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, 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. */
106 char *start; /* Start of input for compile */
107 char *end; /* End of input for compile */
108 char *parse; /* Input-scan pointer. */
109 I32 whilem_seen; /* number of WHILEM in this expr */
110 regnode *emit_start; /* Start of emitted-code area */
111 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
112 I32 naughty; /* How bad is this pattern? */
113 I32 sawback; /* Did we see \1, ...? */
115 I32 size; /* Code size. */
116 I32 npar; /* () count. */
121 HV *charnames; /* cache of named sequences */
123 char *starttry; /* -Dr: where regtry was called. */
124 #define RExC_starttry (pRExC_state->starttry)
127 const char *lastparse;
129 #define RExC_lastparse (pRExC_state->lastparse)
130 #define RExC_lastnum (pRExC_state->lastnum)
134 #define RExC_flags (pRExC_state->flags)
135 #define RExC_precomp (pRExC_state->precomp)
136 #define RExC_rx (pRExC_state->rx)
137 #define RExC_start (pRExC_state->start)
138 #define RExC_end (pRExC_state->end)
139 #define RExC_parse (pRExC_state->parse)
140 #define RExC_whilem_seen (pRExC_state->whilem_seen)
141 #define RExC_offsets (pRExC_state->rx->offsets) /* I am not like the others */
142 #define RExC_emit (pRExC_state->emit)
143 #define RExC_emit_start (pRExC_state->emit_start)
144 #define RExC_naughty (pRExC_state->naughty)
145 #define RExC_sawback (pRExC_state->sawback)
146 #define RExC_seen (pRExC_state->seen)
147 #define RExC_size (pRExC_state->size)
148 #define RExC_npar (pRExC_state->npar)
149 #define RExC_extralen (pRExC_state->extralen)
150 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
151 #define RExC_seen_evals (pRExC_state->seen_evals)
152 #define RExC_utf8 (pRExC_state->utf8)
153 #define RExC_charnames (pRExC_state->charnames)
155 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
156 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
157 ((*s) == '{' && regcurly(s)))
160 #undef SPSTART /* dratted cpp namespace... */
163 * Flags to be passed up and down.
165 #define WORST 0 /* Worst case. */
166 #define HASWIDTH 0x1 /* Known to match non-null strings. */
167 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
168 #define SPSTART 0x4 /* Starts with * or +. */
169 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
171 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
173 /* whether trie related optimizations are enabled */
174 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
175 #define TRIE_STUDY_OPT
176 #define FULL_TRIE_STUDY
181 /* About scan_data_t.
183 During optimisation we recurse through the regexp program performing
184 various inplace (keyhole style) optimisations. In addition study_chunk
185 and scan_commit populate this data structure with information about
186 what strings MUST appear in the pattern. We look for the longest
187 string that must appear for at a fixed location, and we look for the
188 longest string that may appear at a floating location. So for instance
193 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
194 strings (because they follow a .* construct). study_chunk will identify
195 both FOO and BAR as being the longest fixed and floating strings respectively.
197 The strings can be composites, for instance
201 will result in a composite fixed substring 'foo'.
203 For each string some basic information is maintained:
205 - offset or min_offset
206 This is the position the string must appear at, or not before.
207 It also implicitly (when combined with minlenp) tells us how many
208 character must match before the string we are searching.
209 Likewise when combined with minlenp and the length of the string
210 tells us how many characters must appear after the string we have
214 Only used for floating strings. This is the rightmost point that
215 the string can appear at. Ifset to I32 max it indicates that the
216 string can occur infinitely far to the right.
219 A pointer to the minimum length of the pattern that the string
220 was found inside. This is important as in the case of positive
221 lookahead or positive lookbehind we can have multiple patterns
226 The minimum length of the pattern overall is 3, the minimum length
227 of the lookahead part is 3, but the minimum length of the part that
228 will actually match is 1. So 'FOO's minimum length is 3, but the
229 minimum length for the F is 1. This is important as the minimum length
230 is used to determine offsets in front of and behind the string being
231 looked for. Since strings can be composites this is the length of the
232 pattern at the time it was commited with a scan_commit. Note that
233 the length is calculated by study_chunk, so that the minimum lengths
234 are not known until the full pattern has been compiled, thus the
235 pointer to the value.
239 In the case of lookbehind the string being searched for can be
240 offset past the start point of the final matching string.
241 If this value was just blithely removed from the min_offset it would
242 invalidate some of the calculations for how many chars must match
243 before or after (as they are derived from min_offset and minlen and
244 the length of the string being searched for).
245 When the final pattern is compiled and the data is moved from the
246 scan_data_t structure into the regexp structure the information
247 about lookbehind is factored in, with the information that would
248 have been lost precalculated in the end_shift field for the
251 The fields pos_min and pos_delta are used to store the minimum offset
252 and the delta to the maximum offset at the current point in the pattern.
256 typedef struct scan_data_t {
257 /*I32 len_min; unused */
258 /*I32 len_delta; unused */
262 I32 last_end; /* min value, <0 unless valid. */
265 SV **longest; /* Either &l_fixed, or &l_float. */
266 SV *longest_fixed; /* longest fixed string found in pattern */
267 I32 offset_fixed; /* offset where it starts */
268 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
269 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
270 SV *longest_float; /* longest floating string found in pattern */
271 I32 offset_float_min; /* earliest point in string it can appear */
272 I32 offset_float_max; /* latest point in string it can appear */
273 I32 *minlen_float; /* pointer to the minlen relevent to the string */
274 I32 lookbehind_float; /* is the position of the string modified by LB */
278 struct regnode_charclass_class *start_class;
282 * Forward declarations for pregcomp()'s friends.
285 static const scan_data_t zero_scan_data =
286 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
288 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
289 #define SF_BEFORE_SEOL 0x0001
290 #define SF_BEFORE_MEOL 0x0002
291 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
292 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
295 # define SF_FIX_SHIFT_EOL (0+2)
296 # define SF_FL_SHIFT_EOL (0+4)
298 # define SF_FIX_SHIFT_EOL (+2)
299 # define SF_FL_SHIFT_EOL (+4)
302 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
303 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
305 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
306 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
307 #define SF_IS_INF 0x0040
308 #define SF_HAS_PAR 0x0080
309 #define SF_IN_PAR 0x0100
310 #define SF_HAS_EVAL 0x0200
311 #define SCF_DO_SUBSTR 0x0400
312 #define SCF_DO_STCLASS_AND 0x0800
313 #define SCF_DO_STCLASS_OR 0x1000
314 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
315 #define SCF_WHILEM_VISITED_POS 0x2000
317 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
320 #define UTF (RExC_utf8 != 0)
321 #define LOC ((RExC_flags & PMf_LOCALE) != 0)
322 #define FOLD ((RExC_flags & PMf_FOLD) != 0)
324 #define OOB_UNICODE 12345678
325 #define OOB_NAMEDCLASS -1
327 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
328 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
331 /* length of regex to show in messages that don't mark a position within */
332 #define RegexLengthToShowInErrorMessages 127
335 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
336 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
337 * op/pragma/warn/regcomp.
339 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
340 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
342 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
345 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
346 * arg. Show regex, up to a maximum length. If it's too long, chop and add
349 #define FAIL(msg) STMT_START { \
350 const char *ellipses = ""; \
351 IV len = RExC_end - RExC_precomp; \
354 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
355 if (len > RegexLengthToShowInErrorMessages) { \
356 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
357 len = RegexLengthToShowInErrorMessages - 10; \
360 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
361 msg, (int)len, RExC_precomp, ellipses); \
365 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
367 #define Simple_vFAIL(m) STMT_START { \
368 const IV offset = RExC_parse - RExC_precomp; \
369 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
370 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
374 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
376 #define vFAIL(m) STMT_START { \
378 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
383 * Like Simple_vFAIL(), but accepts two arguments.
385 #define Simple_vFAIL2(m,a1) STMT_START { \
386 const IV offset = RExC_parse - RExC_precomp; \
387 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
388 (int)offset, RExC_precomp, RExC_precomp + offset); \
392 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
394 #define vFAIL2(m,a1) STMT_START { \
396 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
397 Simple_vFAIL2(m, a1); \
402 * Like Simple_vFAIL(), but accepts three arguments.
404 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
405 const IV offset = RExC_parse - RExC_precomp; \
406 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
407 (int)offset, RExC_precomp, RExC_precomp + offset); \
411 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
413 #define vFAIL3(m,a1,a2) STMT_START { \
415 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
416 Simple_vFAIL3(m, a1, a2); \
420 * Like Simple_vFAIL(), but accepts four arguments.
422 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
423 const IV offset = RExC_parse - RExC_precomp; \
424 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
425 (int)offset, RExC_precomp, RExC_precomp + offset); \
428 #define vWARN(loc,m) STMT_START { \
429 const IV offset = loc - RExC_precomp; \
430 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
431 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
434 #define vWARNdep(loc,m) STMT_START { \
435 const IV offset = loc - RExC_precomp; \
436 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
437 "%s" REPORT_LOCATION, \
438 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
442 #define vWARN2(loc, m, a1) STMT_START { \
443 const IV offset = loc - RExC_precomp; \
444 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
445 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
448 #define vWARN3(loc, m, a1, a2) STMT_START { \
449 const IV offset = loc - RExC_precomp; \
450 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
451 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
454 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
455 const IV offset = loc - RExC_precomp; \
456 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
457 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
460 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
461 const IV offset = loc - RExC_precomp; \
462 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
463 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
467 /* Allow for side effects in s */
468 #define REGC(c,s) STMT_START { \
469 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
472 /* Macros for recording node offsets. 20001227 mjd@plover.com
473 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
474 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
475 * Element 0 holds the number n.
476 * Position is 1 indexed.
