utf8( "", 0, UTF ); \
1257 av_push( trie_words, tmp ); \
1262 if ( noper_next < tail ) { \
1264 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1265 trie->jump[curword] = (U16)(noper_next - convert); \
1267 jumper = noper_next; \
1269 nextbranch= regnext(cur); \
1273 /* So it's a dupe. This means we need to maintain a */\
1274 /* linked-list from the first to the next. */\
1275 /* we only allocate the nextword buffer when there */\
1276 /* a dupe, so first time we have to do the allocation */\
1277 if (!trie->nextword) \
1278 trie->nextword = (U16 *) \
1279 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1280 while ( trie->nextword[dupe] ) \
1281 dupe= trie->nextword[dupe]; \
1282 trie->nextword[dupe]= curword; \
1284 /* we haven't inserted this word yet. */ \
1285 trie->states[ state ].wordnum = curword; \
1290 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1291 ( ( base + charid >= ucharcount \
1292 && base + charid < ubound \
1293 && state == trie->trans[ base - ucharcount + charid ].check \
1294 && trie->trans[ base - ucharcount + charid ].next ) \
1295 ? trie->trans[ base - ucharcount + charid ].next \
1296 : ( state==1 ? special : 0 ) \
1300 #define MADE_JUMP_TRIE 2
1301 #define MADE_EXACT_TRIE 4
1304 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1307 /* first pass, loop through and scan words */
1308 reg_trie_data *trie;
1309 HV *widecharmap = NULL;
1310 AV *revcharmap = newAV();
1312 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1317 regnode *jumper = NULL;
1318 regnode *nextbranch = NULL;
1319 regnode *convert = NULL;
1320 /* we just use folder as a flag in utf8 */
1321 const U8 * const folder = ( flags == EXACTF
1323 : ( flags == EXACTFL
1330 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1331 AV *trie_words = NULL;
1332 /* along with revcharmap, this only used during construction but both are
1333 * useful during debugging so we store them in the struct when debugging.
1336 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1337 STRLEN trie_charcount=0;
1339 SV *re_trie_maxbuff;
1340 GET_RE_DEBUG_FLAGS_DECL;
1342 PERL_ARGS_ASSERT_MAKE_TRIE;
1344 PERL_UNUSED_ARG(depth);
1347 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1349 trie->startstate = 1;
1350 trie->wordcount = word_count;
1351 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1352 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1353 if (!(UTF && folder))
1354 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1356 trie_words = newAV();
1359 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1360 if (!SvIOK(re_trie_maxbuff)) {
1361 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1364 PerlIO_printf( Perl_debug_log,
1365 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1366 (int)depth * 2 + 2, "",
1367 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1368 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1372 /* Find the node we are going to overwrite */
1373 if ( first == startbranch && OP( last ) != BRANCH ) {
1374 /* whole branch chain */
1377 /* branch sub-chain */
1378 convert = NEXTOPER( first );
1381 /* -- First loop and Setup --
1383 We first traverse the branches and scan each word to determine if it
1384 contains widechars, and how many unique chars there are, this is
1385 important as we have to build a table with at least as many columns as we
1388 We use an array of integers to represent the character codes 0..255
1389 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1390 native representation of the character value as the key and IV's for the
1393 *TODO* If we keep track of how many times each character is used we can
1394 remap the columns so that the table compression later on is more
1395 efficient in terms of memory by ensuring most common value is in the
1396 middle and the least common are on the outside. IMO this would be better
1397 than a most to least common mapping as theres a decent chance the most
1398 common letter will share a node with the least common, meaning the node
1399 will not be compressable. With a middle is most common approach the worst
1400 case is when we have the least common nodes twice.
1404 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1405 regnode * const noper = NEXTOPER( cur );
1406 const U8 *uc = (U8*)STRING( noper );
1407 const U8 * const e = uc + STR_LEN( noper );
1409 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1410 const U8 *scan = (U8*)NULL;
1411 U32 wordlen = 0; /* required init */
1413 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1415 if (OP(noper) == NOTHING) {
1419 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1420 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1421 regardless of encoding */
1423 for ( ; uc < e ; uc += len ) {
1424 TRIE_CHARCOUNT(trie)++;
1428 if ( !trie->charmap[ uvc ] ) {
1429 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1431 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1435 /* store the codepoint in the bitmap, and if its ascii
1436 also store its folded equivelent. */
1437 TRIE_BITMAP_SET(trie,uvc);
1439 /* store the folded codepoint */
1440 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1443 /* store first byte of utf8 representation of
1444 codepoints in the 127 < uvc < 256 range */
1445 if (127 < uvc && uvc < 192) {
1446 TRIE_BITMAP_SET(trie,194);
1447 } else if (191 < uvc ) {
1448 TRIE_BITMAP_SET(trie,195);
1449 /* && uvc < 256 -- we know uvc is < 256 already */
1452 set_bit = 0; /* We've done our bit :-) */
1457 widecharmap = newHV();
1459 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1462 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1464 if ( !SvTRUE( *svpp ) ) {
1465 sv_setiv( *svpp, ++trie->uniquecharcount );
1470 if( cur == first ) {
1473 } else if (chars < trie->minlen) {
1475 } else if (chars > trie->maxlen) {
1479 } /* end first pass */
1480 DEBUG_TRIE_COMPILE_r(
1481 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1482 (int)depth * 2 + 2,"",
1483 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1484 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1485 (int)trie->minlen, (int)trie->maxlen )
1487 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1490 We now know what we are dealing with in terms of unique chars and
1491 string sizes so we can calculate how much memory a naive
1492 representation using a flat table will take. If it's over a reasonable
1493 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1494 conservative but potentially much slower representation using an array
1497 At the end we convert both representations into the same compressed
1498 form that will be used in regexec.c for matching with. The latter
1499 is a form that cannot be used to construct with but has memory
1500 properties similar to the list form and access properties similar
1501 to the table form making it both suitable for fast searches and
1502 small enough that its feasable to store for the duration of a program.
1504 See the comment in the code where the compressed table is produced
1505 inplace from the flat tabe representation for an explanation of how
1506 the compression works.
1511 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1513 Second Pass -- Array Of Lists Representation
1515 Each state will be represented by a list of charid:state records
1516 (reg_trie_trans_le) the first such element holds the CUR and LEN
1517 points of the allocated array. (See defines above).
1519 We build the initial structure using the lists, and then convert
1520 it into the compressed table form which allows faster lookups
1521 (but cant be modified once converted).
1524 STRLEN transcount = 1;
1526 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1527 "%*sCompiling trie using list compiler\n",
1528 (int)depth * 2 + 2, ""));
1530 trie->states = (reg_trie_state *)
1531 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1532 sizeof(reg_trie_state) );
1536 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1538 regnode * const noper = NEXTOPER( cur );
1539 U8 *uc = (U8*)STRING( noper );
1540 const U8 * const e = uc + STR_LEN( noper );
1541 U32 state = 1; /* required init */
1542 U16 charid = 0; /* sanity init */
1543 U8 *scan = (U8*)NULL; /* sanity init */
1544 STRLEN foldlen = 0; /* required init */
1545 U32 wordlen = 0; /* required init */
1546 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1548 if (OP(noper) != NOTHING) {
1549 for ( ; uc < e ; uc += len ) {
1554 charid = trie->charmap[ uvc ];
1556 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1560 charid=(U16)SvIV( *svpp );
1563 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1570 if ( !trie->states[ state ].trans.list ) {
1571 TRIE_LIST_NEW( state );
1573 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1574 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1575 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1580 newstate = next_alloc++;
1581 TRIE_LIST_PUSH( state, charid, newstate );
1586 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1590 TRIE_HANDLE_WORD(state);
1592 } /* end second pass */
1594 /* next alloc is the NEXT state to be allocated */
1595 trie->statecount = next_alloc;
1596 trie->states = (reg_trie_state *)
1597 PerlMemShared_realloc( trie->states,
1599 * sizeof(reg_trie_state) );
1601 /* and now dump it out before we compress it */
1602 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1603 revcharmap, next_alloc,
1607 trie->trans = (reg_trie_trans *)
1608 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1615 for( state=1 ; state < next_alloc ; state ++ ) {
1619 DEBUG_TRIE_COMPILE_MORE_r(
1620 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1624 if (trie->states[state].trans.list) {
1625 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1629 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1630 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1631 if ( forid < minid ) {
1633 } else if ( forid > maxid ) {
1637 if ( transcount < tp + maxid - minid + 1) {
1639 trie->trans = (reg_trie_trans *)
1640 PerlMemShared_realloc( trie->trans,
1642 * sizeof(reg_trie_trans) );
1643 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1645 base = trie->uniquecharcount + tp - minid;
1646 if ( maxid == minid ) {
1648 for ( ; zp < tp ; zp++ ) {
1649 if ( ! trie->trans[ zp ].next ) {
1650 base = trie->uniquecharcount + zp - minid;
1651 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1652 trie->trans[ zp ].check = state;
1658 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1659 trie->trans[ tp ].check = state;
1664 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1665 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1666 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1667 trie->trans[ tid ].check = state;
1669 tp += ( maxid - minid + 1 );
1671 Safefree(trie->states[ state ].trans.list);
1674 DEBUG_TRIE_COMPILE_MORE_r(
1675 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1678 trie->states[ state ].trans.base=base;
1680 trie->lasttrans = tp + 1;
1684 Second Pass -- Flat Table Representation.
1686 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1687 We know that we will need Charcount+1 trans at most to store the data
1688 (one row per char at worst case) So we preallocate both structures
1689 assuming worst case.
1691 We then construct the trie using only the .next slots of the entry
1694 We use the .check field of the first entry of the node temporarily to
1695 make compression both faster and easier by keeping track of how many non
1696 zero fields are in the node.
1698 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1701 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1702 number representing the first entry of the node, and state as a
1703 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1704 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1705 are 2 entrys per node. eg:
1713 The table is internally in the right hand, idx form. However as we also
1714 have to deal with the states array which is indexed by nodenum we have to
1715 use TRIE_NODENUM() to convert.
1718 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1719 "%*sCompiling trie using table compiler\n",
1720 (int)depth * 2 + 2, ""));
1722 trie->trans = (reg_trie_trans *)
1723 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1724 * trie->uniquecharcount + 1,
1725 sizeof(reg_trie_trans) );
1726 trie->states = (reg_trie_state *)
1727 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1728 sizeof(reg_trie_state) );
1729 next_alloc = trie->uniquecharcount + 1;
1732 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1734 regnode * const noper = NEXTOPER( cur );
1735 const U8 *uc = (U8*)STRING( noper );
1736 const U8 * const e = uc + STR_LEN( noper );
1738 U32 state = 1; /* required init */
1740 U16 charid = 0; /* sanity init */
1741 U32 accept_state = 0; /* sanity init */
1742 U8 *scan = (U8*)NULL; /* sanity init */
1744 STRLEN foldlen = 0; /* required init */
1745 U32 wordlen = 0; /* required init */
1746 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1748 if ( OP(noper) != NOTHING ) {
1749 for ( ; uc < e ; uc += len ) {
1754 charid = trie->charmap[ uvc ];
1756 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1757 charid = svpp ? (U16)SvIV(*svpp) : 0;
1761 if ( !trie->trans[ state + charid ].next ) {
1762 trie->trans[ state + charid ].next = next_alloc;
1763 trie->trans[ state ].check++;
1764 next_alloc += trie->uniquecharcount;
1766 state = trie->trans[ state + charid ].next;
1768 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1770 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1773 accept_state = TRIE_NODENUM( state );
1774 TRIE_HANDLE_WORD(accept_state);
1776 } /* end second pass */
1778 /* and now dump it out before we compress it */
1779 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1781 next_alloc, depth+1));
1785 * Inplace compress the table.*
1787 For sparse data sets the table constructed by the trie algorithm will
1788 be mostly 0/FAIL transitions or to put it another way mostly empty.
1789 (Note that leaf nodes will not contain any transitions.)
1791 This algorithm compresses the tables by eliminating most such
1792 transitions, at the cost of a modest bit of extra work during lookup:
1794 - Each states[] entry contains a .base field which indicates the
1795 index in the state[] array wheres its transition data is stored.
1797 - If .base is 0 there are no valid transitions from that node.
1799 - If .base is nonzero then charid is added to it to find an entry in
1802 -If trans[states[state].base+charid].check!=state then the
1803 transition is taken to be a 0/Fail transition. Thus if there are fail
1804 transitions at the front of the node then the .base offset will point
1805 somewhere inside the previous nodes data (or maybe even into a node
1806 even earlier), but the .check field determines if the transition is
1810 The following process inplace converts the table to the compressed
1811 table: We first do not compress the root node 1,and mark its all its
1812 .check pointers as 1 and set its .base pointer as 1 as well. This
1813 allows to do a DFA construction from the compressed table later, and
1814 ensures that any .base pointers we calculate later are greater than
1817 - We set 'pos' to indicate the first entry of the second node.
1819 - We then iterate over the columns of the node, finding the first and
1820 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1821 and set the .check pointers accordingly, and advance pos
1822 appropriately and repreat for the next node. Note that when we copy
1823 the next pointers we have to convert them from the original
1824 NODEIDX form to NODENUM form as the former is not valid post
1827 - If a node has no transitions used we mark its base as 0 and do not
1828 advance the pos pointer.
1830 - If a node only has one transition we use a second pointer into the
1831 structure to fill in allocated fail transitions from other states.
1832 This pointer is independent of the main pointer and scans forward
1833 looking for null transitions that are allocated to a state. When it
1834 finds one it writes the single transition into the "hole". If the
1835 pointer doesnt find one the single transition is appended as normal.
1837 - Once compressed we can Renew/realloc the structures to release the
1840 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1841 specifically Fig 3.47 and the associated pseudocode.
1845 const U32 laststate = TRIE_NODENUM( next_alloc );
1848 trie->statecount = laststate;
1850 for ( state = 1 ; state < laststate ; state++ ) {
1852 const U32 stateidx = TRIE_NODEIDX( state );
1853 const U32 o_used = trie->trans[ stateidx ].check;
1854 U32 used = trie->trans[ stateidx ].check;
1855 trie->trans[ stateidx ].check = 0;
1857 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1858 if ( flag || trie->trans[ stateidx + charid ].next ) {
1859 if ( trie->trans[ stateidx + charid ].next ) {
1861 for ( ; zp < pos ; zp++ ) {
1862 if ( ! trie->trans[ zp ].next ) {
1866 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1867 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1868 trie->trans[ zp ].check = state;
1869 if ( ++zp > pos ) pos = zp;
1876 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1878 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1879 trie->trans[ pos ].check = state;
1884 trie->lasttrans = pos + 1;
1885 trie->states = (reg_trie_state *)
1886 PerlMemShared_realloc( trie->states, laststate
1887 * sizeof(reg_trie_state) );
1888 DEBUG_TRIE_COMPILE_MORE_r(
1889 PerlIO_printf( Perl_debug_log,
1890 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1891 (int)depth * 2 + 2,"",
1892 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1895 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1898 } /* end table compress */
1900 DEBUG_TRIE_COMPILE_MORE_r(
1901 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1902 (int)depth * 2 + 2, "",
1903 (UV)trie->statecount,
1904 (UV)trie->lasttrans)
1906 /* resize the trans array to remove unused space */
1907 trie->trans = (reg_trie_trans *)
1908 PerlMemShared_realloc( trie->trans, trie->lasttrans
1909 * sizeof(reg_trie_trans) );
1911 /* and now dump out the compressed format */
1912 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1914 { /* Modify the program and insert the new TRIE node*/
1915 U8 nodetype =(U8)(flags & 0xFF);
1919 regnode *optimize = NULL;
1920 #ifdef RE_TRACK_PATTERN_OFFSETS
1923 U32 mjd_nodelen = 0;
1924 #endif /* RE_TRACK_PATTERN_OFFSETS */
1925 #endif /* DEBUGGING */
1927 This means we convert either the first branch or the first Exact,
1928 depending on whether the thing following (in 'last') is a branch
1929 or not and whther first is the startbranch (ie is it a sub part of
1930 the alternation or is it the whole thing.)
1931 Assuming its a sub part we conver the EXACT otherwise we convert
1932 the whole branch sequence, including the first.
1934 /* Find the node we are going to overwrite */
1935 if ( first != startbranch || OP( last ) == BRANCH ) {
1936 /* branch sub-chain */
1937 NEXT_OFF( first ) = (U16)(last - first);
1938 #ifdef RE_TRACK_PATTERN_OFFSETS
1940 mjd_offset= Node_Offset((convert));
1941 mjd_nodelen= Node_Length((convert));
1944 /* whole branch chain */
1946 #ifdef RE_TRACK_PATTERN_OFFSETS
1949 const regnode *nop = NEXTOPER( convert );
1950 mjd_offset= Node_Offset((nop));
1951 mjd_nodelen= Node_Length((nop));
1955 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1956 (int)depth * 2 + 2, "",
1957 (UV)mjd_offset, (UV)mjd_nodelen)
1960 /* But first we check to see if there is a common prefix we can
1961 split out as an EXACT and put in front of the TRIE node. */
1962 trie->startstate= 1;
1963 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1965 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1969 const U32 base = trie->states[ state ].trans.base;
1971 if ( trie->states[state].wordnum )
1974 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1975 if ( ( base + ofs >= trie->uniquecharcount ) &&
1976 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1977 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1979 if ( ++count > 1 ) {
1980 SV **tmp = av_fetch( revcharmap, ofs, 0);
1981 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1982 if ( state == 1 ) break;
1984 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1986 PerlIO_printf(Perl_debug_log,
1987 "%*sNew Start State=%"UVuf" Class: [",
1988 (int)depth * 2 + 2, "",
1991 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1992 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1994 TRIE_BITMAP_SET(trie,*ch);
1996 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1998 PerlIO_printf(Perl_debug_log, (char*)ch)
2002 TRIE_BITMAP_SET(trie,*ch);
2004 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2005 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2011 SV **tmp = av_fetch( revcharmap, idx, 0);
2013 char *ch = SvPV( *tmp, len );
2015 SV *sv=sv_newmortal();
2016 PerlIO_printf( Perl_debug_log,
2017 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2018 (int)depth * 2 + 2, "",
2020 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2021 PL_colors[0], PL_colors[1],
2022 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2023 PERL_PV_ESCAPE_FIRSTCHAR
2028 OP( convert ) = nodetype;
2029 str=STRING(convert);
2032 STR_LEN(convert) += len;
2038 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2044 regnode *n = convert+NODE_SZ_STR(convert);
2045 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2046 trie->startstate = state;
2047 trie->minlen -= (state - 1);
2048 trie->maxlen -= (state - 1);
2050 /* At least the UNICOS C compiler choked on this
2051 * being argument to DEBUG_r(), so let's just have
2054 #ifdef PERL_EXT_RE_BUILD
2060 regnode *fix = convert;
2061 U32 word = trie->wordcount;
2063 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2064 while( ++fix < n ) {
2065 Set_Node_Offset_Length(fix, 0, 0);
2068 SV ** const tmp = av_fetch( trie_words, word, 0 );
2070 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2071 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2073 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2081 NEXT_OFF(convert) = (U16)(tail - convert);
2082 DEBUG_r(optimize= n);
2088 if ( trie->maxlen ) {
2089 NEXT_OFF( convert ) = (U16)(tail - convert);
2090 ARG_SET( convert, data_slot );
2091 /* Store the offset to the first unabsorbed branch in
2092 jump[0], which is otherwise unused by the jump logic.
2093 We use this when dumping a trie and during optimisation. */
2095 trie->jump[0] = (U16)(nextbranch - convert);
2098 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2099 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2101 OP( convert ) = TRIEC;
2102 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2103 PerlMemShared_free(trie->bitmap);
2106 OP( convert ) = TRIE;
2108 /* store the type in the flags */
2109 convert->flags = nodetype;
2113 + regarglen[ OP( convert ) ];
2115 /* XXX We really should free up the resource in trie now,
2116 as we won't use them - (which resources?) dmq */
2118 /* needed for dumping*/
2119 DEBUG_r(if (optimize) {
2120 regnode *opt = convert;
2122 while ( ++opt < optimize) {
2123 Set_Node_Offset_Length(opt,0,0);
2126 Try to clean up some of the debris left after the
2129 while( optimize < jumper ) {
2130 mjd_nodelen += Node_Length((optimize));
2131 OP( optimize ) = OPTIMIZED;
2132 Set_Node_Offset_Length(optimize,0,0);
2135 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2137 } /* end node insert */
2138 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2140 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2141 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2143 SvREFCNT_dec(revcharmap);
2147 : trie->startstate>1
2153 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2155 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2157 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2158 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2161 We find the fail state for each state in the trie, this state is the longest proper
2162 suffix of the current states 'word' that is also a proper prefix of another word in our
2163 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2164 the DFA not to have to restart after its tried and failed a word at a given point, it
2165 simply continues as though it had been matching the other word in the first place.
2167 'abcdgu'=~/abcdefg|cdgu/
2168 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2169 fail, which would bring use to the state representing 'd' in the second word where we would
2170 try 'g' and succeed, prodceding to match 'cdgu'.
2172 /* add a fail transition */
2173 const U32 trie_offset = ARG(source);
2174 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2176 const U32 ucharcount = trie->uniquecharcount;
2177 const U32 numstates = trie->statecount;
2178 const U32 ubound = trie->lasttrans + ucharcount;
2182 U32 base = trie->states[ 1 ].trans.base;
2185 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2186 GET_RE_DEBUG_FLAGS_DECL;
2188 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2190 PERL_UNUSED_ARG(depth);
2194 ARG_SET( stclass, data_slot );
2195 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2196 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2197 aho->trie=trie_offset;
2198 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2199 Copy( trie->states, aho->states, numstates, reg_trie_state );
2200 Newxz( q, numstates, U32);
2201 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2204 /* initialize fail[0..1] to be 1 so that we always have
2205 a valid final fail state */
2206 fail[ 0 ] = fail[ 1 ] = 1;
2208 for ( charid = 0; charid < ucharcount ; charid++ ) {
2209 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2211 q[ q_write ] = newstate;
2212 /* set to point at the root */
2213 fail[ q[ q_write++ ] ]=1;
2216 while ( q_read < q_write) {
2217 const U32 cur = q[ q_read++ % numstates ];
2218 base = trie->states[ cur ].trans.base;
2220 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2221 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2223 U32 fail_state = cur;
2226 fail_state = fail[ fail_state ];
2227 fail_base = aho->states[ fail_state ].trans.base;
2228 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2230 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2231 fail[ ch_state ] = fail_state;
2232 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2234 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2236 q[ q_write++ % numstates] = ch_state;
2240 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2241 when we fail in state 1, this allows us to use the
2242 charclass scan to find a valid start char. This is based on the principle
2243 that theres a good chance the string being searched contains lots of stuff
2244 that cant be a start char.
2246 fail[ 0 ] = fail[ 1 ] = 0;
2247 DEBUG_TRIE_COMPILE_r({
2248 PerlIO_printf(Perl_debug_log,
2249 "%*sStclass Failtable (%"UVuf" states): 0",
2250 (int)(depth * 2), "", (UV)numstates
2252 for( q_read=1; q_read<numstates; q_read++ ) {
2253 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2255 PerlIO_printf(Perl_debug_log, "\n");
2258 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2263 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2264 * These need to be revisited when a newer toolchain becomes available.
2266 #if defined(__sparc64__) && defined(__GNUC__)
2267 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2268 # undef SPARC64_GCC_WORKAROUND
2269 # define SPARC64_GCC_WORKAROUND 1
2273 #define DEBUG_PEEP(str,scan,depth) \
2274 DEBUG_OPTIMISE_r({if (scan){ \
2275 SV * const mysv=sv_newmortal(); \
2276 regnode *Next = regnext(scan); \
2277 regprop(RExC_rx, mysv, scan); \
2278 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2279 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2280 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2287 #define JOIN_EXACT(scan,min,flags) \
2288 if (PL_regkind[OP(scan)] == EXACT) \
2289 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2292 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2293 /* Merge several consecutive EXACTish nodes into one. */
2294 regnode *n = regnext(scan);
2296 regnode *next = scan + NODE_SZ_STR(scan);
2300 regnode *stop = scan;
2301 GET_RE_DEBUG_FLAGS_DECL;
2303 PERL_UNUSED_ARG(depth);
2306 PERL_ARGS_ASSERT_JOIN_EXACT;
2307 #ifndef EXPERIMENTAL_INPLACESCAN
2308 PERL_UNUSED_ARG(flags);
2309 PERL_UNUSED_ARG(val);
2311 DEBUG_PEEP("join",scan,depth);
2313 /* Skip NOTHING, merge EXACT*. */
2315 ( PL_regkind[OP(n)] == NOTHING ||
2316 (stringok && (OP(n) == OP(scan))))
2318 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2320 if (OP(n) == TAIL || n > next)
2322 if (PL_regkind[OP(n)] == NOTHING) {
2323 DEBUG_PEEP("skip:",n,depth);
2324 NEXT_OFF(scan) += NEXT_OFF(n);
2325 next = n + NODE_STEP_REGNODE;
2332 else if (stringok) {
2333 const unsigned int oldl = STR_LEN(scan);
2334 regnode * const nnext = regnext(n);
2336 DEBUG_PEEP("merg",n,depth);
2339 if (oldl + STR_LEN(n) > U8_MAX)
2341 NEXT_OFF(scan) += NEXT_OFF(n);
2342 STR_LEN(scan) += STR_LEN(n);
2343 next = n + NODE_SZ_STR(n);
2344 /* Now we can overwrite *n : */
2345 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2353 #ifdef EXPERIMENTAL_INPLACESCAN
2354 if (flags && !NEXT_OFF(n)) {
2355 DEBUG_PEEP("atch", val, depth);
2356 if (reg_off_by_arg[OP(n)]) {
2357 ARG_SET(n, val - n);
2360 NEXT_OFF(n) = val - n;
2367 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2369 Two problematic code points in Unicode casefolding of EXACT nodes:
2371 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2372 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2378 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2379 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2381 This means that in case-insensitive matching (or "loose matching",
2382 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2383 length of the above casefolded versions) can match a target string
2384 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2385 This would rather mess up the minimum length computation.
