5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
8 /* This file contains functions for compiling a regular expression. See
9 * also regexec.c which funnily enough, contains functions for executing
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_comp.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
57 **** Alterations to Henry's code are...
59 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
60 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 by Larry Wall and others
62 **** You may distribute under the terms of either the GNU General Public
63 **** License or the Artistic License, as specified in the README file.
66 * Beware that some of this code is subtly aware of the way operator
67 * precedence is structured in regular expressions. Serious changes in
68 * regular-expression syntax might require a total rethink.
71 #define PERL_IN_REGCOMP_C
74 #ifndef PERL_IN_XSUB_RE
79 #ifdef PERL_IN_XSUB_RE
90 # if defined(BUGGY_MSC6)
91 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
92 # pragma optimize("a",off)
93 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
94 # pragma optimize("w",on )
95 # endif /* BUGGY_MSC6 */
102 typedef struct RExC_state_t {
103 U32 flags; /* are we folding, multilining? */
104 char *precomp; /* uncompiled string. */
105 regexp *rx; /* perl core regexp structure */
106 regexp_internal *rxi; /* internal data for regexp object pprivate field */
107 char *start; /* Start of input for compile */
108 char *end; /* End of input for compile */
109 char *parse; /* Input-scan pointer. */
110 I32 whilem_seen; /* number of WHILEM in this expr */
111 regnode *emit_start; /* Start of emitted-code area */
112 regnode *emit_bound; /* First regnode outside of the allocated space */
113 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
114 I32 naughty; /* How bad is this pattern? */
115 I32 sawback; /* Did we see \1, ...? */
117 I32 size; /* Code size. */
118 I32 npar; /* Capture buffer count, (OPEN). */
119 I32 cpar; /* Capture buffer count, (CLOSE). */
120 I32 nestroot; /* root parens we are in - used by accept */
124 regnode **open_parens; /* pointers to open parens */
125 regnode **close_parens; /* pointers to close parens */
126 regnode *opend; /* END node in program */
128 HV *charnames; /* cache of named sequences */
129 HV *paren_names; /* Paren names */
131 regnode **recurse; /* Recurse regops */
132 I32 recurse_count; /* Number of recurse regops */
134 char *starttry; /* -Dr: where regtry was called. */
135 #define RExC_starttry (pRExC_state->starttry)
138 const char *lastparse;
140 AV *paren_name_list; /* idx -> name */
141 #define RExC_lastparse (pRExC_state->lastparse)
142 #define RExC_lastnum (pRExC_state->lastnum)
143 #define RExC_paren_name_list (pRExC_state->paren_name_list)
147 #define RExC_flags (pRExC_state->flags)
148 #define RExC_precomp (pRExC_state->precomp)
149 #define RExC_rx (pRExC_state->rx)
150 #define RExC_rxi (pRExC_state->rxi)
151 #define RExC_start (pRExC_state->start)
152 #define RExC_end (pRExC_state->end)
153 #define RExC_parse (pRExC_state->parse)
154 #define RExC_whilem_seen (pRExC_state->whilem_seen)
155 #ifdef RE_TRACK_PATTERN_OFFSETS
156 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
158 #define RExC_emit (pRExC_state->emit)
159 #define RExC_emit_start (pRExC_state->emit_start)
160 #define RExC_emit_bound (pRExC_state->emit_bound)
161 #define RExC_naughty (pRExC_state->naughty)
162 #define RExC_sawback (pRExC_state->sawback)
163 #define RExC_seen (pRExC_state->seen)
164 #define RExC_size (pRExC_state->size)
165 #define RExC_npar (pRExC_state->npar)
166 #define RExC_nestroot (pRExC_state->nestroot)
167 #define RExC_extralen (pRExC_state->extralen)
168 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
169 #define RExC_seen_evals (pRExC_state->seen_evals)
170 #define RExC_utf8 (pRExC_state->utf8)
171 #define RExC_charnames (pRExC_state->charnames)
172 #define RExC_open_parens (pRExC_state->open_parens)
173 #define RExC_close_parens (pRExC_state->close_parens)
174 #define RExC_opend (pRExC_state->opend)
175 #define RExC_paren_names (pRExC_state->paren_names)
176 #define RExC_recurse (pRExC_state->recurse)
177 #define RExC_recurse_count (pRExC_state->recurse_count)
180 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
181 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
182 ((*s) == '{' && regcurly(s)))
185 #undef SPSTART /* dratted cpp namespace... */
188 * Flags to be passed up and down.
190 #define WORST 0 /* Worst case. */
191 #define HASWIDTH 0x01 /* Known to match non-null strings. */
192 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
193 #define SPSTART 0x04 /* Starts with * or +. */
194 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
195 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
197 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
199 /* whether trie related optimizations are enabled */
200 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
201 #define TRIE_STUDY_OPT
202 #define FULL_TRIE_STUDY
208 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
209 #define PBITVAL(paren) (1 << ((paren) & 7))
210 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
211 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
212 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
215 /* About scan_data_t.
217 During optimisation we recurse through the regexp program performing
218 various inplace (keyhole style) optimisations. In addition study_chunk
219 and scan_commit populate this data structure with information about
220 what strings MUST appear in the pattern. We look for the longest
221 string that must appear for at a fixed location, and we look for the
222 longest string that may appear at a floating location. So for instance
227 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
228 strings (because they follow a .* construct). study_chunk will identify
229 both FOO and BAR as being the longest fixed and floating strings respectively.
231 The strings can be composites, for instance
235 will result in a composite fixed substring 'foo'.
237 For each string some basic information is maintained:
239 - offset or min_offset
240 This is the position the string must appear at, or not before.
241 It also implicitly (when combined with minlenp) tells us how many
242 character must match before the string we are searching.
243 Likewise when combined with minlenp and the length of the string
244 tells us how many characters must appear after the string we have
248 Only used for floating strings. This is the rightmost point that
249 the string can appear at. Ifset to I32 max it indicates that the
250 string can occur infinitely far to the right.
253 A pointer to the minimum length of the pattern that the string
254 was found inside. This is important as in the case of positive
255 lookahead or positive lookbehind we can have multiple patterns
260 The minimum length of the pattern overall is 3, the minimum length
261 of the lookahead part is 3, but the minimum length of the part that
262 will actually match is 1. So 'FOO's minimum length is 3, but the
263 minimum length for the F is 1. This is important as the minimum length
264 is used to determine offsets in front of and behind the string being
265 looked for. Since strings can be composites this is the length of the
266 pattern at the time it was commited with a scan_commit. Note that
267 the length is calculated by study_chunk, so that the minimum lengths
268 are not known until the full pattern has been compiled, thus the
269 pointer to the value.
273 In the case of lookbehind the string being searched for can be
274 offset past the start point of the final matching string.
275 If this value was just blithely removed from the min_offset it would
276 invalidate some of the calculations for how many chars must match
277 before or after (as they are derived from min_offset and minlen and
278 the length of the string being searched for).
279 When the final pattern is compiled and the data is moved from the
280 scan_data_t structure into the regexp structure the information
281 about lookbehind is factored in, with the information that would
282 have been lost precalculated in the end_shift field for the
285 The fields pos_min and pos_delta are used to store the minimum offset
286 and the delta to the maximum offset at the current point in the pattern.
290 typedef struct scan_data_t {
291 /*I32 len_min; unused */
292 /*I32 len_delta; unused */
296 I32 last_end; /* min value, <0 unless valid. */
299 SV **longest; /* Either &l_fixed, or &l_float. */
300 SV *longest_fixed; /* longest fixed string found in pattern */
301 I32 offset_fixed; /* offset where it starts */
302 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
303 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
304 SV *longest_float; /* longest floating string found in pattern */
305 I32 offset_float_min; /* earliest point in string it can appear */
306 I32 offset_float_max; /* latest point in string it can appear */
307 I32 *minlen_float; /* pointer to the minlen relevent to the string */
308 I32 lookbehind_float; /* is the position of the string modified by LB */
312 struct regnode_charclass_class *start_class;
316 * Forward declarations for pregcomp()'s friends.
319 static const scan_data_t zero_scan_data =
320 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
322 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
323 #define SF_BEFORE_SEOL 0x0001
324 #define SF_BEFORE_MEOL 0x0002
325 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
326 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
329 # define SF_FIX_SHIFT_EOL (0+2)
330 # define SF_FL_SHIFT_EOL (0+4)
332 # define SF_FIX_SHIFT_EOL (+2)
333 # define SF_FL_SHIFT_EOL (+4)
336 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
337 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
339 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
340 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
341 #define SF_IS_INF 0x0040
342 #define SF_HAS_PAR 0x0080
343 #define SF_IN_PAR 0x0100
344 #define SF_HAS_EVAL 0x0200
345 #define SCF_DO_SUBSTR 0x0400
346 #define SCF_DO_STCLASS_AND 0x0800
347 #define SCF_DO_STCLASS_OR 0x1000
348 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
349 #define SCF_WHILEM_VISITED_POS 0x2000
351 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
352 #define SCF_SEEN_ACCEPT 0x8000
354 #define UTF (RExC_utf8 != 0)
355 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
356 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
358 #define OOB_UNICODE 12345678
359 #define OOB_NAMEDCLASS -1
361 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
362 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
365 /* length of regex to show in messages that don't mark a position within */
366 #define RegexLengthToShowInErrorMessages 127
369 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
370 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
371 * op/pragma/warn/regcomp.
373 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
374 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
376 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
379 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
380 * arg. Show regex, up to a maximum length. If it's too long, chop and add
383 #define _FAIL(code) STMT_START { \
384 const char *ellipses = ""; \
385 IV len = RExC_end - RExC_precomp; \
388 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
389 if (len > RegexLengthToShowInErrorMessages) { \
390 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
391 len = RegexLengthToShowInErrorMessages - 10; \
397 #define FAIL(msg) _FAIL( \
398 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
399 msg, (int)len, RExC_precomp, ellipses))
401 #define FAIL2(msg,arg) _FAIL( \
402 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
403 arg, (int)len, RExC_precomp, ellipses))
406 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
408 #define Simple_vFAIL(m) STMT_START { \
409 const IV offset = RExC_parse - RExC_precomp; \
410 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
411 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
415 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
417 #define vFAIL(m) STMT_START { \
419 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
424 * Like Simple_vFAIL(), but accepts two arguments.
426 #define Simple_vFAIL2(m,a1) STMT_START { \
427 const IV offset = RExC_parse - RExC_precomp; \
428 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
429 (int)offset, RExC_precomp, RExC_precomp + offset); \
433 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
435 #define vFAIL2(m,a1) STMT_START { \
437 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
438 Simple_vFAIL2(m, a1); \
443 * Like Simple_vFAIL(), but accepts three arguments.
445 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
446 const IV offset = RExC_parse - RExC_precomp; \
447 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
448 (int)offset, RExC_precomp, RExC_precomp + offset); \
452 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
454 #define vFAIL3(m,a1,a2) STMT_START { \
456 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
457 Simple_vFAIL3(m, a1, a2); \
461 * Like Simple_vFAIL(), but accepts four arguments.
463 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
464 const IV offset = RExC_parse - RExC_precomp; \
465 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
466 (int)offset, RExC_precomp, RExC_precomp + offset); \
469 #define vWARN(loc,m) STMT_START { \
470 const IV offset = loc - RExC_precomp; \
471 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
472 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
475 #define vWARNdep(loc,m) STMT_START { \
476 const IV offset = loc - RExC_precomp; \
477 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
478 "%s" REPORT_LOCATION, \
479 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
483 #define vWARN2(loc, m, a1) STMT_START { \
484 const IV offset = loc - RExC_precomp; \
485 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
486 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
489 #define vWARN3(loc, m, a1, a2) STMT_START { \
490 const IV offset = loc - RExC_precomp; \
491 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
492 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
495 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
496 const IV offset = loc - RExC_precomp; \
497 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
498 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
501 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
502 const IV offset = loc - RExC_precomp; \
503 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
504 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
508 /* Allow for side effects in s */
509 #define REGC(c,s) STMT_START { \
510 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
513 /* Macros for recording node offsets. 20001227 mjd@plover.com
514 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
515 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
516 * Element 0 holds the number n.
517 * Position is 1 indexed.
519 #ifndef RE_TRACK_PATTERN_OFFSETS
520 #define Set_Node_Offset_To_R(node,byte)
521 #define Set_Node_Offset(node,byte)
522 #define Set_Cur_Node_Offset
523 #define Set_Node_Length_To_R(node,len)
524 #define Set_Node_Length(node,len)
525 #define Set_Node_Cur_Length(node)
526 #define Node_Offset(n)
527 #define Node_Length(n)
528 #define Set_Node_Offset_Length(node,offset,len)
529 #define ProgLen(ri) ri->u.proglen
530 #define SetProgLen(ri,x) ri->u.proglen = x
532 #define ProgLen(ri) ri->u.offsets[0]
533 #define SetProgLen(ri,x) ri->u.offsets[0] = x
534 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
536 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
537 __LINE__, (int)(node), (int)(byte))); \
539 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
541 RExC_offsets[2*(node)-1] = (byte); \
546 #define Set_Node_Offset(node,byte) \
547 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
548 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
550 #define Set_Node_Length_To_R(node,len) STMT_START { \
552 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
553 __LINE__, (int)(node), (int)(len))); \
555 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
557 RExC_offsets[2*(node)] = (len); \
562 #define Set_Node_Length(node,len) \
563 Set_Node_Length_To_R((node)-RExC_emit_start, len)
564 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
565 #define Set_Node_Cur_Length(node) \
566 Set_Node_Length(node, RExC_parse - parse_start)
568 /* Get offsets and lengths */
569 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
570 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
572 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
573 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
574 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
578 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
579 #define EXPERIMENTAL_INPLACESCAN
580 #endif /*RE_TRACK_PATTERN_OFFSETS*/
582 #define DEBUG_STUDYDATA(str,data,depth) \
583 DEBUG_OPTIMISE_MORE_r(if(data){ \
584 PerlIO_printf(Perl_debug_log, \
585 "%*s" str "Pos:%"IVdf"/%"IVdf \
586 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
587 (int)(depth)*2, "", \
588 (IV)((data)->pos_min), \
589 (IV)((data)->pos_delta), \
590 (UV)((data)->flags), \
591 (IV)((data)->whilem_c), \
592 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
593 is_inf ? "INF " : "" \
595 if ((data)->last_found) \
596 PerlIO_printf(Perl_debug_log, \
597 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
598 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
599 SvPVX_const((data)->last_found), \
600 (IV)((data)->last_end), \
601 (IV)((data)->last_start_min), \
602 (IV)((data)->last_start_max), \
603 ((data)->longest && \
604 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
605 SvPVX_const((data)->longest_fixed), \
606 (IV)((data)->offset_fixed), \
607 ((data)->longest && \
608 (data)->longest==&((data)->longest_float)) ? "*" : "", \
609 SvPVX_const((data)->longest_float), \
610 (IV)((data)->offset_float_min), \
611 (IV)((data)->offset_float_max) \
613 PerlIO_printf(Perl_debug_log,"\n"); \
616 static void clear_re(pTHX_ void *r);
618 /* Mark that we cannot extend a found fixed substring at this point.
619 Update the longest found anchored substring and the longest found
620 floating substrings if needed. */
623 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
625 const STRLEN l = CHR_SVLEN(data->last_found);
626 const STRLEN old_l = CHR_SVLEN(*data->longest);
627 GET_RE_DEBUG_FLAGS_DECL;
629 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
630 SvSetMagicSV(*data->longest, data->last_found);
631 if (*data->longest == data->longest_fixed) {
632 data->offset_fixed = l ? data->last_start_min : data->pos_min;
633 if (data->flags & SF_BEFORE_EOL)
635 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
637 data->flags &= ~SF_FIX_BEFORE_EOL;
638 data->minlen_fixed=minlenp;
639 data->lookbehind_fixed=0;
641 else { /* *data->longest == data->longest_float */
642 data->offset_float_min = l ? data->last_start_min : data->pos_min;
643 data->offset_float_max = (l
644 ? data->last_start_max
645 : data->pos_min + data->pos_delta);
646 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
647 data->offset_float_max = I32_MAX;
648 if (data->flags & SF_BEFORE_EOL)
650 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
652 data->flags &= ~SF_FL_BEFORE_EOL;
653 data->minlen_float=minlenp;
654 data->lookbehind_float=0;
657 SvCUR_set(data->last_found, 0);
659 SV * const sv = data->last_found;
660 if (SvUTF8(sv) && SvMAGICAL(sv)) {
661 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
667 data->flags &= ~SF_BEFORE_EOL;
668 DEBUG_STUDYDATA("commit: ",data,0);
671 /* Can match anything (initialization) */
673 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
675 ANYOF_CLASS_ZERO(cl);
676 ANYOF_BITMAP_SETALL(cl);
677 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
679 cl->flags |= ANYOF_LOCALE;
682 /* Can match anything (initialization) */
684 S_cl_is_anything(const struct regnode_charclass_class *cl)
688 for (value = 0; value <= ANYOF_MAX; value += 2)
689 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
691 if (!(cl->flags & ANYOF_UNICODE_ALL))
693 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
698 /* Can match anything (initialization) */
700 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
702 Zero(cl, 1, struct regnode_charclass_class);
704 cl_anything(pRExC_state, cl);
708 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
710 Zero(cl, 1, struct regnode_charclass_class);
712 cl_anything(pRExC_state, cl);
714 cl->flags |= ANYOF_LOCALE;
717 /* 'And' a given class with another one. Can create false positives */
718 /* We assume that cl is not inverted */
720 S_cl_and(struct regnode_charclass_class *cl,
721 const struct regnode_charclass_class *and_with)
724 assert(and_with->type == ANYOF);
725 if (!(and_with->flags & ANYOF_CLASS)
726 && !(cl->flags & ANYOF_CLASS)
727 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
728 && !(and_with->flags & ANYOF_FOLD)
729 && !(cl->flags & ANYOF_FOLD)) {
732 if (and_with->flags & ANYOF_INVERT)
733 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
734 cl->bitmap[i] &= ~and_with->bitmap[i];
736 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
737 cl->bitmap[i] &= and_with->bitmap[i];
738 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
739 if (!(and_with->flags & ANYOF_EOS))
740 cl->flags &= ~ANYOF_EOS;
742 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
743 !(and_with->flags & ANYOF_INVERT)) {
744 cl->flags &= ~ANYOF_UNICODE_ALL;
745 cl->flags |= ANYOF_UNICODE;
746 ARG_SET(cl, ARG(and_with));
748 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
749 !(and_with->flags & ANYOF_INVERT))
750 cl->flags &= ~ANYOF_UNICODE_ALL;
751 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
752 !(and_with->flags & ANYOF_INVERT))
753 cl->flags &= ~ANYOF_UNICODE;
756 /* 'OR' a given class with another one. Can create false positives */
757 /* We assume that cl is not inverted */
759 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
761 if (or_with->flags & ANYOF_INVERT) {
763 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
764 * <= (B1 | !B2) | (CL1 | !CL2)
765 * which is wasteful if CL2 is small, but we ignore CL2:
766 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
767 * XXXX Can we handle case-fold? Unclear:
768 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
769 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
771 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
772 && !(or_with->flags & ANYOF_FOLD)
773 && !(cl->flags & ANYOF_FOLD) ) {
776 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
777 cl->bitmap[i] |= ~or_with->bitmap[i];
778 } /* XXXX: logic is complicated otherwise */
780 cl_anything(pRExC_state, cl);
783 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
784 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
785 && (!(or_with->flags & ANYOF_FOLD)
786 || (cl->flags & ANYOF_FOLD)) ) {
789 /* OR char bitmap and class bitmap separately */
790 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
791 cl->bitmap[i] |= or_with->bitmap[i];
792 if (or_with->flags & ANYOF_CLASS) {
793 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
794 cl->classflags[i] |= or_with->classflags[i];
795 cl->flags |= ANYOF_CLASS;
798 else { /* XXXX: logic is complicated, leave it along for a moment. */
799 cl_anything(pRExC_state, cl);
802 if (or_with->flags & ANYOF_EOS)
803 cl->flags |= ANYOF_EOS;
805 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
806 ARG(cl) != ARG(or_with)) {
807 cl->flags |= ANYOF_UNICODE_ALL;
808 cl->flags &= ~ANYOF_UNICODE;
810 if (or_with->flags & ANYOF_UNICODE_ALL) {
811 cl->flags |= ANYOF_UNICODE_ALL;
812 cl->flags &= ~ANYOF_UNICODE;
816 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
817 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
818 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
819 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
824 dump_trie(trie,widecharmap,revcharmap)
825 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
826 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
828 These routines dump out a trie in a somewhat readable format.
829 The _interim_ variants are used for debugging the interim
830 tables that are used to generate the final compressed
831 representation which is what dump_trie expects.
833 Part of the reason for their existance is to provide a form
834 of documentation as to how the different representations function.
839 Dumps the final compressed table form of the trie to Perl_debug_log.
840 Used for debugging make_trie().
844 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
845 AV *revcharmap, U32 depth)
848 SV *sv=sv_newmortal();
849 int colwidth= widecharmap ? 6 : 4;
850 GET_RE_DEBUG_FLAGS_DECL;
853 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
854 (int)depth * 2 + 2,"",
855 "Match","Base","Ofs" );
857 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
858 SV ** const tmp = av_fetch( revcharmap, state, 0);
860 PerlIO_printf( Perl_debug_log, "%*s",
862 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
863 PL_colors[0], PL_colors[1],
864 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
865 PERL_PV_ESCAPE_FIRSTCHAR
870 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
871 (int)depth * 2 + 2,"");
873 for( state = 0 ; state < trie->uniquecharcount ; state++ )
874 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
875 PerlIO_printf( Perl_debug_log, "\n");
877 for( state = 1 ; state < trie->statecount ; state++ ) {
878 const U32 base = trie->states[ state ].trans.base;
880 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
882 if ( trie->states[ state ].wordnum ) {
883 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
885 PerlIO_printf( Perl_debug_log, "%6s", "" );
888 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
893 while( ( base + ofs < trie->uniquecharcount ) ||
894 ( base + ofs - trie->uniquecharcount < trie->lasttrans
895 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
898 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
900 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
901 if ( ( base + ofs >= trie->uniquecharcount ) &&
902 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
903 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
905 PerlIO_printf( Perl_debug_log, "%*"UVXf,
907 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
909 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
913 PerlIO_printf( Perl_debug_log, "]");
916 PerlIO_printf( Perl_debug_log, "\n" );
920 Dumps a fully constructed but uncompressed trie in list form.
921 List tries normally only are used for construction when the number of
922 possible chars (trie->uniquecharcount) is very high.
923 Used for debugging make_trie().
926 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
927 HV *widecharmap, AV *revcharmap, U32 next_alloc,
931 SV *sv=sv_newmortal();
932 int colwidth= widecharmap ? 6 : 4;
933 GET_RE_DEBUG_FLAGS_DECL;
934 /* print out the table precompression. */
935 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
936 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
937 "------:-----+-----------------\n" );
939 for( state=1 ; state < next_alloc ; state ++ ) {
942 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
943 (int)depth * 2 + 2,"", (UV)state );
944 if ( ! trie->states[ state ].wordnum ) {
945 PerlIO_printf( Perl_debug_log, "%5s| ","");
947 PerlIO_printf( Perl_debug_log, "W%4x| ",
948 trie->states[ state ].wordnum
951 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
952 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
954 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
956 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
957 PL_colors[0], PL_colors[1],
958 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
959 PERL_PV_ESCAPE_FIRSTCHAR
961 TRIE_LIST_ITEM(state,charid).forid,
962 (UV)TRIE_LIST_ITEM(state,charid).newstate
965 PerlIO_printf(Perl_debug_log, "\n%*s| ",
966 (int)((depth * 2) + 14), "");
969 PerlIO_printf( Perl_debug_log, "\n");
974 Dumps a fully constructed but uncompressed trie in table form.
975 This is the normal DFA style state transition table, with a few
976 twists to facilitate compression later.
977 Used for debugging make_trie().
980 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
981 HV *widecharmap, AV *revcharmap, U32 next_alloc,
986 SV *sv=sv_newmortal();
987 int colwidth= widecharmap ? 6 : 4;
988 GET_RE_DEBUG_FLAGS_DECL;
991 print out the table precompression so that we can do a visual check
992 that they are identical.
995 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
997 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
998 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1000 PerlIO_printf( Perl_debug_log, "%*s",
1002 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1003 PL_colors[0], PL_colors[1],
1004 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1005 PERL_PV_ESCAPE_FIRSTCHAR
1011 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1013 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1014 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1017 PerlIO_printf( Perl_debug_log, "\n" );
1019 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1021 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1022 (int)depth * 2 + 2,"",
1023 (UV)TRIE_NODENUM( state ) );
1025 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1026 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1028 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1030 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1032 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1033 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1035 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1036 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1043 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1044 startbranch: the first branch in the whole branch sequence
1045 first : start branch of sequence of branch-exact nodes.
1046 May be the same as startbranch
1047 last : Thing following the last branch.
1048 May be the same as tail.
1049 tail : item following the branch sequence
1050 count : words in the sequence
1051 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1052 depth : indent depth
1054 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1056 A trie is an N'ary tree where the branches are determined by digital
1057 decomposition of the key. IE, at the root node you look up the 1st character and
1058 follow that branch repeat until you find the end of the branches. Nodes can be
1059 marked as "accepting" meaning they represent a complete word. Eg:
1063 would convert into the following structure. Numbers represent states, letters
1064 following numbers represent valid transitions on the letter from that state, if
1065 the number is in square brackets it represents an accepting state, otherwise it
1066 will be in parenthesis.
1068 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1072 (1) +-i->(6)-+-s->[7]
1074 +-s->(3)-+-h->(4)-+-e->[5]
1076 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1078 This shows that when matching against the string 'hers' we will begin at state 1
1079 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1080 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1081 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1082 single traverse. We store a mapping from accepting to state to which word was
1083 matched, and then when we have multiple possibilities we try to complete the
1084 rest of the regex in the order in which they occured in the alternation.
1086 The only prior NFA like behaviour that would be changed by the TRIE support is
1087 the silent ignoring of duplicate alternations which are of the form:
1089 / (DUPE|DUPE) X? (?{ ... }) Y /x
1091 Thus EVAL blocks follwing a trie may be called a different number of times with
1092 and without the optimisation. With the optimisations dupes will be silently
1093 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1094 the following demonstrates:
1096 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1098 which prints out 'word' three times, but
1100 'words'=~/(word|word|word)(?{ print $1 })S/
1102 which doesnt print it out at all. This is due to other optimisations kicking in.