479 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
481 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
482 __LINE__, (node), (int)(byte))); \
484 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
486 RExC_offsets[2*(node)-1] = (byte); \
491 #define Set_Node_Offset(node,byte) \
492 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
493 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
495 #define Set_Node_Length_To_R(node,len) STMT_START { \
497 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
498 __LINE__, (int)(node), (int)(len))); \
500 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
502 RExC_offsets[2*(node)] = (len); \
507 #define Set_Node_Length(node,len) \
508 Set_Node_Length_To_R((node)-RExC_emit_start, len)
509 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
510 #define Set_Node_Cur_Length(node) \
511 Set_Node_Length(node, RExC_parse - parse_start)
513 /* Get offsets and lengths */
514 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
515 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
517 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
518 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
519 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
523 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
524 #define EXPERIMENTAL_INPLACESCAN
527 #define DEBUG_STUDYDATA(data,depth) \
528 DEBUG_OPTIMISE_r(if(data){ \
529 PerlIO_printf(Perl_debug_log, \
530 "%*s"/* Len:%"IVdf"/%"IVdf" */" Pos:%"IVdf"/%"IVdf \
531 " Flags: %"IVdf" Whilem_c: %"IVdf" Lcp: %"IVdf" ", \
532 (int)(depth)*2, "", \
533 (IV)((data)->pos_min), \
534 (IV)((data)->pos_delta), \
535 (IV)((data)->flags), \
536 (IV)((data)->whilem_c), \
537 (IV)((data)->last_closep ? *((data)->last_closep) : -1) \
539 if ((data)->last_found) \
540 PerlIO_printf(Perl_debug_log, \
541 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
542 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
543 SvPVX_const((data)->last_found), \
544 (IV)((data)->last_end), \
545 (IV)((data)->last_start_min), \
546 (IV)((data)->last_start_max), \
547 ((data)->longest && \
548 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
549 SvPVX_const((data)->longest_fixed), \
550 (IV)((data)->offset_fixed), \
551 ((data)->longest && \
552 (data)->longest==&((data)->longest_float)) ? "*" : "", \
553 SvPVX_const((data)->longest_float), \
554 (IV)((data)->offset_float_min), \
555 (IV)((data)->offset_float_max) \
557 PerlIO_printf(Perl_debug_log,"\n"); \
560 static void clear_re(pTHX_ void *r);
562 /* Mark that we cannot extend a found fixed substring at this point.
563 Update the longest found anchored substring and the longest found
564 floating substrings if needed. */
567 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp)
569 const STRLEN l = CHR_SVLEN(data->last_found);
570 const STRLEN old_l = CHR_SVLEN(*data->longest);
571 GET_RE_DEBUG_FLAGS_DECL;
573 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
574 SvSetMagicSV(*data->longest, data->last_found);
575 if (*data->longest == data->longest_fixed) {
576 data->offset_fixed = l ? data->last_start_min : data->pos_min;
577 if (data->flags & SF_BEFORE_EOL)
579 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
581 data->flags &= ~SF_FIX_BEFORE_EOL;
582 data->minlen_fixed=minlenp;
583 data->lookbehind_fixed=0;
586 data->offset_float_min = l ? data->last_start_min : data->pos_min;
587 data->offset_float_max = (l
588 ? data->last_start_max
589 : data->pos_min + data->pos_delta);
590 if ((U32)data->offset_float_max > (U32)I32_MAX)
591 data->offset_float_max = I32_MAX;
592 if (data->flags & SF_BEFORE_EOL)
594 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
596 data->flags &= ~SF_FL_BEFORE_EOL;
597 data->minlen_float=minlenp;
598 data->lookbehind_float=0;
601 SvCUR_set(data->last_found, 0);
603 SV * const sv = data->last_found;
604 if (SvUTF8(sv) && SvMAGICAL(sv)) {
605 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
611 data->flags &= ~SF_BEFORE_EOL;
612 DEBUG_STUDYDATA(data,0);
615 /* Can match anything (initialization) */
617 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
619 ANYOF_CLASS_ZERO(cl);
620 ANYOF_BITMAP_SETALL(cl);
621 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
623 cl->flags |= ANYOF_LOCALE;
626 /* Can match anything (initialization) */
628 S_cl_is_anything(const struct regnode_charclass_class *cl)
632 for (value = 0; value <= ANYOF_MAX; value += 2)
633 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
635 if (!(cl->flags & ANYOF_UNICODE_ALL))
637 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
642 /* Can match anything (initialization) */
644 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
646 Zero(cl, 1, struct regnode_charclass_class);
648 cl_anything(pRExC_state, cl);
652 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
654 Zero(cl, 1, struct regnode_charclass_class);
656 cl_anything(pRExC_state, cl);
658 cl->flags |= ANYOF_LOCALE;
661 /* 'And' a given class with another one. Can create false positives */
662 /* We assume that cl is not inverted */
664 S_cl_and(struct regnode_charclass_class *cl,
665 const struct regnode_charclass_class *and_with)
667 if (!(and_with->flags & ANYOF_CLASS)
668 && !(cl->flags & ANYOF_CLASS)
669 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
670 && !(and_with->flags & ANYOF_FOLD)
671 && !(cl->flags & ANYOF_FOLD)) {
674 if (and_with->flags & ANYOF_INVERT)
675 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
676 cl->bitmap[i] &= ~and_with->bitmap[i];
678 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
679 cl->bitmap[i] &= and_with->bitmap[i];
680 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
681 if (!(and_with->flags & ANYOF_EOS))
682 cl->flags &= ~ANYOF_EOS;
684 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
685 !(and_with->flags & ANYOF_INVERT)) {
686 cl->flags &= ~ANYOF_UNICODE_ALL;
687 cl->flags |= ANYOF_UNICODE;
688 ARG_SET(cl, ARG(and_with));
690 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
691 !(and_with->flags & ANYOF_INVERT))
692 cl->flags &= ~ANYOF_UNICODE_ALL;
693 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
694 !(and_with->flags & ANYOF_INVERT))
695 cl->flags &= ~ANYOF_UNICODE;
698 /* 'OR' a given class with another one. Can create false positives */
699 /* We assume that cl is not inverted */
701 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
703 if (or_with->flags & ANYOF_INVERT) {
705 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
706 * <= (B1 | !B2) | (CL1 | !CL2)
707 * which is wasteful if CL2 is small, but we ignore CL2:
708 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
709 * XXXX Can we handle case-fold? Unclear:
710 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
711 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
713 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
714 && !(or_with->flags & ANYOF_FOLD)
715 && !(cl->flags & ANYOF_FOLD) ) {
718 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
719 cl->bitmap[i] |= ~or_with->bitmap[i];
720 } /* XXXX: logic is complicated otherwise */
722 cl_anything(pRExC_state, cl);
725 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
726 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
727 && (!(or_with->flags & ANYOF_FOLD)
728 || (cl->flags & ANYOF_FOLD)) ) {
731 /* OR char bitmap and class bitmap separately */
732 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
733 cl->bitmap[i] |= or_with->bitmap[i];
734 if (or_with->flags & ANYOF_CLASS) {
735 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
736 cl->classflags[i] |= or_with->classflags[i];
737 cl->flags |= ANYOF_CLASS;
740 else { /* XXXX: logic is complicated, leave it along for a moment. */
741 cl_anything(pRExC_state, cl);
744 if (or_with->flags & ANYOF_EOS)
745 cl->flags |= ANYOF_EOS;
747 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
748 ARG(cl) != ARG(or_with)) {
749 cl->flags |= ANYOF_UNICODE_ALL;
750 cl->flags &= ~ANYOF_UNICODE;
752 if (or_with->flags & ANYOF_UNICODE_ALL) {
753 cl->flags |= ANYOF_UNICODE_ALL;
754 cl->flags &= ~ANYOF_UNICODE;
758 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
759 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
760 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
761 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
767 dump_trie_interim_list(trie,next_alloc)
768 dump_trie_interim_table(trie,next_alloc)
770 These routines dump out a trie in a somewhat readable format.
771 The _interim_ variants are used for debugging the interim
772 tables that are used to generate the final compressed
773 representation which is what dump_trie expects.
775 Part of the reason for their existance is to provide a form
776 of documentation as to how the different representations function.
782 Dumps the final compressed table form of the trie to Perl_debug_log.
783 Used for debugging make_trie().
787 S_dump_trie(pTHX_ const struct _reg_trie_data *trie,U32 depth)
790 SV *sv=sv_newmortal();
791 int colwidth= trie->widecharmap ? 6 : 4;
792 GET_RE_DEBUG_FLAGS_DECL;
795 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
796 (int)depth * 2 + 2,"",
797 "Match","Base","Ofs" );
799 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
800 SV ** const tmp = av_fetch( trie->revcharmap, state, 0);
802 PerlIO_printf( Perl_debug_log, "%*s",
804 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
805 PL_colors[0], PL_colors[1],
806 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
807 PERL_PV_ESCAPE_FIRSTCHAR
812 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
813 (int)depth * 2 + 2,"");
815 for( state = 0 ; state < trie->uniquecharcount ; state++ )
816 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
817 PerlIO_printf( Perl_debug_log, "\n");
819 for( state = 1 ; state < trie->laststate ; state++ ) {
820 const U32 base = trie->states[ state ].trans.base;
822 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
824 if ( trie->states[ state ].wordnum ) {
825 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
827 PerlIO_printf( Perl_debug_log, "%6s", "" );
830 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
835 while( ( base + ofs < trie->uniquecharcount ) ||
836 ( base + ofs - trie->uniquecharcount < trie->lasttrans
837 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
840 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
842 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
843 if ( ( base + ofs >= trie->uniquecharcount ) &&
844 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
845 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
847 PerlIO_printf( Perl_debug_log, "%*"UVXf,
849 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
851 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
855 PerlIO_printf( Perl_debug_log, "]");
858 PerlIO_printf( Perl_debug_log, "\n" );
862 dump_trie_interim_list(trie,next_alloc)
863 Dumps a fully constructed but uncompressed trie in list form.
864 List tries normally only are used for construction when the number of
865 possible chars (trie->uniquecharcount) is very high.
866 Used for debugging make_trie().
869 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie, U32 next_alloc,U32 depth)
872 SV *sv=sv_newmortal();
873 int colwidth= trie->widecharmap ? 6 : 4;
874 GET_RE_DEBUG_FLAGS_DECL;
875 /* print out the table precompression. */
876 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
877 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
878 "------:-----+-----------------\n" );
880 for( state=1 ; state < next_alloc ; state ++ ) {
883 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
884 (int)depth * 2 + 2,"", (UV)state );
885 if ( ! trie->states[ state ].wordnum ) {
886 PerlIO_printf( Perl_debug_log, "%5s| ","");
888 PerlIO_printf( Perl_debug_log, "W%4x| ",
889 trie->states[ state ].wordnum
892 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
893 SV ** const tmp = av_fetch( trie->revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
895 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
897 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
898 PL_colors[0], PL_colors[1],
899 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
900 PERL_PV_ESCAPE_FIRSTCHAR
902 TRIE_LIST_ITEM(state,charid).forid,
903 (UV)TRIE_LIST_ITEM(state,charid).newstate
907 PerlIO_printf( Perl_debug_log, "\n");
912 dump_trie_interim_table(trie,next_alloc)
913 Dumps a fully constructed but uncompressed trie in table form.
914 This is the normal DFA style state transition table, with a few
915 twists to facilitate compression later.