2387 What we'll do is to look for the tail four bytes, and then peek
2388 at the preceding two bytes to see whether we need to decrease
2389 the minimum length by four (six minus two).
2391 Thanks to the design of UTF-8, there cannot be false matches:
2392 A sequence of valid UTF-8 bytes cannot be a subsequence of
2393 another valid sequence of UTF-8 bytes.
2396 char * const s0 = STRING(scan), *s, *t;
2397 char * const s1 = s0 + STR_LEN(scan) - 1;
2398 char * const s2 = s1 - 4;
2399 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2400 const char t0[] = "\xaf\x49\xaf\x42";
2402 const char t0[] = "\xcc\x88\xcc\x81";
2404 const char * const t1 = t0 + 3;
2407 s < s2 && (t = ninstr(s, s1, t0, t1));
2410 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2411 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2413 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2414 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2422 n = scan + NODE_SZ_STR(scan);
2424 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2431 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2435 /* REx optimizer. Converts nodes into quickier variants "in place".
2436 Finds fixed substrings. */
2438 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2439 to the position after last scanned or to NULL. */
2441 #define INIT_AND_WITHP \
2442 assert(!and_withp); \
2443 Newx(and_withp,1,struct regnode_charclass_class); \
2444 SAVEFREEPV(and_withp)
2446 /* this is a chain of data about sub patterns we are processing that
2447 need to be handled seperately/specially in study_chunk. Its so
2448 we can simulate recursion without losing state. */
2450 typedef struct scan_frame {
2451 regnode *last; /* last node to process in this frame */
2452 regnode *next; /* next node to process when last is reached */
2453 struct scan_frame *prev; /*previous frame*/
2454 I32 stop; /* what stopparen do we use */
2458 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2460 #define CASE_SYNST_FNC(nAmE) \
2462 if (flags & SCF_DO_STCLASS_AND) { \
2463 for (value = 0; value < 256; value++) \
2464 if (!is_ ## nAmE ## _cp(value)) \
2465 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2468 for (value = 0; value < 256; value++) \
2469 if (is_ ## nAmE ## _cp(value)) \
2470 ANYOF_BITMAP_SET(data->start_class, value); \
2474 if (flags & SCF_DO_STCLASS_AND) { \
2475 for (value = 0; value < 256; value++) \
2476 if (is_ ## nAmE ## _cp(value)) \
2477 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2480 for (value = 0; value < 256; value++) \
2481 if (!is_ ## nAmE ## _cp(value)) \
2482 ANYOF_BITMAP_SET(data->start_class, value); \
2489 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2490 I32 *minlenp, I32 *deltap,
2495 struct regnode_charclass_class *and_withp,
2496 U32 flags, U32 depth)
2497 /* scanp: Start here (read-write). */
2498 /* deltap: Write maxlen-minlen here. */
2499 /* last: Stop before this one. */
2500 /* data: string data about the pattern */
2501 /* stopparen: treat close N as END */
2502 /* recursed: which subroutines have we recursed into */
2503 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2506 I32 min = 0, pars = 0, code;
2507 regnode *scan = *scanp, *next;
2509 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2510 int is_inf_internal = 0; /* The studied chunk is infinite */
2511 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2512 scan_data_t data_fake;
2513 SV *re_trie_maxbuff = NULL;
2514 regnode *first_non_open = scan;
2515 I32 stopmin = I32_MAX;
2516 scan_frame *frame = NULL;
2517 GET_RE_DEBUG_FLAGS_DECL;
2519 PERL_ARGS_ASSERT_STUDY_CHUNK;
2522 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2526 while (first_non_open && OP(first_non_open) == OPEN)
2527 first_non_open=regnext(first_non_open);
2532 while ( scan && OP(scan) != END && scan < last ){
2533 /* Peephole optimizer: */
2534 DEBUG_STUDYDATA("Peep:", data,depth);
2535 DEBUG_PEEP("Peep",scan,depth);
2536 JOIN_EXACT(scan,&min,0);
2538 /* Follow the next-chain of the current node and optimize
2539 away all the NOTHINGs from it. */
2540 if (OP(scan) != CURLYX) {
2541 const int max = (reg_off_by_arg[OP(scan)]
2543 /* I32 may be smaller than U16 on CRAYs! */
2544 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2545 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2549 /* Skip NOTHING and LONGJMP. */
2550 while ((n = regnext(n))
2551 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2552 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2553 && off + noff < max)
2555 if (reg_off_by_arg[OP(scan)])
2558 NEXT_OFF(scan) = off;
2563 /* The principal pseudo-switch. Cannot be a switch, since we
2564 look into several different things. */
2565 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2566 || OP(scan) == IFTHEN) {
2567 next = regnext(scan);
2569 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2571 if (OP(next) == code || code == IFTHEN) {
2572 /* NOTE - There is similar code to this block below for handling
2573 TRIE nodes on a re-study. If you change stuff here check there
2575 I32 max1 = 0, min1 = I32_MAX, num = 0;
2576 struct regnode_charclass_class accum;
2577 regnode * const startbranch=scan;
2579 if (flags & SCF_DO_SUBSTR)
2580 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2581 if (flags & SCF_DO_STCLASS)
2582 cl_init_zero(pRExC_state, &accum);
2584 while (OP(scan) == code) {
2585 I32 deltanext, minnext, f = 0, fake;
2586 struct regnode_charclass_class this_class;
2589 data_fake.flags = 0;
2591 data_fake.whilem_c = data->whilem_c;
2592 data_fake.last_closep = data->last_closep;
2595 data_fake.last_closep = &fake;
2597 data_fake.pos_delta = delta;
2598 next = regnext(scan);
2599 scan = NEXTOPER(scan);
2601 scan = NEXTOPER(scan);
2602 if (flags & SCF_DO_STCLASS) {
2603 cl_init(pRExC_state, &this_class);
2604 data_fake.start_class = &this_class;
2605 f = SCF_DO_STCLASS_AND;
2607 if (flags & SCF_WHILEM_VISITED_POS)
2608 f |= SCF_WHILEM_VISITED_POS;
2610 /* we suppose the run is continuous, last=next...*/
2611 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2613 stopparen, recursed, NULL, f,depth+1);
2616 if (max1 < minnext + deltanext)
2617 max1 = minnext + deltanext;
2618 if (deltanext == I32_MAX)
2619 is_inf = is_inf_internal = 1;
2621 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2623 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2624 if ( stopmin > minnext)
2625 stopmin = min + min1;
2626 flags &= ~SCF_DO_SUBSTR;
2628 data->flags |= SCF_SEEN_ACCEPT;
2631 if (data_fake.flags & SF_HAS_EVAL)
2632 data->flags |= SF_HAS_EVAL;
2633 data->whilem_c = data_fake.whilem_c;
2635 if (flags & SCF_DO_STCLASS)
2636 cl_or(pRExC_state, &accum, &this_class);
2638 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2640 if (flags & SCF_DO_SUBSTR) {
2641 data->pos_min += min1;
2642 data->pos_delta += max1 - min1;
2643 if (max1 != min1 || is_inf)
2644 data->longest = &(data->longest_float);
2647 delta += max1 - min1;
2648 if (flags & SCF_DO_STCLASS_OR) {
2649 cl_or(pRExC_state, data->start_class, &accum);
2651 cl_and(data->start_class, and_withp);
2652 flags &= ~SCF_DO_STCLASS;
2655 else if (flags & SCF_DO_STCLASS_AND) {
2657 cl_and(data->start_class, &accum);
2658 flags &= ~SCF_DO_STCLASS;
2661 /* Switch to OR mode: cache the old value of
2662 * data->start_class */
2664 StructCopy(data->start_class, and_withp,
2665 struct regnode_charclass_class);
2666 flags &= ~SCF_DO_STCLASS_AND;
2667 StructCopy(&accum, data->start_class,
2668 struct regnode_charclass_class);
2669 flags |= SCF_DO_STCLASS_OR;
2670 data->start_class->flags |= ANYOF_EOS;
2674 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2677 Assuming this was/is a branch we are dealing with: 'scan' now
2678 points at the item that follows the branch sequence, whatever
2679 it is. We now start at the beginning of the sequence and look
2686 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2688 If we can find such a subseqence we need to turn the first
2689 element into a trie and then add the subsequent branch exact
2690 strings to the trie.
2694 1. patterns where the whole set of branch can be converted.
2696 2. patterns where only a subset can be converted.
2698 In case 1 we can replace the whole set with a single regop
2699 for the trie. In case 2 we need to keep the start and end
2702 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2703 becomes BRANCH TRIE; BRANCH X;
2705 There is an additional case, that being where there is a
2706 common prefix, which gets split out into an EXACT like node
2707 preceding the TRIE node.
2709 If x(1..n)==tail then we can do a simple trie, if not we make
2710 a "jump" trie, such that when we match the appropriate word
2711 we "jump" to the appopriate tail node. Essentailly we turn
2712 a nested if into a case structure of sorts.
2717 if (!re_trie_maxbuff) {
2718 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2719 if (!SvIOK(re_trie_maxbuff))
2720 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2722 if ( SvIV(re_trie_maxbuff)>=0 ) {
2724 regnode *first = (regnode *)NULL;
2725 regnode *last = (regnode *)NULL;
2726 regnode *tail = scan;
2731 SV * const mysv = sv_newmortal(); /* for dumping */
2733 /* var tail is used because there may be a TAIL
2734 regop in the way. Ie, the exacts will point to the
2735 thing following the TAIL, but the last branch will
2736 point at the TAIL. So we advance tail. If we
2737 have nested (?:) we may have to move through several
2741 while ( OP( tail ) == TAIL ) {
2742 /* this is the TAIL generated by (?:) */
2743 tail = regnext( tail );
2748 regprop(RExC_rx, mysv, tail );
2749 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2750 (int)depth * 2 + 2, "",
2751 "Looking for TRIE'able sequences. Tail node is: ",
2752 SvPV_nolen_const( mysv )
2758 step through the branches, cur represents each
2759 branch, noper is the first thing to be matched
2760 as part of that branch and noper_next is the
2761 regnext() of that node. if noper is an EXACT
2762 and noper_next is the same as scan (our current
2763 position in the regex) then the EXACT branch is
2764 a possible optimization target. Once we have
2765 two or more consequetive such branches we can
2766 create a trie of the EXACT's contents and stich
2767 it in place. If the sequence represents all of
2768 the branches we eliminate the whole thing and
2769 replace it with a single TRIE. If it is a
2770 subsequence then we need to stitch it in. This
2771 means the first branch has to remain, and needs
2772 to be repointed at the item on the branch chain
2773 following the last branch optimized. This could
2774 be either a BRANCH, in which case the
2775 subsequence is internal, or it could be the
2776 item following the branch sequence in which
2777 case the subsequence is at the end.
2781 /* dont use tail as the end marker for this traverse */
2782 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2783 regnode * const noper = NEXTOPER( cur );
2784 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2785 regnode * const noper_next = regnext( noper );
2789 regprop(RExC_rx, mysv, cur);
2790 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2791 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2793 regprop(RExC_rx, mysv, noper);
2794 PerlIO_printf( Perl_debug_log, " -> %s",
2795 SvPV_nolen_const(mysv));
2798 regprop(RExC_rx, mysv, noper_next );
2799 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2800 SvPV_nolen_const(mysv));
2802 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2803 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2805 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2806 : PL_regkind[ OP( noper ) ] == EXACT )
2807 || OP(noper) == NOTHING )
2809 && noper_next == tail
2814 if ( !first || optype == NOTHING ) {
2815 if (!first) first = cur;
2816 optype = OP( noper );
2822 Currently we assume that the trie can handle unicode and ascii
2823 matches fold cased matches. If this proves true then the following
2824 define will prevent tries in this situation.
2826 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2828 #define TRIE_TYPE_IS_SAFE 1
2829 if ( last && TRIE_TYPE_IS_SAFE ) {
2830 make_trie( pRExC_state,
2831 startbranch, first, cur, tail, count,
2834 if ( PL_regkind[ OP( noper ) ] == EXACT
2836 && noper_next == tail
2841 optype = OP( noper );
2851 regprop(RExC_rx, mysv, cur);
2852 PerlIO_printf( Perl_debug_log,
2853 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2854 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2858 if ( last && TRIE_TYPE_IS_SAFE ) {
2859 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2860 #ifdef TRIE_STUDY_OPT
2861 if ( ((made == MADE_EXACT_TRIE &&
2862 startbranch == first)
2863 || ( first_non_open == first )) &&
2865 flags |= SCF_TRIE_RESTUDY;
2866 if ( startbranch == first
2869 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2879 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2880 scan = NEXTOPER(NEXTOPER(scan));
2881 } else /* single branch is optimized. */
2882 scan = NEXTOPER(scan);
2884 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2885 scan_frame *newframe = NULL;
2890 if (OP(scan) != SUSPEND) {
2891 /* set the pointer */
2892 if (OP(scan) == GOSUB) {
2894 RExC_recurse[ARG2L(scan)] = scan;
2895 start = RExC_open_parens[paren-1];
2896 end = RExC_close_parens[paren-1];
2899 start = RExC_rxi->program + 1;
2903 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2904 SAVEFREEPV(recursed);
2906 if (!PAREN_TEST(recursed,paren+1)) {
2907 PAREN_SET(recursed,paren+1);
2908 Newx(newframe,1,scan_frame);
2910 if (flags & SCF_DO_SUBSTR) {
2911 SCAN_COMMIT(pRExC_state,data,minlenp);
2912 data->longest = &(data->longest_float);
2914 is_inf = is_inf_internal = 1;
2915 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2916 cl_anything(pRExC_state, data->start_class);
2917 flags &= ~SCF_DO_STCLASS;
2920 Newx(newframe,1,scan_frame);
2923 end = regnext(scan);
2928 SAVEFREEPV(newframe);
2929 newframe->next = regnext(scan);
2930 newframe->last = last;
2931 newframe->stop = stopparen;
2932 newframe->prev = frame;
2942 else if (OP(scan) == EXACT) {
2943 I32 l = STR_LEN(scan);
2946 const U8 * const s = (U8*)STRING(scan);
2947 l = utf8_length(s, s + l);
2948 uc = utf8_to_uvchr(s, NULL);
2950 uc = *((U8*)STRING(scan));
2953 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2954 /* The code below prefers earlier match for fixed
2955 offset, later match for variable offset. */
2956 if (data->last_end == -1) { /* Update the start info. */
2957 data->last_start_min = data->pos_min;
2958 data->last_start_max = is_inf
2959 ? I32_MAX : data->pos_min + data->pos_delta;
2961 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2963 SvUTF8_on(data->last_found);
2965 SV * const sv = data->last_found;
2966 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2967 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2968 if (mg && mg->mg_len >= 0)
2969 mg->mg_len += utf8_length((U8*)STRING(scan),
2970 (U8*)STRING(scan)+STR_LEN(scan));
2972 data->last_end = data->pos_min + l;
2973 data->pos_min += l; /* As in the first entry. */
2974 data->flags &= ~SF_BEFORE_EOL;
2976 if (flags & SCF_DO_STCLASS_AND) {
2977 /* Check whether it is compatible with what we know already! */
2981 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2982 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2983 && (!(data->start_class->flags & ANYOF_FOLD)
2984 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2987 ANYOF_CLASS_ZERO(data->start_class);
2988 ANYOF_BITMAP_ZERO(data->start_class);
2990 ANYOF_BITMAP_SET(data->start_class, uc);
2991 data->start_class->flags &= ~ANYOF_EOS;
2993 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2995 else if (flags & SCF_DO_STCLASS_OR) {
2996 /* false positive possible if the class is case-folded */
2998 ANYOF_BITMAP_SET(data->start_class, uc);
3000 data->start_class->flags |= ANYOF_UNICODE_ALL;
3001 data->start_class->flags &= ~ANYOF_EOS;
3002 cl_and(data->start_class, and_withp);
3004 flags &= ~SCF_DO_STCLASS;
3006 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3007 I32 l = STR_LEN(scan);
3008 UV uc = *((U8*)STRING(scan));
3010 /* Search for fixed substrings supports EXACT only. */
3011 if (flags & SCF_DO_SUBSTR) {
3013 SCAN_COMMIT(pRExC_state, data, minlenp);
3016 const U8 * const s = (U8 *)STRING(scan);
3017 l = utf8_length(s, s + l);
3018 uc = utf8_to_uvchr(s, NULL);
3021 if (flags & SCF_DO_SUBSTR)
3023 if (flags & SCF_DO_STCLASS_AND) {
3024 /* Check whether it is compatible with what we know already! */
3028 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3029 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3030 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3032 ANYOF_CLASS_ZERO(data->start_class);
3033 ANYOF_BITMAP_ZERO(data->start_class);
3035 ANYOF_BITMAP_SET(data->start_class, uc);
3036 data->start_class->flags &= ~ANYOF_EOS;
3037 data->start_class->flags |= ANYOF_FOLD;
3038 if (OP(scan) == EXACTFL)
3039 data->start_class->flags |= ANYOF_LOCALE;
3042 else if (flags & SCF_DO_STCLASS_OR) {
3043 if (data->start_class->flags & ANYOF_FOLD) {
3044 /* false positive possible if the class is case-folded.
3045 Assume that the locale settings are the same... */
3047 ANYOF_BITMAP_SET(data->start_class, uc);
3048 data->start_class->flags &= ~ANYOF_EOS;
3050 cl_and(data->start_class, and_withp);
3052 flags &= ~SCF_DO_STCLASS;
3054 else if (strchr((const char*)PL_varies,OP(scan))) {
3055 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3056 I32 f = flags, pos_before = 0;
3057 regnode * const oscan = scan;
3058 struct regnode_charclass_class this_class;
3059 struct regnode_charclass_class *oclass = NULL;
3060 I32 next_is_eval = 0;
3062 switch (PL_regkind[OP(scan)]) {
3063 case WHILEM: /* End of (?:...)* . */
3064 scan = NEXTOPER(scan);
3067 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3068 next = NEXTOPER(scan);
3069 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3071 maxcount = REG_INFTY;
3072 next = regnext(scan);
3073 scan = NEXTOPER(scan);
3077 if (flags & SCF_DO_SUBSTR)
3082 if (flags & SCF_DO_STCLASS) {
3084 maxcount = REG_INFTY;
3085 next = regnext(scan);
3086 scan = NEXTOPER(scan);
3089 is_inf = is_inf_internal = 1;
3090 scan = regnext(scan);
3091 if (flags & SCF_DO_SUBSTR) {
3092 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3093 data->longest = &(data->longest_float);
3095 goto optimize_curly_tail;
3097 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3098 && (scan->flags == stopparen))
3103 mincount = ARG1(scan);
3104 maxcount = ARG2(scan);
3106 next = regnext(scan);
3107 if (OP(scan) == CURLYX) {
3108 I32 lp = (data ? *(data->last_closep) : 0);
3109 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3111 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3112 next_is_eval = (OP(scan) == EVAL);
3114 if (flags & SCF_DO_SUBSTR) {
3115 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3116 pos_before = data->pos_min;
3120 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3122 data->flags |= SF_IS_INF;
3124 if (flags & SCF_DO_STCLASS) {
3125 cl_init(pRExC_state, &this_class);
3126 oclass = data->start_class;
3127 data->start_class = &this_class;
3128 f |= SCF_DO_STCLASS_AND;
3129 f &= ~SCF_DO_STCLASS_OR;
3131 /* These are the cases when once a subexpression
3132 fails at a particular position, it cannot succeed
3133 even after backtracking at the enclosing scope.
3135 XXXX what if minimal match and we are at the
3136 initial run of {n,m}? */
3137 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3138 f &= ~SCF_WHILEM_VISITED_POS;
3140 /* This will finish on WHILEM, setting scan, or on NULL: */
3141 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3142 last, data, stopparen, recursed, NULL,
3144 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3146 if (flags & SCF_DO_STCLASS)
3147 data->start_class = oclass;
3148 if (mincount == 0 || minnext == 0) {
3149 if (flags & SCF_DO_STCLASS_OR) {
3150 cl_or(pRExC_state, data->start_class, &this_class);
3152 else if (flags & SCF_DO_STCLASS_AND) {
3153 /* Switch to OR mode: cache the old value of
3154 * data->start_class */
3156 StructCopy(data->start_class, and_withp,
3157 struct regnode_charclass_class);
3158 flags &= ~SCF_DO_STCLASS_AND;
3159 StructCopy(&this_class, data->start_class,
3160 struct regnode_charclass_class);
3161 flags |= SCF_DO_STCLASS_OR;
3162 data->start_class->flags |= ANYOF_EOS;
3164 } else { /* Non-zero len */
3165 if (flags & SCF_DO_STCLASS_OR) {
3166 cl_or(pRExC_state, data->start_class, &this_class);
3167 cl_and(data->start_class, and_withp);
3169 else if (flags & SCF_DO_STCLASS_AND)
3170 cl_and(data->start_class, &this_class);
3171 flags &= ~SCF_DO_STCLASS;
3173 if (!scan) /* It was not CURLYX, but CURLY. */
3175 if ( /* ? quantifier ok, except for (?{ ... }) */
3176 (next_is_eval || !(mincount == 0 && maxcount == 1))
3177 && (minnext == 0) && (deltanext == 0)
3178 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3179 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3180 && ckWARN(WARN_REGEXP))
3183 "Quantifier unexpected on zero-length expression");
3186 min += minnext * mincount;
3187 is_inf_internal |= ((maxcount == REG_INFTY
3188 && (minnext + deltanext) > 0)
3189 || deltanext == I32_MAX);
3190 is_inf |= is_inf_internal;
3191 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3193 /* Try powerful optimization CURLYX => CURLYN. */
3194 if ( OP(oscan) == CURLYX && data
3195 && data->flags & SF_IN_PAR
3196 && !(data->flags & SF_HAS_EVAL)
3197 && !deltanext && minnext == 1 ) {
3198 /* Try to optimize to CURLYN. */
3199 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3200 regnode * const nxt1 = nxt;
3207 if (!strchr((const char*)PL_simple,OP(nxt))
3208 && !(PL_regkind[OP(nxt)] == EXACT
3209 && STR_LEN(nxt) == 1))
3215 if (OP(nxt) != CLOSE)
3217 if (RExC_open_parens) {
3218 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3219 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3221 /* Now we know that nxt2 is the only contents: */
3222 oscan->flags = (U8)ARG(nxt);
3224 OP(nxt1) = NOTHING; /* was OPEN. */
3227 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3228 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3229 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3230 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3231 OP(nxt + 1) = OPTIMIZED; /* was count. */
3232 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3237 /* Try optimization CURLYX => CURLYM. */
3238 if ( OP(oscan) == CURLYX && data
3239 && !(data->flags & SF_HAS_PAR)
3240 && !(data->flags & SF_HAS_EVAL)
3241 && !deltanext /* atom is fixed width */
3242 && minnext != 0 /* CURLYM can't handle zero width */
3244 /* XXXX How to optimize if data == 0? */
3245 /* Optimize to a simpler form. */
3246 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3250 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3251 && (OP(nxt2) != WHILEM))
3253 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3254 /* Need to optimize away parenths. */
3255 if (data->flags & SF_IN_PAR) {
3256 /* Set the parenth number. */
3257 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3259 if (OP(nxt) != CLOSE)
3260 FAIL("Panic opt close");
3261 oscan->flags = (U8)ARG(nxt);
3262 if (RExC_open_parens) {
3263 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3264 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3266 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3267 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3270 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3271 OP(nxt + 1) = OPTIMIZED; /* was count. */
3272 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3273 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3276 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3277 regnode *nnxt = regnext(nxt1);
3280 if (reg_off_by_arg[OP(nxt1)])
3281 ARG_SET(nxt1, nxt2 - nxt1);
3282 else if (nxt2 - nxt1 < U16_MAX)
3283 NEXT_OFF(nxt1) = nxt2 - nxt1;
3285 OP(nxt) = NOTHING; /* Cannot beautify */
3290 /* Optimize again: */
3291 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3292 NULL, stopparen, recursed, NULL, 0,depth+1);
3297 else if ((OP(oscan) == CURLYX)
3298 && (flags & SCF_WHILEM_VISITED_POS)
3299 /* See the comment on a similar expression above.