1104 Example of what happens on a structural level:
1106 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1108 1: CURLYM[1] {1,32767}(18)
1119 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1120 and should turn into:
1122 1: CURLYM[1] {1,32767}(18)
1124 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1132 Cases where tail != last would be like /(?foo|bar)baz/:
1142 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1143 and would end up looking like:
1146 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1153 d = uvuni_to_utf8_flags(d, uv, 0);
1155 is the recommended Unicode-aware way of saying
1160 #define TRIE_STORE_REVCHAR \
1162 SV *tmp = newSVpvs(""); \
1163 if (UTF) SvUTF8_on(tmp); \
1164 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1165 av_push( revcharmap, tmp ); \
1168 #define TRIE_READ_CHAR STMT_START { \
1172 if ( foldlen > 0 ) { \
1173 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1178 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1179 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1180 foldlen -= UNISKIP( uvc ); \
1181 scan = foldbuf + UNISKIP( uvc ); \
1184 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1194 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1195 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1196 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1197 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1199 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1200 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1201 TRIE_LIST_CUR( state )++; \
1204 #define TRIE_LIST_NEW(state) STMT_START { \
1205 Newxz( trie->states[ state ].trans.list, \
1206 4, reg_trie_trans_le ); \
1207 TRIE_LIST_CUR( state ) = 1; \
1208 TRIE_LIST_LEN( state ) = 4; \
1211 #define TRIE_HANDLE_WORD(state) STMT_START { \
1212 U16 dupe= trie->states[ state ].wordnum; \
1213 regnode * const noper_next = regnext( noper ); \
1215 if (trie->wordlen) \
1216 trie->wordlen[ curword ] = wordlen; \
1218 /* store the word for dumping */ \
1220 if (OP(noper) != NOTHING) \
1221 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1223 tmp = newSVpvn( "", 0 ); \
1224 if ( UTF ) SvUTF8_on( tmp ); \
1225 av_push( trie_words, tmp ); \
1230 if ( noper_next < tail ) { \
1232 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1233 trie->jump[curword] = (U16)(noper_next - convert); \
1235 jumper = noper_next; \
1237 nextbranch= regnext(cur); \
1241 /* So it's a dupe. This means we need to maintain a */\
1242 /* linked-list from the first to the next. */\
1243 /* we only allocate the nextword buffer when there */\
1244 /* a dupe, so first time we have to do the allocation */\
1245 if (!trie->nextword) \
1246 trie->nextword = (U16 *) \
1247 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1248 while ( trie->nextword[dupe] ) \
1249 dupe= trie->nextword[dupe]; \
1250 trie->nextword[dupe]= curword; \
1252 /* we haven't inserted this word yet. */ \
1253 trie->states[ state ].wordnum = curword; \
1258 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1259 ( ( base + charid >= ucharcount \
1260 && base + charid < ubound \
1261 && state == trie->trans[ base - ucharcount + charid ].check \
1262 && trie->trans[ base - ucharcount + charid ].next ) \
1263 ? trie->trans[ base - ucharcount + charid ].next \
1264 : ( state==1 ? special : 0 ) \
1268 #define MADE_JUMP_TRIE 2
1269 #define MADE_EXACT_TRIE 4
1272 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1275 /* first pass, loop through and scan words */
1276 reg_trie_data *trie;
1277 HV *widecharmap = NULL;
1278 AV *revcharmap = newAV();
1280 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1285 regnode *jumper = NULL;
1286 regnode *nextbranch = NULL;
1287 regnode *convert = NULL;
1288 /* we just use folder as a flag in utf8 */
1289 const U8 * const folder = ( flags == EXACTF
1291 : ( flags == EXACTFL
1298 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1299 AV *trie_words = NULL;
1300 /* along with revcharmap, this only used during construction but both are
1301 * useful during debugging so we store them in the struct when debugging.
1304 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1305 STRLEN trie_charcount=0;
1307 SV *re_trie_maxbuff;
1308 GET_RE_DEBUG_FLAGS_DECL;
1310 PERL_UNUSED_ARG(depth);
1313 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1315 trie->startstate = 1;
1316 trie->wordcount = word_count;
1317 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1318 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1319 if (!(UTF && folder))
1320 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1322 trie_words = newAV();
1325 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1326 if (!SvIOK(re_trie_maxbuff)) {
1327 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1330 PerlIO_printf( Perl_debug_log,
1331 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1332 (int)depth * 2 + 2, "",
1333 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1334 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1338 /* Find the node we are going to overwrite */
1339 if ( first == startbranch && OP( last ) != BRANCH ) {
1340 /* whole branch chain */
1343 /* branch sub-chain */
1344 convert = NEXTOPER( first );
1347 /* -- First loop and Setup --
1349 We first traverse the branches and scan each word to determine if it
1350 contains widechars, and how many unique chars there are, this is
1351 important as we have to build a table with at least as many columns as we
1354 We use an array of integers to represent the character codes 0..255
1355 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1356 native representation of the character value as the key and IV's for the
1359 *TODO* If we keep track of how many times each character is used we can
1360 remap the columns so that the table compression later on is more
1361 efficient in terms of memory by ensuring most common value is in the
1362 middle and the least common are on the outside. IMO this would be better
1363 than a most to least common mapping as theres a decent chance the most
1364 common letter will share a node with the least common, meaning the node
1365 will not be compressable. With a middle is most common approach the worst
1366 case is when we have the least common nodes twice.
1370 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1371 regnode * const noper = NEXTOPER( cur );
1372 const U8 *uc = (U8*)STRING( noper );
1373 const U8 * const e = uc + STR_LEN( noper );
1375 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1376 const U8 *scan = (U8*)NULL;
1377 U32 wordlen = 0; /* required init */
1380 if (OP(noper) == NOTHING) {
1385 TRIE_BITMAP_SET(trie,*uc);
1386 if ( folder ) TRIE_BITMAP_SET(trie,folder[ *uc ]);
1388 for ( ; uc < e ; uc += len ) {
1389 TRIE_CHARCOUNT(trie)++;
1393 if ( !trie->charmap[ uvc ] ) {
1394 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1396 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1402 widecharmap = newHV();
1404 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1407 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1409 if ( !SvTRUE( *svpp ) ) {
1410 sv_setiv( *svpp, ++trie->uniquecharcount );
1415 if( cur == first ) {
1418 } else if (chars < trie->minlen) {
1420 } else if (chars > trie->maxlen) {
1424 } /* end first pass */
1425 DEBUG_TRIE_COMPILE_r(
1426 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1427 (int)depth * 2 + 2,"",
1428 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1429 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1430 (int)trie->minlen, (int)trie->maxlen )
1432 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1435 We now know what we are dealing with in terms of unique chars and
1436 string sizes so we can calculate how much memory a naive
1437 representation using a flat table will take. If it's over a reasonable
1438 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1439 conservative but potentially much slower representation using an array
1442 At the end we convert both representations into the same compressed
1443 form that will be used in regexec.c for matching with. The latter
1444 is a form that cannot be used to construct with but has memory
1445 properties similar to the list form and access properties similar
1446 to the table form making it both suitable for fast searches and
1447 small enough that its feasable to store for the duration of a program.
1449 See the comment in the code where the compressed table is produced
1450 inplace from the flat tabe representation for an explanation of how
1451 the compression works.
1456 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1458 Second Pass -- Array Of Lists Representation
1460 Each state will be represented by a list of charid:state records
1461 (reg_trie_trans_le) the first such element holds the CUR and LEN
1462 points of the allocated array. (See defines above).
1464 We build the initial structure using the lists, and then convert
1465 it into the compressed table form which allows faster lookups
1466 (but cant be modified once converted).
1469 STRLEN transcount = 1;
1471 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1472 "%*sCompiling trie using list compiler\n",
1473 (int)depth * 2 + 2, ""));
1475 trie->states = (reg_trie_state *)
1476 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1477 sizeof(reg_trie_state) );
1481 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1483 regnode * const noper = NEXTOPER( cur );
1484 U8 *uc = (U8*)STRING( noper );
1485 const U8 * const e = uc + STR_LEN( noper );
1486 U32 state = 1; /* required init */
1487 U16 charid = 0; /* sanity init */
1488 U8 *scan = (U8*)NULL; /* sanity init */
1489 STRLEN foldlen = 0; /* required init */
1490 U32 wordlen = 0; /* required init */
1491 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1493 if (OP(noper) != NOTHING) {
1494 for ( ; uc < e ; uc += len ) {
1499 charid = trie->charmap[ uvc ];
1501 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1505 charid=(U16)SvIV( *svpp );
1508 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1515 if ( !trie->states[ state ].trans.list ) {
1516 TRIE_LIST_NEW( state );
1518 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1519 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1520 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1525 newstate = next_alloc++;
1526 TRIE_LIST_PUSH( state, charid, newstate );
1531 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1535 TRIE_HANDLE_WORD(state);
1537 } /* end second pass */
1539 /* next alloc is the NEXT state to be allocated */
1540 trie->statecount = next_alloc;
1541 trie->states = (reg_trie_state *)
1542 PerlMemShared_realloc( trie->states,
1544 * sizeof(reg_trie_state) );
1546 /* and now dump it out before we compress it */
1547 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1548 revcharmap, next_alloc,
1552 trie->trans = (reg_trie_trans *)
1553 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1560 for( state=1 ; state < next_alloc ; state ++ ) {
1564 DEBUG_TRIE_COMPILE_MORE_r(
1565 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1569 if (trie->states[state].trans.list) {
1570 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1574 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1575 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1576 if ( forid < minid ) {
1578 } else if ( forid > maxid ) {
1582 if ( transcount < tp + maxid - minid + 1) {
1584 trie->trans = (reg_trie_trans *)
1585 PerlMemShared_realloc( trie->trans,
1587 * sizeof(reg_trie_trans) );
1588 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1590 base = trie->uniquecharcount + tp - minid;
1591 if ( maxid == minid ) {
1593 for ( ; zp < tp ; zp++ ) {
1594 if ( ! trie->trans[ zp ].next ) {
1595 base = trie->uniquecharcount + zp - minid;
1596 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1597 trie->trans[ zp ].check = state;
1603 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1604 trie->trans[ tp ].check = state;
1609 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1610 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1611 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1612 trie->trans[ tid ].check = state;
1614 tp += ( maxid - minid + 1 );
1616 Safefree(trie->states[ state ].trans.list);
1619 DEBUG_TRIE_COMPILE_MORE_r(
1620 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1623 trie->states[ state ].trans.base=base;
1625 trie->lasttrans = tp + 1;
1629 Second Pass -- Flat Table Representation.
1631 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1632 We know that we will need Charcount+1 trans at most to store the data
1633 (one row per char at worst case) So we preallocate both structures
1634 assuming worst case.
1636 We then construct the trie using only the .next slots of the entry
1639 We use the .check field of the first entry of the node temporarily to
1640 make compression both faster and easier by keeping track of how many non
1641 zero fields are in the node.
1643 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1646 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1647 number representing the first entry of the node, and state as a
1648 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1649 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1650 are 2 entrys per node. eg:
1658 The table is internally in the right hand, idx form. However as we also
1659 have to deal with the states array which is indexed by nodenum we have to
1660 use TRIE_NODENUM() to convert.
1663 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1664 "%*sCompiling trie using table compiler\n",
1665 (int)depth * 2 + 2, ""));
1667 trie->trans = (reg_trie_trans *)
1668 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1669 * trie->uniquecharcount + 1,
1670 sizeof(reg_trie_trans) );
1671 trie->states = (reg_trie_state *)
1672 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1673 sizeof(reg_trie_state) );
1674 next_alloc = trie->uniquecharcount + 1;
1677 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1679 regnode * const noper = NEXTOPER( cur );
1680 const U8 *uc = (U8*)STRING( noper );
1681 const U8 * const e = uc + STR_LEN( noper );
1683 U32 state = 1; /* required init */
1685 U16 charid = 0; /* sanity init */
1686 U32 accept_state = 0; /* sanity init */
1687 U8 *scan = (U8*)NULL; /* sanity init */
1689 STRLEN foldlen = 0; /* required init */
1690 U32 wordlen = 0; /* required init */
1691 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1693 if ( OP(noper) != NOTHING ) {
1694 for ( ; uc < e ; uc += len ) {
1699 charid = trie->charmap[ uvc ];
1701 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1702 charid = svpp ? (U16)SvIV(*svpp) : 0;
1706 if ( !trie->trans[ state + charid ].next ) {
1707 trie->trans[ state + charid ].next = next_alloc;
1708 trie->trans[ state ].check++;
1709 next_alloc += trie->uniquecharcount;
1711 state = trie->trans[ state + charid ].next;
1713 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1715 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1718 accept_state = TRIE_NODENUM( state );
1719 TRIE_HANDLE_WORD(accept_state);
1721 } /* end second pass */
1723 /* and now dump it out before we compress it */
1724 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1726 next_alloc, depth+1));
1730 * Inplace compress the table.*
1732 For sparse data sets the table constructed by the trie algorithm will
1733 be mostly 0/FAIL transitions or to put it another way mostly empty.
1734 (Note that leaf nodes will not contain any transitions.)
1736 This algorithm compresses the tables by eliminating most such
1737 transitions, at the cost of a modest bit of extra work during lookup:
1739 - Each states[] entry contains a .base field which indicates the
1740 index in the state[] array wheres its transition data is stored.
1742 - If .base is 0 there are no valid transitions from that node.
1744 - If .base is nonzero then charid is added to it to find an entry in
1747 -If trans[states[state].base+charid].check!=state then the
1748 transition is taken to be a 0/Fail transition. Thus if there are fail
1749 transitions at the front of the node then the .base offset will point
1750 somewhere inside the previous nodes data (or maybe even into a node
1751 even earlier), but the .check field determines if the transition is
1755 The following process inplace converts the table to the compressed
1756 table: We first do not compress the root node 1,and mark its all its
1757 .check pointers as 1 and set its .base pointer as 1 as well. This
1758 allows to do a DFA construction from the compressed table later, and
1759 ensures that any .base pointers we calculate later are greater than
1762 - We set 'pos' to indicate the first entry of the second node.
1764 - We then iterate over the columns of the node, finding the first and
1765 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1766 and set the .check pointers accordingly, and advance pos
1767 appropriately and repreat for the next node. Note that when we copy
1768 the next pointers we have to convert them from the original
1769 NODEIDX form to NODENUM form as the former is not valid post
1772 - If a node has no transitions used we mark its base as 0 and do not
1773 advance the pos pointer.
1775 - If a node only has one transition we use a second pointer into the
1776 structure to fill in allocated fail transitions from other states.
1777 This pointer is independent of the main pointer and scans forward
1778 looking for null transitions that are allocated to a state. When it
1779 finds one it writes the single transition into the "hole". If the
1780 pointer doesnt find one the single transition is appended as normal.
1782 - Once compressed we can Renew/realloc the structures to release the
1785 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1786 specifically Fig 3.47 and the associated pseudocode.
1790 const U32 laststate = TRIE_NODENUM( next_alloc );
1793 trie->statecount = laststate;
1795 for ( state = 1 ; state < laststate ; state++ ) {
1797 const U32 stateidx = TRIE_NODEIDX( state );
1798 const U32 o_used = trie->trans[ stateidx ].check;
1799 U32 used = trie->trans[ stateidx ].check;
1800 trie->trans[ stateidx ].check = 0;
1802 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1803 if ( flag || trie->trans[ stateidx + charid ].next ) {
1804 if ( trie->trans[ stateidx + charid ].next ) {
1806 for ( ; zp < pos ; zp++ ) {
1807 if ( ! trie->trans[ zp ].next ) {
1811 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1812 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1813 trie->trans[ zp ].check = state;
1814 if ( ++zp > pos ) pos = zp;
1821 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1823 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1824 trie->trans[ pos ].check = state;
1829 trie->lasttrans = pos + 1;
1830 trie->states = (reg_trie_state *)
1831 PerlMemShared_realloc( trie->states, laststate
1832 * sizeof(reg_trie_state) );
1833 DEBUG_TRIE_COMPILE_MORE_r(
1834 PerlIO_printf( Perl_debug_log,
1835 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1836 (int)depth * 2 + 2,"",
1837 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1840 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1843 } /* end table compress */
1845 DEBUG_TRIE_COMPILE_MORE_r(
1846 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1847 (int)depth * 2 + 2, "",
1848 (UV)trie->statecount,
1849 (UV)trie->lasttrans)
1851 /* resize the trans array to remove unused space */
1852 trie->trans = (reg_trie_trans *)
1853 PerlMemShared_realloc( trie->trans, trie->lasttrans
1854 * sizeof(reg_trie_trans) );
1856 /* and now dump out the compressed format */
1857 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1859 { /* Modify the program and insert the new TRIE node*/
1860 U8 nodetype =(U8)(flags & 0xFF);
1864 regnode *optimize = NULL;
1865 #ifdef RE_TRACK_PATTERN_OFFSETS
1868 U32 mjd_nodelen = 0;
1869 #endif /* RE_TRACK_PATTERN_OFFSETS */
1870 #endif /* DEBUGGING */
1872 This means we convert either the first branch or the first Exact,
1873 depending on whether the thing following (in 'last') is a branch
1874 or not and whther first is the startbranch (ie is it a sub part of
1875 the alternation or is it the whole thing.)
1876 Assuming its a sub part we conver the EXACT otherwise we convert
1877 the whole branch sequence, including the first.
1879 /* Find the node we are going to overwrite */
1880 if ( first != startbranch || OP( last ) == BRANCH ) {
1881 /* branch sub-chain */
1882 NEXT_OFF( first ) = (U16)(last - first);
1883 #ifdef RE_TRACK_PATTERN_OFFSETS
1885 mjd_offset= Node_Offset((convert));
1886 mjd_nodelen= Node_Length((convert));
1889 /* whole branch chain */
1891 #ifdef RE_TRACK_PATTERN_OFFSETS
1894 const regnode *nop = NEXTOPER( convert );
1895 mjd_offset= Node_Offset((nop));
1896 mjd_nodelen= Node_Length((nop));
1900 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1901 (int)depth * 2 + 2, "",
1902 (UV)mjd_offset, (UV)mjd_nodelen)
1905 /* But first we check to see if there is a common prefix we can
1906 split out as an EXACT and put in front of the TRIE node. */
1907 trie->startstate= 1;
1908 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1910 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1914 const U32 base = trie->states[ state ].trans.base;
1916 if ( trie->states[state].wordnum )
1919 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1920 if ( ( base + ofs >= trie->uniquecharcount ) &&
1921 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1922 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1924 if ( ++count > 1 ) {
1925 SV **tmp = av_fetch( revcharmap, ofs, 0);
1926 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1927 if ( state == 1 ) break;
1929 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1931 PerlIO_printf(Perl_debug_log,
1932 "%*sNew Start State=%"UVuf" Class: [",
1933 (int)depth * 2 + 2, "",
1936 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1937 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1939 TRIE_BITMAP_SET(trie,*ch);
1941 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1943 PerlIO_printf(Perl_debug_log, (char*)ch)
1947 TRIE_BITMAP_SET(trie,*ch);
1949 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1950 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1956 SV **tmp = av_fetch( revcharmap, idx, 0);
1957 char *ch = SvPV_nolen( *tmp );
1959 SV *sv=sv_newmortal();
1960 PerlIO_printf( Perl_debug_log,
1961 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1962 (int)depth * 2 + 2, "",
1964 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1965 PL_colors[0], PL_colors[1],
1966 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1967 PERL_PV_ESCAPE_FIRSTCHAR
1972 OP( convert ) = nodetype;
1973 str=STRING(convert);
1984 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
1990 regnode *n = convert+NODE_SZ_STR(convert);
1991 NEXT_OFF(convert) = NODE_SZ_STR(convert);
1992 trie->startstate = state;
1993 trie->minlen -= (state - 1);
1994 trie->maxlen -= (state - 1);
1996 regnode *fix = convert;
1997 U32 word = trie->wordcount;
1999 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2000 while( ++fix < n ) {
2001 Set_Node_Offset_Length(fix, 0, 0);
2004 SV ** const tmp = av_fetch( trie_words, word, 0 );
2006 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2007 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2009 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2016 NEXT_OFF(convert) = (U16)(tail - convert);
2017 DEBUG_r(optimize= n);
2023 if ( trie->maxlen ) {
2024 NEXT_OFF( convert ) = (U16)(tail - convert);
2025 ARG_SET( convert, data_slot );
2026 /* Store the offset to the first unabsorbed branch in
2027 jump[0], which is otherwise unused by the jump logic.
2028 We use this when dumping a trie and during optimisation. */
2030 trie->jump[0] = (U16)(nextbranch - convert);
2033 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2034 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2036 OP( convert ) = TRIEC;
2037 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2038 PerlMemShared_free(trie->bitmap);
2041 OP( convert ) = TRIE;
2043 /* store the type in the flags */
2044 convert->flags = nodetype;
2048 + regarglen[ OP( convert ) ];
2050 /* XXX We really should free up the resource in trie now,
2051 as we won't use them - (which resources?) dmq */
2053 /* needed for dumping*/
2054 DEBUG_r(if (optimize) {
2055 regnode *opt = convert;
2057 while ( ++opt < optimize) {
2058 Set_Node_Offset_Length(opt,0,0);
2061 Try to clean up some of the debris left after the
2064 while( optimize < jumper ) {
2065 mjd_nodelen += Node_Length((optimize));
2066 OP( optimize ) = OPTIMIZED;
2067 Set_Node_Offset_Length(optimize,0,0);
2070 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2072 } /* end node insert */
2073 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2075 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2076 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2078 SvREFCNT_dec(revcharmap);
2082 : trie->startstate>1
2088 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2090 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2092 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2093 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2096 We find the fail state for each state in the trie, this state is the longest proper
2097 suffix of the current states 'word' that is also a proper prefix of another word in our
2098 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2099 the DFA not to have to restart after its tried and failed a word at a given point, it
2100 simply continues as though it had been matching the other word in the first place.
2102 'abcdgu'=~/abcdefg|cdgu/
2103 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2104 fail, which would bring use to the state representing 'd' in the second word where we would
2105 try 'g' and succeed, prodceding to match 'cdgu'.
2107 /* add a fail transition */
2108 const U32 trie_offset = ARG(source);
2109 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2111 const U32 ucharcount = trie->uniquecharcount;
2112 const U32 numstates = trie->statecount;
2113 const U32 ubound = trie->lasttrans + ucharcount;
2117 U32 base = trie->states[ 1 ].trans.base;
2120 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2121 GET_RE_DEBUG_FLAGS_DECL;
2123 PERL_UNUSED_ARG(depth);
2127 ARG_SET( stclass, data_slot );
2128 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2129 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2130 aho->trie=trie_offset;
2131 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2132 Copy( trie->states, aho->states, numstates, reg_trie_state );
2133 Newxz( q, numstates, U32);
2134 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2137 /* initialize fail[0..1] to be 1 so that we always have
2138 a valid final fail state */
2139 fail[ 0 ] = fail[ 1 ] = 1;
2141 for ( charid = 0; charid < ucharcount ; charid++ ) {
2142 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2144 q[ q_write ] = newstate;
2145 /* set to point at the root */
2146 fail[ q[ q_write++ ] ]=1;
2149 while ( q_read < q_write) {
2150 const U32 cur = q[ q_read++ % numstates ];
2151 base = trie->states[ cur ].trans.base;
2153 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2154 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2156 U32 fail_state = cur;
2159 fail_state = fail[ fail_state ];
2160 fail_base = aho->states[ fail_state ].trans.base;
2161 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2163 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2164 fail[ ch_state ] = fail_state;
2165 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2167 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2169 q[ q_write++ % numstates] = ch_state;
2173 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2174 when we fail in state 1, this allows us to use the
2175 charclass scan to find a valid start char. This is based on the principle
2176 that theres a good chance the string being searched contains lots of stuff
2177 that cant be a start char.
2179 fail[ 0 ] = fail[ 1 ] = 0;
2180 DEBUG_TRIE_COMPILE_r({
2181 PerlIO_printf(Perl_debug_log,
2182 "%*sStclass Failtable (%"UVuf" states): 0",
2183 (int)(depth * 2), "", (UV)numstates
2185 for( q_read=1; q_read<numstates; q_read++ ) {
2186 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2188 PerlIO_printf(Perl_debug_log, "\n");
2191 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2196 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2197 * These need to be revisited when a newer toolchain becomes available.
2199 #if defined(__sparc64__) && defined(__GNUC__)
2200 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2201 # undef SPARC64_GCC_WORKAROUND
2202 # define SPARC64_GCC_WORKAROUND 1
2206 #define DEBUG_PEEP(str,scan,depth) \
2207 DEBUG_OPTIMISE_r({if (scan){ \
2208 SV * const mysv=sv_newmortal(); \
2209 regnode *Next = regnext(scan); \
2210 regprop(RExC_rx, mysv, scan); \
2211 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2212 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2213 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2220 #define JOIN_EXACT(scan,min,flags) \
2221 if (PL_regkind[OP(scan)] == EXACT) \
2222 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2225 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2226 /* Merge several consecutive EXACTish nodes into one. */
2227 regnode *n = regnext(scan);
2229 regnode *next = scan + NODE_SZ_STR(scan);
2233 regnode *stop = scan;
2234 GET_RE_DEBUG_FLAGS_DECL;
2236 PERL_UNUSED_ARG(depth);
2238 #ifndef EXPERIMENTAL_INPLACESCAN
2239 PERL_UNUSED_ARG(flags);
2240 PERL_UNUSED_ARG(val);
2242 DEBUG_PEEP("join",scan,depth);
2244 /* Skip NOTHING, merge EXACT*. */
2246 ( PL_regkind[OP(n)] == NOTHING ||
2247 (stringok && (OP(n) == OP(scan))))
2249 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2251 if (OP(n) == TAIL || n > next)
2253 if (PL_regkind[OP(n)] == NOTHING) {
2254 DEBUG_PEEP("skip:",n,depth);
2255 NEXT_OFF(scan) += NEXT_OFF(n);
2256 next = n + NODE_STEP_REGNODE;
2263 else if (stringok) {
2264 const unsigned int oldl = STR_LEN(scan);
2265 regnode * const nnext = regnext(n);
2267 DEBUG_PEEP("merg",n,depth);
2270 if (oldl + STR_LEN(n) > U8_MAX)
2272 NEXT_OFF(scan) += NEXT_OFF(n);
2273 STR_LEN(scan) += STR_LEN(n);
2274 next = n + NODE_SZ_STR(n);
2275 /* Now we can overwrite *n : */
2276 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2284 #ifdef EXPERIMENTAL_INPLACESCAN
2285 if (flags && !NEXT_OFF(n)) {
2286 DEBUG_PEEP("atch", val, depth);
2287 if (reg_off_by_arg[OP(n)]) {
2288 ARG_SET(n, val - n);
2291 NEXT_OFF(n) = val - n;
2298 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2300 Two problematic code points in Unicode casefolding of EXACT nodes:
2302 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2303 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2309 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2310 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2312 This means that in case-insensitive matching (or "loose matching",
2313 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2314 length of the above casefolded versions) can match a target string
2315 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2316 This would rather mess up the minimum length computation.
2318 What we'll do is to look for the tail four bytes, and then peek
2319 at the preceding two bytes to see whether we need to decrease
2320 the minimum length by four (six minus two).
2322 Thanks to the design of UTF-8, there cannot be false matches:
2323 A sequence of valid UTF-8 bytes cannot be a subsequence of
2324 another valid sequence of UTF-8 bytes.