916 Used for debugging make_trie().
919 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie, U32 next_alloc, U32 depth)
923 SV *sv=sv_newmortal();
924 int colwidth= trie->widecharmap ? 6 : 4;
925 GET_RE_DEBUG_FLAGS_DECL;
928 print out the table precompression so that we can do a visual check
929 that they are identical.
932 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
934 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
935 SV ** const tmp = av_fetch( trie->revcharmap, charid, 0);
937 PerlIO_printf( Perl_debug_log, "%*s",
939 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
940 PL_colors[0], PL_colors[1],
941 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
942 PERL_PV_ESCAPE_FIRSTCHAR
948 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
950 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
951 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
954 PerlIO_printf( Perl_debug_log, "\n" );
956 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
958 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
959 (int)depth * 2 + 2,"",
960 (UV)TRIE_NODENUM( state ) );
962 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
963 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
965 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
967 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
969 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
970 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
972 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
973 trie->states[ TRIE_NODENUM( state ) ].wordnum );
980 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
981 startbranch: the first branch in the whole branch sequence
982 first : start branch of sequence of branch-exact nodes.
983 May be the same as startbranch
984 last : Thing following the last branch.
985 May be the same as tail.
986 tail : item following the branch sequence
987 count : words in the sequence
988 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
991 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
993 A trie is an N'ary tree where the branches are determined by digital
994 decomposition of the key. IE, at the root node you look up the 1st character and
995 follow that branch repeat until you find the end of the branches. Nodes can be
996 marked as "accepting" meaning they represent a complete word. Eg:
1000 would convert into the following structure. Numbers represent states, letters
1001 following numbers represent valid transitions on the letter from that state, if
1002 the number is in square brackets it represents an accepting state, otherwise it
1003 will be in parenthesis.
1005 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1009 (1) +-i->(6)-+-s->[7]
1011 +-s->(3)-+-h->(4)-+-e->[5]
1013 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1015 This shows that when matching against the string 'hers' we will begin at state 1
1016 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1017 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1018 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1019 single traverse. We store a mapping from accepting to state to which word was
1020 matched, and then when we have multiple possibilities we try to complete the
1021 rest of the regex in the order in which they occured in the alternation.
1023 The only prior NFA like behaviour that would be changed by the TRIE support is
1024 the silent ignoring of duplicate alternations which are of the form:
1026 / (DUPE|DUPE) X? (?{ ... }) Y /x
1028 Thus EVAL blocks follwing a trie may be called a different number of times with
1029 and without the optimisation. With the optimisations dupes will be silently
1030 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1031 the following demonstrates:
1033 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1035 which prints out 'word' three times, but
1037 'words'=~/(word|word|word)(?{ print $1 })S/
1039 which doesnt print it out at all. This is due to other optimisations kicking in.
1041 Example of what happens on a structural level:
1043 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1045 1: CURLYM[1] {1,32767}(18)
1056 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1057 and should turn into:
1059 1: CURLYM[1] {1,32767}(18)
1061 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1069 Cases where tail != last would be like /(?foo|bar)baz/:
1079 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1080 and would end up looking like:
1083 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1090 d = uvuni_to_utf8_flags(d, uv, 0);
1092 is the recommended Unicode-aware way of saying
1097 #define TRIE_STORE_REVCHAR \
1099 SV *tmp = Perl_newSVpvf_nocontext( "%c", (int)uvc ); \
1100 if (UTF) SvUTF8_on(tmp); \
1101 av_push( TRIE_REVCHARMAP(trie), tmp ); \
1104 #define TRIE_READ_CHAR STMT_START { \
1108 if ( foldlen > 0 ) { \
1109 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1114 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1115 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1116 foldlen -= UNISKIP( uvc ); \
1117 scan = foldbuf + UNISKIP( uvc ); \
1120 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1130 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1131 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1132 TRIE_LIST_LEN( state ) *= 2; \
1133 Renew( trie->states[ state ].trans.list, \
1134 TRIE_LIST_LEN( state ), reg_trie_trans_le ); \
1136 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1137 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1138 TRIE_LIST_CUR( state )++; \
1141 #define TRIE_LIST_NEW(state) STMT_START { \
1142 Newxz( trie->states[ state ].trans.list, \
1143 4, reg_trie_trans_le ); \
1144 TRIE_LIST_CUR( state ) = 1; \
1145 TRIE_LIST_LEN( state ) = 4; \
1148 #define TRIE_HANDLE_WORD(state) STMT_START { \
1149 U16 dupe= trie->states[ state ].wordnum; \
1150 regnode * const noper_next = regnext( noper ); \
1152 if (trie->wordlen) \
1153 trie->wordlen[ curword ] = wordlen; \
1155 /* store the word for dumping */ \
1157 if (OP(noper) != NOTHING) \
1158 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1160 tmp = newSVpvn( "", 0 ); \
1161 if ( UTF ) SvUTF8_on( tmp ); \
1162 av_push( trie->words, tmp ); \
1167 if ( noper_next < tail ) { \
1169 Newxz( trie->jump, word_count + 1, U16); \
1170 trie->jump[curword] = (U16)(tail - noper_next); \
1172 jumper = noper_next; \
1174 nextbranch= regnext(cur); \
1178 /* So it's a dupe. This means we need to maintain a */\
1179 /* linked-list from the first to the next. */\
1180 /* we only allocate the nextword buffer when there */\
1181 /* a dupe, so first time we have to do the allocation */\
1182 if (!trie->nextword) \
1183 Newxz( trie->nextword, word_count + 1, U16); \
1184 while ( trie->nextword[dupe] ) \
1185 dupe= trie->nextword[dupe]; \
1186 trie->nextword[dupe]= curword; \
1188 /* we haven't inserted this word yet. */ \
1189 trie->states[ state ].wordnum = curword; \
1194 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1195 ( ( base + charid >= ucharcount \
1196 && base + charid < ubound \
1197 && state == trie->trans[ base - ucharcount + charid ].check \
1198 && trie->trans[ base - ucharcount + charid ].next ) \
1199 ? trie->trans[ base - ucharcount + charid ].next \
1200 : ( state==1 ? special : 0 ) \
1204 #define MADE_JUMP_TRIE 2
1205 #define MADE_EXACT_TRIE 4
1208 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1211 /* first pass, loop through and scan words */
1212 reg_trie_data *trie;
1214 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1219 regnode *jumper = NULL;
1220 regnode *nextbranch = NULL;
1221 /* we just use folder as a flag in utf8 */
1222 const U8 * const folder = ( flags == EXACTF
1224 : ( flags == EXACTFL
1230 const U32 data_slot = add_data( pRExC_state, 1, "t" );
1231 SV *re_trie_maxbuff;
1233 /* these are only used during construction but are useful during
1234 * debugging so we store them in the struct when debugging.
1236 STRLEN trie_charcount=0;
1237 AV *trie_revcharmap;
1239 GET_RE_DEBUG_FLAGS_DECL;
1241 PERL_UNUSED_ARG(depth);
1244 Newxz( trie, 1, reg_trie_data );
1246 trie->startstate = 1;
1247 trie->wordcount = word_count;
1248 RExC_rx->data->data[ data_slot ] = (void*)trie;
1249 Newxz( trie->charmap, 256, U16 );
1250 if (!(UTF && folder))
1251 Newxz( trie->bitmap, ANYOF_BITMAP_SIZE, char );
1253 trie->words = newAV();
1255 TRIE_REVCHARMAP(trie) = newAV();
1257 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1258 if (!SvIOK(re_trie_maxbuff)) {
1259 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1262 PerlIO_printf( Perl_debug_log,
1263 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1264 (int)depth * 2 + 2, "",
1265 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1266 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1269 /* -- First loop and Setup --
1271 We first traverse the branches and scan each word to determine if it
1272 contains widechars, and how many unique chars there are, this is
1273 important as we have to build a table with at least as many columns as we
1276 We use an array of integers to represent the character codes 0..255
1277 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1278 native representation of the character value as the key and IV's for the
1281 *TODO* If we keep track of how many times each character is used we can
1282 remap the columns so that the table compression later on is more
1283 efficient in terms of memory by ensuring most common value is in the
1284 middle and the least common are on the outside. IMO this would be better
1285 than a most to least common mapping as theres a decent chance the most
1286 common letter will share a node with the least common, meaning the node
1287 will not be compressable. With a middle is most common approach the worst
1288 case is when we have the least common nodes twice.
1292 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1293 regnode * const noper = NEXTOPER( cur );
1294 const U8 *uc = (U8*)STRING( noper );
1295 const U8 * const e = uc + STR_LEN( noper );
1297 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1298 const U8 *scan = (U8*)NULL;
1299 U32 wordlen = 0; /* required init */
1302 if (OP(noper) == NOTHING) {
1307 TRIE_BITMAP_SET(trie,*uc);
1308 if ( folder ) TRIE_BITMAP_SET(trie,folder[ *uc ]);
1310 for ( ; uc < e ; uc += len ) {
1311 TRIE_CHARCOUNT(trie)++;
1315 if ( !trie->charmap[ uvc ] ) {
1316 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1318 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1323 if ( !trie->widecharmap )
1324 trie->widecharmap = newHV();
1326 svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1329 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1331 if ( !SvTRUE( *svpp ) ) {
1332 sv_setiv( *svpp, ++trie->uniquecharcount );
1337 if( cur == first ) {
1340 } else if (chars < trie->minlen) {
1342 } else if (chars > trie->maxlen) {
1346 } /* end first pass */
1347 DEBUG_TRIE_COMPILE_r(
1348 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1349 (int)depth * 2 + 2,"",
1350 ( trie->widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1351 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1352 (int)trie->minlen, (int)trie->maxlen )
1354 Newxz( trie->wordlen, word_count, U32 );
1357 We now know what we are dealing with in terms of unique chars and
1358 string sizes so we can calculate how much memory a naive
1359 representation using a flat table will take. If it's over a reasonable
1360 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1361 conservative but potentially much slower representation using an array
1364 At the end we convert both representations into the same compressed
1365 form that will be used in regexec.c for matching with. The latter
1366 is a form that cannot be used to construct with but has memory
1367 properties similar to the list form and access properties similar
1368 to the table form making it both suitable for fast searches and
1369 small enough that its feasable to store for the duration of a program.
1371 See the comment in the code where the compressed table is produced
1372 inplace from the flat tabe representation for an explanation of how
1373 the compression works.
1378 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1380 Second Pass -- Array Of Lists Representation
1382 Each state will be represented by a list of charid:state records
1383 (reg_trie_trans_le) the first such element holds the CUR and LEN
1384 points of the allocated array. (See defines above).