3300 However, this time it not a subexpression
3301 we care about, but the expression itself. */
3302 && (maxcount == REG_INFTY)
3303 && data && ++data->whilem_c < 16) {
3304 /* This stays as CURLYX, we can put the count/of pair. */
3305 /* Find WHILEM (as in regexec.c) */
3306 regnode *nxt = oscan + NEXT_OFF(oscan);
3308 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3310 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3311 | (RExC_whilem_seen << 4)); /* On WHILEM */
3313 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3315 if (flags & SCF_DO_SUBSTR) {
3316 SV *last_str = NULL;
3317 int counted = mincount != 0;
3319 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3320 #if defined(SPARC64_GCC_WORKAROUND)
3323 const char *s = NULL;
3326 if (pos_before >= data->last_start_min)
3329 b = data->last_start_min;
3332 s = SvPV_const(data->last_found, l);
3333 old = b - data->last_start_min;
3336 I32 b = pos_before >= data->last_start_min
3337 ? pos_before : data->last_start_min;
3339 const char * const s = SvPV_const(data->last_found, l);
3340 I32 old = b - data->last_start_min;
3344 old = utf8_hop((U8*)s, old) - (U8*)s;
3347 /* Get the added string: */
3348 last_str = newSVpvn_utf8(s + old, l, UTF);
3349 if (deltanext == 0 && pos_before == b) {
3350 /* What was added is a constant string */
3352 SvGROW(last_str, (mincount * l) + 1);
3353 repeatcpy(SvPVX(last_str) + l,
3354 SvPVX_const(last_str), l, mincount - 1);
3355 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3356 /* Add additional parts. */
3357 SvCUR_set(data->last_found,
3358 SvCUR(data->last_found) - l);
3359 sv_catsv(data->last_found, last_str);
3361 SV * sv = data->last_found;
3363 SvUTF8(sv) && SvMAGICAL(sv) ?
3364 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3365 if (mg && mg->mg_len >= 0)
3366 mg->mg_len += CHR_SVLEN(last_str) - l;
3368 data->last_end += l * (mincount - 1);
3371 /* start offset must point into the last copy */
3372 data->last_start_min += minnext * (mincount - 1);
3373 data->last_start_max += is_inf ? I32_MAX
3374 : (maxcount - 1) * (minnext + data->pos_delta);
3377 /* It is counted once already... */
3378 data->pos_min += minnext * (mincount - counted);
3379 data->pos_delta += - counted * deltanext +
3380 (minnext + deltanext) * maxcount - minnext * mincount;
3381 if (mincount != maxcount) {
3382 /* Cannot extend fixed substrings found inside
3384 SCAN_COMMIT(pRExC_state,data,minlenp);
3385 if (mincount && last_str) {
3386 SV * const sv = data->last_found;
3387 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3388 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3392 sv_setsv(sv, last_str);
3393 data->last_end = data->pos_min;
3394 data->last_start_min =
3395 data->pos_min - CHR_SVLEN(last_str);
3396 data->last_start_max = is_inf
3398 : data->pos_min + data->pos_delta
3399 - CHR_SVLEN(last_str);
3401 data->longest = &(data->longest_float);
3403 SvREFCNT_dec(last_str);
3405 if (data && (fl & SF_HAS_EVAL))
3406 data->flags |= SF_HAS_EVAL;
3407 optimize_curly_tail:
3408 if (OP(oscan) != CURLYX) {
3409 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3411 NEXT_OFF(oscan) += NEXT_OFF(next);
3414 default: /* REF and CLUMP only? */
3415 if (flags & SCF_DO_SUBSTR) {
3416 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3417 data->longest = &(data->longest_float);
3419 is_inf = is_inf_internal = 1;
3420 if (flags & SCF_DO_STCLASS_OR)
3421 cl_anything(pRExC_state, data->start_class);
3422 flags &= ~SCF_DO_STCLASS;
3426 else if (OP(scan) == LNBREAK) {
3427 if (flags & SCF_DO_STCLASS) {
3429 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3430 if (flags & SCF_DO_STCLASS_AND) {
3431 for (value = 0; value < 256; value++)
3432 if (!is_VERTWS_cp(value))
3433 ANYOF_BITMAP_CLEAR(data->start_class, value);
3436 for (value = 0; value < 256; value++)
3437 if (is_VERTWS_cp(value))
3438 ANYOF_BITMAP_SET(data->start_class, value);
3440 if (flags & SCF_DO_STCLASS_OR)
3441 cl_and(data->start_class, and_withp);
3442 flags &= ~SCF_DO_STCLASS;
3446 if (flags & SCF_DO_SUBSTR) {
3447 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3449 data->pos_delta += 1;
3450 data->longest = &(data->longest_float);
3454 else if (OP(scan) == FOLDCHAR) {
3455 int d = ARG(scan)==0xDF ? 1 : 2;
3456 flags &= ~SCF_DO_STCLASS;
3459 if (flags & SCF_DO_SUBSTR) {
3460 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3462 data->pos_delta += d;
3463 data->longest = &(data->longest_float);
3466 else if (strchr((const char*)PL_simple,OP(scan))) {
3469 if (flags & SCF_DO_SUBSTR) {
3470 SCAN_COMMIT(pRExC_state,data,minlenp);
3474 if (flags & SCF_DO_STCLASS) {
3475 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3477 /* Some of the logic below assumes that switching
3478 locale on will only add false positives. */
3479 switch (PL_regkind[OP(scan)]) {
3483 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3484 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3485 cl_anything(pRExC_state, data->start_class);
3488 if (OP(scan) == SANY)
3490 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3491 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3492 || (data->start_class->flags & ANYOF_CLASS));
3493 cl_anything(pRExC_state, data->start_class);
3495 if (flags & SCF_DO_STCLASS_AND || !value)
3496 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3499 if (flags & SCF_DO_STCLASS_AND)
3500 cl_and(data->start_class,
3501 (struct regnode_charclass_class*)scan);
3503 cl_or(pRExC_state, data->start_class,
3504 (struct regnode_charclass_class*)scan);
3507 if (flags & SCF_DO_STCLASS_AND) {
3508 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3509 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3510 for (value = 0; value < 256; value++)
3511 if (!isALNUM(value))
3512 ANYOF_BITMAP_CLEAR(data->start_class, value);
3516 if (data->start_class->flags & ANYOF_LOCALE)
3517 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3519 for (value = 0; value < 256; value++)
3521 ANYOF_BITMAP_SET(data->start_class, value);
3526 if (flags & SCF_DO_STCLASS_AND) {
3527 if (data->start_class->flags & ANYOF_LOCALE)
3528 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3531 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3532 data->start_class->flags |= ANYOF_LOCALE;
3536 if (flags & SCF_DO_STCLASS_AND) {
3537 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3538 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3539 for (value = 0; value < 256; value++)
3541 ANYOF_BITMAP_CLEAR(data->start_class, value);
3545 if (data->start_class->flags & ANYOF_LOCALE)
3546 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3548 for (value = 0; value < 256; value++)
3549 if (!isALNUM(value))
3550 ANYOF_BITMAP_SET(data->start_class, value);
3555 if (flags & SCF_DO_STCLASS_AND) {
3556 if (data->start_class->flags & ANYOF_LOCALE)
3557 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3560 data->start_class->flags |= ANYOF_LOCALE;
3561 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3565 if (flags & SCF_DO_STCLASS_AND) {
3566 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3567 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3568 for (value = 0; value < 256; value++)
3569 if (!isSPACE(value))
3570 ANYOF_BITMAP_CLEAR(data->start_class, value);
3574 if (data->start_class->flags & ANYOF_LOCALE)
3575 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3577 for (value = 0; value < 256; value++)
3579 ANYOF_BITMAP_SET(data->start_class, value);
3584 if (flags & SCF_DO_STCLASS_AND) {
3585 if (data->start_class->flags & ANYOF_LOCALE)
3586 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3589 data->start_class->flags |= ANYOF_LOCALE;
3590 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3594 if (flags & SCF_DO_STCLASS_AND) {
3595 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3596 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3597 for (value = 0; value < 256; value++)
3599 ANYOF_BITMAP_CLEAR(data->start_class, value);
3603 if (data->start_class->flags & ANYOF_LOCALE)
3604 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3606 for (value = 0; value < 256; value++)
3607 if (!isSPACE(value))
3608 ANYOF_BITMAP_SET(data->start_class, value);
3613 if (flags & SCF_DO_STCLASS_AND) {
3614 if (data->start_class->flags & ANYOF_LOCALE) {
3615 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3616 for (value = 0; value < 256; value++)
3617 if (!isSPACE(value))
3618 ANYOF_BITMAP_CLEAR(data->start_class, value);
3622 data->start_class->flags |= ANYOF_LOCALE;
3623 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3627 if (flags & SCF_DO_STCLASS_AND) {
3628 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3629 for (value = 0; value < 256; value++)
3630 if (!isDIGIT(value))
3631 ANYOF_BITMAP_CLEAR(data->start_class, value);
3634 if (data->start_class->flags & ANYOF_LOCALE)
3635 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3637 for (value = 0; value < 256; value++)
3639 ANYOF_BITMAP_SET(data->start_class, value);
3644 if (flags & SCF_DO_STCLASS_AND) {
3645 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3646 for (value = 0; value < 256; value++)
3648 ANYOF_BITMAP_CLEAR(data->start_class, value);
3651 if (data->start_class->flags & ANYOF_LOCALE)
3652 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3654 for (value = 0; value < 256; value++)
3655 if (!isDIGIT(value))
3656 ANYOF_BITMAP_SET(data->start_class, value);
3660 CASE_SYNST_FNC(VERTWS);
3661 CASE_SYNST_FNC(HORIZWS);
3664 if (flags & SCF_DO_STCLASS_OR)
3665 cl_and(data->start_class, and_withp);
3666 flags &= ~SCF_DO_STCLASS;
3669 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3670 data->flags |= (OP(scan) == MEOL
3674 else if ( PL_regkind[OP(scan)] == BRANCHJ
3675 /* Lookbehind, or need to calculate parens/evals/stclass: */
3676 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3677 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3678 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3679 || OP(scan) == UNLESSM )
3681 /* Negative Lookahead/lookbehind
3682 In this case we can't do fixed string optimisation.
3685 I32 deltanext, minnext, fake = 0;
3687 struct regnode_charclass_class intrnl;
3690 data_fake.flags = 0;
3692 data_fake.whilem_c = data->whilem_c;
3693 data_fake.last_closep = data->last_closep;
3696 data_fake.last_closep = &fake;
3697 data_fake.pos_delta = delta;
3698 if ( flags & SCF_DO_STCLASS && !scan->flags
3699 && OP(scan) == IFMATCH ) { /* Lookahead */
3700 cl_init(pRExC_state, &intrnl);
3701 data_fake.start_class = &intrnl;
3702 f |= SCF_DO_STCLASS_AND;
3704 if (flags & SCF_WHILEM_VISITED_POS)
3705 f |= SCF_WHILEM_VISITED_POS;
3706 next = regnext(scan);
3707 nscan = NEXTOPER(NEXTOPER(scan));
3708 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3709 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3712 FAIL("Variable length lookbehind not implemented");
3714 else if (minnext > (I32)U8_MAX) {
3715 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3717 scan->flags = (U8)minnext;
3720 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3722 if (data_fake.flags & SF_HAS_EVAL)
3723 data->flags |= SF_HAS_EVAL;
3724 data->whilem_c = data_fake.whilem_c;
3726 if (f & SCF_DO_STCLASS_AND) {
3727 const int was = (data->start_class->flags & ANYOF_EOS);
3729 cl_and(data->start_class, &intrnl);
3731 data->start_class->flags |= ANYOF_EOS;
3734 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3736 /* Positive Lookahead/lookbehind
3737 In this case we can do fixed string optimisation,
3738 but we must be careful about it. Note in the case of
3739 lookbehind the positions will be offset by the minimum
3740 length of the pattern, something we won't know about
3741 until after the recurse.
3743 I32 deltanext, fake = 0;
3745 struct regnode_charclass_class intrnl;
3747 /* We use SAVEFREEPV so that when the full compile
3748 is finished perl will clean up the allocated
3749 minlens when its all done. This was we don't
3750 have to worry about freeing them when we know
3751 they wont be used, which would be a pain.
3754 Newx( minnextp, 1, I32 );
3755 SAVEFREEPV(minnextp);
3758 StructCopy(data, &data_fake, scan_data_t);
3759 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3762 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3763 data_fake.last_found=newSVsv(data->last_found);
3767 data_fake.last_closep = &fake;
3768 data_fake.flags = 0;
3769 data_fake.pos_delta = delta;
3771 data_fake.flags |= SF_IS_INF;
3772 if ( flags & SCF_DO_STCLASS && !scan->flags
3773 && OP(scan) == IFMATCH ) { /* Lookahead */
3774 cl_init(pRExC_state, &intrnl);
3775 data_fake.start_class = &intrnl;
3776 f |= SCF_DO_STCLASS_AND;
3778 if (flags & SCF_WHILEM_VISITED_POS)
3779 f |= SCF_WHILEM_VISITED_POS;
3780 next = regnext(scan);
3781 nscan = NEXTOPER(NEXTOPER(scan));
3783 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3784 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3787 FAIL("Variable length lookbehind not implemented");
3789 else if (*minnextp > (I32)U8_MAX) {
3790 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3792 scan->flags = (U8)*minnextp;
3797 if (f & SCF_DO_STCLASS_AND) {
3798 const int was = (data->start_class->flags & ANYOF_EOS);
3800 cl_and(data->start_class, &intrnl);
3802 data->start_class->flags |= ANYOF_EOS;
3805 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3807 if (data_fake.flags & SF_HAS_EVAL)
3808 data->flags |= SF_HAS_EVAL;
3809 data->whilem_c = data_fake.whilem_c;
3810 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3811 if (RExC_rx->minlen<*minnextp)
3812 RExC_rx->minlen=*minnextp;
3813 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3814 SvREFCNT_dec(data_fake.last_found);
3816 if ( data_fake.minlen_fixed != minlenp )
3818 data->offset_fixed= data_fake.offset_fixed;
3819 data->minlen_fixed= data_fake.minlen_fixed;
3820 data->lookbehind_fixed+= scan->flags;
3822 if ( data_fake.minlen_float != minlenp )
3824 data->minlen_float= data_fake.minlen_float;
3825 data->offset_float_min=data_fake.offset_float_min;
3826 data->offset_float_max=data_fake.offset_float_max;
3827 data->lookbehind_float+= scan->flags;
3836 else if (OP(scan) == OPEN) {
3837 if (stopparen != (I32)ARG(scan))
3840 else if (OP(scan) == CLOSE) {
3841 if (stopparen == (I32)ARG(scan)) {
3844 if ((I32)ARG(scan) == is_par) {
3845 next = regnext(scan);
3847 if ( next && (OP(next) != WHILEM) && next < last)
3848 is_par = 0; /* Disable optimization */
3851 *(data->last_closep) = ARG(scan);
3853 else if (OP(scan) == EVAL) {
3855 data->flags |= SF_HAS_EVAL;
3857 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3858 if (flags & SCF_DO_SUBSTR) {
3859 SCAN_COMMIT(pRExC_state,data,minlenp);
3860 flags &= ~SCF_DO_SUBSTR;
3862 if (data && OP(scan)==ACCEPT) {
3863 data->flags |= SCF_SEEN_ACCEPT;
3868 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3870 if (flags & SCF_DO_SUBSTR) {
3871 SCAN_COMMIT(pRExC_state,data,minlenp);
3872 data->longest = &(data->longest_float);
3874 is_inf = is_inf_internal = 1;
3875 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3876 cl_anything(pRExC_state, data->start_class);
3877 flags &= ~SCF_DO_STCLASS;
3879 else if (OP(scan) == GPOS) {
3880 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3881 !(delta || is_inf || (data && data->pos_delta)))
3883 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3884 RExC_rx->extflags |= RXf_ANCH_GPOS;
3885 if (RExC_rx->gofs < (U32)min)
3886 RExC_rx->gofs = min;
3888 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3892 #ifdef TRIE_STUDY_OPT
3893 #ifdef FULL_TRIE_STUDY
3894 else if (PL_regkind[OP(scan)] == TRIE) {
3895 /* NOTE - There is similar code to this block above for handling
3896 BRANCH nodes on the initial study. If you change stuff here
3898 regnode *trie_node= scan;
3899 regnode *tail= regnext(scan);
3900 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3901 I32 max1 = 0, min1 = I32_MAX;
3902 struct regnode_charclass_class accum;
3904 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3905 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3906 if (flags & SCF_DO_STCLASS)
3907 cl_init_zero(pRExC_state, &accum);
3913 const regnode *nextbranch= NULL;
3916 for ( word=1 ; word <= trie->wordcount ; word++)
3918 I32 deltanext=0, minnext=0, f = 0, fake;
3919 struct regnode_charclass_class this_class;
3921 data_fake.flags = 0;
3923 data_fake.whilem_c = data->whilem_c;
3924 data_fake.last_closep = data->last_closep;
3927 data_fake.last_closep = &fake;
3928 data_fake.pos_delta = delta;
3929 if (flags & SCF_DO_STCLASS) {
3930 cl_init(pRExC_state, &this_class);
3931 data_fake.start_class = &this_class;
3932 f = SCF_DO_STCLASS_AND;
3934 if (flags & SCF_WHILEM_VISITED_POS)
3935 f |= SCF_WHILEM_VISITED_POS;
3937 if (trie->jump[word]) {
3939 nextbranch = trie_node + trie->jump[0];
3940 scan= trie_node + trie->jump[word];
3941 /* We go from the jump point to the branch that follows
3942 it. Note this means we need the vestigal unused branches
3943 even though they arent otherwise used.
3945 minnext = study_chunk(pRExC_state, &scan, minlenp,
3946 &deltanext, (regnode *)nextbranch, &data_fake,
3947 stopparen, recursed, NULL, f,depth+1);
3949 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3950 nextbranch= regnext((regnode*)nextbranch);
3952 if (min1 > (I32)(minnext + trie->minlen))
3953 min1 = minnext + trie->minlen;
3954 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3955 max1 = minnext + deltanext + trie->maxlen;
3956 if (deltanext == I32_MAX)
3957 is_inf = is_inf_internal = 1;
3959 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3961 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3962 if ( stopmin > min + min1)
3963 stopmin = min + min1;
3964 flags &= ~SCF_DO_SUBSTR;
3966 data->flags |= SCF_SEEN_ACCEPT;
3969 if (data_fake.flags & SF_HAS_EVAL)
3970 data->flags |= SF_HAS_EVAL;
3971 data->whilem_c = data_fake.whilem_c;
3973 if (flags & SCF_DO_STCLASS)
3974 cl_or(pRExC_state, &accum, &this_class);
3977 if (flags & SCF_DO_SUBSTR) {
3978 data->pos_min += min1;
3979 data->pos_delta += max1 - min1;
3980 if (max1 != min1 || is_inf)
3981 data->longest = &(data->longest_float);
3984 delta += max1 - min1;
3985 if (flags & SCF_DO_STCLASS_OR) {
3986 cl_or(pRExC_state, data->start_class, &accum);
3988 cl_and(data->start_class, and_withp);
3989 flags &= ~SCF_DO_STCLASS;
3992 else if (flags & SCF_DO_STCLASS_AND) {
3994 cl_and(data->start_class, &accum);
3995 flags &= ~SCF_DO_STCLASS;
3998 /* Switch to OR mode: cache the old value of
3999 * data->start_class */
4001 StructCopy(data->start_class, and_withp,
4002 struct regnode_charclass_class);
4003 flags &= ~SCF_DO_STCLASS_AND;
4004 StructCopy(&accum, data->start_class,
4005 struct regnode_charclass_class);
4006 flags |= SCF_DO_STCLASS_OR;
4007 data->start_class->flags |= ANYOF_EOS;
4014 else if (PL_regkind[OP(scan)] == TRIE) {
4015 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4018 min += trie->minlen;
4019 delta += (trie->maxlen - trie->minlen);
4020 flags &= ~SCF_DO_STCLASS; /* xxx */
4021 if (flags & SCF_DO_SUBSTR) {
4022 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4023 data->pos_min += trie->minlen;
4024 data->pos_delta += (trie->maxlen - trie->minlen);
4025 if (trie->maxlen != trie->minlen)
4026 data->longest = &(data->longest_float);
4028 if (trie->jump) /* no more substrings -- for now /grr*/
4029 flags &= ~SCF_DO_SUBSTR;
4031 #endif /* old or new */
4032 #endif /* TRIE_STUDY_OPT */
4034 /* Else: zero-length, ignore. */
4035 scan = regnext(scan);
4040 stopparen = frame->stop;
4041 frame = frame->prev;
4042 goto fake_study_recurse;
4047 DEBUG_STUDYDATA("pre-fin:",data,depth);
4050 *deltap = is_inf_internal ? I32_MAX : delta;
4051 if (flags & SCF_DO_SUBSTR && is_inf)
4052 data->pos_delta = I32_MAX - data->pos_min;
4053 if (is_par > (I32)U8_MAX)
4055 if (is_par && pars==1 && data) {
4056 data->flags |= SF_IN_PAR;
4057 data->flags &= ~SF_HAS_PAR;
4059 else if (pars && data) {
4060 data->flags |= SF_HAS_PAR;
4061 data->flags &= ~SF_IN_PAR;
4063 if (flags & SCF_DO_STCLASS_OR)
4064 cl_and(data->start_class, and_withp);
4065 if (flags & SCF_TRIE_RESTUDY)
4066 data->flags |= SCF_TRIE_RESTUDY;
4068 DEBUG_STUDYDATA("post-fin:",data,depth);
4070 return min < stopmin ? min : stopmin;
4074 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4076 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4078 PERL_ARGS_ASSERT_ADD_DATA;
4080 Renewc(RExC_rxi->data,
4081 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4082 char, struct reg_data);
4084 Renew(RExC_rxi->data->what, count + n, U8);
4086 Newx(RExC_rxi->data->what, n, U8);
4087 RExC_rxi->data->count = count + n;
4088 Copy(s, RExC_rxi->data->what + count, n, U8);
4092 /*XXX: todo make this not included in a non debugging perl */
4093 #ifndef PERL_IN_XSUB_RE
4095 Perl_reginitcolors(pTHX)
4098 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4100 char *t = savepv(s);
4104 t = strchr(t, '\t');
4110 PL_colors[i] = t = (char *)"";
4115 PL_colors[i++] = (char *)"";
4122 #ifdef TRIE_STUDY_OPT
4123 #define CHECK_RESTUDY_GOTO \
4125 (data.flags & SCF_TRIE_RESTUDY) \
4129 #define CHECK_RESTUDY_GOTO
4133 - pregcomp - compile a regular expression into internal code
4135 * We can't allocate space until we know how big the compiled form will be,
4136 * but we can't compile it (and thus know how big it is) until we've got a
4137 * place to put the code. So we cheat: we compile it twice, once with code
4138 * generation turned off and size counting turned on, and once "for real".
4139 * This also means that we don't allocate space until we are sure that the
4140 * thing really will compile successfully, and we never have to move the
4141 * code and thus invalidate pointers into it. (Note that it has to be in
4142 * one piece because free() must be able to free it all.) [NB: not true in perl]
4144 * Beware that the optimization-preparation code in here knows about some
4145 * of the structure of the compiled regexp. [I'll say.]
4150 #ifndef PERL_IN_XSUB_RE
4151 #define RE_ENGINE_PTR &PL_core_reg_engine
4153 extern const struct regexp_engine my_reg_engine;
4154 #define RE_ENGINE_PTR &my_reg_engine
4157 #ifndef PERL_IN_XSUB_RE
4159 Perl_pregcomp(pTHX_ const SV * const pattern, const U32 flags)
4162 HV * const table = GvHV(PL_hintgv);
4164 PERL_ARGS_ASSERT_PREGCOMP;
4166 /* Dispatch a request to compile a regexp to correct
4169 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4170 GET_RE_DEBUG_FLAGS_DECL;
4171 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4172 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4174 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4177 return CALLREGCOMP_ENG(eng, pattern, flags);
4180 return Perl_re_compile(aTHX_ pattern, flags);
4185 Perl_re_compile(pTHX_ const SV * const pattern, U32 pm_flags)
4190 register regexp_internal *ri;
4192 char* exp = SvPV((SV*)pattern, plen);
4193 char* xend = exp + plen;
4200 RExC_state_t RExC_state;
4201 RExC_state_t * const pRExC_state = &RExC_state;
4202 #ifdef TRIE_STUDY_OPT
4204 RExC_state_t copyRExC_state;
4206 GET_RE_DEBUG_FLAGS_DECL;
4208 PERL_ARGS_ASSERT_RE_COMPILE;
4210 DEBUG_r(if (!PL_colorset) reginitcolors());
4212 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4215 SV *dsv= sv_newmortal();
4216 RE_PV_QUOTED_DECL(s, RExC_utf8,
4217 dsv, exp, plen, 60);
4218 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4219 PL_colors[4],PL_colors[5],s);
4224 RExC_flags = pm_flags;
4228 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4229 RExC_seen_evals = 0;
4232 /* First pass: determine size, legality. */
4240 RExC_emit = &PL_regdummy;
4241 RExC_whilem_seen = 0;
4242 RExC_charnames = NULL;
4243 RExC_open_parens = NULL;
4244 RExC_close_parens = NULL;
4246 RExC_paren_names = NULL;
4248 RExC_paren_name_list = NULL;
4250 RExC_recurse = NULL;
4251 RExC_recurse_count = 0;
4253 #if 0 /* REGC() is (currently) a NOP at the first pass.