2327 char * const s0 = STRING(scan), *s, *t;
2328 char * const s1 = s0 + STR_LEN(scan) - 1;
2329 char * const s2 = s1 - 4;
2330 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2331 const char t0[] = "\xaf\x49\xaf\x42";
2333 const char t0[] = "\xcc\x88\xcc\x81";
2335 const char * const t1 = t0 + 3;
2338 s < s2 && (t = ninstr(s, s1, t0, t1));
2341 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2342 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2344 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2345 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2353 n = scan + NODE_SZ_STR(scan);
2355 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2362 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2366 /* REx optimizer. Converts nodes into quickier variants "in place".
2367 Finds fixed substrings. */
2369 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2370 to the position after last scanned or to NULL. */
2372 #define INIT_AND_WITHP \
2373 assert(!and_withp); \
2374 Newx(and_withp,1,struct regnode_charclass_class); \
2375 SAVEFREEPV(and_withp)
2377 /* this is a chain of data about sub patterns we are processing that
2378 need to be handled seperately/specially in study_chunk. Its so
2379 we can simulate recursion without losing state. */
2381 typedef struct scan_frame {
2382 regnode *last; /* last node to process in this frame */
2383 regnode *next; /* next node to process when last is reached */
2384 struct scan_frame *prev; /*previous frame*/
2385 I32 stop; /* what stopparen do we use */
2389 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2392 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2393 I32 *minlenp, I32 *deltap,
2398 struct regnode_charclass_class *and_withp,
2399 U32 flags, U32 depth)
2400 /* scanp: Start here (read-write). */
2401 /* deltap: Write maxlen-minlen here. */
2402 /* last: Stop before this one. */
2403 /* data: string data about the pattern */
2404 /* stopparen: treat close N as END */
2405 /* recursed: which subroutines have we recursed into */
2406 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2409 I32 min = 0, pars = 0, code;
2410 regnode *scan = *scanp, *next;
2412 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2413 int is_inf_internal = 0; /* The studied chunk is infinite */
2414 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2415 scan_data_t data_fake;
2416 SV *re_trie_maxbuff = NULL;
2417 regnode *first_non_open = scan;
2418 I32 stopmin = I32_MAX;
2419 scan_frame *frame = NULL;
2421 GET_RE_DEBUG_FLAGS_DECL;
2424 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2428 while (first_non_open && OP(first_non_open) == OPEN)
2429 first_non_open=regnext(first_non_open);
2434 while ( scan && OP(scan) != END && scan < last ){
2435 /* Peephole optimizer: */
2436 DEBUG_STUDYDATA("Peep:", data,depth);
2437 DEBUG_PEEP("Peep",scan,depth);
2438 JOIN_EXACT(scan,&min,0);
2440 /* Follow the next-chain of the current node and optimize
2441 away all the NOTHINGs from it. */
2442 if (OP(scan) != CURLYX) {
2443 const int max = (reg_off_by_arg[OP(scan)]
2445 /* I32 may be smaller than U16 on CRAYs! */
2446 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2447 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2451 /* Skip NOTHING and LONGJMP. */
2452 while ((n = regnext(n))
2453 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2454 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2455 && off + noff < max)
2457 if (reg_off_by_arg[OP(scan)])
2460 NEXT_OFF(scan) = off;
2465 /* The principal pseudo-switch. Cannot be a switch, since we
2466 look into several different things. */
2467 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2468 || OP(scan) == IFTHEN) {
2469 next = regnext(scan);
2471 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2473 if (OP(next) == code || code == IFTHEN) {
2474 /* NOTE - There is similar code to this block below for handling
2475 TRIE nodes on a re-study. If you change stuff here check there
2477 I32 max1 = 0, min1 = I32_MAX, num = 0;
2478 struct regnode_charclass_class accum;
2479 regnode * const startbranch=scan;
2481 if (flags & SCF_DO_SUBSTR)
2482 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2483 if (flags & SCF_DO_STCLASS)
2484 cl_init_zero(pRExC_state, &accum);
2486 while (OP(scan) == code) {
2487 I32 deltanext, minnext, f = 0, fake;
2488 struct regnode_charclass_class this_class;
2491 data_fake.flags = 0;
2493 data_fake.whilem_c = data->whilem_c;
2494 data_fake.last_closep = data->last_closep;
2497 data_fake.last_closep = &fake;
2499 data_fake.pos_delta = delta;
2500 next = regnext(scan);
2501 scan = NEXTOPER(scan);
2503 scan = NEXTOPER(scan);
2504 if (flags & SCF_DO_STCLASS) {
2505 cl_init(pRExC_state, &this_class);
2506 data_fake.start_class = &this_class;
2507 f = SCF_DO_STCLASS_AND;
2509 if (flags & SCF_WHILEM_VISITED_POS)
2510 f |= SCF_WHILEM_VISITED_POS;
2512 /* we suppose the run is continuous, last=next...*/
2513 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2515 stopparen, recursed, NULL, f,depth+1);
2518 if (max1 < minnext + deltanext)
2519 max1 = minnext + deltanext;
2520 if (deltanext == I32_MAX)
2521 is_inf = is_inf_internal = 1;
2523 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2525 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2526 if ( stopmin > minnext)
2527 stopmin = min + min1;
2528 flags &= ~SCF_DO_SUBSTR;
2530 data->flags |= SCF_SEEN_ACCEPT;
2533 if (data_fake.flags & SF_HAS_EVAL)
2534 data->flags |= SF_HAS_EVAL;
2535 data->whilem_c = data_fake.whilem_c;
2537 if (flags & SCF_DO_STCLASS)
2538 cl_or(pRExC_state, &accum, &this_class);
2540 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2542 if (flags & SCF_DO_SUBSTR) {
2543 data->pos_min += min1;
2544 data->pos_delta += max1 - min1;
2545 if (max1 != min1 || is_inf)
2546 data->longest = &(data->longest_float);
2549 delta += max1 - min1;
2550 if (flags & SCF_DO_STCLASS_OR) {
2551 cl_or(pRExC_state, data->start_class, &accum);
2553 cl_and(data->start_class, and_withp);
2554 flags &= ~SCF_DO_STCLASS;
2557 else if (flags & SCF_DO_STCLASS_AND) {
2559 cl_and(data->start_class, &accum);
2560 flags &= ~SCF_DO_STCLASS;
2563 /* Switch to OR mode: cache the old value of
2564 * data->start_class */
2566 StructCopy(data->start_class, and_withp,
2567 struct regnode_charclass_class);
2568 flags &= ~SCF_DO_STCLASS_AND;
2569 StructCopy(&accum, data->start_class,
2570 struct regnode_charclass_class);
2571 flags |= SCF_DO_STCLASS_OR;
2572 data->start_class->flags |= ANYOF_EOS;
2576 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2579 Assuming this was/is a branch we are dealing with: 'scan' now
2580 points at the item that follows the branch sequence, whatever
2581 it is. We now start at the beginning of the sequence and look
2588 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2590 If we can find such a subseqence we need to turn the first
2591 element into a trie and then add the subsequent branch exact
2592 strings to the trie.
2596 1. patterns where the whole set of branch can be converted.
2598 2. patterns where only a subset can be converted.
2600 In case 1 we can replace the whole set with a single regop
2601 for the trie. In case 2 we need to keep the start and end
2604 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2605 becomes BRANCH TRIE; BRANCH X;
2607 There is an additional case, that being where there is a
2608 common prefix, which gets split out into an EXACT like node
2609 preceding the TRIE node.
2611 If x(1..n)==tail then we can do a simple trie, if not we make
2612 a "jump" trie, such that when we match the appropriate word
2613 we "jump" to the appopriate tail node. Essentailly we turn
2614 a nested if into a case structure of sorts.
2619 if (!re_trie_maxbuff) {
2620 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2621 if (!SvIOK(re_trie_maxbuff))
2622 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2624 if ( SvIV(re_trie_maxbuff)>=0 ) {
2626 regnode *first = (regnode *)NULL;
2627 regnode *last = (regnode *)NULL;
2628 regnode *tail = scan;
2633 SV * const mysv = sv_newmortal(); /* for dumping */
2635 /* var tail is used because there may be a TAIL
2636 regop in the way. Ie, the exacts will point to the
2637 thing following the TAIL, but the last branch will
2638 point at the TAIL. So we advance tail. If we
2639 have nested (?:) we may have to move through several
2643 while ( OP( tail ) == TAIL ) {
2644 /* this is the TAIL generated by (?:) */
2645 tail = regnext( tail );
2650 regprop(RExC_rx, mysv, tail );
2651 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2652 (int)depth * 2 + 2, "",
2653 "Looking for TRIE'able sequences. Tail node is: ",
2654 SvPV_nolen_const( mysv )
2660 step through the branches, cur represents each
2661 branch, noper is the first thing to be matched
2662 as part of that branch and noper_next is the
2663 regnext() of that node. if noper is an EXACT
2664 and noper_next is the same as scan (our current
2665 position in the regex) then the EXACT branch is
2666 a possible optimization target. Once we have
2667 two or more consequetive such branches we can
2668 create a trie of the EXACT's contents and stich
2669 it in place. If the sequence represents all of
2670 the branches we eliminate the whole thing and
2671 replace it with a single TRIE. If it is a
2672 subsequence then we need to stitch it in. This
2673 means the first branch has to remain, and needs
2674 to be repointed at the item on the branch chain
2675 following the last branch optimized. This could
2676 be either a BRANCH, in which case the
2677 subsequence is internal, or it could be the
2678 item following the branch sequence in which
2679 case the subsequence is at the end.
2683 /* dont use tail as the end marker for this traverse */
2684 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2685 regnode * const noper = NEXTOPER( cur );
2686 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2687 regnode * const noper_next = regnext( noper );
2691 regprop(RExC_rx, mysv, cur);
2692 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2693 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2695 regprop(RExC_rx, mysv, noper);
2696 PerlIO_printf( Perl_debug_log, " -> %s",
2697 SvPV_nolen_const(mysv));
2700 regprop(RExC_rx, mysv, noper_next );
2701 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2702 SvPV_nolen_const(mysv));
2704 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2705 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2707 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2708 : PL_regkind[ OP( noper ) ] == EXACT )
2709 || OP(noper) == NOTHING )
2711 && noper_next == tail
2716 if ( !first || optype == NOTHING ) {
2717 if (!first) first = cur;
2718 optype = OP( noper );
2724 make_trie( pRExC_state,
2725 startbranch, first, cur, tail, count,
2728 if ( PL_regkind[ OP( noper ) ] == EXACT
2730 && noper_next == tail
2735 optype = OP( noper );
2745 regprop(RExC_rx, mysv, cur);
2746 PerlIO_printf( Perl_debug_log,
2747 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2748 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2752 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2753 #ifdef TRIE_STUDY_OPT
2754 if ( ((made == MADE_EXACT_TRIE &&
2755 startbranch == first)
2756 || ( first_non_open == first )) &&
2758 flags |= SCF_TRIE_RESTUDY;
2759 if ( startbranch == first
2762 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2772 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2773 scan = NEXTOPER(NEXTOPER(scan));
2774 } else /* single branch is optimized. */
2775 scan = NEXTOPER(scan);
2777 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2778 scan_frame *newframe = NULL;
2783 if (OP(scan) != SUSPEND) {
2784 /* set the pointer */
2785 if (OP(scan) == GOSUB) {
2787 RExC_recurse[ARG2L(scan)] = scan;
2788 start = RExC_open_parens[paren-1];
2789 end = RExC_close_parens[paren-1];
2792 start = RExC_rxi->program + 1;
2796 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2797 SAVEFREEPV(recursed);
2799 if (!PAREN_TEST(recursed,paren+1)) {
2800 PAREN_SET(recursed,paren+1);
2801 Newx(newframe,1,scan_frame);
2803 if (flags & SCF_DO_SUBSTR) {
2804 SCAN_COMMIT(pRExC_state,data,minlenp);
2805 data->longest = &(data->longest_float);
2807 is_inf = is_inf_internal = 1;
2808 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2809 cl_anything(pRExC_state, data->start_class);
2810 flags &= ~SCF_DO_STCLASS;
2813 Newx(newframe,1,scan_frame);
2816 end = regnext(scan);
2821 SAVEFREEPV(newframe);
2822 newframe->next = regnext(scan);
2823 newframe->last = last;
2824 newframe->stop = stopparen;
2825 newframe->prev = frame;
2835 else if (OP(scan) == EXACT) {
2836 I32 l = STR_LEN(scan);
2839 const U8 * const s = (U8*)STRING(scan);
2840 l = utf8_length(s, s + l);
2841 uc = utf8_to_uvchr(s, NULL);
2843 uc = *((U8*)STRING(scan));
2846 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2847 /* The code below prefers earlier match for fixed
2848 offset, later match for variable offset. */
2849 if (data->last_end == -1) { /* Update the start info. */
2850 data->last_start_min = data->pos_min;
2851 data->last_start_max = is_inf
2852 ? I32_MAX : data->pos_min + data->pos_delta;
2854 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2856 SvUTF8_on(data->last_found);
2858 SV * const sv = data->last_found;
2859 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2860 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2861 if (mg && mg->mg_len >= 0)
2862 mg->mg_len += utf8_length((U8*)STRING(scan),
2863 (U8*)STRING(scan)+STR_LEN(scan));
2865 data->last_end = data->pos_min + l;
2866 data->pos_min += l; /* As in the first entry. */
2867 data->flags &= ~SF_BEFORE_EOL;
2869 if (flags & SCF_DO_STCLASS_AND) {
2870 /* Check whether it is compatible with what we know already! */
2874 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2875 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2876 && (!(data->start_class->flags & ANYOF_FOLD)
2877 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2880 ANYOF_CLASS_ZERO(data->start_class);
2881 ANYOF_BITMAP_ZERO(data->start_class);
2883 ANYOF_BITMAP_SET(data->start_class, uc);
2884 data->start_class->flags &= ~ANYOF_EOS;
2886 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2888 else if (flags & SCF_DO_STCLASS_OR) {
2889 /* false positive possible if the class is case-folded */
2891 ANYOF_BITMAP_SET(data->start_class, uc);
2893 data->start_class->flags |= ANYOF_UNICODE_ALL;
2894 data->start_class->flags &= ~ANYOF_EOS;
2895 cl_and(data->start_class, and_withp);
2897 flags &= ~SCF_DO_STCLASS;
2899 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2900 I32 l = STR_LEN(scan);
2901 UV uc = *((U8*)STRING(scan));
2903 /* Search for fixed substrings supports EXACT only. */
2904 if (flags & SCF_DO_SUBSTR) {
2906 SCAN_COMMIT(pRExC_state, data, minlenp);
2909 const U8 * const s = (U8 *)STRING(scan);
2910 l = utf8_length(s, s + l);
2911 uc = utf8_to_uvchr(s, NULL);
2914 if (flags & SCF_DO_SUBSTR)
2916 if (flags & SCF_DO_STCLASS_AND) {
2917 /* Check whether it is compatible with what we know already! */
2921 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2922 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2923 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2925 ANYOF_CLASS_ZERO(data->start_class);
2926 ANYOF_BITMAP_ZERO(data->start_class);
2928 ANYOF_BITMAP_SET(data->start_class, uc);
2929 data->start_class->flags &= ~ANYOF_EOS;
2930 data->start_class->flags |= ANYOF_FOLD;
2931 if (OP(scan) == EXACTFL)
2932 data->start_class->flags |= ANYOF_LOCALE;
2935 else if (flags & SCF_DO_STCLASS_OR) {
2936 if (data->start_class->flags & ANYOF_FOLD) {
2937 /* false positive possible if the class is case-folded.
2938 Assume that the locale settings are the same... */
2940 ANYOF_BITMAP_SET(data->start_class, uc);
2941 data->start_class->flags &= ~ANYOF_EOS;
2943 cl_and(data->start_class, and_withp);
2945 flags &= ~SCF_DO_STCLASS;
2947 else if (strchr((const char*)PL_varies,OP(scan))) {
2948 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2949 I32 f = flags, pos_before = 0;
2950 regnode * const oscan = scan;
2951 struct regnode_charclass_class this_class;
2952 struct regnode_charclass_class *oclass = NULL;
2953 I32 next_is_eval = 0;
2955 switch (PL_regkind[OP(scan)]) {
2956 case WHILEM: /* End of (?:...)* . */
2957 scan = NEXTOPER(scan);
2960 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2961 next = NEXTOPER(scan);
2962 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2964 maxcount = REG_INFTY;
2965 next = regnext(scan);
2966 scan = NEXTOPER(scan);
2970 if (flags & SCF_DO_SUBSTR)
2975 if (flags & SCF_DO_STCLASS) {
2977 maxcount = REG_INFTY;
2978 next = regnext(scan);
2979 scan = NEXTOPER(scan);
2982 is_inf = is_inf_internal = 1;
2983 scan = regnext(scan);
2984 if (flags & SCF_DO_SUBSTR) {
2985 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2986 data->longest = &(data->longest_float);
2988 goto optimize_curly_tail;
2990 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
2991 && (scan->flags == stopparen))
2996 mincount = ARG1(scan);
2997 maxcount = ARG2(scan);
2999 next = regnext(scan);
3000 if (OP(scan) == CURLYX) {
3001 I32 lp = (data ? *(data->last_closep) : 0);
3002 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3004 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3005 next_is_eval = (OP(scan) == EVAL);
3007 if (flags & SCF_DO_SUBSTR) {
3008 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3009 pos_before = data->pos_min;
3013 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3015 data->flags |= SF_IS_INF;
3017 if (flags & SCF_DO_STCLASS) {
3018 cl_init(pRExC_state, &this_class);
3019 oclass = data->start_class;
3020 data->start_class = &this_class;
3021 f |= SCF_DO_STCLASS_AND;
3022 f &= ~SCF_DO_STCLASS_OR;
3024 /* These are the cases when once a subexpression
3025 fails at a particular position, it cannot succeed
3026 even after backtracking at the enclosing scope.
3028 XXXX what if minimal match and we are at the
3029 initial run of {n,m}? */
3030 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3031 f &= ~SCF_WHILEM_VISITED_POS;
3033 /* This will finish on WHILEM, setting scan, or on NULL: */
3034 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3035 last, data, stopparen, recursed, NULL,
3037 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3039 if (flags & SCF_DO_STCLASS)
3040 data->start_class = oclass;
3041 if (mincount == 0 || minnext == 0) {
3042 if (flags & SCF_DO_STCLASS_OR) {
3043 cl_or(pRExC_state, data->start_class, &this_class);
3045 else if (flags & SCF_DO_STCLASS_AND) {
3046 /* Switch to OR mode: cache the old value of
3047 * data->start_class */
3049 StructCopy(data->start_class, and_withp,
3050 struct regnode_charclass_class);
3051 flags &= ~SCF_DO_STCLASS_AND;
3052 StructCopy(&this_class, data->start_class,
3053 struct regnode_charclass_class);
3054 flags |= SCF_DO_STCLASS_OR;
3055 data->start_class->flags |= ANYOF_EOS;
3057 } else { /* Non-zero len */
3058 if (flags & SCF_DO_STCLASS_OR) {
3059 cl_or(pRExC_state, data->start_class, &this_class);
3060 cl_and(data->start_class, and_withp);
3062 else if (flags & SCF_DO_STCLASS_AND)
3063 cl_and(data->start_class, &this_class);
3064 flags &= ~SCF_DO_STCLASS;
3066 if (!scan) /* It was not CURLYX, but CURLY. */
3068 if ( /* ? quantifier ok, except for (?{ ... }) */
3069 (next_is_eval || !(mincount == 0 && maxcount == 1))
3070 && (minnext == 0) && (deltanext == 0)
3071 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3072 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3073 && ckWARN(WARN_REGEXP))
3076 "Quantifier unexpected on zero-length expression");
3079 min += minnext * mincount;
3080 is_inf_internal |= ((maxcount == REG_INFTY
3081 && (minnext + deltanext) > 0)
3082 || deltanext == I32_MAX);
3083 is_inf |= is_inf_internal;
3084 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3086 /* Try powerful optimization CURLYX => CURLYN. */
3087 if ( OP(oscan) == CURLYX && data
3088 && data->flags & SF_IN_PAR
3089 && !(data->flags & SF_HAS_EVAL)
3090 && !deltanext && minnext == 1 ) {
3091 /* Try to optimize to CURLYN. */
3092 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3093 regnode * const nxt1 = nxt;
3100 if (!strchr((const char*)PL_simple,OP(nxt))
3101 && !(PL_regkind[OP(nxt)] == EXACT
3102 && STR_LEN(nxt) == 1))
3108 if (OP(nxt) != CLOSE)
3110 if (RExC_open_parens) {
3111 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3112 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3114 /* Now we know that nxt2 is the only contents: */
3115 oscan->flags = (U8)ARG(nxt);
3117 OP(nxt1) = NOTHING; /* was OPEN. */
3120 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3121 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3122 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3123 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3124 OP(nxt + 1) = OPTIMIZED; /* was count. */
3125 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3130 /* Try optimization CURLYX => CURLYM. */
3131 if ( OP(oscan) == CURLYX && data
3132 && !(data->flags & SF_HAS_PAR)
3133 && !(data->flags & SF_HAS_EVAL)
3134 && !deltanext /* atom is fixed width */
3135 && minnext != 0 /* CURLYM can't handle zero width */
3137 /* XXXX How to optimize if data == 0? */
3138 /* Optimize to a simpler form. */
3139 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3143 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3144 && (OP(nxt2) != WHILEM))
3146 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3147 /* Need to optimize away parenths. */
3148 if (data->flags & SF_IN_PAR) {
3149 /* Set the parenth number. */
3150 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3152 if (OP(nxt) != CLOSE)
3153 FAIL("Panic opt close");
3154 oscan->flags = (U8)ARG(nxt);
3155 if (RExC_open_parens) {
3156 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3157 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3159 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3160 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3163 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3164 OP(nxt + 1) = OPTIMIZED; /* was count. */
3165 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3166 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3169 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3170 regnode *nnxt = regnext(nxt1);
3173 if (reg_off_by_arg[OP(nxt1)])
3174 ARG_SET(nxt1, nxt2 - nxt1);
3175 else if (nxt2 - nxt1 < U16_MAX)
3176 NEXT_OFF(nxt1) = nxt2 - nxt1;
3178 OP(nxt) = NOTHING; /* Cannot beautify */
3183 /* Optimize again: */
3184 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3185 NULL, stopparen, recursed, NULL, 0,depth+1);
3190 else if ((OP(oscan) == CURLYX)
3191 && (flags & SCF_WHILEM_VISITED_POS)
3192 /* See the comment on a similar expression above.
3193 However, this time it not a subexpression
3194 we care about, but the expression itself. */
3195 && (maxcount == REG_INFTY)
3196 && data && ++data->whilem_c < 16) {
3197 /* This stays as CURLYX, we can put the count/of pair. */
3198 /* Find WHILEM (as in regexec.c) */
3199 regnode *nxt = oscan + NEXT_OFF(oscan);
3201 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3203 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3204 | (RExC_whilem_seen << 4)); /* On WHILEM */
3206 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3208 if (flags & SCF_DO_SUBSTR) {
3209 SV *last_str = NULL;
3210 int counted = mincount != 0;
3212 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3213 #if defined(SPARC64_GCC_WORKAROUND)
3216 const char *s = NULL;
3219 if (pos_before >= data->last_start_min)
3222 b = data->last_start_min;
3225 s = SvPV_const(data->last_found, l);
3226 old = b - data->last_start_min;
3229 I32 b = pos_before >= data->last_start_min
3230 ? pos_before : data->last_start_min;
3232 const char * const s = SvPV_const(data->last_found, l);
3233 I32 old = b - data->last_start_min;
3237 old = utf8_hop((U8*)s, old) - (U8*)s;
3240 /* Get the added string: */
3241 last_str = newSVpvn(s + old, l);
3243 SvUTF8_on(last_str);
3244 if (deltanext == 0 && pos_before == b) {
3245 /* What was added is a constant string */
3247 SvGROW(last_str, (mincount * l) + 1);
3248 repeatcpy(SvPVX(last_str) + l,
3249 SvPVX_const(last_str), l, mincount - 1);
3250 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3251 /* Add additional parts. */
3252 SvCUR_set(data->last_found,
3253 SvCUR(data->last_found) - l);
3254 sv_catsv(data->last_found, last_str);