1386 We build the initial structure using the lists, and then convert
1387 it into the compressed table form which allows faster lookups
1388 (but cant be modified once converted).
1391 STRLEN transcount = 1;
1393 Newxz( trie->states, TRIE_CHARCOUNT(trie) + 2, reg_trie_state );
1397 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1399 regnode * const noper = NEXTOPER( cur );
1400 U8 *uc = (U8*)STRING( noper );
1401 const U8 * const e = uc + STR_LEN( noper );
1402 U32 state = 1; /* required init */
1403 U16 charid = 0; /* sanity init */
1404 U8 *scan = (U8*)NULL; /* sanity init */
1405 STRLEN foldlen = 0; /* required init */
1406 U32 wordlen = 0; /* required init */
1407 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1409 if (OP(noper) != NOTHING) {
1410 for ( ; uc < e ; uc += len ) {
1415 charid = trie->charmap[ uvc ];
1417 SV** const svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0);
1421 charid=(U16)SvIV( *svpp );
1424 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1431 if ( !trie->states[ state ].trans.list ) {
1432 TRIE_LIST_NEW( state );
1434 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1435 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1436 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1441 newstate = next_alloc++;
1442 TRIE_LIST_PUSH( state, charid, newstate );
1447 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1451 TRIE_HANDLE_WORD(state);
1453 } /* end second pass */
1455 trie->laststate = next_alloc;
1456 Renew( trie->states, next_alloc, reg_trie_state );
1458 /* and now dump it out before we compress it */
1459 DEBUG_TRIE_COMPILE_MORE_r(
1460 dump_trie_interim_list(trie,next_alloc,depth+1)
1463 Newxz( trie->trans, transcount ,reg_trie_trans );
1470 for( state=1 ; state < next_alloc ; state ++ ) {
1474 DEBUG_TRIE_COMPILE_MORE_r(
1475 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1479 if (trie->states[state].trans.list) {
1480 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1484 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1485 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1486 if ( forid < minid ) {
1488 } else if ( forid > maxid ) {
1492 if ( transcount < tp + maxid - minid + 1) {
1494 Renew( trie->trans, transcount, reg_trie_trans );
1495 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1497 base = trie->uniquecharcount + tp - minid;
1498 if ( maxid == minid ) {
1500 for ( ; zp < tp ; zp++ ) {
1501 if ( ! trie->trans[ zp ].next ) {
1502 base = trie->uniquecharcount + zp - minid;
1503 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1504 trie->trans[ zp ].check = state;
1510 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1511 trie->trans[ tp ].check = state;
1516 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1517 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1518 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1519 trie->trans[ tid ].check = state;
1521 tp += ( maxid - minid + 1 );
1523 Safefree(trie->states[ state ].trans.list);
1526 DEBUG_TRIE_COMPILE_MORE_r(
1527 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1530 trie->states[ state ].trans.base=base;
1532 trie->lasttrans = tp + 1;
1536 Second Pass -- Flat Table Representation.
1538 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1539 We know that we will need Charcount+1 trans at most to store the data
1540 (one row per char at worst case) So we preallocate both structures
1541 assuming worst case.
1543 We then construct the trie using only the .next slots of the entry
1546 We use the .check field of the first entry of the node temporarily to
1547 make compression both faster and easier by keeping track of how many non
1548 zero fields are in the node.
1550 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1553 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1554 number representing the first entry of the node, and state as a
1555 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1556 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1557 are 2 entrys per node. eg:
1565 The table is internally in the right hand, idx form. However as we also
1566 have to deal with the states array which is indexed by nodenum we have to
1567 use TRIE_NODENUM() to convert.
1572 Newxz( trie->trans, ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1,
1574 Newxz( trie->states, TRIE_CHARCOUNT(trie) + 2, reg_trie_state );
1575 next_alloc = trie->uniquecharcount + 1;
1578 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1580 regnode * const noper = NEXTOPER( cur );
1581 const U8 *uc = (U8*)STRING( noper );
1582 const U8 * const e = uc + STR_LEN( noper );
1584 U32 state = 1; /* required init */
1586 U16 charid = 0; /* sanity init */
1587 U32 accept_state = 0; /* sanity init */
1588 U8 *scan = (U8*)NULL; /* sanity init */
1590 STRLEN foldlen = 0; /* required init */
1591 U32 wordlen = 0; /* required init */
1592 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1594 if ( OP(noper) != NOTHING ) {
1595 for ( ; uc < e ; uc += len ) {
1600 charid = trie->charmap[ uvc ];
1602 SV* const * const svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0);
1603 charid = svpp ? (U16)SvIV(*svpp) : 0;
1607 if ( !trie->trans[ state + charid ].next ) {
1608 trie->trans[ state + charid ].next = next_alloc;
1609 trie->trans[ state ].check++;
1610 next_alloc += trie->uniquecharcount;
1612 state = trie->trans[ state + charid ].next;
1614 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1616 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1619 accept_state = TRIE_NODENUM( state );
1620 TRIE_HANDLE_WORD(accept_state);
1622 } /* end second pass */
1624 /* and now dump it out before we compress it */
1625 DEBUG_TRIE_COMPILE_MORE_r(
1626 dump_trie_interim_table(trie,next_alloc,depth+1)
1631 * Inplace compress the table.*
1633 For sparse data sets the table constructed by the trie algorithm will
1634 be mostly 0/FAIL transitions or to put it another way mostly empty.
1635 (Note that leaf nodes will not contain any transitions.)
1637 This algorithm compresses the tables by eliminating most such
1638 transitions, at the cost of a modest bit of extra work during lookup:
1640 - Each states[] entry contains a .base field which indicates the
1641 index in the state[] array wheres its transition data is stored.
1643 - If .base is 0 there are no valid transitions from that node.
1645 - If .base is nonzero then charid is added to it to find an entry in
1648 -If trans[states[state].base+charid].check!=state then the
1649 transition is taken to be a 0/Fail transition. Thus if there are fail
1650 transitions at the front of the node then the .base offset will point
1651 somewhere inside the previous nodes data (or maybe even into a node
1652 even earlier), but the .check field determines if the transition is
1656 The following process inplace converts the table to the compressed
1657 table: We first do not compress the root node 1,and mark its all its
1658 .check pointers as 1 and set its .base pointer as 1 as well. This
1659 allows to do a DFA construction from the compressed table later, and
1660 ensures that any .base pointers we calculate later are greater than
1663 - We set 'pos' to indicate the first entry of the second node.
1665 - We then iterate over the columns of the node, finding the first and
1666 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1667 and set the .check pointers accordingly, and advance pos
1668 appropriately and repreat for the next node. Note that when we copy
1669 the next pointers we have to convert them from the original
1670 NODEIDX form to NODENUM form as the former is not valid post
1673 - If a node has no transitions used we mark its base as 0 and do not
1674 advance the pos pointer.
1676 - If a node only has one transition we use a second pointer into the
1677 structure to fill in allocated fail transitions from other states.
1678 This pointer is independent of the main pointer and scans forward
1679 looking for null transitions that are allocated to a state. When it
1680 finds one it writes the single transition into the "hole". If the
1681 pointer doesnt find one the single transition is appended as normal.
1683 - Once compressed we can Renew/realloc the structures to release the
1686 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1687 specifically Fig 3.47 and the associated pseudocode.
1691 const U32 laststate = TRIE_NODENUM( next_alloc );
1694 trie->laststate = laststate;
1696 for ( state = 1 ; state < laststate ; state++ ) {
1698 const U32 stateidx = TRIE_NODEIDX( state );
1699 const U32 o_used = trie->trans[ stateidx ].check;
1700 U32 used = trie->trans[ stateidx ].check;
1701 trie->trans[ stateidx ].check = 0;
1703 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1704 if ( flag || trie->trans[ stateidx + charid ].next ) {
1705 if ( trie->trans[ stateidx + charid ].next ) {
1707 for ( ; zp < pos ; zp++ ) {
1708 if ( ! trie->trans[ zp ].next ) {
1712 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1713 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1714 trie->trans[ zp ].check = state;
1715 if ( ++zp > pos ) pos = zp;
1722 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1724 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1725 trie->trans[ pos ].check = state;
1730 trie->lasttrans = pos + 1;
1731 Renew( trie->states, laststate + 1, reg_trie_state);
1732 DEBUG_TRIE_COMPILE_MORE_r(
1733 PerlIO_printf( Perl_debug_log,
1734 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1735 (int)depth * 2 + 2,"",
1736 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1739 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1742 } /* end table compress */
1744 /* resize the trans array to remove unused space */
1745 Renew( trie->trans, trie->lasttrans, reg_trie_trans);
1747 /* and now dump out the compressed format */
1748 DEBUG_TRIE_COMPILE_r(
1749 dump_trie(trie,depth+1)
1752 { /* Modify the program and insert the new TRIE node*/
1754 U8 nodetype =(U8)(flags & 0xFF);
1760 U32 mjd_nodelen = 0;
1763 This means we convert either the first branch or the first Exact,
1764 depending on whether the thing following (in 'last') is a branch
1765 or not and whther first is the startbranch (ie is it a sub part of
1766 the alternation or is it the whole thing.)
1767 Assuming its a sub part we conver the EXACT otherwise we convert
1768 the whole branch sequence, including the first.