4254 * Clever compilers notice this and complain. --jhi */
4255 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4257 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4258 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4259 RExC_precomp = NULL;
4262 if (RExC_utf8 && !RExC_orig_utf8) {
4263 /* It's possible to write a regexp in ascii that represents Unicode
4264 codepoints outside of the byte range, such as via \x{100}. If we
4265 detect such a sequence we have to convert the entire pattern to utf8
4266 and then recompile, as our sizing calculation will have been based
4267 on 1 byte == 1 character, but we will need to use utf8 to encode
4268 at least some part of the pattern, and therefore must convert the whole
4270 XXX: somehow figure out how to make this less expensive...
4273 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4274 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4275 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4277 RExC_orig_utf8 = RExC_utf8;
4279 goto redo_first_pass;
4282 PerlIO_printf(Perl_debug_log,
4283 "Required size %"IVdf" nodes\n"
4284 "Starting second pass (creation)\n",
4287 RExC_lastparse=NULL;
4289 /* Small enough for pointer-storage convention?
4290 If extralen==0, this means that we will not need long jumps. */
4291 if (RExC_size >= 0x10000L && RExC_extralen)
4292 RExC_size += RExC_extralen;
4295 if (RExC_whilem_seen > 15)
4296 RExC_whilem_seen = 15;
4298 /* Allocate space and zero-initialize. Note, the two step process
4299 of zeroing when in debug mode, thus anything assigned has to
4300 happen after that */
4301 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4302 r = (struct regexp*)SvANY(rx);
4303 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4304 char, regexp_internal);
4305 if ( r == NULL || ri == NULL )
4306 FAIL("Regexp out of space");
4308 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4309 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4311 /* bulk initialize base fields with 0. */
4312 Zero(ri, sizeof(regexp_internal), char);
4315 /* non-zero initialization begins here */
4317 r->engine= RE_ENGINE_PTR;
4318 r->extflags = pm_flags;
4320 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4321 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4322 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4323 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4324 >> RXf_PMf_STD_PMMOD_SHIFT);
4325 const char *fptr = STD_PAT_MODS; /*"msix"*/
4327 const STRLEN wraplen = plen + has_minus + has_p + has_runon
4328 + (sizeof(STD_PAT_MODS) - 1)
4329 + (sizeof("(?:)") - 1);
4331 p = sv_grow((SV *)rx, wraplen + 1);
4332 SvCUR_set(rx, wraplen);
4334 SvFLAGS(rx) |= SvUTF8(pattern);
4337 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4339 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4340 char *colon = r + 1;
4343 while((ch = *fptr++)) {
4357 Copy(RExC_precomp, p, plen, char);
4358 assert ((RX_WRAPPED(rx) - p) < 16);
4359 r->pre_prefix = p - RX_WRAPPED(rx);
4368 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4370 if (RExC_seen & REG_SEEN_RECURSE) {
4371 Newxz(RExC_open_parens, RExC_npar,regnode *);
4372 SAVEFREEPV(RExC_open_parens);
4373 Newxz(RExC_close_parens,RExC_npar,regnode *);
4374 SAVEFREEPV(RExC_close_parens);
4377 /* Useful during FAIL. */
4378 #ifdef RE_TRACK_PATTERN_OFFSETS
4379 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4380 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4381 "%s %"UVuf" bytes for offset annotations.\n",
4382 ri->u.offsets ? "Got" : "Couldn't get",
4383 (UV)((2*RExC_size+1) * sizeof(U32))));
4385 SetProgLen(ri,RExC_size);
4390 /* Second pass: emit code. */
4391 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4396 RExC_emit_start = ri->program;
4397 RExC_emit = ri->program;
4398 RExC_emit_bound = ri->program + RExC_size + 1;
4400 /* Store the count of eval-groups for security checks: */
4401 RExC_rx->seen_evals = RExC_seen_evals;
4402 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4403 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4407 /* XXXX To minimize changes to RE engine we always allocate
4408 3-units-long substrs field. */
4409 Newx(r->substrs, 1, struct reg_substr_data);
4410 if (RExC_recurse_count) {
4411 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4412 SAVEFREEPV(RExC_recurse);
4416 r->minlen = minlen = sawplus = sawopen = 0;
4417 Zero(r->substrs, 1, struct reg_substr_data);
4419 #ifdef TRIE_STUDY_OPT
4421 StructCopy(&zero_scan_data, &data, scan_data_t);
4422 copyRExC_state = RExC_state;
4425 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4427 RExC_state = copyRExC_state;
4428 if (seen & REG_TOP_LEVEL_BRANCHES)
4429 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4431 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4432 if (data.last_found) {
4433 SvREFCNT_dec(data.longest_fixed);
4434 SvREFCNT_dec(data.longest_float);
4435 SvREFCNT_dec(data.last_found);
4437 StructCopy(&zero_scan_data, &data, scan_data_t);
4440 StructCopy(&zero_scan_data, &data, scan_data_t);
4443 /* Dig out information for optimizations. */
4444 r->extflags = RExC_flags; /* was pm_op */
4445 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4448 SvUTF8_on(rx); /* Unicode in it? */
4449 ri->regstclass = NULL;
4450 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4451 r->intflags |= PREGf_NAUGHTY;
4452 scan = ri->program + 1; /* First BRANCH. */
4454 /* testing for BRANCH here tells us whether there is "must appear"
4455 data in the pattern. If there is then we can use it for optimisations */
4456 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4458 STRLEN longest_float_length, longest_fixed_length;
4459 struct regnode_charclass_class ch_class; /* pointed to by data */
4461 I32 last_close = 0; /* pointed to by data */
4462 regnode *first= scan;
4463 regnode *first_next= regnext(first);
4466 * Skip introductions and multiplicators >= 1
4467 * so that we can extract the 'meat' of the pattern that must
4468 * match in the large if() sequence following.
4469 * NOTE that EXACT is NOT covered here, as it is normally
4470 * picked up by the optimiser separately.
4472 * This is unfortunate as the optimiser isnt handling lookahead
4473 * properly currently.
4476 while ((OP(first) == OPEN && (sawopen = 1)) ||
4477 /* An OR of *one* alternative - should not happen now. */
4478 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4479 /* for now we can't handle lookbehind IFMATCH*/
4480 (OP(first) == IFMATCH && !first->flags) ||
4481 (OP(first) == PLUS) ||
4482 (OP(first) == MINMOD) ||
4483 /* An {n,m} with n>0 */
4484 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4485 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4488 * the only op that could be a regnode is PLUS, all the rest
4489 * will be regnode_1 or regnode_2.
4492 if (OP(first) == PLUS)
4495 first += regarglen[OP(first)];
4497 first = NEXTOPER(first);
4498 first_next= regnext(first);
4501 /* Starting-point info. */
4503 DEBUG_PEEP("first:",first,0);
4504 /* Ignore EXACT as we deal with it later. */
4505 if (PL_regkind[OP(first)] == EXACT) {
4506 if (OP(first) == EXACT)
4507 NOOP; /* Empty, get anchored substr later. */
4508 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4509 ri->regstclass = first;
4512 else if (PL_regkind[OP(first)] == TRIE &&
4513 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4516 /* this can happen only on restudy */
4517 if ( OP(first) == TRIE ) {
4518 struct regnode_1 *trieop = (struct regnode_1 *)
4519 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4520 StructCopy(first,trieop,struct regnode_1);
4521 trie_op=(regnode *)trieop;
4523 struct regnode_charclass *trieop = (struct regnode_charclass *)
4524 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4525 StructCopy(first,trieop,struct regnode_charclass);
4526 trie_op=(regnode *)trieop;
4529 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4530 ri->regstclass = trie_op;
4533 else if (strchr((const char*)PL_simple,OP(first)))
4534 ri->regstclass = first;
4535 else if (PL_regkind[OP(first)] == BOUND ||
4536 PL_regkind[OP(first)] == NBOUND)
4537 ri->regstclass = first;
4538 else if (PL_regkind[OP(first)] == BOL) {
4539 r->extflags |= (OP(first) == MBOL
4541 : (OP(first) == SBOL
4544 first = NEXTOPER(first);
4547 else if (OP(first) == GPOS) {
4548 r->extflags |= RXf_ANCH_GPOS;
4549 first = NEXTOPER(first);
4552 else if ((!sawopen || !RExC_sawback) &&
4553 (OP(first) == STAR &&
4554 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4555 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4557 /* turn .* into ^.* with an implied $*=1 */
4559 (OP(NEXTOPER(first)) == REG_ANY)
4562 r->extflags |= type;
4563 r->intflags |= PREGf_IMPLICIT;
4564 first = NEXTOPER(first);
4567 if (sawplus && (!sawopen || !RExC_sawback)
4568 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4569 /* x+ must match at the 1st pos of run of x's */
4570 r->intflags |= PREGf_SKIP;
4572 /* Scan is after the zeroth branch, first is atomic matcher. */
4573 #ifdef TRIE_STUDY_OPT
4576 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4577 (IV)(first - scan + 1))
4581 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4582 (IV)(first - scan + 1))
4588 * If there's something expensive in the r.e., find the
4589 * longest literal string that must appear and make it the
4590 * regmust. Resolve ties in favor of later strings, since
4591 * the regstart check works with the beginning of the r.e.
4592 * and avoiding duplication strengthens checking. Not a
4593 * strong reason, but sufficient in the absence of others.
4594 * [Now we resolve ties in favor of the earlier string if
4595 * it happens that c_offset_min has been invalidated, since the
4596 * earlier string may buy us something the later one won't.]
4599 data.longest_fixed = newSVpvs("");
4600 data.longest_float = newSVpvs("");
4601 data.last_found = newSVpvs("");
4602 data.longest = &(data.longest_fixed);
4604 if (!ri->regstclass) {
4605 cl_init(pRExC_state, &ch_class);
4606 data.start_class = &ch_class;
4607 stclass_flag = SCF_DO_STCLASS_AND;
4608 } else /* XXXX Check for BOUND? */
4610 data.last_closep = &last_close;
4612 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4613 &data, -1, NULL, NULL,
4614 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4620 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4621 && data.last_start_min == 0 && data.last_end > 0
4622 && !RExC_seen_zerolen
4623 && !(RExC_seen & REG_SEEN_VERBARG)
4624 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4625 r->extflags |= RXf_CHECK_ALL;
4626 scan_commit(pRExC_state, &data,&minlen,0);
4627 SvREFCNT_dec(data.last_found);
4629 /* Note that code very similar to this but for anchored string
4630 follows immediately below, changes may need to be made to both.
4633 longest_float_length = CHR_SVLEN(data.longest_float);
4634 if (longest_float_length
4635 || (data.flags & SF_FL_BEFORE_EOL
4636 && (!(data.flags & SF_FL_BEFORE_MEOL)
4637 || (RExC_flags & RXf_PMf_MULTILINE))))
4641 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4642 && data.offset_fixed == data.offset_float_min
4643 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4644 goto remove_float; /* As in (a)+. */
4646 /* copy the information about the longest float from the reg_scan_data
4647 over to the program. */
4648 if (SvUTF8(data.longest_float)) {
4649 r->float_utf8 = data.longest_float;
4650 r->float_substr = NULL;
4652 r->float_substr = data.longest_float;
4653 r->float_utf8 = NULL;
4655 /* float_end_shift is how many chars that must be matched that
4656 follow this item. We calculate it ahead of time as once the
4657 lookbehind offset is added in we lose the ability to correctly
4659 ml = data.minlen_float ? *(data.minlen_float)
4660 : (I32)longest_float_length;
4661 r->float_end_shift = ml - data.offset_float_min
4662 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4663 + data.lookbehind_float;
4664 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4665 r->float_max_offset = data.offset_float_max;
4666 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4667 r->float_max_offset -= data.lookbehind_float;
4669 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4670 && (!(data.flags & SF_FL_BEFORE_MEOL)
4671 || (RExC_flags & RXf_PMf_MULTILINE)));
4672 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4676 r->float_substr = r->float_utf8 = NULL;
4677 SvREFCNT_dec(data.longest_float);
4678 longest_float_length = 0;
4681 /* Note that code very similar to this but for floating string
4682 is immediately above, changes may need to be made to both.
4685 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4686 if (longest_fixed_length
4687 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4688 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4689 || (RExC_flags & RXf_PMf_MULTILINE))))
4693 /* copy the information about the longest fixed
4694 from the reg_scan_data over to the program. */
4695 if (SvUTF8(data.longest_fixed)) {
4696 r->anchored_utf8 = data.longest_fixed;
4697 r->anchored_substr = NULL;
4699 r->anchored_substr = data.longest_fixed;
4700 r->anchored_utf8 = NULL;
4702 /* fixed_end_shift is how many chars that must be matched that
4703 follow this item. We calculate it ahead of time as once the
4704 lookbehind offset is added in we lose the ability to correctly
4706 ml = data.minlen_fixed ? *(data.minlen_fixed)
4707 : (I32)longest_fixed_length;
4708 r->anchored_end_shift = ml - data.offset_fixed
4709 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4710 + data.lookbehind_fixed;
4711 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4713 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4714 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4715 || (RExC_flags & RXf_PMf_MULTILINE)));
4716 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4719 r->anchored_substr = r->anchored_utf8 = NULL;
4720 SvREFCNT_dec(data.longest_fixed);
4721 longest_fixed_length = 0;
4724 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4725 ri->regstclass = NULL;
4726 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4728 && !(data.start_class->flags & ANYOF_EOS)
4729 && !cl_is_anything(data.start_class))
4731 const U32 n = add_data(pRExC_state, 1, "f");
4733 Newx(RExC_rxi->data->data[n], 1,
4734 struct regnode_charclass_class);
4735 StructCopy(data.start_class,
4736 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4737 struct regnode_charclass_class);
4738 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4739 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4740 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4741 regprop(r, sv, (regnode*)data.start_class);
4742 PerlIO_printf(Perl_debug_log,
4743 "synthetic stclass \"%s\".\n",
4744 SvPVX_const(sv));});
4747 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4748 if (longest_fixed_length > longest_float_length) {
4749 r->check_end_shift = r->anchored_end_shift;
4750 r->check_substr = r->anchored_substr;
4751 r->check_utf8 = r->anchored_utf8;
4752 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4753 if (r->extflags & RXf_ANCH_SINGLE)
4754 r->extflags |= RXf_NOSCAN;
4757 r->check_end_shift = r->float_end_shift;
4758 r->check_substr = r->float_substr;
4759 r->check_utf8 = r->float_utf8;
4760 r->check_offset_min = r->float_min_offset;
4761 r->check_offset_max = r->float_max_offset;
4763 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4764 This should be changed ASAP! */
4765 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4766 r->extflags |= RXf_USE_INTUIT;
4767 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4768 r->extflags |= RXf_INTUIT_TAIL;
4770 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4771 if ( (STRLEN)minlen < longest_float_length )
4772 minlen= longest_float_length;
4773 if ( (STRLEN)minlen < longest_fixed_length )
4774 minlen= longest_fixed_length;
4778 /* Several toplevels. Best we can is to set minlen. */
4780 struct regnode_charclass_class ch_class;
4783 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4785 scan = ri->program + 1;
4786 cl_init(pRExC_state, &ch_class);
4787 data.start_class = &ch_class;
4788 data.last_closep = &last_close;
4791 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4792 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4796 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4797 = r->float_substr = r->float_utf8 = NULL;
4798 if (!(data.start_class->flags & ANYOF_EOS)
4799 && !cl_is_anything(data.start_class))
4801 const U32 n = add_data(pRExC_state, 1, "f");
4803 Newx(RExC_rxi->data->data[n], 1,
4804 struct regnode_charclass_class);
4805 StructCopy(data.start_class,
4806 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4807 struct regnode_charclass_class);
4808 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4809 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4810 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4811 regprop(r, sv, (regnode*)data.start_class);
4812 PerlIO_printf(Perl_debug_log,
4813 "synthetic stclass \"%s\".\n",
4814 SvPVX_const(sv));});
4818 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4819 the "real" pattern. */
4821 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4822 (IV)minlen, (IV)r->minlen);
4824 r->minlenret = minlen;
4825 if (r->minlen < minlen)
4828 if (RExC_seen & REG_SEEN_GPOS)
4829 r->extflags |= RXf_GPOS_SEEN;
4830 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4831 r->extflags |= RXf_LOOKBEHIND_SEEN;
4832 if (RExC_seen & REG_SEEN_EVAL)
4833 r->extflags |= RXf_EVAL_SEEN;
4834 if (RExC_seen & REG_SEEN_CANY)
4835 r->extflags |= RXf_CANY_SEEN;
4836 if (RExC_seen & REG_SEEN_VERBARG)
4837 r->intflags |= PREGf_VERBARG_SEEN;
4838 if (RExC_seen & REG_SEEN_CUTGROUP)
4839 r->intflags |= PREGf_CUTGROUP_SEEN;
4840 if (RExC_paren_names)
4841 RXp_PAREN_NAMES(r) = (HV*)SvREFCNT_inc(RExC_paren_names);
4843 RXp_PAREN_NAMES(r) = NULL;
4845 #ifdef STUPID_PATTERN_CHECKS
4846 if (RX_PRELEN(rx) == 0)
4847 r->extflags |= RXf_NULL;
4848 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4849 /* XXX: this should happen BEFORE we compile */
4850 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4851 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
4852 r->extflags |= RXf_WHITE;
4853 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
4854 r->extflags |= RXf_START_ONLY;
4856 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4857 /* XXX: this should happen BEFORE we compile */
4858 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4860 regnode *first = ri->program + 1;
4862 U8 nop = OP(NEXTOPER(first));
4864 if (PL_regkind[fop] == NOTHING && nop == END)
4865 r->extflags |= RXf_NULL;
4866 else if (PL_regkind[fop] == BOL && nop == END)
4867 r->extflags |= RXf_START_ONLY;
4868 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4869 r->extflags |= RXf_WHITE;
4873 if (RExC_paren_names) {
4874 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4875 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4878 ri->name_list_idx = 0;
4880 if (RExC_recurse_count) {
4881 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4882 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4883 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4886 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4887 /* assume we don't need to swap parens around before we match */
4890 PerlIO_printf(Perl_debug_log,"Final program:\n");
4893 #ifdef RE_TRACK_PATTERN_OFFSETS
4894 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4895 const U32 len = ri->u.offsets[0];
4897 GET_RE_DEBUG_FLAGS_DECL;
4898 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4899 for (i = 1; i <= len; i++) {
4900 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4901 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4902 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4904 PerlIO_printf(Perl_debug_log, "\n");
4910 #undef RE_ENGINE_PTR
4914 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4917 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
4919 PERL_UNUSED_ARG(value);
4921 if (flags & RXapif_FETCH) {
4922 return reg_named_buff_fetch(rx, key, flags);
4923 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
4924 Perl_croak(aTHX_ PL_no_modify);
4926 } else if (flags & RXapif_EXISTS) {
4927 return reg_named_buff_exists(rx, key, flags)
4930 } else if (flags & RXapif_REGNAMES) {
4931 return reg_named_buff_all(rx, flags);
4932 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
4933 return reg_named_buff_scalar(rx, flags);
4935 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
4941 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
4944 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
4945 PERL_UNUSED_ARG(lastkey);
4947 if (flags & RXapif_FIRSTKEY)
4948 return reg_named_buff_firstkey(rx, flags);
4949 else if (flags & RXapif_NEXTKEY)
4950 return reg_named_buff_nextkey(rx, flags);
4952 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
4958 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
4961 AV *retarray = NULL;
4963 struct regexp *const rx = (struct regexp *)SvANY(r);
4965 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
4967 if (flags & RXapif_ALL)
4970 if (rx && RXp_PAREN_NAMES(rx)) {
4971 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
4974 SV* sv_dat=HeVAL(he_str);
4975 I32 *nums=(I32*)SvPVX(sv_dat);
4976 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4977 if ((I32)(rx->nparens) >= nums[i]
4978 && rx->offs[nums[i]].start != -1
4979 && rx->offs[nums[i]].end != -1)
4982 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
4986 ret = newSVsv(&PL_sv_undef);
4989 SvREFCNT_inc_simple_void(ret);
4990 av_push(retarray, ret);
4994 return newRV((SV*)retarray);
5001 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5004 struct regexp *const rx = (struct regexp *)SvANY(r);
5006 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5008 if (rx && RXp_PAREN_NAMES(rx)) {
5009 if (flags & RXapif_ALL) {
5010 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5012 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5026 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5028 struct regexp *const rx = (struct regexp *)SvANY(r);
5030 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5032 if ( rx && RXp_PAREN_NAMES(rx) ) {
5033 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5035 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5042 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5044 struct regexp *const rx = (struct regexp *)SvANY(r);
5045 GET_RE_DEBUG_FLAGS_DECL;
5047 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5049 if (rx && RXp_PAREN_NAMES(rx)) {
5050 HV *hv = RXp_PAREN_NAMES(rx);
5052 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5055 SV* sv_dat = HeVAL(temphe);
5056 I32 *nums = (I32*)SvPVX(sv_dat);
5057 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5058 if ((I32)(rx->lastparen) >= nums[i] &&
5059 rx->offs[nums[i]].start != -1 &&
5060 rx->offs[nums[i]].end != -1)
5066 if (parno || flags & RXapif_ALL) {
5067 return newSVhek(HeKEY_hek(temphe));
5075 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5080 struct regexp *const rx = (struct regexp *)SvANY(r);
5082 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5084 if (rx && RXp_PAREN_NAMES(rx)) {
5085 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5086 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5087 } else if (flags & RXapif_ONE) {
5088 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5089 av = (AV*)SvRV(ret);
5090 length = av_len(av);
5091 return newSViv(length + 1);
5093 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5097 return &PL_sv_undef;
5101 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5103 struct regexp *const rx = (struct regexp *)SvANY(r);
5106 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5108 if (rx && RXp_PAREN_NAMES(rx)) {
5109 HV *hv= RXp_PAREN_NAMES(rx);
5111 (void)hv_iterinit(hv);
5112 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5115 SV* sv_dat = HeVAL(temphe);
5116 I32 *nums = (I32*)SvPVX(sv_dat);
5117 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5118 if ((I32)(rx->lastparen) >= nums[i] &&
5119 rx->offs[nums[i]].start != -1 &&
5120 rx->offs[nums[i]].end != -1)
5126 if (parno || flags & RXapif_ALL) {
5127 av_push(av, newSVhek(HeKEY_hek(temphe)));
5132 return newRV((SV*)av);
5136 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5139 struct regexp *const rx = (struct regexp *)SvANY(r);
5144 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5147 sv_setsv(sv,&PL_sv_undef);
5151 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5153 i = rx->offs[0].start;
5157 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5159 s = rx->subbeg + rx->offs[0].end;
5160 i = rx->sublen - rx->offs[0].end;
5163 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5164 (s1 = rx->offs[paren].start) != -1 &&
5165 (t1 = rx->offs[paren].end) != -1)
5169 s = rx->subbeg + s1;
5171 sv_setsv(sv,&PL_sv_undef);
5174 assert(rx->sublen >= (s - rx->subbeg) + i );
5176 const int oldtainted = PL_tainted;
5178 sv_setpvn(sv, s, i);
5179 PL_tainted = oldtainted;
5180 if ( (rx->extflags & RXf_CANY_SEEN)
5181 ? (RXp_MATCH_UTF8(rx)
5182 && (!i || is_utf8_string((U8*)s, i)))
5183 : (RXp_MATCH_UTF8(rx)) )
5190 if (RXp_MATCH_TAINTED(rx)) {
5191 if (SvTYPE(sv) >= SVt_PVMG) {
5192 MAGIC* const mg = SvMAGIC(sv);
5195 SvMAGIC_set(sv, mg->mg_moremagic);
5197 if ((mgt = SvMAGIC(sv))) {
5198 mg->mg_moremagic = mgt;
5199 SvMAGIC_set(sv, mg);
5209 sv_setsv(sv,&PL_sv_undef);
5215 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5216 SV const * const value)
5218 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5220 PERL_UNUSED_ARG(rx);
5221 PERL_UNUSED_ARG(paren);
5222 PERL_UNUSED_ARG(value);
5225 Perl_croak(aTHX_ PL_no_modify);
5229 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5232 struct regexp *const rx = (struct regexp *)SvANY(r);
5236 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5238 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5240 /* $` / ${^PREMATCH} */
5241 case RX_BUFF_IDX_PREMATCH:
5242 if (rx->offs[0].start != -1) {
5243 i = rx->offs[0].start;
5251 /* $' / ${^POSTMATCH} */
5252 case RX_BUFF_IDX_POSTMATCH:
5253 if (rx->offs[0].end != -1) {
5254 i = rx->sublen - rx->offs[0].end;
5256 s1 = rx->offs[0].end;
5262 /* $& / ${^MATCH}, $1, $2, ... */
5264 if (paren <= (I32)rx->nparens &&
5265 (s1 = rx->offs[paren].start) != -1 &&
5266 (t1 = rx->offs[paren].end) != -1)
5271 if (ckWARN(WARN_UNINITIALIZED))
5272 report_uninit((SV*)sv);
5277 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5278 const char * const s = rx->subbeg + s1;
5283 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5290 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5292 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5293 PERL_UNUSED_ARG(rx);
5297 return newSVpvs("Regexp");
5300 /* Scans the name of a named buffer from the pattern.
5301 * If flags is REG_RSN_RETURN_NULL returns null.
5302 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5303 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5304 * to the parsed name as looked up in the RExC_paren_names hash.
5305 * If there is an error throws a vFAIL().. type exception.