3256 SV * sv = data->last_found;
3258 SvUTF8(sv) && SvMAGICAL(sv) ?
3259 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3260 if (mg && mg->mg_len >= 0)
3261 mg->mg_len += CHR_SVLEN(last_str);
3263 data->last_end += l * (mincount - 1);
3266 /* start offset must point into the last copy */
3267 data->last_start_min += minnext * (mincount - 1);
3268 data->last_start_max += is_inf ? I32_MAX
3269 : (maxcount - 1) * (minnext + data->pos_delta);
3272 /* It is counted once already... */
3273 data->pos_min += minnext * (mincount - counted);
3274 data->pos_delta += - counted * deltanext +
3275 (minnext + deltanext) * maxcount - minnext * mincount;
3276 if (mincount != maxcount) {
3277 /* Cannot extend fixed substrings found inside
3279 SCAN_COMMIT(pRExC_state,data,minlenp);
3280 if (mincount && last_str) {
3281 SV * const sv = data->last_found;
3282 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3283 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3287 sv_setsv(sv, last_str);
3288 data->last_end = data->pos_min;
3289 data->last_start_min =
3290 data->pos_min - CHR_SVLEN(last_str);
3291 data->last_start_max = is_inf
3293 : data->pos_min + data->pos_delta
3294 - CHR_SVLEN(last_str);
3296 data->longest = &(data->longest_float);
3298 SvREFCNT_dec(last_str);
3300 if (data && (fl & SF_HAS_EVAL))
3301 data->flags |= SF_HAS_EVAL;
3302 optimize_curly_tail:
3303 if (OP(oscan) != CURLYX) {
3304 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3306 NEXT_OFF(oscan) += NEXT_OFF(next);
3309 default: /* REF and CLUMP only? */
3310 if (flags & SCF_DO_SUBSTR) {
3311 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3312 data->longest = &(data->longest_float);
3314 is_inf = is_inf_internal = 1;
3315 if (flags & SCF_DO_STCLASS_OR)
3316 cl_anything(pRExC_state, data->start_class);
3317 flags &= ~SCF_DO_STCLASS;
3321 else if (strchr((const char*)PL_simple,OP(scan))) {
3324 if (flags & SCF_DO_SUBSTR) {
3325 SCAN_COMMIT(pRExC_state,data,minlenp);
3329 if (flags & SCF_DO_STCLASS) {
3330 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3332 /* Some of the logic below assumes that switching
3333 locale on will only add false positives. */
3334 switch (PL_regkind[OP(scan)]) {
3338 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3339 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3340 cl_anything(pRExC_state, data->start_class);
3343 if (OP(scan) == SANY)
3345 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3346 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3347 || (data->start_class->flags & ANYOF_CLASS));
3348 cl_anything(pRExC_state, data->start_class);
3350 if (flags & SCF_DO_STCLASS_AND || !value)
3351 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3354 if (flags & SCF_DO_STCLASS_AND)
3355 cl_and(data->start_class,
3356 (struct regnode_charclass_class*)scan);
3358 cl_or(pRExC_state, data->start_class,
3359 (struct regnode_charclass_class*)scan);
3362 if (flags & SCF_DO_STCLASS_AND) {
3363 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3364 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3365 for (value = 0; value < 256; value++)
3366 if (!isALNUM(value))
3367 ANYOF_BITMAP_CLEAR(data->start_class, value);
3371 if (data->start_class->flags & ANYOF_LOCALE)
3372 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3374 for (value = 0; value < 256; value++)
3376 ANYOF_BITMAP_SET(data->start_class, value);
3381 if (flags & SCF_DO_STCLASS_AND) {
3382 if (data->start_class->flags & ANYOF_LOCALE)
3383 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3386 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3387 data->start_class->flags |= ANYOF_LOCALE;
3391 if (flags & SCF_DO_STCLASS_AND) {
3392 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3393 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3394 for (value = 0; value < 256; value++)
3396 ANYOF_BITMAP_CLEAR(data->start_class, value);
3400 if (data->start_class->flags & ANYOF_LOCALE)
3401 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3403 for (value = 0; value < 256; value++)
3404 if (!isALNUM(value))
3405 ANYOF_BITMAP_SET(data->start_class, value);
3410 if (flags & SCF_DO_STCLASS_AND) {
3411 if (data->start_class->flags & ANYOF_LOCALE)
3412 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3415 data->start_class->flags |= ANYOF_LOCALE;
3416 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3420 if (flags & SCF_DO_STCLASS_AND) {
3421 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3422 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3423 for (value = 0; value < 256; value++)
3424 if (!isSPACE(value))
3425 ANYOF_BITMAP_CLEAR(data->start_class, value);
3429 if (data->start_class->flags & ANYOF_LOCALE)
3430 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3432 for (value = 0; value < 256; value++)
3434 ANYOF_BITMAP_SET(data->start_class, value);
3439 if (flags & SCF_DO_STCLASS_AND) {
3440 if (data->start_class->flags & ANYOF_LOCALE)
3441 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3444 data->start_class->flags |= ANYOF_LOCALE;
3445 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3449 if (flags & SCF_DO_STCLASS_AND) {
3450 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3451 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3452 for (value = 0; value < 256; value++)
3454 ANYOF_BITMAP_CLEAR(data->start_class, value);
3458 if (data->start_class->flags & ANYOF_LOCALE)
3459 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3461 for (value = 0; value < 256; value++)
3462 if (!isSPACE(value))
3463 ANYOF_BITMAP_SET(data->start_class, value);
3468 if (flags & SCF_DO_STCLASS_AND) {
3469 if (data->start_class->flags & ANYOF_LOCALE) {
3470 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3471 for (value = 0; value < 256; value++)
3472 if (!isSPACE(value))
3473 ANYOF_BITMAP_CLEAR(data->start_class, value);
3477 data->start_class->flags |= ANYOF_LOCALE;
3478 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3482 if (flags & SCF_DO_STCLASS_AND) {
3483 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3484 for (value = 0; value < 256; value++)
3485 if (!isDIGIT(value))
3486 ANYOF_BITMAP_CLEAR(data->start_class, value);
3489 if (data->start_class->flags & ANYOF_LOCALE)
3490 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3492 for (value = 0; value < 256; value++)
3494 ANYOF_BITMAP_SET(data->start_class, value);
3499 if (flags & SCF_DO_STCLASS_AND) {
3500 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3501 for (value = 0; value < 256; value++)
3503 ANYOF_BITMAP_CLEAR(data->start_class, value);
3506 if (data->start_class->flags & ANYOF_LOCALE)
3507 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3509 for (value = 0; value < 256; value++)
3510 if (!isDIGIT(value))
3511 ANYOF_BITMAP_SET(data->start_class, value);
3516 if (flags & SCF_DO_STCLASS_OR)
3517 cl_and(data->start_class, and_withp);
3518 flags &= ~SCF_DO_STCLASS;
3521 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3522 data->flags |= (OP(scan) == MEOL
3526 else if ( PL_regkind[OP(scan)] == BRANCHJ
3527 /* Lookbehind, or need to calculate parens/evals/stclass: */
3528 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3529 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3530 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3531 || OP(scan) == UNLESSM )
3533 /* Negative Lookahead/lookbehind
3534 In this case we can't do fixed string optimisation.
3537 I32 deltanext, minnext, fake = 0;
3539 struct regnode_charclass_class intrnl;
3542 data_fake.flags = 0;
3544 data_fake.whilem_c = data->whilem_c;
3545 data_fake.last_closep = data->last_closep;
3548 data_fake.last_closep = &fake;
3549 data_fake.pos_delta = delta;
3550 if ( flags & SCF_DO_STCLASS && !scan->flags
3551 && OP(scan) == IFMATCH ) { /* Lookahead */
3552 cl_init(pRExC_state, &intrnl);
3553 data_fake.start_class = &intrnl;
3554 f |= SCF_DO_STCLASS_AND;
3556 if (flags & SCF_WHILEM_VISITED_POS)
3557 f |= SCF_WHILEM_VISITED_POS;
3558 next = regnext(scan);
3559 nscan = NEXTOPER(NEXTOPER(scan));
3560 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3561 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3564 FAIL("Variable length lookbehind not implemented");
3566 else if (minnext > (I32)U8_MAX) {
3567 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3569 scan->flags = (U8)minnext;
3572 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3574 if (data_fake.flags & SF_HAS_EVAL)
3575 data->flags |= SF_HAS_EVAL;
3576 data->whilem_c = data_fake.whilem_c;
3578 if (f & SCF_DO_STCLASS_AND) {
3579 const int was = (data->start_class->flags & ANYOF_EOS);
3581 cl_and(data->start_class, &intrnl);
3583 data->start_class->flags |= ANYOF_EOS;
3586 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3588 /* Positive Lookahead/lookbehind
3589 In this case we can do fixed string optimisation,
3590 but we must be careful about it. Note in the case of
3591 lookbehind the positions will be offset by the minimum
3592 length of the pattern, something we won't know about
3593 until after the recurse.
3595 I32 deltanext, fake = 0;
3597 struct regnode_charclass_class intrnl;
3599 /* We use SAVEFREEPV so that when the full compile
3600 is finished perl will clean up the allocated
3601 minlens when its all done. This was we don't
3602 have to worry about freeing them when we know
3603 they wont be used, which would be a pain.
3606 Newx( minnextp, 1, I32 );
3607 SAVEFREEPV(minnextp);
3610 StructCopy(data, &data_fake, scan_data_t);
3611 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3614 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3615 data_fake.last_found=newSVsv(data->last_found);
3619 data_fake.last_closep = &fake;
3620 data_fake.flags = 0;
3621 data_fake.pos_delta = delta;
3623 data_fake.flags |= SF_IS_INF;
3624 if ( flags & SCF_DO_STCLASS && !scan->flags
3625 && OP(scan) == IFMATCH ) { /* Lookahead */
3626 cl_init(pRExC_state, &intrnl);
3627 data_fake.start_class = &intrnl;
3628 f |= SCF_DO_STCLASS_AND;
3630 if (flags & SCF_WHILEM_VISITED_POS)
3631 f |= SCF_WHILEM_VISITED_POS;
3632 next = regnext(scan);
3633 nscan = NEXTOPER(NEXTOPER(scan));
3635 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3636 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3639 FAIL("Variable length lookbehind not implemented");
3641 else if (*minnextp > (I32)U8_MAX) {
3642 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3644 scan->flags = (U8)*minnextp;
3649 if (f & SCF_DO_STCLASS_AND) {
3650 const int was = (data->start_class->flags & ANYOF_EOS);
3652 cl_and(data->start_class, &intrnl);
3654 data->start_class->flags |= ANYOF_EOS;
3657 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3659 if (data_fake.flags & SF_HAS_EVAL)
3660 data->flags |= SF_HAS_EVAL;
3661 data->whilem_c = data_fake.whilem_c;
3662 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3663 if (RExC_rx->minlen<*minnextp)
3664 RExC_rx->minlen=*minnextp;
3665 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3666 SvREFCNT_dec(data_fake.last_found);
3668 if ( data_fake.minlen_fixed != minlenp )
3670 data->offset_fixed= data_fake.offset_fixed;
3671 data->minlen_fixed= data_fake.minlen_fixed;
3672 data->lookbehind_fixed+= scan->flags;
3674 if ( data_fake.minlen_float != minlenp )
3676 data->minlen_float= data_fake.minlen_float;
3677 data->offset_float_min=data_fake.offset_float_min;
3678 data->offset_float_max=data_fake.offset_float_max;
3679 data->lookbehind_float+= scan->flags;
3688 else if (OP(scan) == OPEN) {
3689 if (stopparen != (I32)ARG(scan))
3692 else if (OP(scan) == CLOSE) {
3693 if (stopparen == (I32)ARG(scan)) {
3696 if ((I32)ARG(scan) == is_par) {
3697 next = regnext(scan);
3699 if ( next && (OP(next) != WHILEM) && next < last)
3700 is_par = 0; /* Disable optimization */
3703 *(data->last_closep) = ARG(scan);
3705 else if (OP(scan) == EVAL) {
3707 data->flags |= SF_HAS_EVAL;
3709 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3710 if (flags & SCF_DO_SUBSTR) {
3711 SCAN_COMMIT(pRExC_state,data,minlenp);
3712 flags &= ~SCF_DO_SUBSTR;
3714 if (data && OP(scan)==ACCEPT) {
3715 data->flags |= SCF_SEEN_ACCEPT;
3720 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3722 if (flags & SCF_DO_SUBSTR) {
3723 SCAN_COMMIT(pRExC_state,data,minlenp);
3724 data->longest = &(data->longest_float);
3726 is_inf = is_inf_internal = 1;
3727 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3728 cl_anything(pRExC_state, data->start_class);
3729 flags &= ~SCF_DO_STCLASS;
3731 else if (OP(scan) == GPOS) {
3732 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3733 !(delta || is_inf || (data && data->pos_delta)))
3735 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3736 RExC_rx->extflags |= RXf_ANCH_GPOS;
3737 if (RExC_rx->gofs < (U32)min)
3738 RExC_rx->gofs = min;
3740 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3744 #ifdef TRIE_STUDY_OPT
3745 #ifdef FULL_TRIE_STUDY
3746 else if (PL_regkind[OP(scan)] == TRIE) {
3747 /* NOTE - There is similar code to this block above for handling
3748 BRANCH nodes on the initial study. If you change stuff here
3750 regnode *trie_node= scan;
3751 regnode *tail= regnext(scan);
3752 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3753 I32 max1 = 0, min1 = I32_MAX;
3754 struct regnode_charclass_class accum;
3756 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3757 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3758 if (flags & SCF_DO_STCLASS)
3759 cl_init_zero(pRExC_state, &accum);
3765 const regnode *nextbranch= NULL;
3768 for ( word=1 ; word <= trie->wordcount ; word++)
3770 I32 deltanext=0, minnext=0, f = 0, fake;
3771 struct regnode_charclass_class this_class;
3773 data_fake.flags = 0;
3775 data_fake.whilem_c = data->whilem_c;
3776 data_fake.last_closep = data->last_closep;
3779 data_fake.last_closep = &fake;
3780 data_fake.pos_delta = delta;
3781 if (flags & SCF_DO_STCLASS) {
3782 cl_init(pRExC_state, &this_class);
3783 data_fake.start_class = &this_class;
3784 f = SCF_DO_STCLASS_AND;
3786 if (flags & SCF_WHILEM_VISITED_POS)
3787 f |= SCF_WHILEM_VISITED_POS;
3789 if (trie->jump[word]) {
3791 nextbranch = trie_node + trie->jump[0];
3792 scan= trie_node + trie->jump[word];
3793 /* We go from the jump point to the branch that follows
3794 it. Note this means we need the vestigal unused branches
3795 even though they arent otherwise used.
3797 minnext = study_chunk(pRExC_state, &scan, minlenp,
3798 &deltanext, (regnode *)nextbranch, &data_fake,
3799 stopparen, recursed, NULL, f,depth+1);
3801 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3802 nextbranch= regnext((regnode*)nextbranch);
3804 if (min1 > (I32)(minnext + trie->minlen))
3805 min1 = minnext + trie->minlen;
3806 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3807 max1 = minnext + deltanext + trie->maxlen;
3808 if (deltanext == I32_MAX)
3809 is_inf = is_inf_internal = 1;
3811 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3813 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3814 if ( stopmin > min + min1)
3815 stopmin = min + min1;
3816 flags &= ~SCF_DO_SUBSTR;
3818 data->flags |= SCF_SEEN_ACCEPT;
3821 if (data_fake.flags & SF_HAS_EVAL)
3822 data->flags |= SF_HAS_EVAL;
3823 data->whilem_c = data_fake.whilem_c;
3825 if (flags & SCF_DO_STCLASS)
3826 cl_or(pRExC_state, &accum, &this_class);
3829 if (flags & SCF_DO_SUBSTR) {
3830 data->pos_min += min1;
3831 data->pos_delta += max1 - min1;
3832 if (max1 != min1 || is_inf)
3833 data->longest = &(data->longest_float);
3836 delta += max1 - min1;
3837 if (flags & SCF_DO_STCLASS_OR) {
3838 cl_or(pRExC_state, data->start_class, &accum);
3840 cl_and(data->start_class, and_withp);
3841 flags &= ~SCF_DO_STCLASS;
3844 else if (flags & SCF_DO_STCLASS_AND) {
3846 cl_and(data->start_class, &accum);
3847 flags &= ~SCF_DO_STCLASS;
3850 /* Switch to OR mode: cache the old value of
3851 * data->start_class */
3853 StructCopy(data->start_class, and_withp,
3854 struct regnode_charclass_class);
3855 flags &= ~SCF_DO_STCLASS_AND;
3856 StructCopy(&accum, data->start_class,
3857 struct regnode_charclass_class);
3858 flags |= SCF_DO_STCLASS_OR;
3859 data->start_class->flags |= ANYOF_EOS;
3866 else if (PL_regkind[OP(scan)] == TRIE) {
3867 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3870 min += trie->minlen;
3871 delta += (trie->maxlen - trie->minlen);
3872 flags &= ~SCF_DO_STCLASS; /* xxx */
3873 if (flags & SCF_DO_SUBSTR) {
3874 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3875 data->pos_min += trie->minlen;
3876 data->pos_delta += (trie->maxlen - trie->minlen);
3877 if (trie->maxlen != trie->minlen)
3878 data->longest = &(data->longest_float);
3880 if (trie->jump) /* no more substrings -- for now /grr*/
3881 flags &= ~SCF_DO_SUBSTR;
3883 #endif /* old or new */
3884 #endif /* TRIE_STUDY_OPT */
3885 /* Else: zero-length, ignore. */
3886 scan = regnext(scan);
3891 stopparen = frame->stop;
3892 frame = frame->prev;
3893 goto fake_study_recurse;
3898 DEBUG_STUDYDATA("pre-fin:",data,depth);
3901 *deltap = is_inf_internal ? I32_MAX : delta;
3902 if (flags & SCF_DO_SUBSTR && is_inf)
3903 data->pos_delta = I32_MAX - data->pos_min;
3904 if (is_par > (I32)U8_MAX)
3906 if (is_par && pars==1 && data) {
3907 data->flags |= SF_IN_PAR;
3908 data->flags &= ~SF_HAS_PAR;
3910 else if (pars && data) {
3911 data->flags |= SF_HAS_PAR;
3912 data->flags &= ~SF_IN_PAR;
3914 if (flags & SCF_DO_STCLASS_OR)
3915 cl_and(data->start_class, and_withp);
3916 if (flags & SCF_TRIE_RESTUDY)
3917 data->flags |= SCF_TRIE_RESTUDY;
3919 DEBUG_STUDYDATA("post-fin:",data,depth);
3921 return min < stopmin ? min : stopmin;
3925 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3927 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3929 Renewc(RExC_rxi->data,
3930 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3931 char, struct reg_data);
3933 Renew(RExC_rxi->data->what, count + n, U8);
3935 Newx(RExC_rxi->data->what, n, U8);
3936 RExC_rxi->data->count = count + n;
3937 Copy(s, RExC_rxi->data->what + count, n, U8);
3941 /*XXX: todo make this not included in a non debugging perl */
3942 #ifndef PERL_IN_XSUB_RE
3944 Perl_reginitcolors(pTHX)
3947 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3949 char *t = savepv(s);
3953 t = strchr(t, '\t');
3959 PL_colors[i] = t = (char *)"";
3964 PL_colors[i++] = (char *)"";
3971 #ifdef TRIE_STUDY_OPT
3972 #define CHECK_RESTUDY_GOTO \
3974 (data.flags & SCF_TRIE_RESTUDY) \
3978 #define CHECK_RESTUDY_GOTO
3982 - pregcomp - compile a regular expression into internal code
3984 * We can't allocate space until we know how big the compiled form will be,
3985 * but we can't compile it (and thus know how big it is) until we've got a
3986 * place to put the code. So we cheat: we compile it twice, once with code
3987 * generation turned off and size counting turned on, and once "for real".
3988 * This also means that we don't allocate space until we are sure that the
3989 * thing really will compile successfully, and we never have to move the
3990 * code and thus invalidate pointers into it. (Note that it has to be in
3991 * one piece because free() must be able to free it all.) [NB: not true in perl]
3993 * Beware that the optimization-preparation code in here knows about some
3994 * of the structure of the compiled regexp. [I'll say.]
3999 #ifndef PERL_IN_XSUB_RE
4000 #define RE_ENGINE_PTR &PL_core_reg_engine
4002 extern const struct regexp_engine my_reg_engine;
4003 #define RE_ENGINE_PTR &my_reg_engine
4006 #ifndef PERL_IN_XSUB_RE
4008 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
4011 HV * const table = GvHV(PL_hintgv);
4012 /* Dispatch a request to compile a regexp to correct
4015 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4016 GET_RE_DEBUG_FLAGS_DECL;
4017 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4018 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4020 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4023 return CALLREGCOMP_ENG(eng, exp, xend, pm);
4026 return Perl_re_compile(aTHX_ exp, xend, pm);
4031 Perl_re_compile(pTHX_ char *exp, char *xend, PMOP *pm)
4035 register regexp_internal *ri;
4043 RExC_state_t RExC_state;
4044 RExC_state_t * const pRExC_state = &RExC_state;
4045 #ifdef TRIE_STUDY_OPT
4047 RExC_state_t copyRExC_state;
4049 GET_RE_DEBUG_FLAGS_DECL;
4050 DEBUG_r(if (!PL_colorset) reginitcolors());
4053 FAIL("NULL regexp argument");
4055 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4059 SV *dsv= sv_newmortal();
4060 RE_PV_QUOTED_DECL(s, RExC_utf8,
4061 dsv, RExC_precomp, (xend - exp), 60);
4062 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4063 PL_colors[4],PL_colors[5],s);
4065 RExC_flags = pm->op_pmflags;
4069 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4070 RExC_seen_evals = 0;
4073 /* First pass: determine size, legality. */
4081 RExC_emit = &PL_regdummy;
4082 RExC_whilem_seen = 0;
4083 RExC_charnames = NULL;
4084 RExC_open_parens = NULL;
4085 RExC_close_parens = NULL;
4087 RExC_paren_names = NULL;
4089 RExC_paren_name_list = NULL;
4091 RExC_recurse = NULL;
4092 RExC_recurse_count = 0;
4094 #if 0 /* REGC() is (currently) a NOP at the first pass.
4095 * Clever compilers notice this and complain. --jhi */
4096 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4098 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4099 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4100 RExC_precomp = NULL;
4104 PerlIO_printf(Perl_debug_log,
4105 "Required size %"IVdf" nodes\n"
4106 "Starting second pass (creation)\n",
4109 RExC_lastparse=NULL;
4111 /* Small enough for pointer-storage convention?
4112 If extralen==0, this means that we will not need long jumps. */
4113 if (RExC_size >= 0x10000L && RExC_extralen)
4114 RExC_size += RExC_extralen;
4117 if (RExC_whilem_seen > 15)
4118 RExC_whilem_seen = 15;
4120 /* Allocate space and zero-initialize. Note, the two step process
4121 of zeroing when in debug mode, thus anything assigned has to
4122 happen after that */
4123 Newxz(r, 1, regexp);
4124 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4125 char, regexp_internal);
4126 if ( r == NULL || ri == NULL )
4127 FAIL("Regexp out of space");
4129 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4130 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4132 /* bulk initialize base fields with 0. */
4133 Zero(ri, sizeof(regexp_internal), char);
4136 /* non-zero initialization begins here */
4138 r->engine= RE_ENGINE_PTR;
4140 r->prelen = xend - exp;
4141 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4143 bool has_k = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4144 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4145 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4146 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) >> 12);
4147 const char *fptr = STD_PAT_MODS; /*"msix"*/
4149 r->wraplen = r->prelen + has_minus + has_k + has_runon
4150 + (sizeof(STD_PAT_MODS) - 1)
4151 + (sizeof("(?:)") - 1);
4153 Newx(r->wrapped, r->wraplen, char );
4157 *p++ = KEEPCOPY_PAT_MOD; /*'k'*/
4159 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4160 char *colon = r + 1;
4163 while((ch = *fptr++)) {
4177 Copy(RExC_precomp, p, r->prelen, char);
4186 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4188 if (RExC_seen & REG_SEEN_RECURSE) {
4189 Newxz(RExC_open_parens, RExC_npar,regnode *);
4190 SAVEFREEPV(RExC_open_parens);
4191 Newxz(RExC_close_parens,RExC_npar,regnode *);
4192 SAVEFREEPV(RExC_close_parens);
4195 /* Useful during FAIL. */
4196 #ifdef RE_TRACK_PATTERN_OFFSETS
4197 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4198 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4199 "%s %"UVuf" bytes for offset annotations.\n",
4200 ri->u.offsets ? "Got" : "Couldn't get",
4201 (UV)((2*RExC_size+1) * sizeof(U32))));
4203 SetProgLen(ri,RExC_size);
4207 /* Second pass: emit code. */
4208 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4213 RExC_emit_start = ri->program;
4214 RExC_emit = ri->program;
4215 RExC_emit_bound = ri->program + RExC_size + 1;
4217 /* Store the count of eval-groups for security checks: */
4218 RExC_rx->seen_evals = RExC_seen_evals;
4219 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4220 if (reg(pRExC_state, 0, &flags,1) == NULL)
4223 /* XXXX To minimize changes to RE engine we always allocate
4224 3-units-long substrs field. */
4225 Newx(r->substrs, 1, struct reg_substr_data);
4226 if (RExC_recurse_count) {
4227 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4228 SAVEFREEPV(RExC_recurse);
4232 r->minlen = minlen = sawplus = sawopen = 0;
4233 Zero(r->substrs, 1, struct reg_substr_data);
4235 #ifdef TRIE_STUDY_OPT
4238 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4240 RExC_state = copyRExC_state;
4241 if (seen & REG_TOP_LEVEL_BRANCHES)
4242 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4244 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4245 if (data.last_found) {
4246 SvREFCNT_dec(data.longest_fixed);
4247 SvREFCNT_dec(data.longest_float);
4248 SvREFCNT_dec(data.last_found);
4250 StructCopy(&zero_scan_data, &data, scan_data_t);
4252 StructCopy(&zero_scan_data, &data, scan_data_t);
4253 copyRExC_state = RExC_state;
4256 StructCopy(&zero_scan_data, &data, scan_data_t);
4259 /* Dig out information for optimizations. */
4260 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4261 pm->op_pmflags = RExC_flags;
4263 r->extflags |= RXf_UTF8; /* Unicode in it? */
4264 ri->regstclass = NULL;
4265 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4266 r->intflags |= PREGf_NAUGHTY;
4267 scan = ri->program + 1; /* First BRANCH. */
4269 /* testing for BRANCH here tells us whether there is "must appear"
4270 data in the pattern. If there is then we can use it for optimisations */
4271 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4273 STRLEN longest_float_length, longest_fixed_length;
4274 struct regnode_charclass_class ch_class; /* pointed to by data */
4276 I32 last_close = 0; /* pointed to by data */
4279 /* Skip introductions and multiplicators >= 1. */
4280 while ((OP(first) == OPEN && (sawopen = 1)) ||
4281 /* An OR of *one* alternative - should not happen now. */
4282 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4283 /* for now we can't handle lookbehind IFMATCH*/
4284 (OP(first) == IFMATCH && !first->flags) ||
4285 (OP(first) == PLUS) ||
4286 (OP(first) == MINMOD) ||
4287 /* An {n,m} with n>0 */
4288 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4291 if (OP(first) == PLUS)
4294 first += regarglen[OP(first)];
4295 if (OP(first) == IFMATCH) {
4296 first = NEXTOPER(first);
4297 first += EXTRA_STEP_2ARGS;
4298 } else /* XXX possible optimisation for /(?=)/ */
4299 first = NEXTOPER(first);
4302 /* Starting-point info. */
4304 DEBUG_PEEP("first:",first,0);
4305 /* Ignore EXACT as we deal with it later. */
4306 if (PL_regkind[OP(first)] == EXACT) {
4307 if (OP(first) == EXACT)
4308 NOOP; /* Empty, get anchored substr later. */
4309 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4310 ri->regstclass = first;
4313 else if (PL_regkind[OP(first)] == TRIE &&
4314 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4317 /* this can happen only on restudy */
4318 if ( OP(first) == TRIE ) {
4319 struct regnode_1 *trieop = (struct regnode_1 *)
4320 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4321 StructCopy(first,trieop,struct regnode_1);
4322 trie_op=(regnode *)trieop;
4324 struct regnode_charclass *trieop = (struct regnode_charclass *)
4325 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4326 StructCopy(first,trieop,struct regnode_charclass);
4327 trie_op=(regnode *)trieop;
4330 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4331 ri->regstclass = trie_op;
4334 else if (strchr((const char*)PL_simple,OP(first)))
4335 ri->regstclass = first;
4336 else if (PL_regkind[OP(first)] == BOUND ||
4337 PL_regkind[OP(first)] == NBOUND)
4338 ri->regstclass = first;
4339 else if (PL_regkind[OP(first)] == BOL) {
4340 r->extflags |= (OP(first) == MBOL
4342 : (OP(first) == SBOL
4345 first = NEXTOPER(first);
4348 else if (OP(first) == GPOS) {
4349 r->extflags |= RXf_ANCH_GPOS;
4350 first = NEXTOPER(first);
4353 else if ((!sawopen || !RExC_sawback) &&
4354 (OP(first) == STAR &&
4355 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4356 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4358 /* turn .* into ^.* with an implied $*=1 */
4360 (OP(NEXTOPER(first)) == REG_ANY)
4363 r->extflags |= type;
4364 r->intflags |= PREGf_IMPLICIT;
4365 first = NEXTOPER(first);
4368 if (sawplus && (!sawopen || !RExC_sawback)
4369 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4370 /* x+ must match at the 1st pos of run of x's */
4371 r->intflags |= PREGf_SKIP;
4373 /* Scan is after the zeroth branch, first is atomic matcher. */
4374 #ifdef TRIE_STUDY_OPT
4377 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4378 (IV)(first - scan + 1))
4382 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4383 (IV)(first - scan + 1))
4389 * If there's something expensive in the r.e., find the
4390 * longest literal string that must appear and make it the
4391 * regmust. Resolve ties in favor of later strings, since
4392 * the regstart check works with the beginning of the r.e.