1770 /* Find the node we are going to overwrite */
1771 if ( first == startbranch && OP( last ) != BRANCH ) {
1772 /* whole branch chain */
1775 const regnode *nop = NEXTOPER( convert );
1776 mjd_offset= Node_Offset((nop));
1777 mjd_nodelen= Node_Length((nop));
1780 /* branch sub-chain */
1781 convert = NEXTOPER( first );
1782 NEXT_OFF( first ) = (U16)(last - first);
1784 mjd_offset= Node_Offset((convert));
1785 mjd_nodelen= Node_Length((convert));
1789 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1790 (int)depth * 2 + 2, "",
1791 (UV)mjd_offset, (UV)mjd_nodelen)
1794 /* But first we check to see if there is a common prefix we can
1795 split out as an EXACT and put in front of the TRIE node. */
1796 trie->startstate= 1;
1797 if ( trie->bitmap && !trie->widecharmap && !trie->jump ) {
1800 PerlIO_printf(Perl_debug_log, "%*sLaststate:%"UVuf"\n",
1801 (int)depth * 2 + 2, "",
1802 (UV)trie->laststate)
1804 for ( state = 1 ; state < trie->laststate-1 ; state++ ) {
1808 const U32 base = trie->states[ state ].trans.base;
1810 if ( trie->states[state].wordnum )
1813 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1814 if ( ( base + ofs >= trie->uniquecharcount ) &&
1815 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1816 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1818 if ( ++count > 1 ) {
1819 SV **tmp = av_fetch( TRIE_REVCHARMAP(trie), ofs, 0);
1820 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1821 if ( state == 1 ) break;
1823 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1825 PerlIO_printf(Perl_debug_log,
1826 "%*sNew Start State=%"UVuf" Class: [",
1827 (int)depth * 2 + 2, "",
1830 SV ** const tmp = av_fetch( TRIE_REVCHARMAP(trie), idx, 0);
1831 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1833 TRIE_BITMAP_SET(trie,*ch);
1835 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1837 PerlIO_printf(Perl_debug_log, (char*)ch)
1841 TRIE_BITMAP_SET(trie,*ch);
1843 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1844 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1850 SV **tmp = av_fetch( TRIE_REVCHARMAP(trie), idx, 0);
1851 const char *ch = SvPV_nolen_const( *tmp );
1853 PerlIO_printf( Perl_debug_log,
1854 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1855 (int)depth * 2 + 2, "",
1856 (UV)state, (UV)idx, ch)
1859 OP( convert ) = nodetype;
1860 str=STRING(convert);
1869 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
1875 regnode *n = convert+NODE_SZ_STR(convert);
1876 NEXT_OFF(convert) = NODE_SZ_STR(convert);
1877 trie->startstate = state;
1878 trie->minlen -= (state - 1);
1879 trie->maxlen -= (state - 1);
1881 regnode *fix = convert;
1883 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
1884 while( ++fix < n ) {
1885 Set_Node_Offset_Length(fix, 0, 0);
1891 NEXT_OFF(convert) = (U16)(tail - convert);
1895 if ( trie->maxlen ) {
1896 NEXT_OFF( convert ) = (U16)(tail - convert);
1897 ARG_SET( convert, data_slot );
1898 /* Store the offset to the first unabsorbed branch in
1899 jump[0], which is otherwise unused by the jump logic.
1900 We use this when dumping a trie and during optimisation. */
1902 trie->jump[0] = (U16)(tail - nextbranch);
1906 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
1907 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
1909 OP( convert ) = TRIEC;
1910 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
1911 Safefree(trie->bitmap);
1914 OP( convert ) = TRIE;
1916 /* store the type in the flags */
1917 convert->flags = nodetype;
1918 /* XXX We really should free up the resource in trie now, as we wont use them */
1920 /* needed for dumping*/
1922 regnode *optimize = convert
1924 + regarglen[ OP( convert ) ];
1925 regnode *opt = convert;
1926 while (++opt<optimize) {
1927 Set_Node_Offset_Length(opt,0,0);
1930 Try to clean up some of the debris left after the
1933 while( optimize < jumper ) {
1934 mjd_nodelen += Node_Length((optimize));
1935 OP( optimize ) = OPTIMIZED;
1936 Set_Node_Offset_Length(optimize,0,0);
1939 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
1941 } /* end node insert */
1943 SvREFCNT_dec(TRIE_REVCHARMAP(trie));
1947 : trie->startstate>1
1953 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
1955 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
1957 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
1958 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
1961 We find the fail state for each state in the trie, this state is the longest proper
1962 suffix of the current states 'word' that is also a proper prefix of another word in our
1963 trie. State 1 represents the word '' and is the thus the default fail state. This allows
1964 the DFA not to have to restart after its tried and failed a word at a given point, it
1965 simply continues as though it had been matching the other word in the first place.
1967 'abcdgu'=~/abcdefg|cdgu/
1968 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
1969 fail, which would bring use to the state representing 'd' in the second word where we would
1970 try 'g' and succeed, prodceding to match 'cdgu'.
1972 /* add a fail transition */
1973 reg_trie_data *trie=(reg_trie_data *)RExC_rx->data->data[ARG(source)];
1975 const U32 ucharcount = trie->uniquecharcount;
1976 const U32 numstates = trie->laststate;
1977 const U32 ubound = trie->lasttrans + ucharcount;
1981 U32 base = trie->states[ 1 ].trans.base;
1984 const U32 data_slot = add_data( pRExC_state, 1, "T" );
1985 GET_RE_DEBUG_FLAGS_DECL;
1987 PERL_UNUSED_ARG(depth);
1991 ARG_SET( stclass, data_slot );
1992 Newxz( aho, 1, reg_ac_data );
1993 RExC_rx->data->data[ data_slot ] = (void*)aho;
1995 aho->states=(reg_trie_state *)savepvn((const char*)trie->states,
1996 (trie->laststate+1)*sizeof(reg_trie_state));
1997 Newxz( q, numstates, U32);
1998 Newxz( aho->fail, numstates, U32 );
2001 /* initialize fail[0..1] to be 1 so that we always have
2002 a valid final fail state */
2003 fail[ 0 ] = fail[ 1 ] = 1;
2005 for ( charid = 0; charid < ucharcount ; charid++ ) {
2006 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2008 q[ q_write ] = newstate;
2009 /* set to point at the root */
2010 fail[ q[ q_write++ ] ]=1;
2013 while ( q_read < q_write) {
2014 const U32 cur = q[ q_read++ % numstates ];
2015 base = trie->states[ cur ].trans.base;
2017 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2018 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2020 U32 fail_state = cur;
2023 fail_state = fail[ fail_state ];
2024 fail_base = aho->states[ fail_state ].trans.base;
2025 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2027 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2028 fail[ ch_state ] = fail_state;
2029 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2031 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2033 q[ q_write++ % numstates] = ch_state;
2037 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2038 when we fail in state 1, this allows us to use the
2039 charclass scan to find a valid start char. This is based on the principle
2040 that theres a good chance the string being searched contains lots of stuff
2041 that cant be a start char.
2043 fail[ 0 ] = fail[ 1 ] = 0;
2044 DEBUG_TRIE_COMPILE_r({
2045 PerlIO_printf(Perl_debug_log, "%*sStclass Failtable: 0", (int)(depth * 2), "");
2046 for( q_read=1; q_read<numstates; q_read++ ) {
2047 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2049 PerlIO_printf(Perl_debug_log, "\n");
2052 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2057 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2058 * These need to be revisited when a newer toolchain becomes available.
2060 #if defined(__sparc64__) && defined(__GNUC__)
2061 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2062 # undef SPARC64_GCC_WORKAROUND
2063 # define SPARC64_GCC_WORKAROUND 1
2067 #define DEBUG_PEEP(str,scan,depth) \
2068 DEBUG_OPTIMISE_r({ \
2069 SV * const mysv=sv_newmortal(); \
2070 regnode *Next = regnext(scan); \
2071 regprop(RExC_rx, mysv, scan); \
2072 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s [%d]\n", \
2073 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2074 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2081 #define JOIN_EXACT(scan,min,flags) \
2082 if (PL_regkind[OP(scan)] == EXACT) \
2083 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2086 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2087 /* Merge several consecutive EXACTish nodes into one. */
2088 regnode *n = regnext(scan);
2090 regnode *next = scan + NODE_SZ_STR(scan);
2094 regnode *stop = scan;
2095 GET_RE_DEBUG_FLAGS_DECL;
2097 PERL_UNUSED_ARG(depth);
2099 #ifndef EXPERIMENTAL_INPLACESCAN
2100 PERL_UNUSED_ARG(flags);
2101 PERL_UNUSED_ARG(val);
2103 DEBUG_PEEP("join",scan,depth);
2105 /* Skip NOTHING, merge EXACT*. */
2107 ( PL_regkind[OP(n)] == NOTHING ||
2108 (stringok && (OP(n) == OP(scan))))
2110 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2112 if (OP(n) == TAIL || n > next)
2114 if (PL_regkind[OP(n)] == NOTHING) {
2115 DEBUG_PEEP("skip:",n,depth);
2116 NEXT_OFF(scan) += NEXT_OFF(n);
2117 next = n + NODE_STEP_REGNODE;
2124 else if (stringok) {
2125 const unsigned int oldl = STR_LEN(scan);
2126 regnode * const nnext = regnext(n);
2128 DEBUG_PEEP("merg",n,depth);
2131 if (oldl + STR_LEN(n) > U8_MAX)
2133 NEXT_OFF(scan) += NEXT_OFF(n);
2134 STR_LEN(scan) += STR_LEN(n);
2135 next = n + NODE_SZ_STR(n);
2136 /* Now we can overwrite *n : */
2137 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2145 #ifdef EXPERIMENTAL_INPLACESCAN
2146 if (flags && !NEXT_OFF(n)) {
2147 DEBUG_PEEP("atch", val, depth);
2148 if (reg_off_by_arg[OP(n)]) {
2149 ARG_SET(n, val - n);
2152 NEXT_OFF(n) = val - n;
2159 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2161 Two problematic code points in Unicode casefolding of EXACT nodes:
2163 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2164 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2170 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2171 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2173 This means that in case-insensitive matching (or "loose matching",
2174 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2175 length of the above casefolded versions) can match a target string
2176 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2177 This would rather mess up the minimum length computation.
2179 What we'll do is to look for the tail four bytes, and then peek
2180 at the preceding two bytes to see whether we need to decrease
2181 the minimum length by four (six minus two).
2183 Thanks to the design of UTF-8, there cannot be false matches:
2184 A sequence of valid UTF-8 bytes cannot be a subsequence of
2185 another valid sequence of UTF-8 bytes.
2188 char * const s0 = STRING(scan), *s, *t;
2189 char * const s1 = s0 + STR_LEN(scan) - 1;
2190 char * const s2 = s1 - 4;
2191 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2192 const char t0[] = "\xaf\x49\xaf\x42";
2194 const char t0[] = "\xcc\x88\xcc\x81";
2196 const char * const t1 = t0 + 3;
2199 s < s2 && (t = ninstr(s, s1, t0, t1));
2202 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2203 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2205 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2206 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2214 n = scan + NODE_SZ_STR(scan);
2216 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2223 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2227 /* REx optimizer. Converts nodes into quickier variants "in place".