5308 #define REG_RSN_RETURN_NULL 0
5309 #define REG_RSN_RETURN_NAME 1
5310 #define REG_RSN_RETURN_DATA 2
5313 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5315 char *name_start = RExC_parse;
5317 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5319 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5320 /* skip IDFIRST by using do...while */
5323 RExC_parse += UTF8SKIP(RExC_parse);
5324 } while (isALNUM_utf8((U8*)RExC_parse));
5328 } while (isALNUM(*RExC_parse));
5333 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5334 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5335 if ( flags == REG_RSN_RETURN_NAME)
5337 else if (flags==REG_RSN_RETURN_DATA) {
5340 if ( ! sv_name ) /* should not happen*/
5341 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5342 if (RExC_paren_names)
5343 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5345 sv_dat = HeVAL(he_str);
5347 vFAIL("Reference to nonexistent named group");
5351 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5358 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5359 int rem=(int)(RExC_end - RExC_parse); \
5368 if (RExC_lastparse!=RExC_parse) \
5369 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5372 iscut ? "..." : "<" \
5375 PerlIO_printf(Perl_debug_log,"%16s",""); \
5378 num = RExC_size + 1; \
5380 num=REG_NODE_NUM(RExC_emit); \
5381 if (RExC_lastnum!=num) \
5382 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5384 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5385 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5386 (int)((depth*2)), "", \
5390 RExC_lastparse=RExC_parse; \
5395 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5396 DEBUG_PARSE_MSG((funcname)); \
5397 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5399 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5400 DEBUG_PARSE_MSG((funcname)); \
5401 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5404 - reg - regular expression, i.e. main body or parenthesized thing
5406 * Caller must absorb opening parenthesis.
5408 * Combining parenthesis handling with the base level of regular expression
5409 * is a trifle forced, but the need to tie the tails of the branches to what
5410 * follows makes it hard to avoid.
5412 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5414 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5416 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5420 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5421 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5424 register regnode *ret; /* Will be the head of the group. */
5425 register regnode *br;
5426 register regnode *lastbr;
5427 register regnode *ender = NULL;
5428 register I32 parno = 0;
5430 U32 oregflags = RExC_flags;
5431 bool have_branch = 0;
5433 I32 freeze_paren = 0;
5434 I32 after_freeze = 0;
5436 /* for (?g), (?gc), and (?o) warnings; warning
5437 about (?c) will warn about (?g) -- japhy */
5439 #define WASTED_O 0x01
5440 #define WASTED_G 0x02
5441 #define WASTED_C 0x04
5442 #define WASTED_GC (0x02|0x04)
5443 I32 wastedflags = 0x00;
5445 char * parse_start = RExC_parse; /* MJD */
5446 char * const oregcomp_parse = RExC_parse;
5448 GET_RE_DEBUG_FLAGS_DECL;
5450 PERL_ARGS_ASSERT_REG;
5451 DEBUG_PARSE("reg ");
5453 *flagp = 0; /* Tentatively. */
5456 /* Make an OPEN node, if parenthesized. */
5458 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5459 char *start_verb = RExC_parse;
5460 STRLEN verb_len = 0;
5461 char *start_arg = NULL;
5462 unsigned char op = 0;
5464 int internal_argval = 0; /* internal_argval is only useful if !argok */
5465 while ( *RExC_parse && *RExC_parse != ')' ) {
5466 if ( *RExC_parse == ':' ) {
5467 start_arg = RExC_parse + 1;
5473 verb_len = RExC_parse - start_verb;
5476 while ( *RExC_parse && *RExC_parse != ')' )
5478 if ( *RExC_parse != ')' )
5479 vFAIL("Unterminated verb pattern argument");
5480 if ( RExC_parse == start_arg )
5483 if ( *RExC_parse != ')' )
5484 vFAIL("Unterminated verb pattern");
5487 switch ( *start_verb ) {
5488 case 'A': /* (*ACCEPT) */
5489 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5491 internal_argval = RExC_nestroot;
5494 case 'C': /* (*COMMIT) */
5495 if ( memEQs(start_verb,verb_len,"COMMIT") )
5498 case 'F': /* (*FAIL) */
5499 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5504 case ':': /* (*:NAME) */
5505 case 'M': /* (*MARK:NAME) */
5506 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5511 case 'P': /* (*PRUNE) */
5512 if ( memEQs(start_verb,verb_len,"PRUNE") )
5515 case 'S': /* (*SKIP) */
5516 if ( memEQs(start_verb,verb_len,"SKIP") )
5519 case 'T': /* (*THEN) */
5520 /* [19:06] <TimToady> :: is then */
5521 if ( memEQs(start_verb,verb_len,"THEN") ) {
5523 RExC_seen |= REG_SEEN_CUTGROUP;
5529 vFAIL3("Unknown verb pattern '%.*s'",
5530 verb_len, start_verb);
5533 if ( start_arg && internal_argval ) {
5534 vFAIL3("Verb pattern '%.*s' may not have an argument",
5535 verb_len, start_verb);
5536 } else if ( argok < 0 && !start_arg ) {
5537 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5538 verb_len, start_verb);
5540 ret = reganode(pRExC_state, op, internal_argval);
5541 if ( ! internal_argval && ! SIZE_ONLY ) {
5543 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5544 ARG(ret) = add_data( pRExC_state, 1, "S" );
5545 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5552 if (!internal_argval)
5553 RExC_seen |= REG_SEEN_VERBARG;
5554 } else if ( start_arg ) {
5555 vFAIL3("Verb pattern '%.*s' may not have an argument",
5556 verb_len, start_verb);
5558 ret = reg_node(pRExC_state, op);
5560 nextchar(pRExC_state);
5563 if (*RExC_parse == '?') { /* (?...) */
5564 bool is_logical = 0;
5565 const char * const seqstart = RExC_parse;
5568 paren = *RExC_parse++;
5569 ret = NULL; /* For look-ahead/behind. */
5572 case 'P': /* (?P...) variants for those used to PCRE/Python */
5573 paren = *RExC_parse++;
5574 if ( paren == '<') /* (?P<...>) named capture */
5576 else if (paren == '>') { /* (?P>name) named recursion */
5577 goto named_recursion;
5579 else if (paren == '=') { /* (?P=...) named backref */
5580 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5581 you change this make sure you change that */
5582 char* name_start = RExC_parse;
5584 SV *sv_dat = reg_scan_name(pRExC_state,
5585 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5586 if (RExC_parse == name_start || *RExC_parse != ')')
5587 vFAIL2("Sequence %.3s... not terminated",parse_start);
5590 num = add_data( pRExC_state, 1, "S" );
5591 RExC_rxi->data->data[num]=(void*)sv_dat;
5592 SvREFCNT_inc_simple_void(sv_dat);
5595 ret = reganode(pRExC_state,
5596 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5600 Set_Node_Offset(ret, parse_start+1);
5601 Set_Node_Cur_Length(ret); /* MJD */
5603 nextchar(pRExC_state);
5607 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5609 case '<': /* (?<...) */
5610 if (*RExC_parse == '!')
5612 else if (*RExC_parse != '=')
5618 case '\'': /* (?'...') */
5619 name_start= RExC_parse;
5620 svname = reg_scan_name(pRExC_state,
5621 SIZE_ONLY ? /* reverse test from the others */
5622 REG_RSN_RETURN_NAME :
5623 REG_RSN_RETURN_NULL);
5624 if (RExC_parse == name_start) {
5626 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5629 if (*RExC_parse != paren)
5630 vFAIL2("Sequence (?%c... not terminated",
5631 paren=='>' ? '<' : paren);
5635 if (!svname) /* shouldnt happen */
5637 "panic: reg_scan_name returned NULL");
5638 if (!RExC_paren_names) {
5639 RExC_paren_names= newHV();
5640 sv_2mortal((SV*)RExC_paren_names);
5642 RExC_paren_name_list= newAV();
5643 sv_2mortal((SV*)RExC_paren_name_list);
5646 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5648 sv_dat = HeVAL(he_str);
5650 /* croak baby croak */
5652 "panic: paren_name hash element allocation failed");
5653 } else if ( SvPOK(sv_dat) ) {
5654 /* (?|...) can mean we have dupes so scan to check
5655 its already been stored. Maybe a flag indicating
5656 we are inside such a construct would be useful,
5657 but the arrays are likely to be quite small, so
5658 for now we punt -- dmq */
5659 IV count = SvIV(sv_dat);
5660 I32 *pv = (I32*)SvPVX(sv_dat);
5662 for ( i = 0 ; i < count ; i++ ) {
5663 if ( pv[i] == RExC_npar ) {
5669 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5670 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5671 pv[count] = RExC_npar;
5675 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5676 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5681 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5682 SvREFCNT_dec(svname);
5685 /*sv_dump(sv_dat);*/
5687 nextchar(pRExC_state);
5689 goto capturing_parens;
5691 RExC_seen |= REG_SEEN_LOOKBEHIND;
5693 case '=': /* (?=...) */
5694 case '!': /* (?!...) */
5695 RExC_seen_zerolen++;
5696 if (*RExC_parse == ')') {
5697 ret=reg_node(pRExC_state, OPFAIL);
5698 nextchar(pRExC_state);
5702 case '|': /* (?|...) */
5703 /* branch reset, behave like a (?:...) except that
5704 buffers in alternations share the same numbers */
5706 after_freeze = freeze_paren = RExC_npar;
5708 case ':': /* (?:...) */
5709 case '>': /* (?>...) */
5711 case '$': /* (?$...) */
5712 case '@': /* (?@...) */
5713 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5715 case '#': /* (?#...) */
5716 while (*RExC_parse && *RExC_parse != ')')
5718 if (*RExC_parse != ')')
5719 FAIL("Sequence (?#... not terminated");
5720 nextchar(pRExC_state);
5723 case '0' : /* (?0) */
5724 case 'R' : /* (?R) */
5725 if (*RExC_parse != ')')
5726 FAIL("Sequence (?R) not terminated");
5727 ret = reg_node(pRExC_state, GOSTART);
5728 *flagp |= POSTPONED;
5729 nextchar(pRExC_state);
5732 { /* named and numeric backreferences */
5734 case '&': /* (?&NAME) */
5735 parse_start = RExC_parse - 1;
5738 SV *sv_dat = reg_scan_name(pRExC_state,
5739 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5740 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5742 goto gen_recurse_regop;
5745 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5747 vFAIL("Illegal pattern");
5749 goto parse_recursion;
5751 case '-': /* (?-1) */
5752 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5753 RExC_parse--; /* rewind to let it be handled later */
5757 case '1': case '2': case '3': case '4': /* (?1) */
5758 case '5': case '6': case '7': case '8': case '9':
5761 num = atoi(RExC_parse);
5762 parse_start = RExC_parse - 1; /* MJD */
5763 if (*RExC_parse == '-')
5765 while (isDIGIT(*RExC_parse))
5767 if (*RExC_parse!=')')
5768 vFAIL("Expecting close bracket");
5771 if ( paren == '-' ) {
5773 Diagram of capture buffer numbering.
5774 Top line is the normal capture buffer numbers
5775 Botton line is the negative indexing as from
5779 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5783 num = RExC_npar + num;
5786 vFAIL("Reference to nonexistent group");
5788 } else if ( paren == '+' ) {
5789 num = RExC_npar + num - 1;
5792 ret = reganode(pRExC_state, GOSUB, num);
5794 if (num > (I32)RExC_rx->nparens) {
5796 vFAIL("Reference to nonexistent group");
5798 ARG2L_SET( ret, RExC_recurse_count++);
5800 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5801 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5805 RExC_seen |= REG_SEEN_RECURSE;
5806 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5807 Set_Node_Offset(ret, parse_start); /* MJD */
5809 *flagp |= POSTPONED;
5810 nextchar(pRExC_state);
5812 } /* named and numeric backreferences */
5815 case '?': /* (??...) */
5817 if (*RExC_parse != '{') {
5819 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5822 *flagp |= POSTPONED;
5823 paren = *RExC_parse++;
5825 case '{': /* (?{...}) */
5830 char *s = RExC_parse;
5832 RExC_seen_zerolen++;
5833 RExC_seen |= REG_SEEN_EVAL;
5834 while (count && (c = *RExC_parse)) {
5845 if (*RExC_parse != ')') {
5847 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5851 OP_4tree *sop, *rop;
5852 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5855 Perl_save_re_context(aTHX);
5856 rop = sv_compile_2op(sv, &sop, "re", &pad);
5857 sop->op_private |= OPpREFCOUNTED;
5858 /* re_dup will OpREFCNT_inc */
5859 OpREFCNT_set(sop, 1);
5862 n = add_data(pRExC_state, 3, "nop");
5863 RExC_rxi->data->data[n] = (void*)rop;
5864 RExC_rxi->data->data[n+1] = (void*)sop;
5865 RExC_rxi->data->data[n+2] = (void*)pad;
5868 else { /* First pass */
5869 if (PL_reginterp_cnt < ++RExC_seen_evals
5871 /* No compiled RE interpolated, has runtime
5872 components ===> unsafe. */
5873 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5874 if (PL_tainting && PL_tainted)
5875 FAIL("Eval-group in insecure regular expression");
5876 #if PERL_VERSION > 8
5877 if (IN_PERL_COMPILETIME)
5882 nextchar(pRExC_state);
5884 ret = reg_node(pRExC_state, LOGICAL);
5887 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5888 /* deal with the length of this later - MJD */
5891 ret = reganode(pRExC_state, EVAL, n);
5892 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5893 Set_Node_Offset(ret, parse_start);
5896 case '(': /* (?(?{...})...) and (?(?=...)...) */
5899 if (RExC_parse[0] == '?') { /* (?(?...)) */
5900 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5901 || RExC_parse[1] == '<'
5902 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5905 ret = reg_node(pRExC_state, LOGICAL);
5908 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5912 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5913 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5915 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5916 char *name_start= RExC_parse++;
5918 SV *sv_dat=reg_scan_name(pRExC_state,
5919 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5920 if (RExC_parse == name_start || *RExC_parse != ch)
5921 vFAIL2("Sequence (?(%c... not terminated",
5922 (ch == '>' ? '<' : ch));
5925 num = add_data( pRExC_state, 1, "S" );
5926 RExC_rxi->data->data[num]=(void*)sv_dat;
5927 SvREFCNT_inc_simple_void(sv_dat);
5929 ret = reganode(pRExC_state,NGROUPP,num);
5930 goto insert_if_check_paren;
5932 else if (RExC_parse[0] == 'D' &&
5933 RExC_parse[1] == 'E' &&
5934 RExC_parse[2] == 'F' &&
5935 RExC_parse[3] == 'I' &&
5936 RExC_parse[4] == 'N' &&
5937 RExC_parse[5] == 'E')
5939 ret = reganode(pRExC_state,DEFINEP,0);
5942 goto insert_if_check_paren;
5944 else if (RExC_parse[0] == 'R') {
5947 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5948 parno = atoi(RExC_parse++);
5949 while (isDIGIT(*RExC_parse))
5951 } else if (RExC_parse[0] == '&') {
5954 sv_dat = reg_scan_name(pRExC_state,
5955 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5956 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5958 ret = reganode(pRExC_state,INSUBP,parno);
5959 goto insert_if_check_paren;
5961 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5964 parno = atoi(RExC_parse++);
5966 while (isDIGIT(*RExC_parse))
5968 ret = reganode(pRExC_state, GROUPP, parno);
5970 insert_if_check_paren:
5971 if ((c = *nextchar(pRExC_state)) != ')')
5972 vFAIL("Switch condition not recognized");
5974 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5975 br = regbranch(pRExC_state, &flags, 1,depth+1);
5977 br = reganode(pRExC_state, LONGJMP, 0);
5979 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5980 c = *nextchar(pRExC_state);
5985 vFAIL("(?(DEFINE)....) does not allow branches");
5986 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5987 regbranch(pRExC_state, &flags, 1,depth+1);
5988 REGTAIL(pRExC_state, ret, lastbr);
5991 c = *nextchar(pRExC_state);
5996 vFAIL("Switch (?(condition)... contains too many branches");
5997 ender = reg_node(pRExC_state, TAIL);
5998 REGTAIL(pRExC_state, br, ender);
6000 REGTAIL(pRExC_state, lastbr, ender);
6001 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6004 REGTAIL(pRExC_state, ret, ender);
6005 RExC_size++; /* XXX WHY do we need this?!!
6006 For large programs it seems to be required
6007 but I can't figure out why. -- dmq*/
6011 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6015 RExC_parse--; /* for vFAIL to print correctly */
6016 vFAIL("Sequence (? incomplete");
6020 parse_flags: /* (?i) */
6022 U32 posflags = 0, negflags = 0;
6023 U32 *flagsp = &posflags;
6025 while (*RExC_parse) {
6026 /* && strchr("iogcmsx", *RExC_parse) */
6027 /* (?g), (?gc) and (?o) are useless here
6028 and must be globally applied -- japhy */
6029 switch (*RExC_parse) {
6030 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6031 case ONCE_PAT_MOD: /* 'o' */
6032 case GLOBAL_PAT_MOD: /* 'g' */
6033 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6034 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6035 if (! (wastedflags & wflagbit) ) {
6036 wastedflags |= wflagbit;
6039 "Useless (%s%c) - %suse /%c modifier",
6040 flagsp == &negflags ? "?-" : "?",
6042 flagsp == &negflags ? "don't " : "",
6049 case CONTINUE_PAT_MOD: /* 'c' */
6050 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6051 if (! (wastedflags & WASTED_C) ) {
6052 wastedflags |= WASTED_GC;
6055 "Useless (%sc) - %suse /gc modifier",
6056 flagsp == &negflags ? "?-" : "?",
6057 flagsp == &negflags ? "don't " : ""
6062 case KEEPCOPY_PAT_MOD: /* 'p' */
6063 if (flagsp == &negflags) {
6064 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
6065 vWARN(RExC_parse + 1,"Useless use of (?-p)");
6067 *flagsp |= RXf_PMf_KEEPCOPY;
6071 if (flagsp == &negflags) {
6073 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6077 wastedflags = 0; /* reset so (?g-c) warns twice */
6083 RExC_flags |= posflags;
6084 RExC_flags &= ~negflags;
6086 oregflags |= posflags;
6087 oregflags &= ~negflags;
6089 nextchar(pRExC_state);
6100 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6105 }} /* one for the default block, one for the switch */
6112 ret = reganode(pRExC_state, OPEN, parno);
6115 RExC_nestroot = parno;
6116 if (RExC_seen & REG_SEEN_RECURSE
6117 && !RExC_open_parens[parno-1])
6119 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6120 "Setting open paren #%"IVdf" to %d\n",
6121 (IV)parno, REG_NODE_NUM(ret)));
6122 RExC_open_parens[parno-1]= ret;
6125 Set_Node_Length(ret, 1); /* MJD */
6126 Set_Node_Offset(ret, RExC_parse); /* MJD */
6134 /* Pick up the branches, linking them together. */
6135 parse_start = RExC_parse; /* MJD */
6136 br = regbranch(pRExC_state, &flags, 1,depth+1);
6137 /* branch_len = (paren != 0); */
6141 if (*RExC_parse == '|') {
6142 if (!SIZE_ONLY && RExC_extralen) {
6143 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6146 reginsert(pRExC_state, BRANCH, br, depth+1);
6147 Set_Node_Length(br, paren != 0);
6148 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6152 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6154 else if (paren == ':') {
6155 *flagp |= flags&SIMPLE;
6157 if (is_open) { /* Starts with OPEN. */
6158 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6160 else if (paren != '?') /* Not Conditional */
6162 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6164 while (*RExC_parse == '|') {
6165 if (!SIZE_ONLY && RExC_extralen) {
6166 ender = reganode(pRExC_state, LONGJMP,0);
6167 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6170 RExC_extralen += 2; /* Account for LONGJMP. */
6171 nextchar(pRExC_state);
6173 if (RExC_npar > after_freeze)
6174 after_freeze = RExC_npar;
6175 RExC_npar = freeze_paren;
6177 br = regbranch(pRExC_state, &flags, 0, depth+1);
6181 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6183 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6186 if (have_branch || paren != ':') {
6187 /* Make a closing node, and hook it on the end. */
6190 ender = reg_node(pRExC_state, TAIL);
6193 ender = reganode(pRExC_state, CLOSE, parno);
6194 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6195 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6196 "Setting close paren #%"IVdf" to %d\n",
6197 (IV)parno, REG_NODE_NUM(ender)));
6198 RExC_close_parens[parno-1]= ender;
6199 if (RExC_nestroot == parno)
6202 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6203 Set_Node_Length(ender,1); /* MJD */
6209 *flagp &= ~HASWIDTH;
6212 ender = reg_node(pRExC_state, SUCCEED);
6215 ender = reg_node(pRExC_state, END);
6217 assert(!RExC_opend); /* there can only be one! */
6222 REGTAIL(pRExC_state, lastbr, ender);
6224 if (have_branch && !SIZE_ONLY) {
6226 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6228 /* Hook the tails of the branches to the closing node. */
6229 for (br = ret; br; br = regnext(br)) {
6230 const U8 op = PL_regkind[OP(br)];
6232 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6234 else if (op == BRANCHJ) {
6235 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6243 static const char parens[] = "=!<,>";
6245 if (paren && (p = strchr(parens, paren))) {
6246 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6247 int flag = (p - parens) > 1;
6250 node = SUSPEND, flag = 0;
6251 reginsert(pRExC_state, node,ret, depth+1);
6252 Set_Node_Cur_Length(ret);
6253 Set_Node_Offset(ret, parse_start + 1);
6255 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6259 /* Check for proper termination. */
6261 RExC_flags = oregflags;
6262 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6263 RExC_parse = oregcomp_parse;
6264 vFAIL("Unmatched (");
6267 else if (!paren && RExC_parse < RExC_end) {
6268 if (*RExC_parse == ')') {
6270 vFAIL("Unmatched )");
6273 FAIL("Junk on end of regexp"); /* "Can't happen". */
6277 RExC_npar = after_freeze;
6282 - regbranch - one alternative of an | operator
6284 * Implements the concatenation operator.
6287 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6290 register regnode *ret;
6291 register regnode *chain = NULL;
6292 register regnode *latest;
6293 I32 flags = 0, c = 0;
6294 GET_RE_DEBUG_FLAGS_DECL;
6296 PERL_ARGS_ASSERT_REGBRANCH;
6298 DEBUG_PARSE("brnc");
6303 if (!SIZE_ONLY && RExC_extralen)
6304 ret = reganode(pRExC_state, BRANCHJ,0);
6306 ret = reg_node(pRExC_state, BRANCH);
6307 Set_Node_Length(ret, 1);
6311 if (!first && SIZE_ONLY)
6312 RExC_extralen += 1; /* BRANCHJ */
6314 *flagp = WORST; /* Tentatively. */
6317 nextchar(pRExC_state);
6318 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6320 latest = regpiece(pRExC_state, &flags,depth+1);
6321 if (latest == NULL) {
6322 if (flags & TRYAGAIN)
6326 else if (ret == NULL)
6328 *flagp |= flags&(HASWIDTH|POSTPONED);
6329 if (chain == NULL) /* First piece. */
6330 *flagp |= flags&SPSTART;
6333 REGTAIL(pRExC_state, chain, latest);
6338 if (chain == NULL) { /* Loop ran zero times. */
6339 chain = reg_node(pRExC_state, NOTHING);
6344 *flagp |= flags&SIMPLE;
6351 - regpiece - something followed by possible [*+?]
6353 * Note that the branching code sequences used for ? and the general cases
6354 * of * and + are somewhat optimized: they use the same NOTHING node as
6355 * both the endmarker for their branch list and the body of the last branch.
6356 * It might seem that this node could be dispensed with entirely, but the
6357 * endmarker role is not redundant.
6360 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6363 register regnode *ret;
6365 register char *next;
6367 const char * const origparse = RExC_parse;
6369 I32 max = REG_INFTY;
6371 const char *maxpos = NULL;
6372 GET_RE_DEBUG_FLAGS_DECL;
6374 PERL_ARGS_ASSERT_REGPIECE;
6376 DEBUG_PARSE("piec");
6378 ret = regatom(pRExC_state, &flags,depth+1);
6380 if (flags & TRYAGAIN)
6387 if (op == '{' && regcurly(RExC_parse)) {
6389 parse_start = RExC_parse; /* MJD */
6390 next = RExC_parse + 1;
6391 while (isDIGIT(*next) || *next == ',') {
6400 if (*next == '}') { /* got one */
6404 min = atoi(RExC_parse);
6408 maxpos = RExC_parse;
6410 if (!max && *maxpos != '0')
6411 max = REG_INFTY; /* meaning "infinity" */
6412 else if (max >= REG_INFTY)
6413 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6415 nextchar(pRExC_state);
6418 if ((flags&SIMPLE)) {
6419 RExC_naughty += 2 + RExC_naughty / 2;
6420 reginsert(pRExC_state, CURLY, ret, depth+1);
6421 Set_Node_Offset(ret, parse_start+1); /* MJD */
6422 Set_Node_Cur_Length(ret);
6425 regnode * const w = reg_node(pRExC_state, WHILEM);
6428 REGTAIL(pRExC_state, ret, w);
6429 if (!SIZE_ONLY && RExC_extralen) {
6430 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6431 reginsert(pRExC_state, NOTHING,ret, depth+1);
6432 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6434 reginsert(pRExC_state, CURLYX,ret, depth+1);
6436 Set_Node_Offset(ret, parse_start+1);
6437 Set_Node_Length(ret,
6438 op == '{' ? (RExC_parse - parse_start) : 1);
6440 if (!SIZE_ONLY && RExC_extralen)
6441 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6442 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6444 RExC_whilem_seen++, RExC_extralen += 3;
6445 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6453 if (max && max < min)
6454 vFAIL("Can't do {n,m} with n > m");
6456 ARG1_SET(ret, (U16)min);
6457 ARG2_SET(ret, (U16)max);
6469 #if 0 /* Now runtime fix should be reliable. */
6471 /* if this is reinstated, don't forget to put this back into perldiag:
6473 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6475 (F) The part of the regexp subject to either the * or + quantifier
6476 could match an empty string. The {#} shows in the regular
6477 expression about where the problem was discovered.