4393 * and avoiding duplication strengthens checking. Not a
4394 * strong reason, but sufficient in the absence of others.
4395 * [Now we resolve ties in favor of the earlier string if
4396 * it happens that c_offset_min has been invalidated, since the
4397 * earlier string may buy us something the later one won't.]
4400 data.longest_fixed = newSVpvs("");
4401 data.longest_float = newSVpvs("");
4402 data.last_found = newSVpvs("");
4403 data.longest = &(data.longest_fixed);
4405 if (!ri->regstclass) {
4406 cl_init(pRExC_state, &ch_class);
4407 data.start_class = &ch_class;
4408 stclass_flag = SCF_DO_STCLASS_AND;
4409 } else /* XXXX Check for BOUND? */
4411 data.last_closep = &last_close;
4413 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4414 &data, -1, NULL, NULL,
4415 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4421 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4422 && data.last_start_min == 0 && data.last_end > 0
4423 && !RExC_seen_zerolen
4424 && !(RExC_seen & REG_SEEN_VERBARG)
4425 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4426 r->extflags |= RXf_CHECK_ALL;
4427 scan_commit(pRExC_state, &data,&minlen,0);
4428 SvREFCNT_dec(data.last_found);
4430 /* Note that code very similar to this but for anchored string
4431 follows immediately below, changes may need to be made to both.
4434 longest_float_length = CHR_SVLEN(data.longest_float);
4435 if (longest_float_length
4436 || (data.flags & SF_FL_BEFORE_EOL
4437 && (!(data.flags & SF_FL_BEFORE_MEOL)
4438 || (RExC_flags & RXf_PMf_MULTILINE))))
4442 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4443 && data.offset_fixed == data.offset_float_min
4444 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4445 goto remove_float; /* As in (a)+. */
4447 /* copy the information about the longest float from the reg_scan_data
4448 over to the program. */
4449 if (SvUTF8(data.longest_float)) {
4450 r->float_utf8 = data.longest_float;
4451 r->float_substr = NULL;
4453 r->float_substr = data.longest_float;
4454 r->float_utf8 = NULL;
4456 /* float_end_shift is how many chars that must be matched that
4457 follow this item. We calculate it ahead of time as once the
4458 lookbehind offset is added in we lose the ability to correctly
4460 ml = data.minlen_float ? *(data.minlen_float)
4461 : (I32)longest_float_length;
4462 r->float_end_shift = ml - data.offset_float_min
4463 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4464 + data.lookbehind_float;
4465 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4466 r->float_max_offset = data.offset_float_max;
4467 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4468 r->float_max_offset -= data.lookbehind_float;
4470 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4471 && (!(data.flags & SF_FL_BEFORE_MEOL)
4472 || (RExC_flags & RXf_PMf_MULTILINE)));
4473 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4477 r->float_substr = r->float_utf8 = NULL;
4478 SvREFCNT_dec(data.longest_float);
4479 longest_float_length = 0;
4482 /* Note that code very similar to this but for floating string
4483 is immediately above, changes may need to be made to both.
4486 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4487 if (longest_fixed_length
4488 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4489 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4490 || (RExC_flags & RXf_PMf_MULTILINE))))
4494 /* copy the information about the longest fixed
4495 from the reg_scan_data over to the program. */
4496 if (SvUTF8(data.longest_fixed)) {
4497 r->anchored_utf8 = data.longest_fixed;
4498 r->anchored_substr = NULL;
4500 r->anchored_substr = data.longest_fixed;
4501 r->anchored_utf8 = NULL;
4503 /* fixed_end_shift is how many chars that must be matched that
4504 follow this item. We calculate it ahead of time as once the
4505 lookbehind offset is added in we lose the ability to correctly
4507 ml = data.minlen_fixed ? *(data.minlen_fixed)
4508 : (I32)longest_fixed_length;
4509 r->anchored_end_shift = ml - data.offset_fixed
4510 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4511 + data.lookbehind_fixed;
4512 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4514 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4515 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4516 || (RExC_flags & RXf_PMf_MULTILINE)));
4517 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4520 r->anchored_substr = r->anchored_utf8 = NULL;
4521 SvREFCNT_dec(data.longest_fixed);
4522 longest_fixed_length = 0;
4525 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4526 ri->regstclass = NULL;
4527 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4529 && !(data.start_class->flags & ANYOF_EOS)
4530 && !cl_is_anything(data.start_class))
4532 const U32 n = add_data(pRExC_state, 1, "f");
4534 Newx(RExC_rxi->data->data[n], 1,
4535 struct regnode_charclass_class);
4536 StructCopy(data.start_class,
4537 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4538 struct regnode_charclass_class);
4539 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4540 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4541 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4542 regprop(r, sv, (regnode*)data.start_class);
4543 PerlIO_printf(Perl_debug_log,
4544 "synthetic stclass \"%s\".\n",
4545 SvPVX_const(sv));});
4548 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4549 if (longest_fixed_length > longest_float_length) {
4550 r->check_end_shift = r->anchored_end_shift;
4551 r->check_substr = r->anchored_substr;
4552 r->check_utf8 = r->anchored_utf8;
4553 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4554 if (r->extflags & RXf_ANCH_SINGLE)
4555 r->extflags |= RXf_NOSCAN;
4558 r->check_end_shift = r->float_end_shift;
4559 r->check_substr = r->float_substr;
4560 r->check_utf8 = r->float_utf8;
4561 r->check_offset_min = r->float_min_offset;
4562 r->check_offset_max = r->float_max_offset;
4564 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4565 This should be changed ASAP! */
4566 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4567 r->extflags |= RXf_USE_INTUIT;
4568 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4569 r->extflags |= RXf_INTUIT_TAIL;
4571 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4572 if ( (STRLEN)minlen < longest_float_length )
4573 minlen= longest_float_length;
4574 if ( (STRLEN)minlen < longest_fixed_length )
4575 minlen= longest_fixed_length;
4579 /* Several toplevels. Best we can is to set minlen. */
4581 struct regnode_charclass_class ch_class;
4584 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4586 scan = ri->program + 1;
4587 cl_init(pRExC_state, &ch_class);
4588 data.start_class = &ch_class;
4589 data.last_closep = &last_close;
4592 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4593 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4597 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4598 = r->float_substr = r->float_utf8 = NULL;
4599 if (!(data.start_class->flags & ANYOF_EOS)
4600 && !cl_is_anything(data.start_class))
4602 const U32 n = add_data(pRExC_state, 1, "f");
4604 Newx(RExC_rxi->data->data[n], 1,
4605 struct regnode_charclass_class);
4606 StructCopy(data.start_class,
4607 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4608 struct regnode_charclass_class);
4609 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4610 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4611 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4612 regprop(r, sv, (regnode*)data.start_class);
4613 PerlIO_printf(Perl_debug_log,
4614 "synthetic stclass \"%s\".\n",
4615 SvPVX_const(sv));});
4619 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4620 the "real" pattern. */
4622 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4623 (IV)minlen, (IV)r->minlen);
4625 r->minlenret = minlen;
4626 if (r->minlen < minlen)
4629 if (RExC_seen & REG_SEEN_GPOS)
4630 r->extflags |= RXf_GPOS_SEEN;
4631 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4632 r->extflags |= RXf_LOOKBEHIND_SEEN;
4633 if (RExC_seen & REG_SEEN_EVAL)
4634 r->extflags |= RXf_EVAL_SEEN;
4635 if (RExC_seen & REG_SEEN_CANY)
4636 r->extflags |= RXf_CANY_SEEN;
4637 if (RExC_seen & REG_SEEN_VERBARG)
4638 r->intflags |= PREGf_VERBARG_SEEN;
4639 if (RExC_seen & REG_SEEN_CUTGROUP)
4640 r->intflags |= PREGf_CUTGROUP_SEEN;
4641 if (RExC_paren_names)
4642 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4644 r->paren_names = NULL;
4645 if (r->prelen == 3 && strEQ("\\s+", r->precomp))
4646 r->extflags |= RXf_WHITE;
4647 else if (r->prelen == 1 && r->precomp[0] == '^')
4648 r->extflags |= RXf_START_ONLY;
4651 if (RExC_paren_names) {
4652 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4653 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4656 ri->name_list_idx = 0;
4658 if (RExC_recurse_count) {
4659 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4660 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4661 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4664 Newxz(r->startp, RExC_npar, I32);
4665 Newxz(r->endp, RExC_npar, I32);
4666 /* assume we don't need to swap parens around before we match */
4669 PerlIO_printf(Perl_debug_log,"Final program:\n");
4672 #ifdef RE_TRACK_PATTERN_OFFSETS
4673 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4674 const U32 len = ri->u.offsets[0];
4676 GET_RE_DEBUG_FLAGS_DECL;
4677 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4678 for (i = 1; i <= len; i++) {
4679 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4680 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4681 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4683 PerlIO_printf(Perl_debug_log, "\n");
4689 #undef RE_ENGINE_PTR
4693 Perl_reg_named_buff_get(pTHX_ const REGEXP * const rx, SV* namesv, U32 flags)
4695 AV *retarray = NULL;
4700 if (rx && rx->paren_names) {
4701 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4704 SV* sv_dat=HeVAL(he_str);
4705 I32 *nums=(I32*)SvPVX(sv_dat);
4706 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4707 if ((I32)(rx->nparens) >= nums[i]
4708 && rx->startp[nums[i]] != -1
4709 && rx->endp[nums[i]] != -1)
4711 ret = CALLREG_NUMBUF(rx,nums[i],NULL);
4715 ret = newSVsv(&PL_sv_undef);
4719 av_push(retarray, ret);
4723 return (SV*)retarray;
4730 Perl_reg_numbered_buff_get(pTHX_ const REGEXP * const rx, I32 paren, SV* usesv)
4735 SV *sv = usesv ? usesv : newSVpvs("");
4738 sv_setsv(sv,&PL_sv_undef);
4742 if (paren == -2 && rx->startp[0] != -1) {
4748 if (paren == -1 && rx->endp[0] != -1) {
4750 s = rx->subbeg + rx->endp[0];
4751 i = rx->sublen - rx->endp[0];
4754 if ( 0 <= paren && paren <= (I32)rx->nparens &&
4755 (s1 = rx->startp[paren]) != -1 &&
4756 (t1 = rx->endp[paren]) != -1)
4760 s = rx->subbeg + s1;
4762 sv_setsv(sv,&PL_sv_undef);
4765 assert(rx->sublen >= (s - rx->subbeg) + i );
4767 const int oldtainted = PL_tainted;
4769 sv_setpvn(sv, s, i);
4770 PL_tainted = oldtainted;
4771 if ( (rx->extflags & RXf_CANY_SEEN)
4772 ? (RX_MATCH_UTF8(rx)
4773 && (!i || is_utf8_string((U8*)s, i)))
4774 : (RX_MATCH_UTF8(rx)) )
4781 if (RX_MATCH_TAINTED(rx)) {
4782 if (SvTYPE(sv) >= SVt_PVMG) {
4783 MAGIC* const mg = SvMAGIC(sv);
4786 SvMAGIC_set(sv, mg->mg_moremagic);
4788 if ((mgt = SvMAGIC(sv))) {
4789 mg->mg_moremagic = mgt;
4790 SvMAGIC_set(sv, mg);
4800 sv_setsv(sv,&PL_sv_undef);
4806 /* Scans the name of a named buffer from the pattern.
4807 * If flags is REG_RSN_RETURN_NULL returns null.
4808 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4809 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4810 * to the parsed name as looked up in the RExC_paren_names hash.
4811 * If there is an error throws a vFAIL().. type exception.
4814 #define REG_RSN_RETURN_NULL 0
4815 #define REG_RSN_RETURN_NAME 1
4816 #define REG_RSN_RETURN_DATA 2
4819 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4820 char *name_start = RExC_parse;
4822 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
4823 /* skip IDFIRST by using do...while */
4826 RExC_parse += UTF8SKIP(RExC_parse);
4827 } while (isALNUM_utf8((U8*)RExC_parse));
4831 } while (isALNUM(*RExC_parse));
4835 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4836 (int)(RExC_parse - name_start)));
4839 if ( flags == REG_RSN_RETURN_NAME)
4841 else if (flags==REG_RSN_RETURN_DATA) {
4844 if ( ! sv_name ) /* should not happen*/
4845 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4846 if (RExC_paren_names)
4847 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4849 sv_dat = HeVAL(he_str);
4851 vFAIL("Reference to nonexistent named group");
4855 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4862 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4863 int rem=(int)(RExC_end - RExC_parse); \
4872 if (RExC_lastparse!=RExC_parse) \
4873 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4876 iscut ? "..." : "<" \
4879 PerlIO_printf(Perl_debug_log,"%16s",""); \
4882 num = RExC_size + 1; \
4884 num=REG_NODE_NUM(RExC_emit); \
4885 if (RExC_lastnum!=num) \
4886 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4888 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4889 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4890 (int)((depth*2)), "", \
4894 RExC_lastparse=RExC_parse; \
4899 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4900 DEBUG_PARSE_MSG((funcname)); \
4901 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4903 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4904 DEBUG_PARSE_MSG((funcname)); \
4905 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4908 - reg - regular expression, i.e. main body or parenthesized thing
4910 * Caller must absorb opening parenthesis.
4912 * Combining parenthesis handling with the base level of regular expression
4913 * is a trifle forced, but the need to tie the tails of the branches to what
4914 * follows makes it hard to avoid.
4916 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4918 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4920 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4923 /* this idea is borrowed from STR_WITH_LEN in handy.h */
4924 #define CHECK_WORD(s,v,l) \
4925 (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
4928 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4929 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4932 register regnode *ret; /* Will be the head of the group. */
4933 register regnode *br;
4934 register regnode *lastbr;
4935 register regnode *ender = NULL;
4936 register I32 parno = 0;
4938 const I32 oregflags = RExC_flags;
4939 bool have_branch = 0;
4941 I32 freeze_paren = 0;
4942 I32 after_freeze = 0;
4944 /* for (?g), (?gc), and (?o) warnings; warning
4945 about (?c) will warn about (?g) -- japhy */
4947 #define WASTED_O 0x01
4948 #define WASTED_G 0x02
4949 #define WASTED_C 0x04
4950 #define WASTED_GC (0x02|0x04)
4951 I32 wastedflags = 0x00;
4953 char * parse_start = RExC_parse; /* MJD */
4954 char * const oregcomp_parse = RExC_parse;
4956 GET_RE_DEBUG_FLAGS_DECL;
4957 DEBUG_PARSE("reg ");
4960 *flagp = 0; /* Tentatively. */
4963 /* Make an OPEN node, if parenthesized. */
4965 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4966 char *start_verb = RExC_parse;
4967 STRLEN verb_len = 0;
4968 char *start_arg = NULL;
4969 unsigned char op = 0;
4971 int internal_argval = 0; /* internal_argval is only useful if !argok */
4972 while ( *RExC_parse && *RExC_parse != ')' ) {
4973 if ( *RExC_parse == ':' ) {
4974 start_arg = RExC_parse + 1;
4980 verb_len = RExC_parse - start_verb;
4983 while ( *RExC_parse && *RExC_parse != ')' )
4985 if ( *RExC_parse != ')' )
4986 vFAIL("Unterminated verb pattern argument");
4987 if ( RExC_parse == start_arg )
4990 if ( *RExC_parse != ')' )
4991 vFAIL("Unterminated verb pattern");
4994 switch ( *start_verb ) {
4995 case 'A': /* (*ACCEPT) */
4996 if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
4998 internal_argval = RExC_nestroot;
5001 case 'C': /* (*COMMIT) */
5002 if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
5005 case 'F': /* (*FAIL) */
5006 if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
5011 case ':': /* (*:NAME) */
5012 case 'M': /* (*MARK:NAME) */
5013 if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
5018 case 'P': /* (*PRUNE) */
5019 if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
5022 case 'S': /* (*SKIP) */
5023 if ( CHECK_WORD("SKIP",start_verb,verb_len) )
5026 case 'T': /* (*THEN) */
5027 /* [19:06] <TimToady> :: is then */
5028 if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
5030 RExC_seen |= REG_SEEN_CUTGROUP;
5036 vFAIL3("Unknown verb pattern '%.*s'",
5037 verb_len, start_verb);
5040 if ( start_arg && internal_argval ) {
5041 vFAIL3("Verb pattern '%.*s' may not have an argument",
5042 verb_len, start_verb);
5043 } else if ( argok < 0 && !start_arg ) {
5044 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5045 verb_len, start_verb);
5047 ret = reganode(pRExC_state, op, internal_argval);
5048 if ( ! internal_argval && ! SIZE_ONLY ) {
5050 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5051 ARG(ret) = add_data( pRExC_state, 1, "S" );
5052 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5059 if (!internal_argval)
5060 RExC_seen |= REG_SEEN_VERBARG;
5061 } else if ( start_arg ) {
5062 vFAIL3("Verb pattern '%.*s' may not have an argument",
5063 verb_len, start_verb);
5065 ret = reg_node(pRExC_state, op);
5067 nextchar(pRExC_state);
5070 if (*RExC_parse == '?') { /* (?...) */
5071 bool is_logical = 0;
5072 const char * const seqstart = RExC_parse;
5075 paren = *RExC_parse++;
5076 ret = NULL; /* For look-ahead/behind. */
5079 case 'P': /* (?P...) variants for those used to PCRE/Python */
5080 paren = *RExC_parse++;
5081 if ( paren == '<') /* (?P<...>) named capture */
5083 else if (paren == '>') { /* (?P>name) named recursion */
5084 goto named_recursion;
5086 else if (paren == '=') { /* (?P=...) named backref */
5087 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5088 you change this make sure you change that */
5089 char* name_start = RExC_parse;
5091 SV *sv_dat = reg_scan_name(pRExC_state,
5092 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5093 if (RExC_parse == name_start || *RExC_parse != ')')
5094 vFAIL2("Sequence %.3s... not terminated",parse_start);
5097 num = add_data( pRExC_state, 1, "S" );
5098 RExC_rxi->data->data[num]=(void*)sv_dat;
5099 SvREFCNT_inc(sv_dat);
5102 ret = reganode(pRExC_state,
5103 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5107 Set_Node_Offset(ret, parse_start+1);
5108 Set_Node_Cur_Length(ret); /* MJD */
5110 nextchar(pRExC_state);
5114 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5116 case '<': /* (?<...) */
5117 if (*RExC_parse == '!')
5119 else if (*RExC_parse != '=')
5125 case '\'': /* (?'...') */
5126 name_start= RExC_parse;
5127 svname = reg_scan_name(pRExC_state,
5128 SIZE_ONLY ? /* reverse test from the others */
5129 REG_RSN_RETURN_NAME :
5130 REG_RSN_RETURN_NULL);
5131 if (RExC_parse == name_start) {
5133 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5136 if (*RExC_parse != paren)
5137 vFAIL2("Sequence (?%c... not terminated",
5138 paren=='>' ? '<' : paren);
5142 if (!svname) /* shouldnt happen */
5144 "panic: reg_scan_name returned NULL");
5145 if (!RExC_paren_names) {
5146 RExC_paren_names= newHV();
5147 sv_2mortal((SV*)RExC_paren_names);
5149 RExC_paren_name_list= newAV();
5150 sv_2mortal((SV*)RExC_paren_name_list);
5153 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5155 sv_dat = HeVAL(he_str);
5157 /* croak baby croak */
5159 "panic: paren_name hash element allocation failed");
5160 } else if ( SvPOK(sv_dat) ) {
5161 /* (?|...) can mean we have dupes so scan to check
5162 its already been stored. Maybe a flag indicating
5163 we are inside such a construct would be useful,
5164 but the arrays are likely to be quite small, so
5165 for now we punt -- dmq */
5166 IV count = SvIV(sv_dat);
5167 I32 *pv = (I32*)SvPVX(sv_dat);
5169 for ( i = 0 ; i < count ; i++ ) {
5170 if ( pv[i] == RExC_npar ) {
5176 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5177 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5178 pv[count] = RExC_npar;
5182 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5183 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5188 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5189 SvREFCNT_dec(svname);
5192 /*sv_dump(sv_dat);*/
5194 nextchar(pRExC_state);
5196 goto capturing_parens;
5198 RExC_seen |= REG_SEEN_LOOKBEHIND;
5200 case '=': /* (?=...) */
5201 case '!': /* (?!...) */
5202 RExC_seen_zerolen++;
5203 if (*RExC_parse == ')') {
5204 ret=reg_node(pRExC_state, OPFAIL);
5205 nextchar(pRExC_state);
5209 case '|': /* (?|...) */
5210 /* branch reset, behave like a (?:...) except that
5211 buffers in alternations share the same numbers */
5213 after_freeze = freeze_paren = RExC_npar;
5215 case ':': /* (?:...) */
5216 case '>': /* (?>...) */
5218 case '$': /* (?$...) */
5219 case '@': /* (?@...) */
5220 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5222 case '#': /* (?#...) */
5223 while (*RExC_parse && *RExC_parse != ')')
5225 if (*RExC_parse != ')')
5226 FAIL("Sequence (?#... not terminated");
5227 nextchar(pRExC_state);
5230 case '0' : /* (?0) */
5231 case 'R' : /* (?R) */
5232 if (*RExC_parse != ')')
5233 FAIL("Sequence (?R) not terminated");
5234 ret = reg_node(pRExC_state, GOSTART);
5235 *flagp |= POSTPONED;
5236 nextchar(pRExC_state);
5239 { /* named and numeric backreferences */
5241 case '&': /* (?&NAME) */
5242 parse_start = RExC_parse - 1;
5245 SV *sv_dat = reg_scan_name(pRExC_state,
5246 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5247 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5249 goto gen_recurse_regop;
5252 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5254 vFAIL("Illegal pattern");
5256 goto parse_recursion;
5258 case '-': /* (?-1) */
5259 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5260 RExC_parse--; /* rewind to let it be handled later */
5264 case '1': case '2': case '3': case '4': /* (?1) */
5265 case '5': case '6': case '7': case '8': case '9':
5268 num = atoi(RExC_parse);
5269 parse_start = RExC_parse - 1; /* MJD */
5270 if (*RExC_parse == '-')
5272 while (isDIGIT(*RExC_parse))
5274 if (*RExC_parse!=')')
5275 vFAIL("Expecting close bracket");
5278 if ( paren == '-' ) {
5280 Diagram of capture buffer numbering.
5281 Top line is the normal capture buffer numbers
5282 Botton line is the negative indexing as from
5286 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5290 num = RExC_npar + num;
5293 vFAIL("Reference to nonexistent group");
5295 } else if ( paren == '+' ) {
5296 num = RExC_npar + num - 1;
5299 ret = reganode(pRExC_state, GOSUB, num);
5301 if (num > (I32)RExC_rx->nparens) {
5303 vFAIL("Reference to nonexistent group");
5305 ARG2L_SET( ret, RExC_recurse_count++);
5307 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5308 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5312 RExC_seen |= REG_SEEN_RECURSE;
5313 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5314 Set_Node_Offset(ret, parse_start); /* MJD */
5316 *flagp |= POSTPONED;
5317 nextchar(pRExC_state);
5319 } /* named and numeric backreferences */
5322 case 'p': /* (?p...) */
5323 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
5324 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
5326 case '?': /* (??...) */
5328 if (*RExC_parse != '{') {
5330 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5333 *flagp |= POSTPONED;
5334 paren = *RExC_parse++;
5336 case '{': /* (?{...}) */
5341 char *s = RExC_parse;
5343 RExC_seen_zerolen++;
5344 RExC_seen |= REG_SEEN_EVAL;
5345 while (count && (c = *RExC_parse)) {
5356 if (*RExC_parse != ')') {
5358 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5362 OP_4tree *sop, *rop;
5363 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5366 Perl_save_re_context(aTHX);
5367 rop = sv_compile_2op(sv, &sop, "re", &pad);
5368 sop->op_private |= OPpREFCOUNTED;
5369 /* re_dup will OpREFCNT_inc */
5370 OpREFCNT_set(sop, 1);
5373 n = add_data(pRExC_state, 3, "nop");
5374 RExC_rxi->data->data[n] = (void*)rop;
5375 RExC_rxi->data->data[n+1] = (void*)sop;
5376 RExC_rxi->data->data[n+2] = (void*)pad;
5379 else { /* First pass */
5380 if (PL_reginterp_cnt < ++RExC_seen_evals
5382 /* No compiled RE interpolated, has runtime
5383 components ===> unsafe. */
5384 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5385 if (PL_tainting && PL_tainted)
5386 FAIL("Eval-group in insecure regular expression");
5387 #if PERL_VERSION > 8
5388 if (IN_PERL_COMPILETIME)
5393 nextchar(pRExC_state);
5395 ret = reg_node(pRExC_state, LOGICAL);
5398 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5399 /* deal with the length of this later - MJD */
5402 ret = reganode(pRExC_state, EVAL, n);
5403 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5404 Set_Node_Offset(ret, parse_start);
5407 case '(': /* (?(?{...})...) and (?(?=...)...) */
5410 if (RExC_parse[0] == '?') { /* (?(?...)) */
5411 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5412 || RExC_parse[1] == '<'
5413 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5416 ret = reg_node(pRExC_state, LOGICAL);
5419 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5423 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5424 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5426 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5427 char *name_start= RExC_parse++;
5429 SV *sv_dat=reg_scan_name(pRExC_state,
5430 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5431 if (RExC_parse == name_start || *RExC_parse != ch)
5432 vFAIL2("Sequence (?(%c... not terminated",
5433 (ch == '>' ? '<' : ch));
5436 num = add_data( pRExC_state, 1, "S" );
5437 RExC_rxi->data->data[num]=(void*)sv_dat;
5438 SvREFCNT_inc(sv_dat);
5440 ret = reganode(pRExC_state,NGROUPP,num);
5441 goto insert_if_check_paren;
5443 else if (RExC_parse[0] == 'D' &&
5444 RExC_parse[1] == 'E' &&
5445 RExC_parse[2] == 'F' &&
5446 RExC_parse[3] == 'I' &&
5447 RExC_parse[4] == 'N' &&
5448 RExC_parse[5] == 'E')
5450 ret = reganode(pRExC_state,DEFINEP,0);
5453 goto insert_if_check_paren;
5455 else if (RExC_parse[0] == 'R') {
5458 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5459 parno = atoi(RExC_parse++);
5460 while (isDIGIT(*RExC_parse))
5462 } else if (RExC_parse[0] == '&') {
5465 sv_dat = reg_scan_name(pRExC_state,
5466 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5467 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5469 ret = reganode(pRExC_state,INSUBP,parno);
5470 goto insert_if_check_paren;
5472 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5475 parno = atoi(RExC_parse++);
5477 while (isDIGIT(*RExC_parse))
5479 ret = reganode(pRExC_state, GROUPP, parno);
5481 insert_if_check_paren:
5482 if ((c = *nextchar(pRExC_state)) != ')')
5483 vFAIL("Switch condition not recognized");
5485 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5486 br = regbranch(pRExC_state, &flags, 1,depth+1);
5488 br = reganode(pRExC_state, LONGJMP, 0);
5490 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5491 c = *nextchar(pRExC_state);
5496 vFAIL("(?(DEFINE)....) does not allow branches");
5497 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5498 regbranch(pRExC_state, &flags, 1,depth+1);
5499 REGTAIL(pRExC_state, ret, lastbr);
5502 c = *nextchar(pRExC_state);
5507 vFAIL("Switch (?(condition)... contains too many branches");
5508 ender = reg_node(pRExC_state, TAIL);
5509 REGTAIL(pRExC_state, br, ender);
5511 REGTAIL(pRExC_state, lastbr, ender);
5512 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5515 REGTAIL(pRExC_state, ret, ender);