2228 Finds fixed substrings. */
2230 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2231 to the position after last scanned or to NULL. */
2236 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2237 I32 *minlenp, I32 *deltap,
2238 regnode *last, scan_data_t *data, U32 flags, U32 depth)
2239 /* scanp: Start here (read-write). */
2240 /* deltap: Write maxlen-minlen here. */
2241 /* last: Stop before this one. */
2244 I32 min = 0, pars = 0, code;
2245 regnode *scan = *scanp, *next;
2247 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2248 int is_inf_internal = 0; /* The studied chunk is infinite */
2249 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2250 scan_data_t data_fake;
2251 struct regnode_charclass_class and_with; /* Valid if flags & SCF_DO_STCLASS_OR */
2252 SV *re_trie_maxbuff = NULL;
2253 regnode *first_non_open = scan;
2256 GET_RE_DEBUG_FLAGS_DECL;
2258 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2261 while (first_non_open && OP(first_non_open) == OPEN)
2262 first_non_open=regnext(first_non_open);
2266 while (scan && OP(scan) != END && scan < last) {
2267 /* Peephole optimizer: */
2268 DEBUG_STUDYDATA(data,depth);
2269 DEBUG_PEEP("Peep",scan,depth);
2270 JOIN_EXACT(scan,&min,0);
2272 /* Follow the next-chain of the current node and optimize
2273 away all the NOTHINGs from it. */
2274 if (OP(scan) != CURLYX) {
2275 const int max = (reg_off_by_arg[OP(scan)]
2277 /* I32 may be smaller than U16 on CRAYs! */
2278 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2279 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2283 /* Skip NOTHING and LONGJMP. */
2284 while ((n = regnext(n))
2285 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2286 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2287 && off + noff < max)
2289 if (reg_off_by_arg[OP(scan)])
2292 NEXT_OFF(scan) = off;
2297 /* The principal pseudo-switch. Cannot be a switch, since we
2298 look into several different things. */
2299 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2300 || OP(scan) == IFTHEN || OP(scan) == SUSPEND) {
2301 next = regnext(scan);
2303 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2305 if (OP(next) == code || code == IFTHEN || code == SUSPEND) {
2306 /* NOTE - There is similar code to this block below for handling
2307 TRIE nodes on a re-study. If you change stuff here check there
2309 I32 max1 = 0, min1 = I32_MAX, num = 0;
2310 struct regnode_charclass_class accum;
2311 regnode * const startbranch=scan;
2313 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
2314 scan_commit(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2315 if (flags & SCF_DO_STCLASS)
2316 cl_init_zero(pRExC_state, &accum);
2318 while (OP(scan) == code) {
2319 I32 deltanext, minnext, f = 0, fake;
2320 struct regnode_charclass_class this_class;
2323 data_fake.flags = 0;
2325 data_fake.whilem_c = data->whilem_c;
2326 data_fake.last_closep = data->last_closep;
2329 data_fake.last_closep = &fake;
2330 next = regnext(scan);
2331 scan = NEXTOPER(scan);
2333 scan = NEXTOPER(scan);
2334 if (flags & SCF_DO_STCLASS) {
2335 cl_init(pRExC_state, &this_class);
2336 data_fake.start_class = &this_class;
2337 f = SCF_DO_STCLASS_AND;
2339 if (flags & SCF_WHILEM_VISITED_POS)
2340 f |= SCF_WHILEM_VISITED_POS;
2342 /* we suppose the run is continuous, last=next...*/
2343 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2344 next, &data_fake, f,depth+1);
2347 if (max1 < minnext + deltanext)
2348 max1 = minnext + deltanext;
2349 if (deltanext == I32_MAX)
2350 is_inf = is_inf_internal = 1;
2352 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2355 if (data_fake.flags & SF_HAS_EVAL)
2356 data->flags |= SF_HAS_EVAL;
2357 data->whilem_c = data_fake.whilem_c;
2359 if (flags & SCF_DO_STCLASS)
2360 cl_or(pRExC_state, &accum, &this_class);
2361 if (code == SUSPEND)
2364 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2366 if (flags & SCF_DO_SUBSTR) {
2367 data->pos_min += min1;
2368 data->pos_delta += max1 - min1;
2369 if (max1 != min1 || is_inf)
2370 data->longest = &(data->longest_float);
2373 delta += max1 - min1;
2374 if (flags & SCF_DO_STCLASS_OR) {
2375 cl_or(pRExC_state, data->start_class, &accum);
2377 cl_and(data->start_class, &and_with);
2378 flags &= ~SCF_DO_STCLASS;
2381 else if (flags & SCF_DO_STCLASS_AND) {
2383 cl_and(data->start_class, &accum);
2384 flags &= ~SCF_DO_STCLASS;
2387 /* Switch to OR mode: cache the old value of
2388 * data->start_class */
2389 StructCopy(data->start_class, &and_with,
2390 struct regnode_charclass_class);
2391 flags &= ~SCF_DO_STCLASS_AND;
2392 StructCopy(&accum, data->start_class,
2393 struct regnode_charclass_class);
2394 flags |= SCF_DO_STCLASS_OR;
2395 data->start_class->flags |= ANYOF_EOS;
2399 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2402 Assuming this was/is a branch we are dealing with: 'scan' now
2403 points at the item that follows the branch sequence, whatever
2404 it is. We now start at the beginning of the sequence and look
2411 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2413 If we can find such a subseqence we need to turn the first
2414 element into a trie and then add the subsequent branch exact
2415 strings to the trie.
2419 1. patterns where the whole set of branch can be converted.
2421 2. patterns where only a subset can be converted.
2423 In case 1 we can replace the whole set with a single regop
2424 for the trie. In case 2 we need to keep the start and end
2427 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2428 becomes BRANCH TRIE; BRANCH X;
2430 There is an additional case, that being where there is a
2431 common prefix, which gets split out into an EXACT like node
2432 preceding the TRIE node.
2434 If x(1..n)==tail then we can do a simple trie, if not we make
2435 a "jump" trie, such that when we match the appropriate word
2436 we "jump" to the appopriate tail node. Essentailly we turn
2437 a nested if into a case structure of sorts.
2442 if (!re_trie_maxbuff) {
2443 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2444 if (!SvIOK(re_trie_maxbuff))
2445 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2447 if ( SvIV(re_trie_maxbuff)>=0 ) {
2449 regnode *first = (regnode *)NULL;
2450 regnode *last = (regnode *)NULL;
2451 regnode *tail = scan;
2456 SV * const mysv = sv_newmortal(); /* for dumping */
2458 /* var tail is used because there may be a TAIL
2459 regop in the way. Ie, the exacts will point to the
2460 thing following the TAIL, but the last branch will
2461 point at the TAIL. So we advance tail. If we
2462 have nested (?:) we may have to move through several
2466 while ( OP( tail ) == TAIL ) {
2467 /* this is the TAIL generated by (?:) */
2468 tail = regnext( tail );
2473 regprop(RExC_rx, mysv, tail );
2474 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2475 (int)depth * 2 + 2, "",
2476 "Looking for TRIE'able sequences. Tail node is: ",
2477 SvPV_nolen_const( mysv )
2483 step through the branches, cur represents each
2484 branch, noper is the first thing to be matched
2485 as part of that branch and noper_next is the
2486 regnext() of that node. if noper is an EXACT
2487 and noper_next is the same as scan (our current
2488 position in the regex) then the EXACT branch is
2489 a possible optimization target. Once we have
2490 two or more consequetive such branches we can
2491 create a trie of the EXACT's contents and stich
2492 it in place. If the sequence represents all of
2493 the branches we eliminate the whole thing and
2494 replace it with a single TRIE. If it is a
2495 subsequence then we need to stitch it in. This
2496 means the first branch has to remain, and needs
2497 to be repointed at the item on the branch chain
2498 following the last branch optimized. This could
2499 be either a BRANCH, in which case the
2500 subsequence is internal, or it could be the
2501 item following the branch sequence in which
2502 case the subsequence is at the end.
2506 /* dont use tail as the end marker for this traverse */
2507 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2508 regnode * const noper = NEXTOPER( cur );
2509 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2510 regnode * const noper_next = regnext( noper );
2514 regprop(RExC_rx, mysv, cur);
2515 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2516 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2518 regprop(RExC_rx, mysv, noper);
2519 PerlIO_printf( Perl_debug_log, " -> %s",
2520 SvPV_nolen_const(mysv));
2523 regprop(RExC_rx, mysv, noper_next );
2524 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2525 SvPV_nolen_const(mysv));
2527 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2528 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2530 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2531 : PL_regkind[ OP( noper ) ] == EXACT )
2532 || OP(noper) == NOTHING )
2534 && noper_next == tail
2539 if ( !first || optype == NOTHING ) {
2540 if (!first) first = cur;
2541 optype = OP( noper );
2547 make_trie( pRExC_state,
2548 startbranch, first, cur, tail, count,
2551 if ( PL_regkind[ OP( noper ) ] == EXACT
2553 && noper_next == tail
2558 optype = OP( noper );
2568 regprop(RExC_rx, mysv, cur);
2569 PerlIO_printf( Perl_debug_log,
2570 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2571 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2575 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2576 #ifdef TRIE_STUDY_OPT
2577 if ( ((made == MADE_EXACT_TRIE &&
2578 startbranch == first)
2579 || ( first_non_open == first )) &&
2581 flags |= SCF_TRIE_RESTUDY;
2589 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2590 scan = NEXTOPER(NEXTOPER(scan));
2591 } else /* single branch is optimized. */
2592 scan = NEXTOPER(scan);
2595 else if (OP(scan) == EXACT) {
2596 I32 l = STR_LEN(scan);
2599 const U8 * const s = (U8*)STRING(scan);
2600 l = utf8_length(s, s + l);
2601 uc = utf8_to_uvchr(s, NULL);
2603 uc = *((U8*)STRING(scan));
2606 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2607 /* The code below prefers earlier match for fixed
2608 offset, later match for variable offset. */
2609 if (data->last_end == -1) { /* Update the start info. */
2610 data->last_start_min = data->pos_min;
2611 data->last_start_max = is_inf
2612 ? I32_MAX : data->pos_min + data->pos_delta;
2614 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2616 SvUTF8_on(data->last_found);
2618 SV * const sv = data->last_found;
2619 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2620 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2621 if (mg && mg->mg_len >= 0)
2622 mg->mg_len += utf8_length((U8*)STRING(scan),
2623 (U8*)STRING(scan)+STR_LEN(scan));
2625 data->last_end = data->pos_min + l;
2626 data->pos_min += l; /* As in the first entry. */
2627 data->flags &= ~SF_BEFORE_EOL;
2629 if (flags & SCF_DO_STCLASS_AND) {
2630 /* Check whether it is compatible with what we know already! */
2634 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2635 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2636 && (!(data->start_class->flags & ANYOF_FOLD)
2637 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2640 ANYOF_CLASS_ZERO(data->start_class);
2641 ANYOF_BITMAP_ZERO(data->start_class);
2643 ANYOF_BITMAP_SET(data->start_class, uc);
2644 data->start_class->flags &= ~ANYOF_EOS;
2646 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2648 else if (flags & SCF_DO_STCLASS_OR) {
2649 /* false positive possible if the class is case-folded */
2651 ANYOF_BITMAP_SET(data->start_class, uc);
2653 data->start_class->flags |= ANYOF_UNICODE_ALL;
2654 data->start_class->flags &= ~ANYOF_EOS;
2655 cl_and(data->start_class, &and_with);
2657 flags &= ~SCF_DO_STCLASS;
2659 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2660 I32 l = STR_LEN(scan);
2661 UV uc = *((U8*)STRING(scan));
2663 /* Search for fixed substrings supports EXACT only. */
2664 if (flags & SCF_DO_SUBSTR) {
2666 scan_commit(pRExC_state, data, minlenp);
2669 const U8 * const s = (U8 *)STRING(scan);
2670 l = utf8_length(s, s + l);
2671 uc = utf8_to_uvchr(s, NULL);
2674 if (flags & SCF_DO_SUBSTR)
2676 if (flags & SCF_DO_STCLASS_AND) {
2677 /* Check whether it is compatible with what we know already! */
2681 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2682 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2683 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2685 ANYOF_CLASS_ZERO(data->start_class);
2686 ANYOF_BITMAP_ZERO(data->start_class);
2688 ANYOF_BITMAP_SET(data->start_class, uc);
2689 data->start_class->flags &= ~ANYOF_EOS;
2690 data->start_class->flags |= ANYOF_FOLD;
2691 if (OP(scan) == EXACTFL)
2692 data->start_class->flags |= ANYOF_LOCALE;
2695 else if (flags & SCF_DO_STCLASS_OR) {
2696 if (data->start_class->flags & ANYOF_FOLD) {
2697 /* false positive possible if the class is case-folded.