6481 if (!(flags&HASWIDTH) && op != '?')
6482 vFAIL("Regexp *+ operand could be empty");
6485 parse_start = RExC_parse;
6486 nextchar(pRExC_state);
6488 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6490 if (op == '*' && (flags&SIMPLE)) {
6491 reginsert(pRExC_state, STAR, ret, depth+1);
6495 else if (op == '*') {
6499 else if (op == '+' && (flags&SIMPLE)) {
6500 reginsert(pRExC_state, PLUS, ret, depth+1);
6504 else if (op == '+') {
6508 else if (op == '?') {
6513 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6515 "%.*s matches null string many times",
6516 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6520 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6521 nextchar(pRExC_state);
6522 reginsert(pRExC_state, MINMOD, ret, depth+1);
6523 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6525 #ifndef REG_ALLOW_MINMOD_SUSPEND
6528 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6530 nextchar(pRExC_state);
6531 ender = reg_node(pRExC_state, SUCCEED);
6532 REGTAIL(pRExC_state, ret, ender);
6533 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6535 ender = reg_node(pRExC_state, TAIL);
6536 REGTAIL(pRExC_state, ret, ender);
6540 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6542 vFAIL("Nested quantifiers");
6549 /* reg_namedseq(pRExC_state,UVp)
6551 This is expected to be called by a parser routine that has
6552 recognized'\N' and needs to handle the rest. RExC_parse is
6553 expected to point at the first char following the N at the time
6556 If valuep is non-null then it is assumed that we are parsing inside
6557 of a charclass definition and the first codepoint in the resolved
6558 string is returned via *valuep and the routine will return NULL.
6559 In this mode if a multichar string is returned from the charnames
6560 handler a warning will be issued, and only the first char in the
6561 sequence will be examined. If the string returned is zero length
6562 then the value of *valuep is undefined and NON-NULL will
6563 be returned to indicate failure. (This will NOT be a valid pointer
6566 If value is null then it is assumed that we are parsing normal text
6567 and inserts a new EXACT node into the program containing the resolved
6568 string and returns a pointer to the new node. If the string is
6569 zerolength a NOTHING node is emitted.
6571 On success RExC_parse is set to the char following the endbrace.
6572 Parsing failures will generate a fatal errorvia vFAIL(...)
6574 NOTE: We cache all results from the charnames handler locally in
6575 the RExC_charnames hash (created on first use) to prevent a charnames
6576 handler from playing silly-buggers and returning a short string and
6577 then a long string for a given pattern. Since the regexp program
6578 size is calculated during an initial parse this would result
6579 in a buffer overrun so we cache to prevent the charname result from
6580 changing during the course of the parse.
6584 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6586 char * name; /* start of the content of the name */
6587 char * endbrace; /* endbrace following the name */
6590 STRLEN len; /* this has various purposes throughout the code */
6591 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6592 regnode *ret = NULL;
6594 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6596 if (*RExC_parse != '{') {
6597 vFAIL("Missing braces on \\N{}");
6599 name = RExC_parse+1;
6600 endbrace = strchr(RExC_parse, '}');
6603 vFAIL("Missing right brace on \\N{}");
6605 RExC_parse = endbrace + 1;
6608 /* RExC_parse points at the beginning brace,
6609 endbrace points at the last */
6610 if ( name[0]=='U' && name[1]=='+' ) {
6611 /* its a "Unicode hex" notation {U+89AB} */
6612 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6613 | PERL_SCAN_DISALLOW_PREFIX
6614 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6617 len = (STRLEN)(endbrace - name - 2);
6618 cp = grok_hex(name + 2, &len, &fl, NULL);
6619 if ( len != (STRLEN)(endbrace - name - 2) ) {
6629 sv_str= newSVpvn(&string, 1);
6631 /* fetch the charnames handler for this scope */
6632 HV * const table = GvHV(PL_hintgv);
6634 hv_fetchs(table, "charnames", FALSE) :
6636 SV *cv= cvp ? *cvp : NULL;
6639 /* create an SV with the name as argument */
6640 sv_name = newSVpvn(name, endbrace - name);
6642 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6643 vFAIL2("Constant(\\N{%s}) unknown: "
6644 "(possibly a missing \"use charnames ...\")",
6647 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6648 vFAIL2("Constant(\\N{%s}): "
6649 "$^H{charnames} is not defined",SvPVX(sv_name));
6654 if (!RExC_charnames) {
6655 /* make sure our cache is allocated */
6656 RExC_charnames = newHV();
6657 sv_2mortal((SV*)RExC_charnames);
6659 /* see if we have looked this one up before */
6660 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6662 sv_str = HeVAL(he_str);
6675 count= call_sv(cv, G_SCALAR);
6677 if (count == 1) { /* XXXX is this right? dmq */
6679 SvREFCNT_inc_simple_void(sv_str);
6687 if ( !sv_str || !SvOK(sv_str) ) {
6688 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6689 "did not return a defined value",SvPVX(sv_name));
6691 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6696 char *p = SvPV(sv_str, len);
6699 if ( SvUTF8(sv_str) ) {
6700 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6704 We have to turn on utf8 for high bit chars otherwise
6705 we get failures with
6707 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6708 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6710 This is different from what \x{} would do with the same
6711 codepoint, where the condition is > 0xFF.
6718 /* warn if we havent used the whole string? */
6720 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6722 "Ignoring excess chars from \\N{%s} in character class",
6726 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6728 "Ignoring zero length \\N{%s} in character class",
6733 SvREFCNT_dec(sv_name);
6735 SvREFCNT_dec(sv_str);
6736 return len ? NULL : (regnode *)&len;
6737 } else if(SvCUR(sv_str)) {
6743 char * parse_start = name-3; /* needed for the offsets */
6745 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6747 ret = reg_node(pRExC_state,
6748 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6751 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6752 sv_utf8_upgrade(sv_str);
6753 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6757 p = SvPV(sv_str, len);
6759 /* len is the length written, charlen is the size the char read */
6760 for ( len = 0; p < pend; p += charlen ) {
6762 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6764 STRLEN foldlen,numlen;
6765 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6766 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6767 /* Emit all the Unicode characters. */
6769 for (foldbuf = tmpbuf;
6773 uvc = utf8_to_uvchr(foldbuf, &numlen);
6775 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6778 /* In EBCDIC the numlen
6779 * and unilen can differ. */
6781 if (numlen >= foldlen)
6785 break; /* "Can't happen." */
6788 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6800 RExC_size += STR_SZ(len);
6803 RExC_emit += STR_SZ(len);
6805 Set_Node_Cur_Length(ret); /* MJD */
6807 nextchar(pRExC_state);
6809 ret = reg_node(pRExC_state,NOTHING);
6812 SvREFCNT_dec(sv_str);
6815 SvREFCNT_dec(sv_name);
6825 * It returns the code point in utf8 for the value in *encp.
6826 * value: a code value in the source encoding
6827 * encp: a pointer to an Encode object
6829 * If the result from Encode is not a single character,
6830 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6833 S_reg_recode(pTHX_ const char value, SV **encp)
6836 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
6837 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6838 const STRLEN newlen = SvCUR(sv);
6839 UV uv = UNICODE_REPLACEMENT;
6841 PERL_ARGS_ASSERT_REG_RECODE;
6845 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6848 if (!newlen || numlen != newlen) {
6849 uv = UNICODE_REPLACEMENT;
6857 - regatom - the lowest level
6859 Try to identify anything special at the start of the pattern. If there
6860 is, then handle it as required. This may involve generating a single regop,
6861 such as for an assertion; or it may involve recursing, such as to
6862 handle a () structure.
6864 If the string doesn't start with something special then we gobble up
6865 as much literal text as we can.
6867 Once we have been able to handle whatever type of thing started the
6868 sequence, we return.
6870 Note: we have to be careful with escapes, as they can be both literal
6871 and special, and in the case of \10 and friends can either, depending
6872 on context. Specifically there are two seperate switches for handling
6873 escape sequences, with the one for handling literal escapes requiring
6874 a dummy entry for all of the special escapes that are actually handled
6879 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6882 register regnode *ret = NULL;
6884 char *parse_start = RExC_parse;
6885 GET_RE_DEBUG_FLAGS_DECL;
6886 DEBUG_PARSE("atom");
6887 *flagp = WORST; /* Tentatively. */
6889 PERL_ARGS_ASSERT_REGATOM;
6892 switch ((U8)*RExC_parse) {
6894 RExC_seen_zerolen++;
6895 nextchar(pRExC_state);
6896 if (RExC_flags & RXf_PMf_MULTILINE)
6897 ret = reg_node(pRExC_state, MBOL);
6898 else if (RExC_flags & RXf_PMf_SINGLELINE)
6899 ret = reg_node(pRExC_state, SBOL);
6901 ret = reg_node(pRExC_state, BOL);
6902 Set_Node_Length(ret, 1); /* MJD */
6905 nextchar(pRExC_state);
6907 RExC_seen_zerolen++;
6908 if (RExC_flags & RXf_PMf_MULTILINE)
6909 ret = reg_node(pRExC_state, MEOL);
6910 else if (RExC_flags & RXf_PMf_SINGLELINE)
6911 ret = reg_node(pRExC_state, SEOL);
6913 ret = reg_node(pRExC_state, EOL);
6914 Set_Node_Length(ret, 1); /* MJD */
6917 nextchar(pRExC_state);
6918 if (RExC_flags & RXf_PMf_SINGLELINE)
6919 ret = reg_node(pRExC_state, SANY);
6921 ret = reg_node(pRExC_state, REG_ANY);
6922 *flagp |= HASWIDTH|SIMPLE;
6924 Set_Node_Length(ret, 1); /* MJD */
6928 char * const oregcomp_parse = ++RExC_parse;
6929 ret = regclass(pRExC_state,depth+1);
6930 if (*RExC_parse != ']') {
6931 RExC_parse = oregcomp_parse;
6932 vFAIL("Unmatched [");
6934 nextchar(pRExC_state);
6935 *flagp |= HASWIDTH|SIMPLE;
6936 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6940 nextchar(pRExC_state);
6941 ret = reg(pRExC_state, 1, &flags,depth+1);
6943 if (flags & TRYAGAIN) {
6944 if (RExC_parse == RExC_end) {
6945 /* Make parent create an empty node if needed. */
6953 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6957 if (flags & TRYAGAIN) {
6961 vFAIL("Internal urp");
6962 /* Supposed to be caught earlier. */
6965 if (!regcurly(RExC_parse)) {
6974 vFAIL("Quantifier follows nothing");
6982 len=0; /* silence a spurious compiler warning */
6983 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
6984 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
6985 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
6986 ret = reganode(pRExC_state, FOLDCHAR, cp);
6987 Set_Node_Length(ret, 1); /* MJD */
6988 nextchar(pRExC_state); /* kill whitespace under /x */
6996 This switch handles escape sequences that resolve to some kind
6997 of special regop and not to literal text. Escape sequnces that
6998 resolve to literal text are handled below in the switch marked
7001 Every entry in this switch *must* have a corresponding entry
7002 in the literal escape switch. However, the opposite is not
7003 required, as the default for this switch is to jump to the
7004 literal text handling code.
7006 switch ((U8)*++RExC_parse) {
7011 /* Special Escapes */
7013 RExC_seen_zerolen++;
7014 ret = reg_node(pRExC_state, SBOL);
7016 goto finish_meta_pat;
7018 ret = reg_node(pRExC_state, GPOS);
7019 RExC_seen |= REG_SEEN_GPOS;
7021 goto finish_meta_pat;
7023 RExC_seen_zerolen++;
7024 ret = reg_node(pRExC_state, KEEPS);
7026 /* XXX:dmq : disabling in-place substitution seems to
7027 * be necessary here to avoid cases of memory corruption, as
7028 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7030 RExC_seen |= REG_SEEN_LOOKBEHIND;
7031 goto finish_meta_pat;
7033 ret = reg_node(pRExC_state, SEOL);
7035 RExC_seen_zerolen++; /* Do not optimize RE away */
7036 goto finish_meta_pat;
7038 ret = reg_node(pRExC_state, EOS);
7040 RExC_seen_zerolen++; /* Do not optimize RE away */
7041 goto finish_meta_pat;
7043 ret = reg_node(pRExC_state, CANY);
7044 RExC_seen |= REG_SEEN_CANY;
7045 *flagp |= HASWIDTH|SIMPLE;
7046 goto finish_meta_pat;
7048 ret = reg_node(pRExC_state, CLUMP);
7050 goto finish_meta_pat;
7052 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
7053 *flagp |= HASWIDTH|SIMPLE;
7054 goto finish_meta_pat;
7056 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
7057 *flagp |= HASWIDTH|SIMPLE;
7058 goto finish_meta_pat;
7060 RExC_seen_zerolen++;
7061 RExC_seen |= REG_SEEN_LOOKBEHIND;
7062 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
7064 goto finish_meta_pat;
7066 RExC_seen_zerolen++;
7067 RExC_seen |= REG_SEEN_LOOKBEHIND;
7068 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
7070 goto finish_meta_pat;
7072 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
7073 *flagp |= HASWIDTH|SIMPLE;
7074 goto finish_meta_pat;
7076 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
7077 *flagp |= HASWIDTH|SIMPLE;
7078 goto finish_meta_pat;
7080 ret = reg_node(pRExC_state, DIGIT);
7081 *flagp |= HASWIDTH|SIMPLE;
7082 goto finish_meta_pat;
7084 ret = reg_node(pRExC_state, NDIGIT);
7085 *flagp |= HASWIDTH|SIMPLE;
7086 goto finish_meta_pat;
7088 ret = reg_node(pRExC_state, LNBREAK);
7089 *flagp |= HASWIDTH|SIMPLE;
7090 goto finish_meta_pat;
7092 ret = reg_node(pRExC_state, HORIZWS);
7093 *flagp |= HASWIDTH|SIMPLE;
7094 goto finish_meta_pat;
7096 ret = reg_node(pRExC_state, NHORIZWS);
7097 *flagp |= HASWIDTH|SIMPLE;
7098 goto finish_meta_pat;
7100 ret = reg_node(pRExC_state, VERTWS);
7101 *flagp |= HASWIDTH|SIMPLE;
7102 goto finish_meta_pat;
7104 ret = reg_node(pRExC_state, NVERTWS);
7105 *flagp |= HASWIDTH|SIMPLE;
7107 nextchar(pRExC_state);
7108 Set_Node_Length(ret, 2); /* MJD */
7113 char* const oldregxend = RExC_end;
7115 char* parse_start = RExC_parse - 2;
7118 if (RExC_parse[1] == '{') {
7119 /* a lovely hack--pretend we saw [\pX] instead */
7120 RExC_end = strchr(RExC_parse, '}');
7122 const U8 c = (U8)*RExC_parse;
7124 RExC_end = oldregxend;
7125 vFAIL2("Missing right brace on \\%c{}", c);
7130 RExC_end = RExC_parse + 2;
7131 if (RExC_end > oldregxend)
7132 RExC_end = oldregxend;
7136 ret = regclass(pRExC_state,depth+1);
7138 RExC_end = oldregxend;
7141 Set_Node_Offset(ret, parse_start + 2);
7142 Set_Node_Cur_Length(ret);
7143 nextchar(pRExC_state);
7144 *flagp |= HASWIDTH|SIMPLE;
7148 /* Handle \N{NAME} here and not below because it can be
7149 multicharacter. join_exact() will join them up later on.
7150 Also this makes sure that things like /\N{BLAH}+/ and
7151 \N{BLAH} being multi char Just Happen. dmq*/
7153 ret= reg_namedseq(pRExC_state, NULL);
7155 case 'k': /* Handle \k<NAME> and \k'NAME' */
7158 char ch= RExC_parse[1];
7159 if (ch != '<' && ch != '\'' && ch != '{') {
7161 vFAIL2("Sequence %.2s... not terminated",parse_start);
7163 /* this pretty much dupes the code for (?P=...) in reg(), if
7164 you change this make sure you change that */
7165 char* name_start = (RExC_parse += 2);
7167 SV *sv_dat = reg_scan_name(pRExC_state,
7168 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7169 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7170 if (RExC_parse == name_start || *RExC_parse != ch)
7171 vFAIL2("Sequence %.3s... not terminated",parse_start);
7174 num = add_data( pRExC_state, 1, "S" );
7175 RExC_rxi->data->data[num]=(void*)sv_dat;
7176 SvREFCNT_inc_simple_void(sv_dat);
7180 ret = reganode(pRExC_state,
7181 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7185 /* override incorrect value set in reganode MJD */
7186 Set_Node_Offset(ret, parse_start+1);
7187 Set_Node_Cur_Length(ret); /* MJD */
7188 nextchar(pRExC_state);
7194 case '1': case '2': case '3': case '4':
7195 case '5': case '6': case '7': case '8': case '9':
7198 bool isg = *RExC_parse == 'g';
7203 if (*RExC_parse == '{') {
7207 if (*RExC_parse == '-') {
7211 if (hasbrace && !isDIGIT(*RExC_parse)) {
7212 if (isrel) RExC_parse--;
7214 goto parse_named_seq;
7216 num = atoi(RExC_parse);
7217 if (isg && num == 0)
7218 vFAIL("Reference to invalid group 0");
7220 num = RExC_npar - num;
7222 vFAIL("Reference to nonexistent or unclosed group");
7224 if (!isg && num > 9 && num >= RExC_npar)
7227 char * const parse_start = RExC_parse - 1; /* MJD */
7228 while (isDIGIT(*RExC_parse))
7230 if (parse_start == RExC_parse - 1)
7231 vFAIL("Unterminated \\g... pattern");
7233 if (*RExC_parse != '}')
7234 vFAIL("Unterminated \\g{...} pattern");
7238 if (num > (I32)RExC_rx->nparens)
7239 vFAIL("Reference to nonexistent group");
7242 ret = reganode(pRExC_state,
7243 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7247 /* override incorrect value set in reganode MJD */
7248 Set_Node_Offset(ret, parse_start+1);
7249 Set_Node_Cur_Length(ret); /* MJD */
7251 nextchar(pRExC_state);
7256 if (RExC_parse >= RExC_end)
7257 FAIL("Trailing \\");
7260 /* Do not generate "unrecognized" warnings here, we fall
7261 back into the quick-grab loop below */
7268 if (RExC_flags & RXf_PMf_EXTENDED) {
7269 if ( reg_skipcomment( pRExC_state ) )
7276 register STRLEN len;
7281 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7283 parse_start = RExC_parse - 1;
7289 ret = reg_node(pRExC_state,
7290 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7292 for (len = 0, p = RExC_parse - 1;
7293 len < 127 && p < RExC_end;
7296 char * const oldp = p;
7298 if (RExC_flags & RXf_PMf_EXTENDED)
7299 p = regwhite( pRExC_state, p );
7304 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7305 goto normal_default;
7315 /* Literal Escapes Switch
7317 This switch is meant to handle escape sequences that
7318 resolve to a literal character.
7320 Every escape sequence that represents something
7321 else, like an assertion or a char class, is handled
7322 in the switch marked 'Special Escapes' above in this
7323 routine, but also has an entry here as anything that
7324 isn't explicitly mentioned here will be treated as
7325 an unescaped equivalent literal.
7329 /* These are all the special escapes. */
7333 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7334 goto normal_default;
7335 case 'A': /* Start assertion */
7336 case 'b': case 'B': /* Word-boundary assertion*/
7337 case 'C': /* Single char !DANGEROUS! */
7338 case 'd': case 'D': /* digit class */
7339 case 'g': case 'G': /* generic-backref, pos assertion */
7340 case 'h': case 'H': /* HORIZWS */
7341 case 'k': case 'K': /* named backref, keep marker */
7342 case 'N': /* named char sequence */
7343 case 'p': case 'P': /* Unicode property */
7344 case 'R': /* LNBREAK */
7345 case 's': case 'S': /* space class */
7346 case 'v': case 'V': /* VERTWS */
7347 case 'w': case 'W': /* word class */
7348 case 'X': /* eXtended Unicode "combining character sequence" */
7349 case 'z': case 'Z': /* End of line/string assertion */
7353 /* Anything after here is an escape that resolves to a
7354 literal. (Except digits, which may or may not)
7373 ender = ASCII_TO_NATIVE('\033');
7377 ender = ASCII_TO_NATIVE('\007');
7382 char* const e = strchr(p, '}');
7386 vFAIL("Missing right brace on \\x{}");
7389 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7390 | PERL_SCAN_DISALLOW_PREFIX;
7391 STRLEN numlen = e - p - 1;
7392 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7399 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7401 ender = grok_hex(p, &numlen, &flags, NULL);
7404 if (PL_encoding && ender < 0x100)
7405 goto recode_encoding;
7409 ender = UCHARAT(p++);
7410 ender = toCTRL(ender);
7412 case '0': case '1': case '2': case '3':case '4':
7413 case '5': case '6': case '7': case '8':case '9':
7415 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7418 ender = grok_oct(p, &numlen, &flags, NULL);
7425 if (PL_encoding && ender < 0x100)
7426 goto recode_encoding;
7430 SV* enc = PL_encoding;
7431 ender = reg_recode((const char)(U8)ender, &enc);
7432 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7433 vWARN(p, "Invalid escape in the specified encoding");
7439 FAIL("Trailing \\");
7442 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
7443 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7444 goto normal_default;
7449 if (UTF8_IS_START(*p) && UTF) {
7451 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7452 &numlen, UTF8_ALLOW_DEFAULT);
7459 if ( RExC_flags & RXf_PMf_EXTENDED)
7460 p = regwhite( pRExC_state, p );
7462 /* Prime the casefolded buffer. */
7463 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7465 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7470 /* Emit all the Unicode characters. */
7472 for (foldbuf = tmpbuf;
7474 foldlen -= numlen) {
7475 ender = utf8_to_uvchr(foldbuf, &numlen);
7477 const STRLEN unilen = reguni(pRExC_state, ender, s);
7480 /* In EBCDIC the numlen
7481 * and unilen can differ. */
7483 if (numlen >= foldlen)
7487 break; /* "Can't happen." */
7491 const STRLEN unilen = reguni(pRExC_state, ender, s);
7500 REGC((char)ender, s++);
7506 /* Emit all the Unicode characters. */
7508 for (foldbuf = tmpbuf;
7510 foldlen -= numlen) {
7511 ender = utf8_to_uvchr(foldbuf, &numlen);
7513 const STRLEN unilen = reguni(pRExC_state, ender, s);
7516 /* In EBCDIC the numlen
7517 * and unilen can differ. */
7519 if (numlen >= foldlen)
7527 const STRLEN unilen = reguni(pRExC_state, ender, s);
7536 REGC((char)ender, s++);
7540 Set_Node_Cur_Length(ret); /* MJD */
7541 nextchar(pRExC_state);
7543 /* len is STRLEN which is unsigned, need to copy to signed */
7546 vFAIL("Internal disaster");
7550 if (len == 1 && UNI_IS_INVARIANT(ender))
7554 RExC_size += STR_SZ(len);
7557 RExC_emit += STR_SZ(len);
7567 S_regwhite( RExC_state_t *pRExC_state, char *p )
7569 const char *e = RExC_end;
7571 PERL_ARGS_ASSERT_REGWHITE;
7576 else if (*p == '#') {
7585 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7593 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7594 Character classes ([:foo:]) can also be negated ([:^foo:]).