5516 RExC_size++; /* XXX WHY do we need this?!!
5517 For large programs it seems to be required
5518 but I can't figure out why. -- dmq*/
5522 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5526 RExC_parse--; /* for vFAIL to print correctly */
5527 vFAIL("Sequence (? incomplete");
5531 parse_flags: /* (?i) */
5533 U32 posflags = 0, negflags = 0;
5534 U32 *flagsp = &posflags;
5536 while (*RExC_parse) {
5537 /* && strchr("iogcmsx", *RExC_parse) */
5538 /* (?g), (?gc) and (?o) are useless here
5539 and must be globally applied -- japhy */
5540 switch (*RExC_parse) {
5541 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5544 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5545 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5546 if (! (wastedflags & wflagbit) ) {
5547 wastedflags |= wflagbit;
5550 "Useless (%s%c) - %suse /%c modifier",
5551 flagsp == &negflags ? "?-" : "?",
5553 flagsp == &negflags ? "don't " : "",
5561 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5562 if (! (wastedflags & WASTED_C) ) {
5563 wastedflags |= WASTED_GC;
5566 "Useless (%sc) - %suse /gc modifier",
5567 flagsp == &negflags ? "?-" : "?",
5568 flagsp == &negflags ? "don't " : ""
5574 if (flagsp == &negflags) {
5575 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5576 vWARN(RExC_parse + 1,"Useless use of (?-k)");
5578 *flagsp |= RXf_PMf_KEEPCOPY;
5582 if (flagsp == &negflags) {
5584 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5588 wastedflags = 0; /* reset so (?g-c) warns twice */
5594 RExC_flags |= posflags;
5595 RExC_flags &= ~negflags;
5596 nextchar(pRExC_state);
5607 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5612 }} /* one for the default block, one for the switch */
5619 ret = reganode(pRExC_state, OPEN, parno);
5622 RExC_nestroot = parno;
5623 if (RExC_seen & REG_SEEN_RECURSE
5624 && !RExC_open_parens[parno-1])
5626 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5627 "Setting open paren #%"IVdf" to %d\n",
5628 (IV)parno, REG_NODE_NUM(ret)));
5629 RExC_open_parens[parno-1]= ret;
5632 Set_Node_Length(ret, 1); /* MJD */
5633 Set_Node_Offset(ret, RExC_parse); /* MJD */
5641 /* Pick up the branches, linking them together. */
5642 parse_start = RExC_parse; /* MJD */
5643 br = regbranch(pRExC_state, &flags, 1,depth+1);
5644 /* branch_len = (paren != 0); */
5648 if (*RExC_parse == '|') {
5649 if (!SIZE_ONLY && RExC_extralen) {
5650 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5653 reginsert(pRExC_state, BRANCH, br, depth+1);
5654 Set_Node_Length(br, paren != 0);
5655 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5659 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5661 else if (paren == ':') {
5662 *flagp |= flags&SIMPLE;
5664 if (is_open) { /* Starts with OPEN. */
5665 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5667 else if (paren != '?') /* Not Conditional */
5669 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5671 while (*RExC_parse == '|') {
5672 if (!SIZE_ONLY && RExC_extralen) {
5673 ender = reganode(pRExC_state, LONGJMP,0);
5674 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5677 RExC_extralen += 2; /* Account for LONGJMP. */
5678 nextchar(pRExC_state);
5680 if (RExC_npar > after_freeze)
5681 after_freeze = RExC_npar;
5682 RExC_npar = freeze_paren;
5684 br = regbranch(pRExC_state, &flags, 0, depth+1);
5688 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5690 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5693 if (have_branch || paren != ':') {
5694 /* Make a closing node, and hook it on the end. */
5697 ender = reg_node(pRExC_state, TAIL);
5700 ender = reganode(pRExC_state, CLOSE, parno);
5701 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5702 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5703 "Setting close paren #%"IVdf" to %d\n",
5704 (IV)parno, REG_NODE_NUM(ender)));
5705 RExC_close_parens[parno-1]= ender;
5706 if (RExC_nestroot == parno)
5709 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5710 Set_Node_Length(ender,1); /* MJD */
5716 *flagp &= ~HASWIDTH;
5719 ender = reg_node(pRExC_state, SUCCEED);
5722 ender = reg_node(pRExC_state, END);
5724 assert(!RExC_opend); /* there can only be one! */
5729 REGTAIL(pRExC_state, lastbr, ender);
5731 if (have_branch && !SIZE_ONLY) {
5733 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5735 /* Hook the tails of the branches to the closing node. */
5736 for (br = ret; br; br = regnext(br)) {
5737 const U8 op = PL_regkind[OP(br)];
5739 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5741 else if (op == BRANCHJ) {
5742 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5750 static const char parens[] = "=!<,>";
5752 if (paren && (p = strchr(parens, paren))) {
5753 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5754 int flag = (p - parens) > 1;
5757 node = SUSPEND, flag = 0;
5758 reginsert(pRExC_state, node,ret, depth+1);
5759 Set_Node_Cur_Length(ret);
5760 Set_Node_Offset(ret, parse_start + 1);
5762 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5766 /* Check for proper termination. */
5768 RExC_flags = oregflags;
5769 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5770 RExC_parse = oregcomp_parse;
5771 vFAIL("Unmatched (");
5774 else if (!paren && RExC_parse < RExC_end) {
5775 if (*RExC_parse == ')') {
5777 vFAIL("Unmatched )");
5780 FAIL("Junk on end of regexp"); /* "Can't happen". */
5784 RExC_npar = after_freeze;
5789 - regbranch - one alternative of an | operator
5791 * Implements the concatenation operator.
5794 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5797 register regnode *ret;
5798 register regnode *chain = NULL;
5799 register regnode *latest;
5800 I32 flags = 0, c = 0;
5801 GET_RE_DEBUG_FLAGS_DECL;
5802 DEBUG_PARSE("brnc");
5806 if (!SIZE_ONLY && RExC_extralen)
5807 ret = reganode(pRExC_state, BRANCHJ,0);
5809 ret = reg_node(pRExC_state, BRANCH);
5810 Set_Node_Length(ret, 1);
5814 if (!first && SIZE_ONLY)
5815 RExC_extralen += 1; /* BRANCHJ */
5817 *flagp = WORST; /* Tentatively. */
5820 nextchar(pRExC_state);
5821 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5823 latest = regpiece(pRExC_state, &flags,depth+1);
5824 if (latest == NULL) {
5825 if (flags & TRYAGAIN)
5829 else if (ret == NULL)
5831 *flagp |= flags&(HASWIDTH|POSTPONED);
5832 if (chain == NULL) /* First piece. */
5833 *flagp |= flags&SPSTART;
5836 REGTAIL(pRExC_state, chain, latest);
5841 if (chain == NULL) { /* Loop ran zero times. */
5842 chain = reg_node(pRExC_state, NOTHING);
5847 *flagp |= flags&SIMPLE;
5854 - regpiece - something followed by possible [*+?]
5856 * Note that the branching code sequences used for ? and the general cases
5857 * of * and + are somewhat optimized: they use the same NOTHING node as
5858 * both the endmarker for their branch list and the body of the last branch.
5859 * It might seem that this node could be dispensed with entirely, but the
5860 * endmarker role is not redundant.
5863 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5866 register regnode *ret;
5868 register char *next;
5870 const char * const origparse = RExC_parse;
5872 I32 max = REG_INFTY;
5874 const char *maxpos = NULL;
5875 GET_RE_DEBUG_FLAGS_DECL;
5876 DEBUG_PARSE("piec");
5878 ret = regatom(pRExC_state, &flags,depth+1);
5880 if (flags & TRYAGAIN)
5887 if (op == '{' && regcurly(RExC_parse)) {
5889 parse_start = RExC_parse; /* MJD */
5890 next = RExC_parse + 1;
5891 while (isDIGIT(*next) || *next == ',') {
5900 if (*next == '}') { /* got one */
5904 min = atoi(RExC_parse);
5908 maxpos = RExC_parse;
5910 if (!max && *maxpos != '0')
5911 max = REG_INFTY; /* meaning "infinity" */
5912 else if (max >= REG_INFTY)
5913 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5915 nextchar(pRExC_state);
5918 if ((flags&SIMPLE)) {
5919 RExC_naughty += 2 + RExC_naughty / 2;
5920 reginsert(pRExC_state, CURLY, ret, depth+1);
5921 Set_Node_Offset(ret, parse_start+1); /* MJD */
5922 Set_Node_Cur_Length(ret);
5925 regnode * const w = reg_node(pRExC_state, WHILEM);
5928 REGTAIL(pRExC_state, ret, w);
5929 if (!SIZE_ONLY && RExC_extralen) {
5930 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5931 reginsert(pRExC_state, NOTHING,ret, depth+1);
5932 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5934 reginsert(pRExC_state, CURLYX,ret, depth+1);
5936 Set_Node_Offset(ret, parse_start+1);
5937 Set_Node_Length(ret,
5938 op == '{' ? (RExC_parse - parse_start) : 1);
5940 if (!SIZE_ONLY && RExC_extralen)
5941 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5942 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5944 RExC_whilem_seen++, RExC_extralen += 3;
5945 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5953 if (max && max < min)
5954 vFAIL("Can't do {n,m} with n > m");
5956 ARG1_SET(ret, (U16)min);
5957 ARG2_SET(ret, (U16)max);
5969 #if 0 /* Now runtime fix should be reliable. */
5971 /* if this is reinstated, don't forget to put this back into perldiag:
5973 =item Regexp *+ operand could be empty at {#} in regex m/%s/
5975 (F) The part of the regexp subject to either the * or + quantifier
5976 could match an empty string. The {#} shows in the regular
5977 expression about where the problem was discovered.
5981 if (!(flags&HASWIDTH) && op != '?')
5982 vFAIL("Regexp *+ operand could be empty");
5985 parse_start = RExC_parse;
5986 nextchar(pRExC_state);
5988 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
5990 if (op == '*' && (flags&SIMPLE)) {
5991 reginsert(pRExC_state, STAR, ret, depth+1);
5995 else if (op == '*') {
5999 else if (op == '+' && (flags&SIMPLE)) {
6000 reginsert(pRExC_state, PLUS, ret, depth+1);
6004 else if (op == '+') {
6008 else if (op == '?') {
6013 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6015 "%.*s matches null string many times",
6016 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6020 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6021 nextchar(pRExC_state);
6022 reginsert(pRExC_state, MINMOD, ret, depth+1);
6023 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6025 #ifndef REG_ALLOW_MINMOD_SUSPEND
6028 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6030 nextchar(pRExC_state);
6031 ender = reg_node(pRExC_state, SUCCEED);
6032 REGTAIL(pRExC_state, ret, ender);
6033 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6035 ender = reg_node(pRExC_state, TAIL);
6036 REGTAIL(pRExC_state, ret, ender);
6040 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6042 vFAIL("Nested quantifiers");
6049 /* reg_namedseq(pRExC_state,UVp)
6051 This is expected to be called by a parser routine that has
6052 recognized'\N' and needs to handle the rest. RExC_parse is
6053 expected to point at the first char following the N at the time
6056 If valuep is non-null then it is assumed that we are parsing inside
6057 of a charclass definition and the first codepoint in the resolved
6058 string is returned via *valuep and the routine will return NULL.
6059 In this mode if a multichar string is returned from the charnames
6060 handler a warning will be issued, and only the first char in the
6061 sequence will be examined. If the string returned is zero length
6062 then the value of *valuep is undefined and NON-NULL will
6063 be returned to indicate failure. (This will NOT be a valid pointer
6066 If value is null then it is assumed that we are parsing normal text
6067 and inserts a new EXACT node into the program containing the resolved
6068 string and returns a pointer to the new node. If the string is
6069 zerolength a NOTHING node is emitted.
6071 On success RExC_parse is set to the char following the endbrace.
6072 Parsing failures will generate a fatal errorvia vFAIL(...)
6074 NOTE: We cache all results from the charnames handler locally in
6075 the RExC_charnames hash (created on first use) to prevent a charnames
6076 handler from playing silly-buggers and returning a short string and
6077 then a long string for a given pattern. Since the regexp program
6078 size is calculated during an initial parse this would result
6079 in a buffer overrun so we cache to prevent the charname result from
6080 changing during the course of the parse.
6084 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6086 char * name; /* start of the content of the name */
6087 char * endbrace; /* endbrace following the name */
6090 STRLEN len; /* this has various purposes throughout the code */
6091 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6092 regnode *ret = NULL;
6094 if (*RExC_parse != '{') {
6095 vFAIL("Missing braces on \\N{}");
6097 name = RExC_parse+1;
6098 endbrace = strchr(RExC_parse, '}');
6101 vFAIL("Missing right brace on \\N{}");
6103 RExC_parse = endbrace + 1;
6106 /* RExC_parse points at the beginning brace,
6107 endbrace points at the last */
6108 if ( name[0]=='U' && name[1]=='+' ) {
6109 /* its a "unicode hex" notation {U+89AB} */
6110 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6111 | PERL_SCAN_DISALLOW_PREFIX
6112 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6114 len = (STRLEN)(endbrace - name - 2);
6115 cp = grok_hex(name + 2, &len, &fl, NULL);
6116 if ( len != (STRLEN)(endbrace - name - 2) ) {
6125 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6127 /* fetch the charnames handler for this scope */
6128 HV * const table = GvHV(PL_hintgv);
6130 hv_fetchs(table, "charnames", FALSE) :
6132 SV *cv= cvp ? *cvp : NULL;
6135 /* create an SV with the name as argument */
6136 sv_name = newSVpvn(name, endbrace - name);
6138 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6139 vFAIL2("Constant(\\N{%s}) unknown: "
6140 "(possibly a missing \"use charnames ...\")",
6143 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6144 vFAIL2("Constant(\\N{%s}): "
6145 "$^H{charnames} is not defined",SvPVX(sv_name));
6150 if (!RExC_charnames) {
6151 /* make sure our cache is allocated */
6152 RExC_charnames = newHV();
6153 sv_2mortal((SV*)RExC_charnames);
6155 /* see if we have looked this one up before */
6156 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6158 sv_str = HeVAL(he_str);
6171 count= call_sv(cv, G_SCALAR);
6173 if (count == 1) { /* XXXX is this right? dmq */
6175 SvREFCNT_inc_simple_void(sv_str);
6183 if ( !sv_str || !SvOK(sv_str) ) {
6184 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6185 "did not return a defined value",SvPVX(sv_name));
6187 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6192 char *p = SvPV(sv_str, len);
6195 if ( SvUTF8(sv_str) ) {
6196 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6200 We have to turn on utf8 for high bit chars otherwise
6201 we get failures with
6203 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6204 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6206 This is different from what \x{} would do with the same
6207 codepoint, where the condition is > 0xFF.
6214 /* warn if we havent used the whole string? */
6216 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6218 "Ignoring excess chars from \\N{%s} in character class",
6222 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6224 "Ignoring zero length \\N{%s} in character class",
6229 SvREFCNT_dec(sv_name);
6231 SvREFCNT_dec(sv_str);
6232 return len ? NULL : (regnode *)&len;
6233 } else if(SvCUR(sv_str)) {
6239 char * parse_start = name-3; /* needed for the offsets */
6241 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6243 ret = reg_node(pRExC_state,
6244 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6247 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6248 sv_utf8_upgrade(sv_str);
6249 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6253 p = SvPV(sv_str, len);
6255 /* len is the length written, charlen is the size the char read */
6256 for ( len = 0; p < pend; p += charlen ) {
6258 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6260 STRLEN foldlen,numlen;
6261 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6262 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6263 /* Emit all the Unicode characters. */
6265 for (foldbuf = tmpbuf;
6269 uvc = utf8_to_uvchr(foldbuf, &numlen);
6271 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6274 /* In EBCDIC the numlen
6275 * and unilen can differ. */
6277 if (numlen >= foldlen)
6281 break; /* "Can't happen." */
6284 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6296 RExC_size += STR_SZ(len);
6299 RExC_emit += STR_SZ(len);
6301 Set_Node_Cur_Length(ret); /* MJD */
6303 nextchar(pRExC_state);
6305 ret = reg_node(pRExC_state,NOTHING);
6308 SvREFCNT_dec(sv_str);
6311 SvREFCNT_dec(sv_name);
6321 * It returns the code point in utf8 for the value in *encp.
6322 * value: a code value in the source encoding
6323 * encp: a pointer to an Encode object
6325 * If the result from Encode is not a single character,
6326 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6329 S_reg_recode(pTHX_ const char value, SV **encp)
6332 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6333 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6335 const STRLEN newlen = SvCUR(sv);
6336 UV uv = UNICODE_REPLACEMENT;
6340 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6343 if (!newlen || numlen != newlen) {
6344 uv = UNICODE_REPLACEMENT;
6353 - regatom - the lowest level
6355 Try to identify anything special at the start of the pattern. If there
6356 is, then handle it as required. This may involve generating a single regop,
6357 such as for an assertion; or it may involve recursing, such as to
6358 handle a () structure.
6360 If the string doesn't start with something special then we gobble up
6361 as much literal text as we can.
6363 Once we have been able to handle whatever type of thing started the
6364 sequence, we return.
6366 Note: we have to be careful with escapes, as they can be both literal
6367 and special, and in the case of \10 and friends can either, depending
6368 on context. Specifically there are two seperate switches for handling
6369 escape sequences, with the one for handling literal escapes requiring
6370 a dummy entry for all of the special escapes that are actually handled
6375 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6378 register regnode *ret = NULL;
6380 char *parse_start = RExC_parse;
6381 GET_RE_DEBUG_FLAGS_DECL;
6382 DEBUG_PARSE("atom");
6383 *flagp = WORST; /* Tentatively. */
6387 switch (*RExC_parse) {
6389 RExC_seen_zerolen++;
6390 nextchar(pRExC_state);
6391 if (RExC_flags & RXf_PMf_MULTILINE)
6392 ret = reg_node(pRExC_state, MBOL);
6393 else if (RExC_flags & RXf_PMf_SINGLELINE)
6394 ret = reg_node(pRExC_state, SBOL);
6396 ret = reg_node(pRExC_state, BOL);
6397 Set_Node_Length(ret, 1); /* MJD */
6400 nextchar(pRExC_state);
6402 RExC_seen_zerolen++;
6403 if (RExC_flags & RXf_PMf_MULTILINE)
6404 ret = reg_node(pRExC_state, MEOL);
6405 else if (RExC_flags & RXf_PMf_SINGLELINE)
6406 ret = reg_node(pRExC_state, SEOL);
6408 ret = reg_node(pRExC_state, EOL);
6409 Set_Node_Length(ret, 1); /* MJD */
6412 nextchar(pRExC_state);
6413 if (RExC_flags & RXf_PMf_SINGLELINE)
6414 ret = reg_node(pRExC_state, SANY);
6416 ret = reg_node(pRExC_state, REG_ANY);
6417 *flagp |= HASWIDTH|SIMPLE;
6419 Set_Node_Length(ret, 1); /* MJD */
6423 char * const oregcomp_parse = ++RExC_parse;
6424 ret = regclass(pRExC_state,depth+1);
6425 if (*RExC_parse != ']') {
6426 RExC_parse = oregcomp_parse;
6427 vFAIL("Unmatched [");
6429 nextchar(pRExC_state);
6430 *flagp |= HASWIDTH|SIMPLE;
6431 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6435 nextchar(pRExC_state);
6436 ret = reg(pRExC_state, 1, &flags,depth+1);
6438 if (flags & TRYAGAIN) {
6439 if (RExC_parse == RExC_end) {
6440 /* Make parent create an empty node if needed. */
6448 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6452 if (flags & TRYAGAIN) {
6456 vFAIL("Internal urp");
6457 /* Supposed to be caught earlier. */
6460 if (!regcurly(RExC_parse)) {
6469 vFAIL("Quantifier follows nothing");
6474 This switch handles escape sequences that resolve to some kind
6475 of special regop and not to literal text. Escape sequnces that
6476 resolve to literal text are handled below in the switch marked
6479 Every entry in this switch *must* have a corresponding entry
6480 in the literal escape switch. However, the opposite is not
6481 required, as the default for this switch is to jump to the
6482 literal text handling code.
6484 switch (*++RExC_parse) {
6485 /* Special Escapes */
6487 RExC_seen_zerolen++;
6488 ret = reg_node(pRExC_state, SBOL);
6490 goto finish_meta_pat;
6492 ret = reg_node(pRExC_state, GPOS);
6493 RExC_seen |= REG_SEEN_GPOS;
6495 goto finish_meta_pat;
6497 RExC_seen_zerolen++;
6498 ret = reg_node(pRExC_state, KEEPS);
6500 goto finish_meta_pat;
6502 ret = reg_node(pRExC_state, SEOL);
6504 RExC_seen_zerolen++; /* Do not optimize RE away */
6505 goto finish_meta_pat;
6507 ret = reg_node(pRExC_state, EOS);
6509 RExC_seen_zerolen++; /* Do not optimize RE away */
6510 goto finish_meta_pat;
6512 ret = reg_node(pRExC_state, CANY);
6513 RExC_seen |= REG_SEEN_CANY;
6514 *flagp |= HASWIDTH|SIMPLE;
6515 goto finish_meta_pat;
6517 ret = reg_node(pRExC_state, CLUMP);
6519 goto finish_meta_pat;
6521 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6522 *flagp |= HASWIDTH|SIMPLE;
6523 goto finish_meta_pat;
6525 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6526 *flagp |= HASWIDTH|SIMPLE;
6527 goto finish_meta_pat;
6529 RExC_seen_zerolen++;
6530 RExC_seen |= REG_SEEN_LOOKBEHIND;
6531 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6533 goto finish_meta_pat;
6535 RExC_seen_zerolen++;
6536 RExC_seen |= REG_SEEN_LOOKBEHIND;
6537 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6539 goto finish_meta_pat;
6541 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6542 *flagp |= HASWIDTH|SIMPLE;
6543 goto finish_meta_pat;
6545 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6546 *flagp |= HASWIDTH|SIMPLE;
6547 goto finish_meta_pat;
6549 ret = reg_node(pRExC_state, DIGIT);
6550 *flagp |= HASWIDTH|SIMPLE;
6551 goto finish_meta_pat;
6553 ret = reg_node(pRExC_state, NDIGIT);
6554 *flagp |= HASWIDTH|SIMPLE;
6555 goto finish_meta_pat;
6557 ret = reganode(pRExC_state, PRUNE, 0);
6560 goto finish_meta_pat;
6562 ret = reganode(pRExC_state, SKIP, 0);
6566 nextchar(pRExC_state);
6567 Set_Node_Length(ret, 2); /* MJD */
6572 char* const oldregxend = RExC_end;
6574 char* parse_start = RExC_parse - 2;
6577 if (RExC_parse[1] == '{') {
6578 /* a lovely hack--pretend we saw [\pX] instead */
6579 RExC_end = strchr(RExC_parse, '}');
6581 const U8 c = (U8)*RExC_parse;
6583 RExC_end = oldregxend;
6584 vFAIL2("Missing right brace on \\%c{}", c);
6589 RExC_end = RExC_parse + 2;
6590 if (RExC_end > oldregxend)
6591 RExC_end = oldregxend;
6595 ret = regclass(pRExC_state,depth+1);
6597 RExC_end = oldregxend;
6600 Set_Node_Offset(ret, parse_start + 2);
6601 Set_Node_Cur_Length(ret);
6602 nextchar(pRExC_state);
6603 *flagp |= HASWIDTH|SIMPLE;
6607 /* Handle \N{NAME} here and not below because it can be
6608 multicharacter. join_exact() will join them up later on.
6609 Also this makes sure that things like /\N{BLAH}+/ and
6610 \N{BLAH} being multi char Just Happen. dmq*/
6612 ret= reg_namedseq(pRExC_state, NULL);
6614 case 'k': /* Handle \k<NAME> and \k'NAME' */
6617 char ch= RExC_parse[1];
6618 if (ch != '<' && ch != '\'' && ch != '{') {
6620 vFAIL2("Sequence %.2s... not terminated",parse_start);
6622 /* this pretty much dupes the code for (?P=...) in reg(), if
6623 you change this make sure you change that */
6624 char* name_start = (RExC_parse += 2);
6626 SV *sv_dat = reg_scan_name(pRExC_state,
6627 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6628 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
6629 if (RExC_parse == name_start || *RExC_parse != ch)
6630 vFAIL2("Sequence %.3s... not terminated",parse_start);
6633 num = add_data( pRExC_state, 1, "S" );
6634 RExC_rxi->data->data[num]=(void*)sv_dat;
6635 SvREFCNT_inc(sv_dat);
6639 ret = reganode(pRExC_state,
6640 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6644 /* override incorrect value set in reganode MJD */
6645 Set_Node_Offset(ret, parse_start+1);
6646 Set_Node_Cur_Length(ret); /* MJD */
6647 nextchar(pRExC_state);
6653 case '1': case '2': case '3': case '4':
6654 case '5': case '6': case '7': case '8': case '9':
6657 bool isg = *RExC_parse == 'g';
6662 if (*RExC_parse == '{') {
6666 if (*RExC_parse == '-') {
6670 if (hasbrace && !isDIGIT(*RExC_parse)) {
6671 if (isrel) RExC_parse--;
6673 goto parse_named_seq;
6675 num = atoi(RExC_parse);
6677 num = RExC_npar - num;
6679 vFAIL("Reference to nonexistent or unclosed group");
6681 if (!isg && num > 9 && num >= RExC_npar)
6684 char * const parse_start = RExC_parse - 1; /* MJD */
6685 while (isDIGIT(*RExC_parse))
6687 if (parse_start == RExC_parse - 1)
6688 vFAIL("Unterminated \\g... pattern");
6690 if (*RExC_parse != '}')
6691 vFAIL("Unterminated \\g{...} pattern");
6695 if (num > (I32)RExC_rx->nparens)
6696 vFAIL("Reference to nonexistent group");
6699 ret = reganode(pRExC_state,
6700 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6704 /* override incorrect value set in reganode MJD */
6705 Set_Node_Offset(ret, parse_start+1);
6706 Set_Node_Cur_Length(ret); /* MJD */
6708 nextchar(pRExC_state);
6713 if (RExC_parse >= RExC_end)
6714 FAIL("Trailing \\");
6717 /* Do not generate "unrecognized" warnings here, we fall
6718 back into the quick-grab loop below */
6725 if (RExC_flags & RXf_PMf_EXTENDED) {
6726 if ( reg_skipcomment( pRExC_state ) )
6732 register STRLEN len;
6737 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6739 parse_start = RExC_parse - 1;
6745 ret = reg_node(pRExC_state,
6746 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6748 for (len = 0, p = RExC_parse - 1;
6749 len < 127 && p < RExC_end;
6752 char * const oldp = p;
6754 if (RExC_flags & RXf_PMf_EXTENDED)
6755 p = regwhite( pRExC_state, p );
6766 /* Literal Escapes Switch
6768 This switch is meant to handle escape sequences that
6769 resolve to a literal character.