2698 Assume that the locale settings are the same... */
2700 ANYOF_BITMAP_SET(data->start_class, uc);
2701 data->start_class->flags &= ~ANYOF_EOS;
2703 cl_and(data->start_class, &and_with);
2705 flags &= ~SCF_DO_STCLASS;
2707 else if (strchr((const char*)PL_varies,OP(scan))) {
2708 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2709 I32 f = flags, pos_before = 0;
2710 regnode * const oscan = scan;
2711 struct regnode_charclass_class this_class;
2712 struct regnode_charclass_class *oclass = NULL;
2713 I32 next_is_eval = 0;
2715 switch (PL_regkind[OP(scan)]) {
2716 case WHILEM: /* End of (?:...)* . */
2717 scan = NEXTOPER(scan);
2720 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2721 next = NEXTOPER(scan);
2722 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2724 maxcount = REG_INFTY;
2725 next = regnext(scan);
2726 scan = NEXTOPER(scan);
2730 if (flags & SCF_DO_SUBSTR)
2735 if (flags & SCF_DO_STCLASS) {
2737 maxcount = REG_INFTY;
2738 next = regnext(scan);
2739 scan = NEXTOPER(scan);
2742 is_inf = is_inf_internal = 1;
2743 scan = regnext(scan);
2744 if (flags & SCF_DO_SUBSTR) {
2745 scan_commit(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2746 data->longest = &(data->longest_float);
2748 goto optimize_curly_tail;
2750 mincount = ARG1(scan);
2751 maxcount = ARG2(scan);
2752 next = regnext(scan);
2753 if (OP(scan) == CURLYX) {
2754 I32 lp = (data ? *(data->last_closep) : 0);
2755 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
2757 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
2758 next_is_eval = (OP(scan) == EVAL);
2760 if (flags & SCF_DO_SUBSTR) {
2761 if (mincount == 0) scan_commit(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
2762 pos_before = data->pos_min;
2766 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
2768 data->flags |= SF_IS_INF;
2770 if (flags & SCF_DO_STCLASS) {
2771 cl_init(pRExC_state, &this_class);
2772 oclass = data->start_class;
2773 data->start_class = &this_class;
2774 f |= SCF_DO_STCLASS_AND;
2775 f &= ~SCF_DO_STCLASS_OR;
2777 /* These are the cases when once a subexpression
2778 fails at a particular position, it cannot succeed
2779 even after backtracking at the enclosing scope.
2781 XXXX what if minimal match and we are at the
2782 initial run of {n,m}? */
2783 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
2784 f &= ~SCF_WHILEM_VISITED_POS;
2786 /* This will finish on WHILEM, setting scan, or on NULL: */
2787 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, last, data,
2789 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
2791 if (flags & SCF_DO_STCLASS)
2792 data->start_class = oclass;
2793 if (mincount == 0 || minnext == 0) {
2794 if (flags & SCF_DO_STCLASS_OR) {
2795 cl_or(pRExC_state, data->start_class, &this_class);
2797 else if (flags & SCF_DO_STCLASS_AND) {
2798 /* Switch to OR mode: cache the old value of
2799 * data->start_class */
2800 StructCopy(data->start_class, &and_with,
2801 struct regnode_charclass_class);
2802 flags &= ~SCF_DO_STCLASS_AND;
2803 StructCopy(&this_class, data->start_class,
2804 struct regnode_charclass_class);
2805 flags |= SCF_DO_STCLASS_OR;
2806 data->start_class->flags |= ANYOF_EOS;
2808 } else { /* Non-zero len */
2809 if (flags & SCF_DO_STCLASS_OR) {
2810 cl_or(pRExC_state, data->start_class, &this_class);
2811 cl_and(data->start_class, &and_with);
2813 else if (flags & SCF_DO_STCLASS_AND)
2814 cl_and(data->start_class, &this_class);
2815 flags &= ~SCF_DO_STCLASS;
2817 if (!scan) /* It was not CURLYX, but CURLY. */
2819 if ( /* ? quantifier ok, except for (?{ ... }) */
2820 (next_is_eval || !(mincount == 0 && maxcount == 1))
2821 && (minnext == 0) && (deltanext == 0)
2822 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
2823 && maxcount <= REG_INFTY/3 /* Complement check for big count */
2824 && ckWARN(WARN_REGEXP))
2827 "Quantifier unexpected on zero-length expression");
2830 min += minnext * mincount;
2831 is_inf_internal |= ((maxcount == REG_INFTY
2832 && (minnext + deltanext) > 0)
2833 || deltanext == I32_MAX);
2834 is_inf |= is_inf_internal;
2835 delta += (minnext + deltanext) * maxcount - minnext * mincount;
2837 /* Try powerful optimization CURLYX => CURLYN. */
2838 if ( OP(oscan) == CURLYX && data
2839 && data->flags & SF_IN_PAR
2840 && !(data->flags & SF_HAS_EVAL)
2841 && !deltanext && minnext == 1 ) {
2842 /* Try to optimize to CURLYN. */
2843 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
2844 regnode * const nxt1 = nxt;
2851 if (!strchr((const char*)PL_simple,OP(nxt))
2852 && !(PL_regkind[OP(nxt)] == EXACT
2853 && STR_LEN(nxt) == 1))
2859 if (OP(nxt) != CLOSE)
2861 /* Now we know that nxt2 is the only contents: */
2862 oscan->flags = (U8)ARG(nxt);
2864 OP(nxt1) = NOTHING; /* was OPEN. */
2866 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
2867 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
2868 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
2869 OP(nxt) = OPTIMIZED; /* was CLOSE. */
2870 OP(nxt + 1) = OPTIMIZED; /* was count. */
2871 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
2876 /* Try optimization CURLYX => CURLYM. */
2877 if ( OP(oscan) == CURLYX && data
2878 && !(data->flags & SF_HAS_PAR)
2879 && !(data->flags & SF_HAS_EVAL)
2880 && !deltanext /* atom is fixed width */
2881 && minnext != 0 /* CURLYM can't handle zero width */
2883 /* XXXX How to optimize if data == 0? */
2884 /* Optimize to a simpler form. */
2885 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
2889 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
2890 && (OP(nxt2) != WHILEM))
2892 OP(nxt2) = SUCCEED; /* Whas WHILEM */
2893 /* Need to optimize away parenths. */
2894 if (data->flags & SF_IN_PAR) {
2895 /* Set the parenth number. */
2896 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
2898 if (OP(nxt) != CLOSE)
2899 FAIL("Panic opt close");
2900 oscan->flags = (U8)ARG(nxt);
2901 OP(nxt1) = OPTIMIZED; /* was OPEN. */
2902 OP(nxt) = OPTIMIZED; /* was CLOSE. */
2904 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
2905 OP(nxt + 1) = OPTIMIZED; /* was count. */
2906 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
2907 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
2910 while ( nxt1 && (OP(nxt1) != WHILEM)) {
2911 regnode *nnxt = regnext(nxt1);
2914 if (reg_off_by_arg[OP(nxt1)])
2915 ARG_SET(nxt1, nxt2 - nxt1);
2916 else if (nxt2 - nxt1 < U16_MAX)
2917 NEXT_OFF(nxt1) = nxt2 - nxt1;
2919 OP(nxt) = NOTHING; /* Cannot beautify */
2924 /* Optimize again: */
2925 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
2931 else if ((OP(oscan) == CURLYX)
2932 && (flags & SCF_WHILEM_VISITED_POS)
2933 /* See the comment on a similar expression above.