7595 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7596 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7597 but trigger failures because they are currently unimplemented. */
7599 #define POSIXCC_DONE(c) ((c) == ':')
7600 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7601 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7604 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7607 I32 namedclass = OOB_NAMEDCLASS;
7609 PERL_ARGS_ASSERT_REGPPOSIXCC;
7611 if (value == '[' && RExC_parse + 1 < RExC_end &&
7612 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7613 POSIXCC(UCHARAT(RExC_parse))) {
7614 const char c = UCHARAT(RExC_parse);
7615 char* const s = RExC_parse++;
7617 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7619 if (RExC_parse == RExC_end)
7620 /* Grandfather lone [:, [=, [. */
7623 const char* const t = RExC_parse++; /* skip over the c */
7626 if (UCHARAT(RExC_parse) == ']') {
7627 const char *posixcc = s + 1;
7628 RExC_parse++; /* skip over the ending ] */
7631 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7632 const I32 skip = t - posixcc;
7634 /* Initially switch on the length of the name. */
7637 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7638 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7641 /* Names all of length 5. */
7642 /* alnum alpha ascii blank cntrl digit graph lower
7643 print punct space upper */
7644 /* Offset 4 gives the best switch position. */
7645 switch (posixcc[4]) {
7647 if (memEQ(posixcc, "alph", 4)) /* alpha */
7648 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7651 if (memEQ(posixcc, "spac", 4)) /* space */
7652 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7655 if (memEQ(posixcc, "grap", 4)) /* graph */
7656 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7659 if (memEQ(posixcc, "asci", 4)) /* ascii */
7660 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7663 if (memEQ(posixcc, "blan", 4)) /* blank */
7664 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7667 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7668 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7671 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7672 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7675 if (memEQ(posixcc, "lowe", 4)) /* lower */
7676 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7677 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7678 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7681 if (memEQ(posixcc, "digi", 4)) /* digit */
7682 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7683 else if (memEQ(posixcc, "prin", 4)) /* print */
7684 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7685 else if (memEQ(posixcc, "punc", 4)) /* punct */
7686 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7691 if (memEQ(posixcc, "xdigit", 6))
7692 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7696 if (namedclass == OOB_NAMEDCLASS)
7697 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7699 assert (posixcc[skip] == ':');
7700 assert (posixcc[skip+1] == ']');
7701 } else if (!SIZE_ONLY) {
7702 /* [[=foo=]] and [[.foo.]] are still future. */
7704 /* adjust RExC_parse so the warning shows after
7706 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7708 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7711 /* Maternal grandfather:
7712 * "[:" ending in ":" but not in ":]" */
7722 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7726 PERL_ARGS_ASSERT_CHECKPOSIXCC;
7728 if (POSIXCC(UCHARAT(RExC_parse))) {
7729 const char *s = RExC_parse;
7730 const char c = *s++;
7734 if (*s && c == *s && s[1] == ']') {
7735 if (ckWARN(WARN_REGEXP))
7737 "POSIX syntax [%c %c] belongs inside character classes",
7740 /* [[=foo=]] and [[.foo.]] are still future. */
7741 if (POSIXCC_NOTYET(c)) {
7742 /* adjust RExC_parse so the error shows after
7744 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7746 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7753 #define _C_C_T_(NAME,TEST,WORD) \
7756 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7758 for (value = 0; value < 256; value++) \
7760 ANYOF_BITMAP_SET(ret, value); \
7765 case ANYOF_N##NAME: \
7767 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7769 for (value = 0; value < 256; value++) \
7771 ANYOF_BITMAP_SET(ret, value); \
7777 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7779 for (value = 0; value < 256; value++) \
7781 ANYOF_BITMAP_SET(ret, value); \
7785 case ANYOF_N##NAME: \
7786 for (value = 0; value < 256; value++) \
7788 ANYOF_BITMAP_SET(ret, value); \
7794 parse a class specification and produce either an ANYOF node that
7795 matches the pattern or if the pattern matches a single char only and
7796 that char is < 256 and we are case insensitive then we produce an
7801 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7804 register UV nextvalue;
7805 register IV prevvalue = OOB_UNICODE;
7806 register IV range = 0;
7807 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7808 register regnode *ret;
7811 char *rangebegin = NULL;
7812 bool need_class = 0;
7815 bool optimize_invert = TRUE;
7816 AV* unicode_alternate = NULL;
7818 UV literal_endpoint = 0;
7820 UV stored = 0; /* number of chars stored in the class */
7822 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7823 case we need to change the emitted regop to an EXACT. */
7824 const char * orig_parse = RExC_parse;
7825 GET_RE_DEBUG_FLAGS_DECL;
7827 PERL_ARGS_ASSERT_REGCLASS;
7829 PERL_UNUSED_ARG(depth);
7832 DEBUG_PARSE("clas");
7834 /* Assume we are going to generate an ANYOF node. */
7835 ret = reganode(pRExC_state, ANYOF, 0);
7838 ANYOF_FLAGS(ret) = 0;
7840 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7844 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7848 RExC_size += ANYOF_SKIP;
7849 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7852 RExC_emit += ANYOF_SKIP;
7854 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7856 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7857 ANYOF_BITMAP_ZERO(ret);
7858 listsv = newSVpvs("# comment\n");
7861 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7863 if (!SIZE_ONLY && POSIXCC(nextvalue))
7864 checkposixcc(pRExC_state);
7866 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7867 if (UCHARAT(RExC_parse) == ']')
7871 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7875 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7878 rangebegin = RExC_parse;
7880 value = utf8n_to_uvchr((U8*)RExC_parse,
7881 RExC_end - RExC_parse,
7882 &numlen, UTF8_ALLOW_DEFAULT);
7883 RExC_parse += numlen;
7886 value = UCHARAT(RExC_parse++);
7888 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7889 if (value == '[' && POSIXCC(nextvalue))
7890 namedclass = regpposixcc(pRExC_state, value);
7891 else if (value == '\\') {
7893 value = utf8n_to_uvchr((U8*)RExC_parse,
7894 RExC_end - RExC_parse,
7895 &numlen, UTF8_ALLOW_DEFAULT);
7896 RExC_parse += numlen;
7899 value = UCHARAT(RExC_parse++);
7900 /* Some compilers cannot handle switching on 64-bit integer
7901 * values, therefore value cannot be an UV. Yes, this will
7902 * be a problem later if we want switch on Unicode.
7903 * A similar issue a little bit later when switching on
7904 * namedclass. --jhi */
7905 switch ((I32)value) {
7906 case 'w': namedclass = ANYOF_ALNUM; break;
7907 case 'W': namedclass = ANYOF_NALNUM; break;
7908 case 's': namedclass = ANYOF_SPACE; break;
7909 case 'S': namedclass = ANYOF_NSPACE; break;
7910 case 'd': namedclass = ANYOF_DIGIT; break;
7911 case 'D': namedclass = ANYOF_NDIGIT; break;
7912 case 'v': namedclass = ANYOF_VERTWS; break;
7913 case 'V': namedclass = ANYOF_NVERTWS; break;
7914 case 'h': namedclass = ANYOF_HORIZWS; break;
7915 case 'H': namedclass = ANYOF_NHORIZWS; break;
7916 case 'N': /* Handle \N{NAME} in class */
7918 /* We only pay attention to the first char of
7919 multichar strings being returned. I kinda wonder
7920 if this makes sense as it does change the behaviour
7921 from earlier versions, OTOH that behaviour was broken
7923 UV v; /* value is register so we cant & it /grrr */
7924 if (reg_namedseq(pRExC_state, &v)) {
7934 if (RExC_parse >= RExC_end)
7935 vFAIL2("Empty \\%c{}", (U8)value);
7936 if (*RExC_parse == '{') {
7937 const U8 c = (U8)value;
7938 e = strchr(RExC_parse++, '}');
7940 vFAIL2("Missing right brace on \\%c{}", c);
7941 while (isSPACE(UCHARAT(RExC_parse)))
7943 if (e == RExC_parse)
7944 vFAIL2("Empty \\%c{}", c);
7946 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7954 if (UCHARAT(RExC_parse) == '^') {
7957 value = value == 'p' ? 'P' : 'p'; /* toggle */
7958 while (isSPACE(UCHARAT(RExC_parse))) {
7963 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7964 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7967 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7968 namedclass = ANYOF_MAX; /* no official name, but it's named */
7971 case 'n': value = '\n'; break;
7972 case 'r': value = '\r'; break;
7973 case 't': value = '\t'; break;
7974 case 'f': value = '\f'; break;
7975 case 'b': value = '\b'; break;
7976 case 'e': value = ASCII_TO_NATIVE('\033');break;
7977 case 'a': value = ASCII_TO_NATIVE('\007');break;
7979 if (*RExC_parse == '{') {
7980 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7981 | PERL_SCAN_DISALLOW_PREFIX;
7982 char * const e = strchr(RExC_parse++, '}');
7984 vFAIL("Missing right brace on \\x{}");
7986 numlen = e - RExC_parse;
7987 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7991 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7993 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7994 RExC_parse += numlen;
7996 if (PL_encoding && value < 0x100)
7997 goto recode_encoding;
8000 value = UCHARAT(RExC_parse++);
8001 value = toCTRL(value);
8003 case '0': case '1': case '2': case '3': case '4':
8004 case '5': case '6': case '7': case '8': case '9':
8008 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8009 RExC_parse += numlen;
8010 if (PL_encoding && value < 0x100)
8011 goto recode_encoding;
8016 SV* enc = PL_encoding;
8017 value = reg_recode((const char)(U8)value, &enc);
8018 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
8020 "Invalid escape in the specified encoding");
8024 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
8026 "Unrecognized escape \\%c in character class passed through",
8030 } /* end of \blah */
8036 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8038 if (!SIZE_ONLY && !need_class)
8039 ANYOF_CLASS_ZERO(ret);
8043 /* a bad range like a-\d, a-[:digit:] ? */
8046 if (ckWARN(WARN_REGEXP)) {
8048 RExC_parse >= rangebegin ?
8049 RExC_parse - rangebegin : 0;
8051 "False [] range \"%*.*s\"",
8054 if (prevvalue < 256) {
8055 ANYOF_BITMAP_SET(ret, prevvalue);
8056 ANYOF_BITMAP_SET(ret, '-');
8059 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8060 Perl_sv_catpvf(aTHX_ listsv,
8061 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8065 range = 0; /* this was not a true range */
8071 const char *what = NULL;
8074 if (namedclass > OOB_NAMEDCLASS)
8075 optimize_invert = FALSE;
8076 /* Possible truncation here but in some 64-bit environments
8077 * the compiler gets heartburn about switch on 64-bit values.
8078 * A similar issue a little earlier when switching on value.
8080 switch ((I32)namedclass) {
8081 case _C_C_T_(ALNUM, isALNUM(value), "Word");
8082 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
8083 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
8084 case _C_C_T_(BLANK, isBLANK(value), "Blank");
8085 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
8086 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
8087 case _C_C_T_(LOWER, isLOWER(value), "Lower");
8088 case _C_C_T_(PRINT, isPRINT(value), "Print");
8089 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
8090 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
8091 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
8092 case _C_C_T_(UPPER, isUPPER(value), "Upper");
8093 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
8094 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8095 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8098 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8101 for (value = 0; value < 128; value++)
8102 ANYOF_BITMAP_SET(ret, value);
8104 for (value = 0; value < 256; value++) {
8106 ANYOF_BITMAP_SET(ret, value);
8115 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8118 for (value = 128; value < 256; value++)
8119 ANYOF_BITMAP_SET(ret, value);
8121 for (value = 0; value < 256; value++) {
8122 if (!isASCII(value))
8123 ANYOF_BITMAP_SET(ret, value);
8132 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8134 /* consecutive digits assumed */
8135 for (value = '0'; value <= '9'; value++)
8136 ANYOF_BITMAP_SET(ret, value);
8143 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8145 /* consecutive digits assumed */
8146 for (value = 0; value < '0'; value++)
8147 ANYOF_BITMAP_SET(ret, value);
8148 for (value = '9' + 1; value < 256; value++)
8149 ANYOF_BITMAP_SET(ret, value);
8155 /* this is to handle \p and \P */
8158 vFAIL("Invalid [::] class");
8162 /* Strings such as "+utf8::isWord\n" */
8163 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8166 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8169 } /* end of namedclass \blah */
8172 if (prevvalue > (IV)value) /* b-a */ {
8173 const int w = RExC_parse - rangebegin;
8174 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8175 range = 0; /* not a valid range */
8179 prevvalue = value; /* save the beginning of the range */
8180 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8181 RExC_parse[1] != ']') {
8184 /* a bad range like \w-, [:word:]- ? */
8185 if (namedclass > OOB_NAMEDCLASS) {
8186 if (ckWARN(WARN_REGEXP)) {
8188 RExC_parse >= rangebegin ?
8189 RExC_parse - rangebegin : 0;
8191 "False [] range \"%*.*s\"",
8195 ANYOF_BITMAP_SET(ret, '-');
8197 range = 1; /* yeah, it's a range! */
8198 continue; /* but do it the next time */
8202 /* now is the next time */
8203 /*stored += (value - prevvalue + 1);*/
8205 if (prevvalue < 256) {
8206 const IV ceilvalue = value < 256 ? value : 255;
8209 /* In EBCDIC [\x89-\x91] should include
8210 * the \x8e but [i-j] should not. */
8211 if (literal_endpoint == 2 &&
8212 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8213 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8215 if (isLOWER(prevvalue)) {
8216 for (i = prevvalue; i <= ceilvalue; i++)
8217 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8219 ANYOF_BITMAP_SET(ret, i);
8222 for (i = prevvalue; i <= ceilvalue; i++)
8223 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8225 ANYOF_BITMAP_SET(ret, i);
8231 for (i = prevvalue; i <= ceilvalue; i++) {
8232 if (!ANYOF_BITMAP_TEST(ret,i)) {
8234 ANYOF_BITMAP_SET(ret, i);
8238 if (value > 255 || UTF) {
8239 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8240 const UV natvalue = NATIVE_TO_UNI(value);
8241 stored+=2; /* can't optimize this class */
8242 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8243 if (prevnatvalue < natvalue) { /* what about > ? */
8244 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8245 prevnatvalue, natvalue);
8247 else if (prevnatvalue == natvalue) {
8248 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8250 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8252 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8254 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8255 if (RExC_precomp[0] == ':' &&
8256 RExC_precomp[1] == '[' &&
8257 (f == 0xDF || f == 0x92)) {
8258 f = NATIVE_TO_UNI(f);
8261 /* If folding and foldable and a single
8262 * character, insert also the folded version
8263 * to the charclass. */
8265 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8266 if ((RExC_precomp[0] == ':' &&
8267 RExC_precomp[1] == '[' &&
8269 (value == 0xFB05 || value == 0xFB06))) ?
8270 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8271 foldlen == (STRLEN)UNISKIP(f) )
8273 if (foldlen == (STRLEN)UNISKIP(f))
8275 Perl_sv_catpvf(aTHX_ listsv,
8278 /* Any multicharacter foldings
8279 * require the following transform:
8280 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8281 * where E folds into "pq" and F folds
8282 * into "rst", all other characters
8283 * fold to single characters. We save
8284 * away these multicharacter foldings,
8285 * to be later saved as part of the
8286 * additional "s" data. */
8289 if (!unicode_alternate)
8290 unicode_alternate = newAV();
8291 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8293 av_push(unicode_alternate, sv);
8297 /* If folding and the value is one of the Greek
8298 * sigmas insert a few more sigmas to make the
8299 * folding rules of the sigmas to work right.
8300 * Note that not all the possible combinations
8301 * are handled here: some of them are handled
8302 * by the standard folding rules, and some of
8303 * them (literal or EXACTF cases) are handled
8304 * during runtime in regexec.c:S_find_byclass(). */
8305 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8306 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8307 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8308 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8309 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8311 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8312 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8313 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8318 literal_endpoint = 0;
8322 range = 0; /* this range (if it was one) is done now */
8326 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8328 RExC_size += ANYOF_CLASS_ADD_SKIP;
8330 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8336 /****** !SIZE_ONLY AFTER HERE *********/
8338 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8339 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8341 /* optimize single char class to an EXACT node
8342 but *only* when its not a UTF/high char */
8343 const char * cur_parse= RExC_parse;
8344 RExC_emit = (regnode *)orig_emit;
8345 RExC_parse = (char *)orig_parse;
8346 ret = reg_node(pRExC_state,
8347 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8348 RExC_parse = (char *)cur_parse;
8349 *STRING(ret)= (char)value;
8351 RExC_emit += STR_SZ(1);
8354 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8355 if ( /* If the only flag is folding (plus possibly inversion). */
8356 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8358 for (value = 0; value < 256; ++value) {
8359 if (ANYOF_BITMAP_TEST(ret, value)) {
8360 UV fold = PL_fold[value];
8363 ANYOF_BITMAP_SET(ret, fold);
8366 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8369 /* optimize inverted simple patterns (e.g. [^a-z]) */
8370 if (optimize_invert &&
8371 /* If the only flag is inversion. */
8372 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8373 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8374 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8375 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8378 AV * const av = newAV();
8380 /* The 0th element stores the character class description
8381 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8382 * to initialize the appropriate swash (which gets stored in
8383 * the 1st element), and also useful for dumping the regnode.
8384 * The 2nd element stores the multicharacter foldings,
8385 * used later (regexec.c:S_reginclass()). */
8386 av_store(av, 0, listsv);
8387 av_store(av, 1, NULL);
8388 av_store(av, 2, (SV*)unicode_alternate);
8389 rv = newRV_noinc((SV*)av);
8390 n = add_data(pRExC_state, 1, "s");
8391 RExC_rxi->data->data[n] = (void*)rv;
8399 /* reg_skipcomment()
8401 Absorbs an /x style # comments from the input stream.
8402 Returns true if there is more text remaining in the stream.
8403 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8404 terminates the pattern without including a newline.
8406 Note its the callers responsibility to ensure that we are
8412 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8416 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8418 while (RExC_parse < RExC_end)
8419 if (*RExC_parse++ == '\n') {
8424 /* we ran off the end of the pattern without ending
8425 the comment, so we have to add an \n when wrapping */
8426 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8434 Advance that parse position, and optionall absorbs
8435 "whitespace" from the inputstream.
8437 Without /x "whitespace" means (?#...) style comments only,
8438 with /x this means (?#...) and # comments and whitespace proper.
8440 Returns the RExC_parse point from BEFORE the scan occurs.
8442 This is the /x friendly way of saying RExC_parse++.
8446 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8448 char* const retval = RExC_parse++;
8450 PERL_ARGS_ASSERT_NEXTCHAR;
8453 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8454 RExC_parse[2] == '#') {
8455 while (*RExC_parse != ')') {
8456 if (RExC_parse == RExC_end)
8457 FAIL("Sequence (?#... not terminated");
8463 if (RExC_flags & RXf_PMf_EXTENDED) {
8464 if (isSPACE(*RExC_parse)) {
8468 else if (*RExC_parse == '#') {
8469 if ( reg_skipcomment( pRExC_state ) )
8478 - reg_node - emit a node
8480 STATIC regnode * /* Location. */
8481 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8484 register regnode *ptr;
8485 regnode * const ret = RExC_emit;
8486 GET_RE_DEBUG_FLAGS_DECL;
8488 PERL_ARGS_ASSERT_REG_NODE;
8491 SIZE_ALIGN(RExC_size);
8495 if (RExC_emit >= RExC_emit_bound)
8496 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8498 NODE_ALIGN_FILL(ret);
8500 FILL_ADVANCE_NODE(ptr, op);
8501 #ifdef RE_TRACK_PATTERN_OFFSETS
8502 if (RExC_offsets) { /* MJD */
8503 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8504 "reg_node", __LINE__,
8506 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8507 ? "Overwriting end of array!\n" : "OK",
8508 (UV)(RExC_emit - RExC_emit_start),
8509 (UV)(RExC_parse - RExC_start),
8510 (UV)RExC_offsets[0]));
8511 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8519 - reganode - emit a node with an argument
8521 STATIC regnode * /* Location. */
8522 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8525 register regnode *ptr;
8526 regnode * const ret = RExC_emit;
8527 GET_RE_DEBUG_FLAGS_DECL;
8529 PERL_ARGS_ASSERT_REGANODE;
8532 SIZE_ALIGN(RExC_size);
8537 assert(2==regarglen[op]+1);
8539 Anything larger than this has to allocate the extra amount.
8540 If we changed this to be:
8542 RExC_size += (1 + regarglen[op]);
8544 then it wouldn't matter. Its not clear what side effect
8545 might come from that so its not done so far.
8550 if (RExC_emit >= RExC_emit_bound)
8551 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8553 NODE_ALIGN_FILL(ret);
8555 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8556 #ifdef RE_TRACK_PATTERN_OFFSETS
8557 if (RExC_offsets) { /* MJD */
8558 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8562 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8563 "Overwriting end of array!\n" : "OK",
8564 (UV)(RExC_emit - RExC_emit_start),
8565 (UV)(RExC_parse - RExC_start),
8566 (UV)RExC_offsets[0]));
8567 Set_Cur_Node_Offset;
8575 - reguni - emit (if appropriate) a Unicode character
8578 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8582 PERL_ARGS_ASSERT_REGUNI;
8584 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8588 - reginsert - insert an operator in front of already-emitted operand
8590 * Means relocating the operand.
8593 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8596 register regnode *src;
8597 register regnode *dst;
8598 register regnode *place;
8599 const int offset = regarglen[(U8)op];
8600 const int size = NODE_STEP_REGNODE + offset;
8601 GET_RE_DEBUG_FLAGS_DECL;
8603 PERL_ARGS_ASSERT_REGINSERT;
8604 PERL_UNUSED_ARG(depth);
8605 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8606 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8615 if (RExC_open_parens) {
8617 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8618 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8619 if ( RExC_open_parens[paren] >= opnd ) {
8620 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8621 RExC_open_parens[paren] += size;
8623 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8625 if ( RExC_close_parens[paren] >= opnd ) {
8626 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8627 RExC_close_parens[paren] += size;
8629 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8634 while (src > opnd) {
8635 StructCopy(--src, --dst, regnode);
8636 #ifdef RE_TRACK_PATTERN_OFFSETS
8637 if (RExC_offsets) { /* MJD 20010112 */
8638 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8642 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8643 ? "Overwriting end of array!\n" : "OK",
8644 (UV)(src - RExC_emit_start),
8645 (UV)(dst - RExC_emit_start),
8646 (UV)RExC_offsets[0]));
8647 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8648 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8654 place = opnd; /* Op node, where operand used to be. */
8655 #ifdef RE_TRACK_PATTERN_OFFSETS
8656 if (RExC_offsets) { /* MJD */
8657 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8661 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8662 ? "Overwriting end of array!\n" : "OK",
8663 (UV)(place - RExC_emit_start),
8664 (UV)(RExC_parse - RExC_start),
8665 (UV)RExC_offsets[0]));
8666 Set_Node_Offset(place, RExC_parse);
8667 Set_Node_Length(place, 1);
8670 src = NEXTOPER(place);
8671 FILL_ADVANCE_NODE(place, op);
8672 Zero(src, offset, regnode);
8676 - regtail - set the next-pointer at the end of a node chain of p to val.
8677 - SEE ALSO: regtail_study
8679 /* TODO: All three parms should be const */
8681 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8684 register regnode *scan;
8685 GET_RE_DEBUG_FLAGS_DECL;
8687 PERL_ARGS_ASSERT_REGTAIL;
8689 PERL_UNUSED_ARG(depth);
8695 /* Find last node. */
8698 regnode * const temp = regnext(scan);
8700 SV * const mysv=sv_newmortal();
8701 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8702 regprop(RExC_rx, mysv, scan);
8703 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8704 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8705 (temp == NULL ? "->" : ""),
8706 (temp == NULL ? PL_reg_name[OP(val)] : "")
8714 if (reg_off_by_arg[OP(scan)]) {
8715 ARG_SET(scan, val - scan);
8718 NEXT_OFF(scan) = val - scan;
8724 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8725 - Look for optimizable sequences at the same time.
8726 - currently only looks for EXACT chains.
8728 This is expermental code. The idea is to use this routine to perform
8729 in place optimizations on branches and groups as they are constructed,
8730 with the long term intention of removing optimization from study_chunk so
8731 that it is purely analytical.
8733 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8734 to control which is which.