6771 Every escape sequence that represents something
6772 else, like an assertion or a char class, is handled
6773 in the switch marked 'Special Escapes' above in this
6774 routine, but also has an entry here as anything that
6775 isn't explicitly mentioned here will be treated as
6776 an unescaped equivalent literal.
6780 /* These are all the special escapes. */
6781 case 'A': /* Start assertion */
6782 case 'b': case 'B': /* Word-boundary assertion*/
6783 case 'C': /* Single char !DANGEROUS! */
6784 case 'd': case 'D': /* digit class */
6785 case 'g': case 'G': /* generic-backref, pos assertion */
6786 case 'k': case 'K': /* named backref, keep marker */
6787 case 'N': /* named char sequence */
6788 case 'p': case 'P': /* unicode property */
6789 case 's': case 'S': /* space class */
6790 case 'v': case 'V': /* (*PRUNE) and (*SKIP) */
6791 case 'w': case 'W': /* word class */
6792 case 'X': /* eXtended Unicode "combining character sequence" */
6793 case 'z': case 'Z': /* End of line/string assertion */
6797 /* Anything after here is an escape that resolves to a
6798 literal. (Except digits, which may or may not)
6817 ender = ASCII_TO_NATIVE('\033');
6821 ender = ASCII_TO_NATIVE('\007');
6826 char* const e = strchr(p, '}');
6830 vFAIL("Missing right brace on \\x{}");
6833 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6834 | PERL_SCAN_DISALLOW_PREFIX;
6835 STRLEN numlen = e - p - 1;
6836 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6843 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6845 ender = grok_hex(p, &numlen, &flags, NULL);
6848 if (PL_encoding && ender < 0x100)
6849 goto recode_encoding;
6853 ender = UCHARAT(p++);
6854 ender = toCTRL(ender);
6856 case '0': case '1': case '2': case '3':case '4':
6857 case '5': case '6': case '7': case '8':case '9':
6859 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6862 ender = grok_oct(p, &numlen, &flags, NULL);
6869 if (PL_encoding && ender < 0x100)
6870 goto recode_encoding;
6874 SV* enc = PL_encoding;
6875 ender = reg_recode((const char)(U8)ender, &enc);
6876 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6877 vWARN(p, "Invalid escape in the specified encoding");
6883 FAIL("Trailing \\");
6886 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6887 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6888 goto normal_default;
6893 if (UTF8_IS_START(*p) && UTF) {
6895 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6896 &numlen, UTF8_ALLOW_DEFAULT);
6903 if ( RExC_flags & RXf_PMf_EXTENDED)
6904 p = regwhite( pRExC_state, p );
6906 /* Prime the casefolded buffer. */
6907 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6909 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
6914 /* Emit all the Unicode characters. */
6916 for (foldbuf = tmpbuf;
6918 foldlen -= numlen) {
6919 ender = utf8_to_uvchr(foldbuf, &numlen);
6921 const STRLEN unilen = reguni(pRExC_state, ender, s);
6924 /* In EBCDIC the numlen
6925 * and unilen can differ. */
6927 if (numlen >= foldlen)
6931 break; /* "Can't happen." */
6935 const STRLEN unilen = reguni(pRExC_state, ender, s);
6944 REGC((char)ender, s++);
6950 /* Emit all the Unicode characters. */
6952 for (foldbuf = tmpbuf;
6954 foldlen -= numlen) {
6955 ender = utf8_to_uvchr(foldbuf, &numlen);
6957 const STRLEN unilen = reguni(pRExC_state, ender, s);
6960 /* In EBCDIC the numlen
6961 * and unilen can differ. */
6963 if (numlen >= foldlen)
6971 const STRLEN unilen = reguni(pRExC_state, ender, s);
6980 REGC((char)ender, s++);
6984 Set_Node_Cur_Length(ret); /* MJD */
6985 nextchar(pRExC_state);
6987 /* len is STRLEN which is unsigned, need to copy to signed */
6990 vFAIL("Internal disaster");
6994 if (len == 1 && UNI_IS_INVARIANT(ender))
6998 RExC_size += STR_SZ(len);
7001 RExC_emit += STR_SZ(len);
7011 S_regwhite( RExC_state_t *pRExC_state, char *p )
7013 const char *e = RExC_end;
7017 else if (*p == '#') {
7026 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7034 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7035 Character classes ([:foo:]) can also be negated ([:^foo:]).
7036 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7037 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7038 but trigger failures because they are currently unimplemented. */
7040 #define POSIXCC_DONE(c) ((c) == ':')
7041 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7042 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7045 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7048 I32 namedclass = OOB_NAMEDCLASS;
7050 if (value == '[' && RExC_parse + 1 < RExC_end &&
7051 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7052 POSIXCC(UCHARAT(RExC_parse))) {
7053 const char c = UCHARAT(RExC_parse);
7054 char* const s = RExC_parse++;
7056 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7058 if (RExC_parse == RExC_end)
7059 /* Grandfather lone [:, [=, [. */
7062 const char* const t = RExC_parse++; /* skip over the c */
7065 if (UCHARAT(RExC_parse) == ']') {
7066 const char *posixcc = s + 1;
7067 RExC_parse++; /* skip over the ending ] */
7070 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7071 const I32 skip = t - posixcc;
7073 /* Initially switch on the length of the name. */
7076 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7077 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7080 /* Names all of length 5. */
7081 /* alnum alpha ascii blank cntrl digit graph lower
7082 print punct space upper */
7083 /* Offset 4 gives the best switch position. */
7084 switch (posixcc[4]) {
7086 if (memEQ(posixcc, "alph", 4)) /* alpha */
7087 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7090 if (memEQ(posixcc, "spac", 4)) /* space */
7091 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7094 if (memEQ(posixcc, "grap", 4)) /* graph */
7095 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7098 if (memEQ(posixcc, "asci", 4)) /* ascii */
7099 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7102 if (memEQ(posixcc, "blan", 4)) /* blank */
7103 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7106 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7107 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7110 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7111 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7114 if (memEQ(posixcc, "lowe", 4)) /* lower */
7115 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7116 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7117 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7120 if (memEQ(posixcc, "digi", 4)) /* digit */
7121 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7122 else if (memEQ(posixcc, "prin", 4)) /* print */
7123 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7124 else if (memEQ(posixcc, "punc", 4)) /* punct */
7125 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7130 if (memEQ(posixcc, "xdigit", 6))
7131 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7135 if (namedclass == OOB_NAMEDCLASS)
7136 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7138 assert (posixcc[skip] == ':');
7139 assert (posixcc[skip+1] == ']');
7140 } else if (!SIZE_ONLY) {
7141 /* [[=foo=]] and [[.foo.]] are still future. */
7143 /* adjust RExC_parse so the warning shows after
7145 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7147 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7150 /* Maternal grandfather:
7151 * "[:" ending in ":" but not in ":]" */
7161 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7164 if (POSIXCC(UCHARAT(RExC_parse))) {
7165 const char *s = RExC_parse;
7166 const char c = *s++;
7170 if (*s && c == *s && s[1] == ']') {
7171 if (ckWARN(WARN_REGEXP))
7173 "POSIX syntax [%c %c] belongs inside character classes",
7176 /* [[=foo=]] and [[.foo.]] are still future. */
7177 if (POSIXCC_NOTYET(c)) {
7178 /* adjust RExC_parse so the error shows after
7180 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7182 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7189 #define _C_C_T_(NAME,TEST,WORD) \
7192 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7194 for (value = 0; value < 256; value++) \
7196 ANYOF_BITMAP_SET(ret, value); \
7201 case ANYOF_N##NAME: \
7203 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7205 for (value = 0; value < 256; value++) \
7207 ANYOF_BITMAP_SET(ret, value); \
7215 parse a class specification and produce either an ANYOF node that
7216 matches the pattern or if the pattern matches a single char only and
7217 that char is < 256 and we are case insensitive then we produce an
7222 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7225 register UV value = 0;
7226 register UV nextvalue;
7227 register IV prevvalue = OOB_UNICODE;
7228 register IV range = 0;
7229 register regnode *ret;
7232 char *rangebegin = NULL;
7233 bool need_class = 0;
7236 bool optimize_invert = TRUE;
7237 AV* unicode_alternate = NULL;
7239 UV literal_endpoint = 0;
7241 UV stored = 0; /* number of chars stored in the class */
7243 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7244 case we need to change the emitted regop to an EXACT. */
7245 const char * orig_parse = RExC_parse;
7246 GET_RE_DEBUG_FLAGS_DECL;
7248 PERL_UNUSED_ARG(depth);
7251 DEBUG_PARSE("clas");
7253 /* Assume we are going to generate an ANYOF node. */
7254 ret = reganode(pRExC_state, ANYOF, 0);
7257 ANYOF_FLAGS(ret) = 0;
7259 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7263 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7267 RExC_size += ANYOF_SKIP;
7268 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7271 RExC_emit += ANYOF_SKIP;
7273 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7275 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7276 ANYOF_BITMAP_ZERO(ret);
7277 listsv = newSVpvs("# comment\n");
7280 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7282 if (!SIZE_ONLY && POSIXCC(nextvalue))
7283 checkposixcc(pRExC_state);
7285 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7286 if (UCHARAT(RExC_parse) == ']')
7290 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7294 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7297 rangebegin = RExC_parse;
7299 value = utf8n_to_uvchr((U8*)RExC_parse,
7300 RExC_end - RExC_parse,
7301 &numlen, UTF8_ALLOW_DEFAULT);
7302 RExC_parse += numlen;
7305 value = UCHARAT(RExC_parse++);
7307 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7308 if (value == '[' && POSIXCC(nextvalue))
7309 namedclass = regpposixcc(pRExC_state, value);
7310 else if (value == '\\') {
7312 value = utf8n_to_uvchr((U8*)RExC_parse,
7313 RExC_end - RExC_parse,
7314 &numlen, UTF8_ALLOW_DEFAULT);
7315 RExC_parse += numlen;
7318 value = UCHARAT(RExC_parse++);
7319 /* Some compilers cannot handle switching on 64-bit integer
7320 * values, therefore value cannot be an UV. Yes, this will
7321 * be a problem later if we want switch on Unicode.
7322 * A similar issue a little bit later when switching on
7323 * namedclass. --jhi */
7324 switch ((I32)value) {
7325 case 'w': namedclass = ANYOF_ALNUM; break;
7326 case 'W': namedclass = ANYOF_NALNUM; break;
7327 case 's': namedclass = ANYOF_SPACE; break;
7328 case 'S': namedclass = ANYOF_NSPACE; break;
7329 case 'd': namedclass = ANYOF_DIGIT; break;
7330 case 'D': namedclass = ANYOF_NDIGIT; break;
7331 case 'N': /* Handle \N{NAME} in class */
7333 /* We only pay attention to the first char of
7334 multichar strings being returned. I kinda wonder
7335 if this makes sense as it does change the behaviour
7336 from earlier versions, OTOH that behaviour was broken
7338 UV v; /* value is register so we cant & it /grrr */
7339 if (reg_namedseq(pRExC_state, &v)) {
7349 if (RExC_parse >= RExC_end)
7350 vFAIL2("Empty \\%c{}", (U8)value);
7351 if (*RExC_parse == '{') {
7352 const U8 c = (U8)value;
7353 e = strchr(RExC_parse++, '}');
7355 vFAIL2("Missing right brace on \\%c{}", c);
7356 while (isSPACE(UCHARAT(RExC_parse)))
7358 if (e == RExC_parse)
7359 vFAIL2("Empty \\%c{}", c);
7361 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7369 if (UCHARAT(RExC_parse) == '^') {
7372 value = value == 'p' ? 'P' : 'p'; /* toggle */
7373 while (isSPACE(UCHARAT(RExC_parse))) {
7378 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7379 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7382 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7383 namedclass = ANYOF_MAX; /* no official name, but it's named */
7386 case 'n': value = '\n'; break;
7387 case 'r': value = '\r'; break;
7388 case 't': value = '\t'; break;
7389 case 'f': value = '\f'; break;
7390 case 'b': value = '\b'; break;
7391 case 'e': value = ASCII_TO_NATIVE('\033');break;
7392 case 'a': value = ASCII_TO_NATIVE('\007');break;
7394 if (*RExC_parse == '{') {
7395 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7396 | PERL_SCAN_DISALLOW_PREFIX;
7397 char * const e = strchr(RExC_parse++, '}');
7399 vFAIL("Missing right brace on \\x{}");
7401 numlen = e - RExC_parse;
7402 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7406 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7408 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7409 RExC_parse += numlen;
7411 if (PL_encoding && value < 0x100)
7412 goto recode_encoding;
7415 value = UCHARAT(RExC_parse++);
7416 value = toCTRL(value);
7418 case '0': case '1': case '2': case '3': case '4':
7419 case '5': case '6': case '7': case '8': case '9':
7423 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7424 RExC_parse += numlen;
7425 if (PL_encoding && value < 0x100)
7426 goto recode_encoding;
7431 SV* enc = PL_encoding;
7432 value = reg_recode((const char)(U8)value, &enc);
7433 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7435 "Invalid escape in the specified encoding");
7439 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7441 "Unrecognized escape \\%c in character class passed through",
7445 } /* end of \blah */
7451 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7453 if (!SIZE_ONLY && !need_class)
7454 ANYOF_CLASS_ZERO(ret);
7458 /* a bad range like a-\d, a-[:digit:] ? */
7461 if (ckWARN(WARN_REGEXP)) {
7463 RExC_parse >= rangebegin ?
7464 RExC_parse - rangebegin : 0;
7466 "False [] range \"%*.*s\"",
7469 if (prevvalue < 256) {
7470 ANYOF_BITMAP_SET(ret, prevvalue);
7471 ANYOF_BITMAP_SET(ret, '-');
7474 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7475 Perl_sv_catpvf(aTHX_ listsv,
7476 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7480 range = 0; /* this was not a true range */
7486 const char *what = NULL;
7489 if (namedclass > OOB_NAMEDCLASS)
7490 optimize_invert = FALSE;
7491 /* Possible truncation here but in some 64-bit environments
7492 * the compiler gets heartburn about switch on 64-bit values.
7493 * A similar issue a little earlier when switching on value.
7495 switch ((I32)namedclass) {
7496 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7497 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7498 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7499 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7500 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7501 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7502 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7503 case _C_C_T_(PRINT, isPRINT(value), "Print");
7504 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7505 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7506 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7507 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7508 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7511 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7514 for (value = 0; value < 128; value++)
7515 ANYOF_BITMAP_SET(ret, value);
7517 for (value = 0; value < 256; value++) {
7519 ANYOF_BITMAP_SET(ret, value);
7528 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7531 for (value = 128; value < 256; value++)
7532 ANYOF_BITMAP_SET(ret, value);
7534 for (value = 0; value < 256; value++) {
7535 if (!isASCII(value))
7536 ANYOF_BITMAP_SET(ret, value);
7545 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7547 /* consecutive digits assumed */
7548 for (value = '0'; value <= '9'; value++)
7549 ANYOF_BITMAP_SET(ret, value);
7556 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7558 /* consecutive digits assumed */
7559 for (value = 0; value < '0'; value++)
7560 ANYOF_BITMAP_SET(ret, value);
7561 for (value = '9' + 1; value < 256; value++)
7562 ANYOF_BITMAP_SET(ret, value);
7568 /* this is to handle \p and \P */
7571 vFAIL("Invalid [::] class");
7575 /* Strings such as "+utf8::isWord\n" */
7576 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7579 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7582 } /* end of namedclass \blah */
7585 if (prevvalue > (IV)value) /* b-a */ {
7586 const int w = RExC_parse - rangebegin;
7587 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7588 range = 0; /* not a valid range */
7592 prevvalue = value; /* save the beginning of the range */
7593 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7594 RExC_parse[1] != ']') {
7597 /* a bad range like \w-, [:word:]- ? */
7598 if (namedclass > OOB_NAMEDCLASS) {
7599 if (ckWARN(WARN_REGEXP)) {
7601 RExC_parse >= rangebegin ?
7602 RExC_parse - rangebegin : 0;
7604 "False [] range \"%*.*s\"",
7608 ANYOF_BITMAP_SET(ret, '-');
7610 range = 1; /* yeah, it's a range! */
7611 continue; /* but do it the next time */
7615 /* now is the next time */
7616 /*stored += (value - prevvalue + 1);*/
7618 if (prevvalue < 256) {
7619 const IV ceilvalue = value < 256 ? value : 255;
7622 /* In EBCDIC [\x89-\x91] should include
7623 * the \x8e but [i-j] should not. */
7624 if (literal_endpoint == 2 &&
7625 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7626 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7628 if (isLOWER(prevvalue)) {
7629 for (i = prevvalue; i <= ceilvalue; i++)
7631 ANYOF_BITMAP_SET(ret, i);
7633 for (i = prevvalue; i <= ceilvalue; i++)
7635 ANYOF_BITMAP_SET(ret, i);
7640 for (i = prevvalue; i <= ceilvalue; i++) {
7641 if (!ANYOF_BITMAP_TEST(ret,i)) {
7643 ANYOF_BITMAP_SET(ret, i);
7647 if (value > 255 || UTF) {
7648 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7649 const UV natvalue = NATIVE_TO_UNI(value);
7650 stored+=2; /* can't optimize this class */
7651 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7652 if (prevnatvalue < natvalue) { /* what about > ? */
7653 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7654 prevnatvalue, natvalue);
7656 else if (prevnatvalue == natvalue) {
7657 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7659 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7661 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7663 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7664 if (RExC_precomp[0] == ':' &&
7665 RExC_precomp[1] == '[' &&
7666 (f == 0xDF || f == 0x92)) {
7667 f = NATIVE_TO_UNI(f);
7670 /* If folding and foldable and a single
7671 * character, insert also the folded version
7672 * to the charclass. */
7674 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7675 if ((RExC_precomp[0] == ':' &&
7676 RExC_precomp[1] == '[' &&
7678 (value == 0xFB05 || value == 0xFB06))) ?
7679 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7680 foldlen == (STRLEN)UNISKIP(f) )
7682 if (foldlen == (STRLEN)UNISKIP(f))
7684 Perl_sv_catpvf(aTHX_ listsv,
7687 /* Any multicharacter foldings
7688 * require the following transform:
7689 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7690 * where E folds into "pq" and F folds
7691 * into "rst", all other characters
7692 * fold to single characters. We save
7693 * away these multicharacter foldings,
7694 * to be later saved as part of the
7695 * additional "s" data. */
7698 if (!unicode_alternate)
7699 unicode_alternate = newAV();
7700 sv = newSVpvn((char*)foldbuf, foldlen);
7702 av_push(unicode_alternate, sv);
7706 /* If folding and the value is one of the Greek
7707 * sigmas insert a few more sigmas to make the
7708 * folding rules of the sigmas to work right.
7709 * Note that not all the possible combinations
7710 * are handled here: some of them are handled
7711 * by the standard folding rules, and some of
7712 * them (literal or EXACTF cases) are handled
7713 * during runtime in regexec.c:S_find_byclass(). */
7714 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7715 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7716 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7717 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7718 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7720 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7721 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7722 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7727 literal_endpoint = 0;
7731 range = 0; /* this range (if it was one) is done now */
7735 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7737 RExC_size += ANYOF_CLASS_ADD_SKIP;
7739 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7745 /****** !SIZE_ONLY AFTER HERE *********/
7747 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
7748 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7750 /* optimize single char class to an EXACT node
7751 but *only* when its not a UTF/high char */
7752 const char * cur_parse= RExC_parse;
7753 RExC_emit = (regnode *)orig_emit;
7754 RExC_parse = (char *)orig_parse;
7755 ret = reg_node(pRExC_state,
7756 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7757 RExC_parse = (char *)cur_parse;
7758 *STRING(ret)= (char)value;
7760 RExC_emit += STR_SZ(1);
7763 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7764 if ( /* If the only flag is folding (plus possibly inversion). */
7765 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7767 for (value = 0; value < 256; ++value) {
7768 if (ANYOF_BITMAP_TEST(ret, value)) {
7769 UV fold = PL_fold[value];
7772 ANYOF_BITMAP_SET(ret, fold);
7775 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7778 /* optimize inverted simple patterns (e.g. [^a-z]) */
7779 if (optimize_invert &&
7780 /* If the only flag is inversion. */
7781 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7782 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7783 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7784 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7787 AV * const av = newAV();
7789 /* The 0th element stores the character class description
7790 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7791 * to initialize the appropriate swash (which gets stored in
7792 * the 1st element), and also useful for dumping the regnode.
7793 * The 2nd element stores the multicharacter foldings,
7794 * used later (regexec.c:S_reginclass()). */
7795 av_store(av, 0, listsv);
7796 av_store(av, 1, NULL);
7797 av_store(av, 2, (SV*)unicode_alternate);
7798 rv = newRV_noinc((SV*)av);
7799 n = add_data(pRExC_state, 1, "s");
7800 RExC_rxi->data->data[n] = (void*)rv;
7808 /* reg_skipcomment()
7810 Absorbs an /x style # comments from the input stream.
7811 Returns true if there is more text remaining in the stream.
7812 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
7813 terminates the pattern without including a newline.
7815 Note its the callers responsibility to ensure that we are
7821 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
7824 while (RExC_parse < RExC_end)
7825 if (*RExC_parse++ == '\n') {
7830 /* we ran off the end of the pattern without ending
7831 the comment, so we have to add an \n when wrapping */
7832 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7840 Advance that parse position, and optionall absorbs
7841 "whitespace" from the inputstream.
7843 Without /x "whitespace" means (?#...) style comments only,
7844 with /x this means (?#...) and # comments and whitespace proper.
7846 Returns the RExC_parse point from BEFORE the scan occurs.
7848 This is the /x friendly way of saying RExC_parse++.
7852 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7854 char* const retval = RExC_parse++;
7857 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7858 RExC_parse[2] == '#') {
7859 while (*RExC_parse != ')') {
7860 if (RExC_parse == RExC_end)
7861 FAIL("Sequence (?#... not terminated");
7867 if (RExC_flags & RXf_PMf_EXTENDED) {
7868 if (isSPACE(*RExC_parse)) {
7872 else if (*RExC_parse == '#') {
7873 if ( reg_skipcomment( pRExC_state ) )
7882 - reg_node - emit a node
7884 STATIC regnode * /* Location. */
7885 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7888 register regnode *ptr;
7889 regnode * const ret = RExC_emit;
7890 GET_RE_DEBUG_FLAGS_DECL;
7893 SIZE_ALIGN(RExC_size);
7897 if (RExC_emit >= RExC_emit_bound)
7898 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
7900 NODE_ALIGN_FILL(ret);
7902 FILL_ADVANCE_NODE(ptr, op);
7903 #ifdef RE_TRACK_PATTERN_OFFSETS
7904 if (RExC_offsets) { /* MJD */
7905 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7906 "reg_node", __LINE__,
7908 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7909 ? "Overwriting end of array!\n" : "OK",
7910 (UV)(RExC_emit - RExC_emit_start),
7911 (UV)(RExC_parse - RExC_start),
7912 (UV)RExC_offsets[0]));
7913 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7921 - reganode - emit a node with an argument
7923 STATIC regnode * /* Location. */
7924 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7927 register regnode *ptr;
7928 regnode * const ret = RExC_emit;
7929 GET_RE_DEBUG_FLAGS_DECL;
7932 SIZE_ALIGN(RExC_size);
7937 assert(2==regarglen[op]+1);
7939 Anything larger than this has to allocate the extra amount.
7940 If we changed this to be:
7942 RExC_size += (1 + regarglen[op]);
7944 then it wouldn't matter. Its not clear what side effect
7945 might come from that so its not done so far.
7950 if (RExC_emit >= RExC_emit_bound)
7951 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
7953 NODE_ALIGN_FILL(ret);
7955 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7956 #ifdef RE_TRACK_PATTERN_OFFSETS
7957 if (RExC_offsets) { /* MJD */
7958 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7962 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7963 "Overwriting end of array!\n" : "OK",
7964 (UV)(RExC_emit - RExC_emit_start),
7965 (UV)(RExC_parse - RExC_start),
7966 (UV)RExC_offsets[0]));
7967 Set_Cur_Node_Offset;
7975 - reguni - emit (if appropriate) a Unicode character
7978 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
7981 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
7985 - reginsert - insert an operator in front of already-emitted operand
7987 * Means relocating the operand.
7990 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
7993 register regnode *src;
7994 register regnode *dst;
7995 register regnode *place;
7996 const int offset = regarglen[(U8)op];
7997 const int size = NODE_STEP_REGNODE + offset;
7998 GET_RE_DEBUG_FLAGS_DECL;
7999 PERL_UNUSED_ARG(depth);
8000 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8001 DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
8010 if (RExC_open_parens) {
8012 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8013 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8014 if ( RExC_open_parens[paren] >= opnd ) {
8015 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8016 RExC_open_parens[paren] += size;
8018 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8020 if ( RExC_close_parens[paren] >= opnd ) {
8021 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8022 RExC_close_parens[paren] += size;
8024 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8029 while (src > opnd) {
8030 StructCopy(--src, --dst, regnode);
8031 #ifdef RE_TRACK_PATTERN_OFFSETS
8032 if (RExC_offsets) { /* MJD 20010112 */
8033 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8037 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8038 ? "Overwriting end of array!\n" : "OK",
8039 (UV)(src - RExC_emit_start),
8040 (UV)(dst - RExC_emit_start),
8041 (UV)RExC_offsets[0]));
8042 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8043 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8049 place = opnd; /* Op node, where operand used to be. */
8050 #ifdef RE_TRACK_PATTERN_OFFSETS
8051 if (RExC_offsets) { /* MJD */
8052 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8056 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8057 ? "Overwriting end of array!\n" : "OK",
8058 (UV)(place - RExC_emit_start),
8059 (UV)(RExC_parse - RExC_start),
8060 (UV)RExC_offsets[0]));
8061 Set_Node_Offset(place, RExC_parse);
8062 Set_Node_Length(place, 1);
8065 src = NEXTOPER(place);
8066 FILL_ADVANCE_NODE(place, op);
8067 Zero(src, offset, regnode);
8071 - regtail - set the next-pointer at the end of a node chain of p to val.