8737 /* TODO: All four parms should be const */
8740 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8743 register regnode *scan;
8745 #ifdef EXPERIMENTAL_INPLACESCAN
8748 GET_RE_DEBUG_FLAGS_DECL;
8750 PERL_ARGS_ASSERT_REGTAIL_STUDY;
8756 /* Find last node. */
8760 regnode * const temp = regnext(scan);
8761 #ifdef EXPERIMENTAL_INPLACESCAN
8762 if (PL_regkind[OP(scan)] == EXACT)
8763 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8771 if( exact == PSEUDO )
8773 else if ( exact != OP(scan) )
8782 SV * const mysv=sv_newmortal();
8783 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8784 regprop(RExC_rx, mysv, scan);
8785 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8786 SvPV_nolen_const(mysv),
8788 PL_reg_name[exact]);
8795 SV * const mysv_val=sv_newmortal();
8796 DEBUG_PARSE_MSG("");
8797 regprop(RExC_rx, mysv_val, val);
8798 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8799 SvPV_nolen_const(mysv_val),
8800 (IV)REG_NODE_NUM(val),
8804 if (reg_off_by_arg[OP(scan)]) {
8805 ARG_SET(scan, val - scan);
8808 NEXT_OFF(scan) = val - scan;
8816 - regcurly - a little FSA that accepts {\d+,?\d*}
8819 S_regcurly(register const char *s)
8821 PERL_ARGS_ASSERT_REGCURLY;
8840 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8844 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
8849 for (bit=0; bit<32; bit++) {
8850 if (flags & (1<<bit)) {
8852 PerlIO_printf(Perl_debug_log, "%s",lead);
8853 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
8858 PerlIO_printf(Perl_debug_log, "\n");
8860 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
8866 Perl_regdump(pTHX_ const regexp *r)
8870 SV * const sv = sv_newmortal();
8871 SV *dsv= sv_newmortal();
8873 GET_RE_DEBUG_FLAGS_DECL;
8875 PERL_ARGS_ASSERT_REGDUMP;
8877 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8879 /* Header fields of interest. */
8880 if (r->anchored_substr) {
8881 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8882 RE_SV_DUMPLEN(r->anchored_substr), 30);
8883 PerlIO_printf(Perl_debug_log,
8884 "anchored %s%s at %"IVdf" ",
8885 s, RE_SV_TAIL(r->anchored_substr),
8886 (IV)r->anchored_offset);
8887 } else if (r->anchored_utf8) {
8888 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8889 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8890 PerlIO_printf(Perl_debug_log,
8891 "anchored utf8 %s%s at %"IVdf" ",
8892 s, RE_SV_TAIL(r->anchored_utf8),
8893 (IV)r->anchored_offset);
8895 if (r->float_substr) {
8896 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8897 RE_SV_DUMPLEN(r->float_substr), 30);
8898 PerlIO_printf(Perl_debug_log,
8899 "floating %s%s at %"IVdf"..%"UVuf" ",
8900 s, RE_SV_TAIL(r->float_substr),
8901 (IV)r->float_min_offset, (UV)r->float_max_offset);
8902 } else if (r->float_utf8) {
8903 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8904 RE_SV_DUMPLEN(r->float_utf8), 30);
8905 PerlIO_printf(Perl_debug_log,
8906 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8907 s, RE_SV_TAIL(r->float_utf8),
8908 (IV)r->float_min_offset, (UV)r->float_max_offset);
8910 if (r->check_substr || r->check_utf8)
8911 PerlIO_printf(Perl_debug_log,
8913 (r->check_substr == r->float_substr
8914 && r->check_utf8 == r->float_utf8
8915 ? "(checking floating" : "(checking anchored"));
8916 if (r->extflags & RXf_NOSCAN)
8917 PerlIO_printf(Perl_debug_log, " noscan");
8918 if (r->extflags & RXf_CHECK_ALL)
8919 PerlIO_printf(Perl_debug_log, " isall");
8920 if (r->check_substr || r->check_utf8)
8921 PerlIO_printf(Perl_debug_log, ") ");
8923 if (ri->regstclass) {
8924 regprop(r, sv, ri->regstclass);
8925 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8927 if (r->extflags & RXf_ANCH) {
8928 PerlIO_printf(Perl_debug_log, "anchored");
8929 if (r->extflags & RXf_ANCH_BOL)
8930 PerlIO_printf(Perl_debug_log, "(BOL)");
8931 if (r->extflags & RXf_ANCH_MBOL)
8932 PerlIO_printf(Perl_debug_log, "(MBOL)");
8933 if (r->extflags & RXf_ANCH_SBOL)
8934 PerlIO_printf(Perl_debug_log, "(SBOL)");
8935 if (r->extflags & RXf_ANCH_GPOS)
8936 PerlIO_printf(Perl_debug_log, "(GPOS)");
8937 PerlIO_putc(Perl_debug_log, ' ');
8939 if (r->extflags & RXf_GPOS_SEEN)
8940 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8941 if (r->intflags & PREGf_SKIP)
8942 PerlIO_printf(Perl_debug_log, "plus ");
8943 if (r->intflags & PREGf_IMPLICIT)
8944 PerlIO_printf(Perl_debug_log, "implicit ");
8945 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8946 if (r->extflags & RXf_EVAL_SEEN)
8947 PerlIO_printf(Perl_debug_log, "with eval ");
8948 PerlIO_printf(Perl_debug_log, "\n");
8949 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
8951 PERL_ARGS_ASSERT_REGDUMP;
8952 PERL_UNUSED_CONTEXT;
8954 #endif /* DEBUGGING */
8958 - regprop - printable representation of opcode
8961 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8966 RXi_GET_DECL(prog,progi);
8967 GET_RE_DEBUG_FLAGS_DECL;
8969 PERL_ARGS_ASSERT_REGPROP;
8971 sv_setpvn(sv, "", 0);
8973 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8974 /* It would be nice to FAIL() here, but this may be called from
8975 regexec.c, and it would be hard to supply pRExC_state. */
8976 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8977 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
8979 k = PL_regkind[OP(o)];
8983 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8984 * is a crude hack but it may be the best for now since
8985 * we have no flag "this EXACTish node was UTF-8"
8987 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
8988 PERL_PV_ESCAPE_UNI_DETECT |
8989 PERL_PV_PRETTY_ELLIPSES |
8990 PERL_PV_PRETTY_LTGT |
8991 PERL_PV_PRETTY_NOCLEAR
8993 } else if (k == TRIE) {
8994 /* print the details of the trie in dumpuntil instead, as
8995 * progi->data isn't available here */
8996 const char op = OP(o);
8997 const U32 n = ARG(o);
8998 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8999 (reg_ac_data *)progi->data->data[n] :
9001 const reg_trie_data * const trie
9002 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9004 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9005 DEBUG_TRIE_COMPILE_r(
9006 Perl_sv_catpvf(aTHX_ sv,
9007 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9008 (UV)trie->startstate,
9009 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9010 (UV)trie->wordcount,
9013 (UV)TRIE_CHARCOUNT(trie),
9014 (UV)trie->uniquecharcount
9017 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9019 int rangestart = -1;
9020 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9022 for (i = 0; i <= 256; i++) {
9023 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9024 if (rangestart == -1)
9026 } else if (rangestart != -1) {
9027 if (i <= rangestart + 3)
9028 for (; rangestart < i; rangestart++)
9029 put_byte(sv, rangestart);
9031 put_byte(sv, rangestart);
9033 put_byte(sv, i - 1);
9041 } else if (k == CURLY) {
9042 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9043 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9044 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9046 else if (k == WHILEM && o->flags) /* Ordinal/of */
9047 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9048 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9049 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9050 if ( RXp_PAREN_NAMES(prog) ) {
9051 if ( k != REF || OP(o) < NREF) {
9052 AV *list= (AV *)progi->data->data[progi->name_list_idx];
9053 SV **name= av_fetch(list, ARG(o), 0 );
9055 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9058 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
9059 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
9060 I32 *nums=(I32*)SvPVX(sv_dat);
9061 SV **name= av_fetch(list, nums[0], 0 );
9064 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9065 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9066 (n ? "," : ""), (IV)nums[n]);
9068 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9072 } else if (k == GOSUB)
9073 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9074 else if (k == VERB) {
9076 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9077 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
9078 } else if (k == LOGICAL)
9079 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9080 else if (k == FOLDCHAR)
9081 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9082 else if (k == ANYOF) {
9083 int i, rangestart = -1;
9084 const U8 flags = ANYOF_FLAGS(o);
9086 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9087 static const char * const anyofs[] = {
9120 if (flags & ANYOF_LOCALE)
9121 sv_catpvs(sv, "{loc}");
9122 if (flags & ANYOF_FOLD)
9123 sv_catpvs(sv, "{i}");
9124 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9125 if (flags & ANYOF_INVERT)
9127 for (i = 0; i <= 256; i++) {
9128 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9129 if (rangestart == -1)
9131 } else if (rangestart != -1) {
9132 if (i <= rangestart + 3)
9133 for (; rangestart < i; rangestart++)
9134 put_byte(sv, rangestart);
9136 put_byte(sv, rangestart);
9138 put_byte(sv, i - 1);
9144 if (o->flags & ANYOF_CLASS)
9145 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9146 if (ANYOF_CLASS_TEST(o,i))
9147 sv_catpv(sv, anyofs[i]);
9149 if (flags & ANYOF_UNICODE)
9150 sv_catpvs(sv, "{unicode}");
9151 else if (flags & ANYOF_UNICODE_ALL)
9152 sv_catpvs(sv, "{unicode_all}");
9156 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9160 U8 s[UTF8_MAXBYTES_CASE+1];
9162 for (i = 0; i <= 256; i++) { /* just the first 256 */
9163 uvchr_to_utf8(s, i);
9165 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9166 if (rangestart == -1)
9168 } else if (rangestart != -1) {
9169 if (i <= rangestart + 3)
9170 for (; rangestart < i; rangestart++) {
9171 const U8 * const e = uvchr_to_utf8(s,rangestart);
9173 for(p = s; p < e; p++)
9177 const U8 *e = uvchr_to_utf8(s,rangestart);
9179 for (p = s; p < e; p++)
9182 e = uvchr_to_utf8(s, i-1);
9183 for (p = s; p < e; p++)
9190 sv_catpvs(sv, "..."); /* et cetera */
9194 char *s = savesvpv(lv);
9195 char * const origs = s;
9197 while (*s && *s != '\n')
9201 const char * const t = ++s;
9219 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9221 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9222 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9224 PERL_UNUSED_CONTEXT;
9225 PERL_UNUSED_ARG(sv);
9227 PERL_UNUSED_ARG(prog);
9228 #endif /* DEBUGGING */
9232 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9233 { /* Assume that RE_INTUIT is set */
9235 struct regexp *const prog = (struct regexp *)SvANY(r);
9236 GET_RE_DEBUG_FLAGS_DECL;
9238 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9239 PERL_UNUSED_CONTEXT;
9243 const char * const s = SvPV_nolen_const(prog->check_substr
9244 ? prog->check_substr : prog->check_utf8);
9246 if (!PL_colorset) reginitcolors();
9247 PerlIO_printf(Perl_debug_log,
9248 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9250 prog->check_substr ? "" : "utf8 ",
9251 PL_colors[5],PL_colors[0],
9254 (strlen(s) > 60 ? "..." : ""));
9257 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9263 handles refcounting and freeing the perl core regexp structure. When
9264 it is necessary to actually free the structure the first thing it
9265 does is call the 'free' method of the regexp_engine associated to to
9266 the regexp, allowing the handling of the void *pprivate; member
9267 first. (This routine is not overridable by extensions, which is why
9268 the extensions free is called first.)
9270 See regdupe and regdupe_internal if you change anything here.
9272 #ifndef PERL_IN_XSUB_RE
9274 Perl_pregfree(pTHX_ REGEXP *r)
9280 Perl_pregfree2(pTHX_ REGEXP *rx)
9283 struct regexp *const r = (struct regexp *)SvANY(rx);
9284 GET_RE_DEBUG_FLAGS_DECL;
9286 PERL_ARGS_ASSERT_PREGFREE2;
9289 ReREFCNT_dec(r->mother_re);
9291 CALLREGFREE_PVT(rx); /* free the private data */
9292 if (RXp_PAREN_NAMES(r))
9293 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9296 if (r->anchored_substr)
9297 SvREFCNT_dec(r->anchored_substr);
9298 if (r->anchored_utf8)
9299 SvREFCNT_dec(r->anchored_utf8);
9300 if (r->float_substr)
9301 SvREFCNT_dec(r->float_substr);
9303 SvREFCNT_dec(r->float_utf8);
9304 Safefree(r->substrs);
9306 RX_MATCH_COPY_FREE(rx);
9307 #ifdef PERL_OLD_COPY_ON_WRITE
9309 SvREFCNT_dec(r->saved_copy);
9317 This is a hacky workaround to the structural issue of match results
9318 being stored in the regexp structure which is in turn stored in
9319 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9320 could be PL_curpm in multiple contexts, and could require multiple
9321 result sets being associated with the pattern simultaneously, such
9322 as when doing a recursive match with (??{$qr})
9324 The solution is to make a lightweight copy of the regexp structure
9325 when a qr// is returned from the code executed by (??{$qr}) this
9326 lightweight copy doesnt actually own any of its data except for
9327 the starp/end and the actual regexp structure itself.
9333 Perl_reg_temp_copy (pTHX_ REGEXP *rx)
9335 REGEXP *ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9336 struct regexp *ret = (struct regexp *)SvANY(ret_x);
9337 struct regexp *const r = (struct regexp *)SvANY(rx);
9338 register const I32 npar = r->nparens+1;
9340 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9342 (void)ReREFCNT_inc(rx);
9343 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9344 by pointing directly at the buffer, but flagging that the allocated
9345 space in the copy is zero. As we've just done a struct copy, it's now
9346 a case of zero-ing that, rather than copying the current length. */
9347 SvPV_set(ret_x, RX_WRAPPED(rx));
9348 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9349 StructCopy(&(r->xpv_cur), &(ret->xpv_cur), struct regexp_allocated);
9350 SvLEN_set(ret_x, 0);
9351 Newx(ret->offs, npar, regexp_paren_pair);
9352 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9354 Newx(ret->substrs, 1, struct reg_substr_data);
9355 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9357 SvREFCNT_inc_void(ret->anchored_substr);
9358 SvREFCNT_inc_void(ret->anchored_utf8);
9359 SvREFCNT_inc_void(ret->float_substr);
9360 SvREFCNT_inc_void(ret->float_utf8);
9362 /* check_substr and check_utf8, if non-NULL, point to either their
9363 anchored or float namesakes, and don't hold a second reference. */
9365 RX_MATCH_COPIED_off(ret_x);
9366 #ifdef PERL_OLD_COPY_ON_WRITE
9367 ret->saved_copy = NULL;
9369 ret->mother_re = rx;
9376 /* regfree_internal()
9378 Free the private data in a regexp. This is overloadable by
9379 extensions. Perl takes care of the regexp structure in pregfree(),
9380 this covers the *pprivate pointer which technically perldoesnt
9381 know about, however of course we have to handle the
9382 regexp_internal structure when no extension is in use.
9384 Note this is called before freeing anything in the regexp
9389 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9392 struct regexp *const r = (struct regexp *)SvANY(rx);
9394 GET_RE_DEBUG_FLAGS_DECL;
9396 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9402 SV *dsv= sv_newmortal();
9403 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9404 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9405 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9406 PL_colors[4],PL_colors[5],s);
9409 #ifdef RE_TRACK_PATTERN_OFFSETS
9411 Safefree(ri->u.offsets); /* 20010421 MJD */
9414 int n = ri->data->count;
9415 PAD* new_comppad = NULL;
9420 /* If you add a ->what type here, update the comment in regcomp.h */
9421 switch (ri->data->what[n]) {
9425 SvREFCNT_dec((SV*)ri->data->data[n]);
9428 Safefree(ri->data->data[n]);
9431 new_comppad = (AV*)ri->data->data[n];
9434 if (new_comppad == NULL)
9435 Perl_croak(aTHX_ "panic: pregfree comppad");
9436 PAD_SAVE_LOCAL(old_comppad,
9437 /* Watch out for global destruction's random ordering. */
9438 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9441 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9444 op_free((OP_4tree*)ri->data->data[n]);
9446 PAD_RESTORE_LOCAL(old_comppad);
9447 SvREFCNT_dec((SV*)new_comppad);
9453 { /* Aho Corasick add-on structure for a trie node.
9454 Used in stclass optimization only */
9456 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9458 refcount = --aho->refcount;
9461 PerlMemShared_free(aho->states);
9462 PerlMemShared_free(aho->fail);
9463 /* do this last!!!! */
9464 PerlMemShared_free(ri->data->data[n]);
9465 PerlMemShared_free(ri->regstclass);
9471 /* trie structure. */
9473 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9475 refcount = --trie->refcount;
9478 PerlMemShared_free(trie->charmap);
9479 PerlMemShared_free(trie->states);
9480 PerlMemShared_free(trie->trans);
9482 PerlMemShared_free(trie->bitmap);
9484 PerlMemShared_free(trie->wordlen);
9486 PerlMemShared_free(trie->jump);
9488 PerlMemShared_free(trie->nextword);
9489 /* do this last!!!! */
9490 PerlMemShared_free(ri->data->data[n]);
9495 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9498 Safefree(ri->data->what);
9505 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
9506 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9507 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9508 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9511 re_dup - duplicate a regexp.
9513 This routine is expected to clone a given regexp structure. It is not
9514 compiler under USE_ITHREADS.
9516 After all of the core data stored in struct regexp is duplicated
9517 the regexp_engine.dupe method is used to copy any private data
9518 stored in the *pprivate pointer. This allows extensions to handle
9519 any duplication it needs to do.
9521 See pregfree() and regfree_internal() if you change anything here.
9523 #if defined(USE_ITHREADS)
9524 #ifndef PERL_IN_XSUB_RE
9526 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9530 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9531 struct regexp *ret = (struct regexp *)SvANY(dstr);
9533 PERL_ARGS_ASSERT_RE_DUP_GUTS;
9535 npar = r->nparens+1;
9536 Newx(ret->offs, npar, regexp_paren_pair);
9537 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9539 /* no need to copy these */
9540 Newx(ret->swap, npar, regexp_paren_pair);
9544 /* Do it this way to avoid reading from *r after the StructCopy().
9545 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9546 cache, it doesn't matter. */
9547 const bool anchored = r->check_substr
9548 ? r->check_substr == r->anchored_substr
9549 : r->check_utf8 == r->anchored_utf8;
9550 Newx(ret->substrs, 1, struct reg_substr_data);
9551 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9553 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9554 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9555 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9556 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9558 /* check_substr and check_utf8, if non-NULL, point to either their
9559 anchored or float namesakes, and don't hold a second reference. */
9561 if (ret->check_substr) {
9563 assert(r->check_utf8 == r->anchored_utf8);
9564 ret->check_substr = ret->anchored_substr;
9565 ret->check_utf8 = ret->anchored_utf8;
9567 assert(r->check_substr == r->float_substr);
9568 assert(r->check_utf8 == r->float_utf8);
9569 ret->check_substr = ret->float_substr;
9570 ret->check_utf8 = ret->float_utf8;
9572 } else if (ret->check_utf8) {
9574 ret->check_utf8 = ret->anchored_utf8;
9576 ret->check_utf8 = ret->float_utf8;
9581 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9584 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9586 if (RX_MATCH_COPIED(dstr))
9587 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9590 #ifdef PERL_OLD_COPY_ON_WRITE
9591 ret->saved_copy = NULL;
9594 ret->mother_re = NULL;
9597 #endif /* PERL_IN_XSUB_RE */
9602 This is the internal complement to regdupe() which is used to copy
9603 the structure pointed to by the *pprivate pointer in the regexp.
9604 This is the core version of the extension overridable cloning hook.
9605 The regexp structure being duplicated will be copied by perl prior
9606 to this and will be provided as the regexp *r argument, however
9607 with the /old/ structures pprivate pointer value. Thus this routine
9608 may override any copying normally done by perl.
9610 It returns a pointer to the new regexp_internal structure.
9614 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
9617 struct regexp *const r = (struct regexp *)SvANY(rx);
9618 regexp_internal *reti;
9622 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
9624 npar = r->nparens+1;
9627 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
9628 Copy(ri->program, reti->program, len+1, regnode);
9631 reti->regstclass = NULL;
9635 const int count = ri->data->count;
9638 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9639 char, struct reg_data);
9640 Newx(d->what, count, U8);
9643 for (i = 0; i < count; i++) {
9644 d->what[i] = ri->data->what[i];
9645 switch (d->what[i]) {
9646 /* legal options are one of: sSfpontTu
9647 see also regcomp.h and pregfree() */
9650 case 'p': /* actually an AV, but the dup function is identical. */
9651 case 'u': /* actually an HV, but the dup function is identical. */
9652 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
9655 /* This is cheating. */
9656 Newx(d->data[i], 1, struct regnode_charclass_class);
9657 StructCopy(ri->data->data[i], d->data[i],
9658 struct regnode_charclass_class);
9659 reti->regstclass = (regnode*)d->data[i];
9662 /* Compiled op trees are readonly and in shared memory,
9663 and can thus be shared without duplication. */
9665 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9669 /* Trie stclasses are readonly and can thus be shared
9670 * without duplication. We free the stclass in pregfree
9671 * when the corresponding reg_ac_data struct is freed.
9673 reti->regstclass= ri->regstclass;
9677 ((reg_trie_data*)ri->data->data[i])->refcount++;
9681 d->data[i] = ri->data->data[i];
9684 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9693 reti->name_list_idx = ri->name_list_idx;
9695 #ifdef RE_TRACK_PATTERN_OFFSETS
9696 if (ri->u.offsets) {
9697 Newx(reti->u.offsets, 2*len+1, U32);
9698 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9701 SetProgLen(reti,len);
9707 #endif /* USE_ITHREADS */
9709 #ifndef PERL_IN_XSUB_RE
9712 - regnext - dig the "next" pointer out of a node
9715 Perl_regnext(pTHX_ register regnode *p)
9718 register I32 offset;
9723 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9732 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9735 STRLEN l1 = strlen(pat1);
9736 STRLEN l2 = strlen(pat2);
9739 const char *message;
9741 PERL_ARGS_ASSERT_RE_CROAK2;
9747 Copy(pat1, buf, l1 , char);
9748 Copy(pat2, buf + l1, l2 , char);
9749 buf[l1 + l2] = '\n';
9750 buf[l1 + l2 + 1] = '\0';
9752 /* ANSI variant takes additional second argument */
9753 va_start(args, pat2);
9757 msv = vmess(buf, &args);
9759 message = SvPV_const(msv,l1);
9762 Copy(message, buf, l1 , char);
9763 buf[l1-1] = '\0'; /* Overwrite \n */
9764 Perl_croak(aTHX_ "%s", buf);
9767 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9769 #ifndef PERL_IN_XSUB_RE
9771 Perl_save_re_context(pTHX)
9775 struct re_save_state *state;
9777 SAVEVPTR(PL_curcop);
9778 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9780 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9781 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9782 SSPUSHINT(SAVEt_RE_STATE);
9784 Copy(&PL_reg_state, state, 1, struct re_save_state);
9786 PL_reg_start_tmp = 0;
9787 PL_reg_start_tmpl = 0;
9788 PL_reg_oldsaved = NULL;
9789 PL_reg_oldsavedlen = 0;
9791 PL_reg_leftiter = 0;
9792 PL_reg_poscache = NULL;
9793 PL_reg_poscache_size = 0;
9794 #ifdef PERL_OLD_COPY_ON_WRITE
9798 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9800 const REGEXP * const rx = PM_GETRE(PL_curpm);
9803 for (i = 1; i <= RX_NPARENS(rx); i++) {
9804 char digits[TYPE_CHARS(long)];
9805 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9806 GV *const *const gvp
9807 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9810 GV * const gv = *gvp;
9811 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9821 clear_re(pTHX_ void *r)
9824 ReREFCNT_dec((REGEXP *)r);
9830 S_put_byte(pTHX_ SV *sv, int c)
9832 PERL_ARGS_ASSERT_PUT_BYTE;
9834 /* Our definition of isPRINT() ignores locales, so only bytes that are
9835 not part of UTF-8 are considered printable. I assume that the same
9836 holds for UTF-EBCDIC.
9837 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
9838 which Wikipedia says:
9840 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
9841 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
9842 identical, to the ASCII delete (DEL) or rubout control character.
9843 ) So the old condition can be simplified to !isPRINT(c) */
9845 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9847 const char string = c;
9848 if (c == '-' || c == ']' || c == '\\' || c == '^')
9849 sv_catpvs(sv, "\\");
9850 sv_catpvn(sv, &string, 1);
9855 #define CLEAR_OPTSTART \
9856 if (optstart) STMT_START { \
9857 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9861 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9863 STATIC const regnode *
9864 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9865 const regnode *last, const regnode *plast,
9866 SV* sv, I32 indent, U32 depth)
9869 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9870 register const regnode *next;
9871 const regnode *optstart= NULL;
9874 GET_RE_DEBUG_FLAGS_DECL;
9876 PERL_ARGS_ASSERT_DUMPUNTIL;
9878 #ifdef DEBUG_DUMPUNTIL
9879 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9880 last ? last-start : 0,plast ? plast-start : 0);
9883 if (plast && plast < last)
9886 while (PL_regkind[op] != END && (!last || node < last)) {
9887 /* While that wasn't END last time... */
9890 if (op == CLOSE || op == WHILEM)
9892 next = regnext((regnode *)node);
9895 if (OP(node) == OPTIMIZED) {
9896 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9903 regprop(r, sv, node);
9904 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9905 (int)(2*indent + 1), "", SvPVX_const(sv));
9907 if (OP(node) != OPTIMIZED) {
9908 if (next == NULL) /* Next ptr. */
9909 PerlIO_printf(Perl_debug_log, " (0)");
9910 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9911 PerlIO_printf(Perl_debug_log, " (FAIL)");
9913 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9914 (void)PerlIO_putc(Perl_debug_log, '\n');
9918 if (PL_regkind[(U8)op] == BRANCHJ) {
9921 register const regnode *nnode = (OP(next) == LONGJMP
9922 ? regnext((regnode *)next)
9924 if (last && nnode > last)
9926 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9929 else if (PL_regkind[(U8)op] == BRANCH) {
9931 DUMPUNTIL(NEXTOPER(node), next);
9933 else if ( PL_regkind[(U8)op] == TRIE ) {
9934 const regnode *this_trie = node;
9935 const char op = OP(node);
9936 const U32 n = ARG(node);
9937 const reg_ac_data * const ac = op>=AHOCORASICK ?
9938 (reg_ac_data *)ri->data->data[n] :
9940 const reg_trie_data * const trie =
9941 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9943 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9945 const regnode *nextbranch= NULL;
9947 sv_setpvn(sv, "", 0);
9948 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9949 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9951 PerlIO_printf(Perl_debug_log, "%*s%s ",
9952 (int)(2*(indent+3)), "",
9953 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9954 PL_colors[0], PL_colors[1],
9955 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9956 PERL_PV_PRETTY_ELLIPSES |
9962 U16 dist= trie->jump[word_idx+1];
9963 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9964 (UV)((dist ? this_trie + dist : next) - start));
9967 nextbranch= this_trie + trie->jump[0];
9968 DUMPUNTIL(this_trie + dist, nextbranch);
9970 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9971 nextbranch= regnext((regnode *)nextbranch);
9973 PerlIO_printf(Perl_debug_log, "\n");
9976 if (last && next > last)
9981 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9982 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9983 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9985 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9987 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9989 else if ( op == PLUS || op == STAR) {
9990 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9992 else if (op == ANYOF) {
9993 /* arglen 1 + class block */
9994 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9995 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9996 node = NEXTOPER(node);
9998 else if (PL_regkind[(U8)op] == EXACT) {
9999 /* Literal string, where present. */
10000 node += NODE_SZ_STR(node) - 1;
10001 node = NEXTOPER(node);
10004 node = NEXTOPER(node);
10005 node += regarglen[(U8)op];
10007 if (op == CURLYX || op == OPEN)
10011 #ifdef DEBUG_DUMPUNTIL
10012 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10017 #endif /* DEBUGGING */
10021 * c-indentation-style: bsd
10022 * c-basic-offset: 4
10023 * indent-tabs-mode: t
10026 * ex: set ts=8 sts=4 sw=4 noet:
https://perl5.git.perl.org / perl5.git / blob