8072 - SEE ALSO: regtail_study
8074 /* TODO: All three parms should be const */
8076 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8079 register regnode *scan;
8080 GET_RE_DEBUG_FLAGS_DECL;
8082 PERL_UNUSED_ARG(depth);
8088 /* Find last node. */
8091 regnode * const temp = regnext(scan);
8093 SV * const mysv=sv_newmortal();
8094 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8095 regprop(RExC_rx, mysv, scan);
8096 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8097 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8098 (temp == NULL ? "->" : ""),
8099 (temp == NULL ? reg_name[OP(val)] : "")
8107 if (reg_off_by_arg[OP(scan)]) {
8108 ARG_SET(scan, val - scan);
8111 NEXT_OFF(scan) = val - scan;
8117 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8118 - Look for optimizable sequences at the same time.
8119 - currently only looks for EXACT chains.
8121 This is expermental code. The idea is to use this routine to perform
8122 in place optimizations on branches and groups as they are constructed,
8123 with the long term intention of removing optimization from study_chunk so
8124 that it is purely analytical.
8126 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8127 to control which is which.
8130 /* TODO: All four parms should be const */
8133 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8136 register regnode *scan;
8138 #ifdef EXPERIMENTAL_INPLACESCAN
8142 GET_RE_DEBUG_FLAGS_DECL;
8148 /* Find last node. */
8152 regnode * const temp = regnext(scan);
8153 #ifdef EXPERIMENTAL_INPLACESCAN
8154 if (PL_regkind[OP(scan)] == EXACT)
8155 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8163 if( exact == PSEUDO )
8165 else if ( exact != OP(scan) )
8174 SV * const mysv=sv_newmortal();
8175 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8176 regprop(RExC_rx, mysv, scan);
8177 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8178 SvPV_nolen_const(mysv),
8187 SV * const mysv_val=sv_newmortal();
8188 DEBUG_PARSE_MSG("");
8189 regprop(RExC_rx, mysv_val, val);
8190 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8191 SvPV_nolen_const(mysv_val),
8192 (IV)REG_NODE_NUM(val),
8196 if (reg_off_by_arg[OP(scan)]) {
8197 ARG_SET(scan, val - scan);
8200 NEXT_OFF(scan) = val - scan;
8208 - regcurly - a little FSA that accepts {\d+,?\d*}
8211 S_regcurly(register const char *s)
8230 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8233 Perl_regdump(pTHX_ const regexp *r)
8237 SV * const sv = sv_newmortal();
8238 SV *dsv= sv_newmortal();
8241 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8243 /* Header fields of interest. */
8244 if (r->anchored_substr) {
8245 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8246 RE_SV_DUMPLEN(r->anchored_substr), 30);
8247 PerlIO_printf(Perl_debug_log,
8248 "anchored %s%s at %"IVdf" ",
8249 s, RE_SV_TAIL(r->anchored_substr),
8250 (IV)r->anchored_offset);
8251 } else if (r->anchored_utf8) {
8252 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8253 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8254 PerlIO_printf(Perl_debug_log,
8255 "anchored utf8 %s%s at %"IVdf" ",
8256 s, RE_SV_TAIL(r->anchored_utf8),
8257 (IV)r->anchored_offset);
8259 if (r->float_substr) {
8260 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8261 RE_SV_DUMPLEN(r->float_substr), 30);
8262 PerlIO_printf(Perl_debug_log,
8263 "floating %s%s at %"IVdf"..%"UVuf" ",
8264 s, RE_SV_TAIL(r->float_substr),
8265 (IV)r->float_min_offset, (UV)r->float_max_offset);
8266 } else if (r->float_utf8) {
8267 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8268 RE_SV_DUMPLEN(r->float_utf8), 30);
8269 PerlIO_printf(Perl_debug_log,
8270 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8271 s, RE_SV_TAIL(r->float_utf8),
8272 (IV)r->float_min_offset, (UV)r->float_max_offset);
8274 if (r->check_substr || r->check_utf8)
8275 PerlIO_printf(Perl_debug_log,
8277 (r->check_substr == r->float_substr
8278 && r->check_utf8 == r->float_utf8
8279 ? "(checking floating" : "(checking anchored"));
8280 if (r->extflags & RXf_NOSCAN)
8281 PerlIO_printf(Perl_debug_log, " noscan");
8282 if (r->extflags & RXf_CHECK_ALL)
8283 PerlIO_printf(Perl_debug_log, " isall");
8284 if (r->check_substr || r->check_utf8)
8285 PerlIO_printf(Perl_debug_log, ") ");
8287 if (ri->regstclass) {
8288 regprop(r, sv, ri->regstclass);
8289 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8291 if (r->extflags & RXf_ANCH) {
8292 PerlIO_printf(Perl_debug_log, "anchored");
8293 if (r->extflags & RXf_ANCH_BOL)
8294 PerlIO_printf(Perl_debug_log, "(BOL)");
8295 if (r->extflags & RXf_ANCH_MBOL)
8296 PerlIO_printf(Perl_debug_log, "(MBOL)");
8297 if (r->extflags & RXf_ANCH_SBOL)
8298 PerlIO_printf(Perl_debug_log, "(SBOL)");
8299 if (r->extflags & RXf_ANCH_GPOS)
8300 PerlIO_printf(Perl_debug_log, "(GPOS)");
8301 PerlIO_putc(Perl_debug_log, ' ');
8303 if (r->extflags & RXf_GPOS_SEEN)
8304 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8305 if (r->intflags & PREGf_SKIP)
8306 PerlIO_printf(Perl_debug_log, "plus ");
8307 if (r->intflags & PREGf_IMPLICIT)
8308 PerlIO_printf(Perl_debug_log, "implicit ");
8309 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8310 if (r->extflags & RXf_EVAL_SEEN)
8311 PerlIO_printf(Perl_debug_log, "with eval ");
8312 PerlIO_printf(Perl_debug_log, "\n");
8314 PERL_UNUSED_CONTEXT;
8316 #endif /* DEBUGGING */
8320 - regprop - printable representation of opcode
8323 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8328 RXi_GET_DECL(prog,progi);
8329 GET_RE_DEBUG_FLAGS_DECL;
8332 sv_setpvn(sv, "", 0);
8334 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8335 /* It would be nice to FAIL() here, but this may be called from
8336 regexec.c, and it would be hard to supply pRExC_state. */
8337 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8338 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
8340 k = PL_regkind[OP(o)];
8343 SV * const dsv = sv_2mortal(newSVpvs(""));
8344 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8345 * is a crude hack but it may be the best for now since
8346 * we have no flag "this EXACTish node was UTF-8"
8348 const char * const s =
8349 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8350 PL_colors[0], PL_colors[1],
8351 PERL_PV_ESCAPE_UNI_DETECT |
8352 PERL_PV_PRETTY_ELIPSES |
8355 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8356 } else if (k == TRIE) {
8357 /* print the details of the trie in dumpuntil instead, as
8358 * progi->data isn't available here */
8359 const char op = OP(o);
8360 const U32 n = ARG(o);
8361 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8362 (reg_ac_data *)progi->data->data[n] :
8364 const reg_trie_data * const trie
8365 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8367 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
8368 DEBUG_TRIE_COMPILE_r(
8369 Perl_sv_catpvf(aTHX_ sv,
8370 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8371 (UV)trie->startstate,
8372 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8373 (UV)trie->wordcount,
8376 (UV)TRIE_CHARCOUNT(trie),
8377 (UV)trie->uniquecharcount
8380 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8382 int rangestart = -1;
8383 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8384 Perl_sv_catpvf(aTHX_ sv, "[");
8385 for (i = 0; i <= 256; i++) {
8386 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8387 if (rangestart == -1)
8389 } else if (rangestart != -1) {
8390 if (i <= rangestart + 3)
8391 for (; rangestart < i; rangestart++)
8392 put_byte(sv, rangestart);
8394 put_byte(sv, rangestart);
8396 put_byte(sv, i - 1);
8401 Perl_sv_catpvf(aTHX_ sv, "]");
8404 } else if (k == CURLY) {
8405 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8406 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8407 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8409 else if (k == WHILEM && o->flags) /* Ordinal/of */
8410 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8411 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8412 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8413 if ( prog->paren_names ) {
8414 if ( k != REF || OP(o) < NREF) {
8415 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8416 SV **name= av_fetch(list, ARG(o), 0 );
8418 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8421 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8422 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8423 I32 *nums=(I32*)SvPVX(sv_dat);
8424 SV **name= av_fetch(list, nums[0], 0 );
8427 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8428 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8429 (n ? "," : ""), (IV)nums[n]);
8431 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8435 } else if (k == GOSUB)
8436 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8437 else if (k == VERB) {
8439 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8440 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8441 } else if (k == LOGICAL)
8442 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8443 else if (k == ANYOF) {
8444 int i, rangestart = -1;
8445 const U8 flags = ANYOF_FLAGS(o);
8447 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8448 static const char * const anyofs[] = {
8481 if (flags & ANYOF_LOCALE)
8482 sv_catpvs(sv, "{loc}");
8483 if (flags & ANYOF_FOLD)
8484 sv_catpvs(sv, "{i}");
8485 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8486 if (flags & ANYOF_INVERT)
8488 for (i = 0; i <= 256; i++) {
8489 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8490 if (rangestart == -1)
8492 } else if (rangestart != -1) {
8493 if (i <= rangestart + 3)
8494 for (; rangestart < i; rangestart++)
8495 put_byte(sv, rangestart);
8497 put_byte(sv, rangestart);
8499 put_byte(sv, i - 1);
8505 if (o->flags & ANYOF_CLASS)
8506 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8507 if (ANYOF_CLASS_TEST(o,i))
8508 sv_catpv(sv, anyofs[i]);
8510 if (flags & ANYOF_UNICODE)
8511 sv_catpvs(sv, "{unicode}");
8512 else if (flags & ANYOF_UNICODE_ALL)
8513 sv_catpvs(sv, "{unicode_all}");
8517 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8521 U8 s[UTF8_MAXBYTES_CASE+1];
8523 for (i = 0; i <= 256; i++) { /* just the first 256 */
8524 uvchr_to_utf8(s, i);
8526 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8527 if (rangestart == -1)
8529 } else if (rangestart != -1) {
8530 if (i <= rangestart + 3)
8531 for (; rangestart < i; rangestart++) {
8532 const U8 * const e = uvchr_to_utf8(s,rangestart);
8534 for(p = s; p < e; p++)
8538 const U8 *e = uvchr_to_utf8(s,rangestart);
8540 for (p = s; p < e; p++)
8543 e = uvchr_to_utf8(s, i-1);
8544 for (p = s; p < e; p++)
8551 sv_catpvs(sv, "..."); /* et cetera */
8555 char *s = savesvpv(lv);
8556 char * const origs = s;
8558 while (*s && *s != '\n')
8562 const char * const t = ++s;
8580 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8582 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8583 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8585 PERL_UNUSED_CONTEXT;
8586 PERL_UNUSED_ARG(sv);
8588 PERL_UNUSED_ARG(prog);
8589 #endif /* DEBUGGING */
8593 Perl_re_intuit_string(pTHX_ regexp *prog)
8594 { /* Assume that RE_INTUIT is set */
8596 GET_RE_DEBUG_FLAGS_DECL;
8597 PERL_UNUSED_CONTEXT;
8601 const char * const s = SvPV_nolen_const(prog->check_substr
8602 ? prog->check_substr : prog->check_utf8);
8604 if (!PL_colorset) reginitcolors();
8605 PerlIO_printf(Perl_debug_log,
8606 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8608 prog->check_substr ? "" : "utf8 ",
8609 PL_colors[5],PL_colors[0],
8612 (strlen(s) > 60 ? "..." : ""));
8615 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8621 handles refcounting and freeing the perl core regexp structure. When
8622 it is necessary to actually free the structure the first thing it
8623 does is call the 'free' method of the regexp_engine associated to to
8624 the regexp, allowing the handling of the void *pprivate; member
8625 first. (This routine is not overridable by extensions, which is why
8626 the extensions free is called first.)
8628 See regdupe and regdupe_internal if you change anything here.
8630 #ifndef PERL_IN_XSUB_RE
8632 Perl_pregfree(pTHX_ struct regexp *r)
8635 GET_RE_DEBUG_FLAGS_DECL;
8637 if (!r || (--r->refcnt > 0))
8640 CALLREGFREE_PVT(r); /* free the private data */
8641 RX_MATCH_COPY_FREE(r);
8642 #ifdef PERL_OLD_COPY_ON_WRITE
8644 SvREFCNT_dec(r->saved_copy);
8647 if (r->anchored_substr)
8648 SvREFCNT_dec(r->anchored_substr);
8649 if (r->anchored_utf8)
8650 SvREFCNT_dec(r->anchored_utf8);
8651 if (r->float_substr)
8652 SvREFCNT_dec(r->float_substr);
8654 SvREFCNT_dec(r->float_utf8);
8655 Safefree(r->substrs);
8658 SvREFCNT_dec(r->paren_names);
8659 Safefree(r->wrapped);
8660 Safefree(r->startp);
8666 /* regfree_internal()
8668 Free the private data in a regexp. This is overloadable by
8669 extensions. Perl takes care of the regexp structure in pregfree(),
8670 this covers the *pprivate pointer which technically perldoesnt
8671 know about, however of course we have to handle the
8672 regexp_internal structure when no extension is in use.
8674 Note this is called before freeing anything in the regexp
8679 Perl_regfree_internal(pTHX_ struct regexp *r)
8683 GET_RE_DEBUG_FLAGS_DECL;
8689 SV *dsv= sv_newmortal();
8690 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8691 dsv, r->precomp, r->prelen, 60);
8692 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8693 PL_colors[4],PL_colors[5],s);
8696 #ifdef RE_TRACK_PATTERN_OFFSETS
8698 Safefree(ri->u.offsets); /* 20010421 MJD */
8701 int n = ri->data->count;
8702 PAD* new_comppad = NULL;
8707 /* If you add a ->what type here, update the comment in regcomp.h */
8708 switch (ri->data->what[n]) {
8712 SvREFCNT_dec((SV*)ri->data->data[n]);
8715 Safefree(ri->data->data[n]);
8718 new_comppad = (AV*)ri->data->data[n];
8721 if (new_comppad == NULL)
8722 Perl_croak(aTHX_ "panic: pregfree comppad");
8723 PAD_SAVE_LOCAL(old_comppad,
8724 /* Watch out for global destruction's random ordering. */
8725 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8728 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8731 op_free((OP_4tree*)ri->data->data[n]);
8733 PAD_RESTORE_LOCAL(old_comppad);
8734 SvREFCNT_dec((SV*)new_comppad);
8740 { /* Aho Corasick add-on structure for a trie node.
8741 Used in stclass optimization only */
8743 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8745 refcount = --aho->refcount;
8748 PerlMemShared_free(aho->states);
8749 PerlMemShared_free(aho->fail);
8750 /* do this last!!!! */
8751 PerlMemShared_free(ri->data->data[n]);
8752 PerlMemShared_free(ri->regstclass);
8758 /* trie structure. */
8760 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8762 refcount = --trie->refcount;
8765 PerlMemShared_free(trie->charmap);
8766 PerlMemShared_free(trie->states);
8767 PerlMemShared_free(trie->trans);
8769 PerlMemShared_free(trie->bitmap);
8771 PerlMemShared_free(trie->wordlen);
8773 PerlMemShared_free(trie->jump);
8775 PerlMemShared_free(trie->nextword);
8776 /* do this last!!!! */
8777 PerlMemShared_free(ri->data->data[n]);
8782 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8785 Safefree(ri->data->what);
8789 Safefree(ri->swap->startp);
8790 Safefree(ri->swap->endp);
8796 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8797 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8798 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8799 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8802 regdupe - duplicate a regexp.
8804 This routine is called by sv.c's re_dup and is expected to clone a
8805 given regexp structure. It is a no-op when not under USE_ITHREADS.
8806 (Originally this *was* re_dup() for change history see sv.c)
8808 After all of the core data stored in struct regexp is duplicated
8809 the regexp_engine.dupe method is used to copy any private data
8810 stored in the *pprivate pointer. This allows extensions to handle
8811 any duplication it needs to do.
8813 See pregfree() and regfree_internal() if you change anything here.
8815 #if defined(USE_ITHREADS)
8816 #ifndef PERL_IN_XSUB_RE
8818 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
8823 struct reg_substr_datum *s;
8826 return (REGEXP *)NULL;
8828 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8832 npar = r->nparens+1;
8833 Newxz(ret, 1, regexp);
8834 Newx(ret->startp, npar, I32);
8835 Copy(r->startp, ret->startp, npar, I32);
8836 Newx(ret->endp, npar, I32);
8837 Copy(r->endp, ret->endp, npar, I32);
8840 Newx(ret->substrs, 1, struct reg_substr_data);
8841 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8842 s->min_offset = r->substrs->data[i].min_offset;
8843 s->max_offset = r->substrs->data[i].max_offset;
8844 s->end_shift = r->substrs->data[i].end_shift;
8845 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8846 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8849 ret->substrs = NULL;
8851 ret->wrapped = SAVEPVN(r->wrapped, r->wraplen);
8852 ret->precomp = ret->wrapped + (r->precomp - r->wrapped);
8853 ret->prelen = r->prelen;
8854 ret->wraplen = r->wraplen;
8856 ret->refcnt = r->refcnt;
8857 ret->minlen = r->minlen;
8858 ret->minlenret = r->minlenret;
8859 ret->nparens = r->nparens;
8860 ret->lastparen = r->lastparen;
8861 ret->lastcloseparen = r->lastcloseparen;
8862 ret->intflags = r->intflags;
8863 ret->extflags = r->extflags;
8865 ret->sublen = r->sublen;
8867 ret->engine = r->engine;
8869 ret->paren_names = hv_dup_inc(r->paren_names, param);
8871 if (RX_MATCH_COPIED(ret))
8872 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8875 #ifdef PERL_OLD_COPY_ON_WRITE
8876 ret->saved_copy = NULL;
8879 ret->pprivate = r->pprivate;
8881 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
8883 ptr_table_store(PL_ptr_table, r, ret);
8886 #endif /* PERL_IN_XSUB_RE */
8891 This is the internal complement to regdupe() which is used to copy
8892 the structure pointed to by the *pprivate pointer in the regexp.
8893 This is the core version of the extension overridable cloning hook.
8894 The regexp structure being duplicated will be copied by perl prior
8895 to this and will be provided as the regexp *r argument, however
8896 with the /old/ structures pprivate pointer value. Thus this routine
8897 may override any copying normally done by perl.
8899 It returns a pointer to the new regexp_internal structure.
8903 Perl_regdupe_internal(pTHX_ const regexp *r, CLONE_PARAMS *param)
8906 regexp_internal *reti;
8910 npar = r->nparens+1;
8913 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8914 Copy(ri->program, reti->program, len+1, regnode);
8917 Newx(reti->swap, 1, regexp_paren_ofs);
8918 /* no need to copy these */
8919 Newx(reti->swap->startp, npar, I32);
8920 Newx(reti->swap->endp, npar, I32);
8925 reti->regstclass = NULL;
8929 const int count = ri->data->count;
8932 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8933 char, struct reg_data);
8934 Newx(d->what, count, U8);
8937 for (i = 0; i < count; i++) {
8938 d->what[i] = ri->data->what[i];
8939 switch (d->what[i]) {
8940 /* legal options are one of: sSfpontTu
8941 see also regcomp.h and pregfree() */
8944 case 'p': /* actually an AV, but the dup function is identical. */
8945 case 'u': /* actually an HV, but the dup function is identical. */
8946 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
8949 /* This is cheating. */
8950 Newx(d->data[i], 1, struct regnode_charclass_class);
8951 StructCopy(ri->data->data[i], d->data[i],
8952 struct regnode_charclass_class);
8953 reti->regstclass = (regnode*)d->data[i];
8956 /* Compiled op trees are readonly and in shared memory,
8957 and can thus be shared without duplication. */
8959 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
8963 /* Trie stclasses are readonly and can thus be shared
8964 * without duplication. We free the stclass in pregfree
8965 * when the corresponding reg_ac_data struct is freed.
8967 reti->regstclass= ri->regstclass;
8971 ((reg_trie_data*)ri->data->data[i])->refcount++;
8975 d->data[i] = ri->data->data[i];
8978 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
8987 reti->name_list_idx = ri->name_list_idx;
8989 #ifdef RE_TRACK_PATTERN_OFFSETS
8990 if (ri->u.offsets) {
8991 Newx(reti->u.offsets, 2*len+1, U32);
8992 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
8995 SetProgLen(reti,len);
9001 #endif /* USE_ITHREADS */
9006 converts a regexp embedded in a MAGIC struct to its stringified form,
9007 caching the converted form in the struct and returns the cached
9010 If lp is nonnull then it is used to return the length of the
9013 If flags is nonnull and the returned string contains UTF8 then
9014 (*flags & 1) will be true.
9016 If haseval is nonnull then it is used to return whether the pattern
9019 Normally called via macro:
9021 CALLREG_STRINGIFY(mg,&len,&utf8);
9025 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9027 See sv_2pv_flags() in sv.c for an example of internal usage.
9030 #ifndef PERL_IN_XSUB_RE
9033 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9035 const regexp * const re = (regexp *)mg->mg_obj;
9037 *haseval = re->seen_evals;
9039 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
9046 - regnext - dig the "next" pointer out of a node
9049 Perl_regnext(pTHX_ register regnode *p)
9052 register I32 offset;
9057 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9066 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9069 STRLEN l1 = strlen(pat1);
9070 STRLEN l2 = strlen(pat2);
9073 const char *message;
9079 Copy(pat1, buf, l1 , char);
9080 Copy(pat2, buf + l1, l2 , char);
9081 buf[l1 + l2] = '\n';
9082 buf[l1 + l2 + 1] = '\0';
9084 /* ANSI variant takes additional second argument */
9085 va_start(args, pat2);
9089 msv = vmess(buf, &args);
9091 message = SvPV_const(msv,l1);
9094 Copy(message, buf, l1 , char);
9095 buf[l1-1] = '\0'; /* Overwrite \n */
9096 Perl_croak(aTHX_ "%s", buf);
9099 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9101 #ifndef PERL_IN_XSUB_RE
9103 Perl_save_re_context(pTHX)
9107 struct re_save_state *state;
9109 SAVEVPTR(PL_curcop);
9110 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9112 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9113 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9114 SSPUSHINT(SAVEt_RE_STATE);
9116 Copy(&PL_reg_state, state, 1, struct re_save_state);
9118 PL_reg_start_tmp = 0;
9119 PL_reg_start_tmpl = 0;
9120 PL_reg_oldsaved = NULL;
9121 PL_reg_oldsavedlen = 0;
9123 PL_reg_leftiter = 0;
9124 PL_reg_poscache = NULL;
9125 PL_reg_poscache_size = 0;
9126 #ifdef PERL_OLD_COPY_ON_WRITE
9130 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9132 const REGEXP * const rx = PM_GETRE(PL_curpm);
9135 for (i = 1; i <= rx->nparens; i++) {
9136 char digits[TYPE_CHARS(long)];
9137 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9138 GV *const *const gvp
9139 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9142 GV * const gv = *gvp;
9143 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9153 clear_re(pTHX_ void *r)
9156 ReREFCNT_dec((regexp *)r);
9162 S_put_byte(pTHX_ SV *sv, int c)
9164 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9165 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9166 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9167 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9169 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9173 #define CLEAR_OPTSTART \
9174 if (optstart) STMT_START { \
9175 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9179 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9181 STATIC const regnode *
9182 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9183 const regnode *last, const regnode *plast,
9184 SV* sv, I32 indent, U32 depth)
9187 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9188 register const regnode *next;
9189 const regnode *optstart= NULL;
9192 GET_RE_DEBUG_FLAGS_DECL;
9194 #ifdef DEBUG_DUMPUNTIL
9195 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9196 last ? last-start : 0,plast ? plast-start : 0);
9199 if (plast && plast < last)
9202 while (PL_regkind[op] != END && (!last || node < last)) {
9203 /* While that wasn't END last time... */
9206 if (op == CLOSE || op == WHILEM)
9208 next = regnext((regnode *)node);
9211 if (OP(node) == OPTIMIZED) {
9212 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9219 regprop(r, sv, node);
9220 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9221 (int)(2*indent + 1), "", SvPVX_const(sv));
9223 if (OP(node) != OPTIMIZED) {
9224 if (next == NULL) /* Next ptr. */
9225 PerlIO_printf(Perl_debug_log, " (0)");
9226 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9227 PerlIO_printf(Perl_debug_log, " (FAIL)");
9229 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9230 (void)PerlIO_putc(Perl_debug_log, '\n');
9234 if (PL_regkind[(U8)op] == BRANCHJ) {
9237 register const regnode *nnode = (OP(next) == LONGJMP
9238 ? regnext((regnode *)next)
9240 if (last && nnode > last)
9242 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9245 else if (PL_regkind[(U8)op] == BRANCH) {
9247 DUMPUNTIL(NEXTOPER(node), next);
9249 else if ( PL_regkind[(U8)op] == TRIE ) {
9250 const regnode *this_trie = node;
9251 const char op = OP(node);
9252 const U32 n = ARG(node);
9253 const reg_ac_data * const ac = op>=AHOCORASICK ?
9254 (reg_ac_data *)ri->data->data[n] :
9256 const reg_trie_data * const trie =
9257 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9259 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9261 const regnode *nextbranch= NULL;
9263 sv_setpvn(sv, "", 0);
9264 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9265 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9267 PerlIO_printf(Perl_debug_log, "%*s%s ",
9268 (int)(2*(indent+3)), "",
9269 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9270 PL_colors[0], PL_colors[1],
9271 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9272 PERL_PV_PRETTY_ELIPSES |
9278 U16 dist= trie->jump[word_idx+1];
9279 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9280 (UV)((dist ? this_trie + dist : next) - start));
9283 nextbranch= this_trie + trie->jump[0];
9284 DUMPUNTIL(this_trie + dist, nextbranch);
9286 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9287 nextbranch= regnext((regnode *)nextbranch);
9289 PerlIO_printf(Perl_debug_log, "\n");
9292 if (last && next > last)
9297 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9298 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9299 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9301 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9303 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9305 else if ( op == PLUS || op == STAR) {
9306 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9308 else if (op == ANYOF) {
9309 /* arglen 1 + class block */
9310 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9311 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9312 node = NEXTOPER(node);
9314 else if (PL_regkind[(U8)op] == EXACT) {
9315 /* Literal string, where present. */
9316 node += NODE_SZ_STR(node) - 1;
9317 node = NEXTOPER(node);
9320 node = NEXTOPER(node);
9321 node += regarglen[(U8)op];
9323 if (op == CURLYX || op == OPEN)
9327 #ifdef DEBUG_DUMPUNTIL
9328 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9333 #endif /* DEBUGGING */
9337 * c-indentation-style: bsd
9339 * indent-tabs-mode: t
9342 * ex: set ts=8 sts=4 sw=4 noet:
projects / p5sagit/p5-mst-13.2.git / blob