2934 However, this time it not a subexpression
2935 we care about, but the expression itself. */
2936 && (maxcount == REG_INFTY)
2937 && data && ++data->whilem_c < 16) {
2938 /* This stays as CURLYX, we can put the count/of pair. */
2939 /* Find WHILEM (as in regexec.c) */
2940 regnode *nxt = oscan + NEXT_OFF(oscan);
2942 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
2944 PREVOPER(nxt)->flags = (U8)(data->whilem_c
2945 | (RExC_whilem_seen << 4)); /* On WHILEM */
2947 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
2949 if (flags & SCF_DO_SUBSTR) {
2950 SV *last_str = NULL;
2951 int counted = mincount != 0;
2953 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
2954 #if defined(SPARC64_GCC_WORKAROUND)
2957 const char *s = NULL;
2960 if (pos_before >= data->last_start_min)
2963 b = data->last_start_min;
2966 s = SvPV_const(data->last_found, l);
2967 old = b - data->last_start_min;
2970 I32 b = pos_before >= data->last_start_min
2971 ? pos_before : data->last_start_min;
2973 const char * const s = SvPV_const(data->last_found, l);
2974 I32 old = b - data->last_start_min;
2978 old = utf8_hop((U8*)s, old) - (U8*)s;
2981 /* Get the added string: */
2982 last_str = newSVpvn(s + old, l);
2984 SvUTF8_on(last_str);
2985 if (deltanext == 0 && pos_before == b) {
2986 /* What was added is a constant string */
2988 SvGROW(last_str, (mincount * l) + 1);
2989 repeatcpy(SvPVX(last_str) + l,
2990 SvPVX_const(last_str), l, mincount - 1);
2991 SvCUR_set(last_str, SvCUR(last_str) * mincount);
2992 /* Add additional parts. */
2993 SvCUR_set(data->last_found,
2994 SvCUR(data->last_found) - l);
2995 sv_catsv(data->last_found, last_str);
2997 SV * sv = data->last_found;
2999 SvUTF8(sv) && SvMAGICAL(sv) ?
3000 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3001 if (mg && mg->mg_len >= 0)
3002 mg->mg_len += CHR_SVLEN(last_str);
3004 data->last_end += l * (mincount - 1);
3007 /* start offset must point into the last copy */
3008 data->last_start_min += minnext * (mincount - 1);
3009 data->last_start_max += is_inf ? I32_MAX
3010 : (maxcount - 1) * (minnext + data->pos_delta);
3013 /* It is counted once already... */
3014 data->pos_min += minnext * (mincount - counted);
3015 data->pos_delta += - counted * deltanext +
3016 (minnext + deltanext) * maxcount - minnext * mincount;
3017 if (mincount != maxcount) {
3018 /* Cannot extend fixed substrings found inside
3020 scan_commit(pRExC_state,data,minlenp);
3021 if (mincount && last_str) {
3022 SV * const sv = data->last_found;
3023 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3024 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3028 sv_setsv(sv, last_str);
3029 data->last_end = data->pos_min;
3030 data->last_start_min =
3031 data->pos_min - CHR_SVLEN(last_str);
3032 data->last_start_max = is_inf
3034 : data->pos_min + data->pos_delta
3035 - CHR_SVLEN(last_str);
3037 data->longest = &(data->longest_float);
3039 SvREFCNT_dec(last_str);
3041 if (data && (fl & SF_HAS_EVAL))
3042 data->flags |= SF_HAS_EVAL;
3043 optimize_curly_tail:
3044 if (OP(oscan) != CURLYX) {
3045 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3047 NEXT_OFF(oscan) += NEXT_OFF(next);
3050 default: /* REF and CLUMP only? */
3051 if (flags & SCF_DO_SUBSTR) {
3052 scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
3053 data->longest = &(data->longest_float);
3055 is_inf = is_inf_internal = 1;
3056 if (flags & SCF_DO_STCLASS_OR)
3057 cl_anything(pRExC_state, data->start_class);
3058 flags &= ~SCF_DO_STCLASS;
3062 else if (strchr((const char*)PL_simple,OP(scan))) {
3065 if (flags & SCF_DO_SUBSTR) {
3066 scan_commit(pRExC_state,data,minlenp);
3070 if (flags & SCF_DO_STCLASS) {
3071 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3073 /* Some of the logic below assumes that switching
3074 locale on will only add false positives. */
3075 switch (PL_regkind[OP(scan)]) {
3079 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3080 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3081 cl_anything(pRExC_state, data->start_class);
3084 if (OP(scan) == SANY)
3086 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3087 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3088 || (data->start_class->flags & ANYOF_CLASS));
3089 cl_anything(pRExC_state, data->start_class);
3091 if (flags & SCF_DO_STCLASS_AND || !value)
3092 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3095 if (flags & SCF_DO_STCLASS_AND)
3096 cl_and(data->start_class,
3097 (struct regnode_charclass_class*)scan);
3099 cl_or(pRExC_state, data->start_class,
3100 (struct regnode_charclass_class*)scan);
3103 if (flags & SCF_DO_STCLASS_AND) {
3104 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3105 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3106 for (value = 0; value < 256; value++)
3107 if (!isALNUM(value))
3108 ANYOF_BITMAP_CLEAR(data->start_class, value);
3112 if (data->start_class->flags & ANYOF_LOCALE)
3113 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3115 for (value = 0; value < 256; value++)
3117 ANYOF_BITMAP_SET(data->start_class, value);
3122 if (flags & SCF_DO_STCLASS_AND) {
3123 if (data->start_class->flags & ANYOF_LOCALE)
3124 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3127 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3128 data->start_class->flags |= ANYOF_LOCALE;
3132 if (flags & SCF_DO_STCLASS_AND) {
3133 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3134 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3135 for (value = 0; value < 256; value++)
3137 ANYOF_BITMAP_CLEAR(data->start_class, value);
3141 if (data->start_class->flags & ANYOF_LOCALE)
3142 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3144 for (value = 0; value < 256; value++)
3145 if (!isALNUM(value))
3146 ANYOF_BITMAP_SET(data->start_class, value);
3151 if (flags & SCF_DO_STCLASS_AND) {
3152 if (data->start_class->flags & ANYOF_LOCALE)
3153 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3156 data->start_class->flags |= ANYOF_LOCALE;
3157 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3161 if (flags & SCF_DO_STCLASS_AND) {
3162 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3163 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3164 for (value = 0; value < 256; value++)
3165 if (!isSPACE(value))
3166 ANYOF_BITMAP_CLEAR(data->start_class, value);
3170 if (data->start_class->flags & ANYOF_LOCALE)
3171 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3173 for (value = 0; value < 256; value++)
3175 ANYOF_BITMAP_SET(data->start_class, value);
3180 if (flags & SCF_DO_STCLASS_AND) {
3181 if (data->start_class->flags & ANYOF_LOCALE)
3182 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3185 data->start_class->flags |= ANYOF_LOCALE;
3186 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3190 if (flags & SCF_DO_STCLASS_AND) {
3191 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3192 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3193 for (value = 0; value < 256; value++)
3195 ANYOF_BITMAP_CLEAR(data->start_class, value);
3199 if (data->start_class->flags & ANYOF_LOCALE)
3200 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3202 for (value = 0; value < 256; value++)
3203 if (!isSPACE(value))
3204 ANYOF_BITMAP_SET(data->start_class, value);
3209 if (flags & SCF_DO_STCLASS_AND) {
3210 if (data->start_class->flags & ANYOF_LOCALE) {
3211 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3212 for (value = 0; value < 256; value++)
3213 if (!isSPACE(value))
3214 ANYOF_BITMAP_CLEAR(data->start_class, value);
3218 data->start_class->flags |= ANYOF_LOCALE;
3219 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3223 if (flags & SCF_DO_STCLASS_AND) {
3224 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3225 for (value = 0; value < 256; value++)
3226 if (!isDIGIT(value))
3227 ANYOF_BITMAP_CLEAR(data->start_class, value);
3230 if (data->start_class->flags & ANYOF_LOCALE)
3231 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3233 for (value = 0; value < 256; value++)
3235 ANYOF_BITMAP_SET(data->start_class, value);
3240 if (flags & SCF_DO_STCLASS_AND) {
3241 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3242 for (value = 0; value < 256; value++)
3244 ANYOF_BITMAP_CLEAR(data->start_class, value);
3247 if (data->start_class->flags & ANYOF_LOCALE)
3248 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3250 for (value = 0; value < 256; value++)
3251 if (!isDIGIT(value))
3252 ANYOF_BITMAP_SET(data->start_class, value);
3257 if (flags & SCF_DO_STCLASS_OR)
3258 cl_and(data->start_class, &and_with);
3259 flags &= ~SCF_DO_STCLASS;
3262 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3263 data->flags |= (OP(scan) == MEOL
3267 else if ( PL_regkind[OP(scan)] == BRANCHJ
3268 /* Lookbehind, or need to calculate parens/evals/stclass: */
3269 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3270 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3271 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3272 || OP(scan) == UNLESSM )
3274 /* Negative Lookahead/lookbehind
3275 In this case we can't do fixed string optimisation.
3278 I32 deltanext, minnext, fake = 0;
3280 struct regnode_charclass_class intrnl;
3283 data_fake.flags = 0;
3285 data_fake.whilem_c = data->whilem_c;
3286 data_fake.last_closep = data->last_closep;
3289 data_fake.last_closep = &fake;
3290 if ( flags & SCF_DO_STCLASS && !scan->flags
3291 && OP(scan) == IFMATCH ) { /* Lookahead */
3292 cl_init(pRExC_state, &intrnl);
3293 data_fake.start_class = &intrnl;
3294 f |= SCF_DO_STCLASS_AND;
3296 if (flags & SCF_WHILEM_VISITED_POS)
3297 f |= SCF_WHILEM_VISITED_POS;
3298 next = regnext(scan);
3299 nscan = NEXTOPER(NEXTOPER(scan));
3300 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, last, &data_fake, f,depth+1);
3303 vFAIL("Variable length lookbehind not implemented");
3305 else if (minnext > (I32)U8_MAX) {
3306 vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3308 scan->flags = (U8)minnext;
3311 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3313 if (data_fake.flags & SF_HAS_EVAL)
3314 data->flags |= SF_HAS_EVAL;
3315 data->whilem_c = data_fake.whilem_c;
3317 if (f & SCF_DO_STCLASS_AND) {
3318 const int was = (data->start_class->flags & ANYOF_EOS);
3320 cl_and(data->start_class, &intrnl);
3322 data->start_class->flags |= ANYOF_EOS;
3325 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3327 /* Positive Lookahead/lookbehind
3328 In this case we can do fixed string optimisation,
3329 but we must be careful about it. Note in the case of
3330 lookbehind the positions will be offset by the minimum
3331 length of the pattern, something we won't know about
3332 until after the recurse.
3334 I32 deltanext, fake = 0;
3336 struct regnode_charclass_class intrnl;
3338 /* We use SAVEFREEPV so that when the full compile
3339 is finished perl will clean up the allocated
3340 minlens when its all done. This was we don't
3341 have to worry about freeing them when we know
3342 they wont be used, which would be a pain.
3345 Newx( minnextp, 1, I32 );
3346 SAVEFREEPV(minnextp);
3349 StructCopy(data, &data_fake, scan_data_t);
3350 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3353 scan_commit(pRExC_state, &data_fake,minlenp);
3354 data_fake.last_found=newSVsv(data->last_found);
3358 data_fake.last_closep = &fake;
3359 data_fake.flags = 0;
3361 data_fake.flags |= SF_IS_INF;
3362 if ( flags & SCF_DO_STCLASS && !scan->flags
3363 && OP(scan) == IFMATCH ) { /* Lookahead */
3364 cl_init(pRExC_state, &intrnl);
3365 data_fake.start_class = &intrnl;
3366 f |= SCF_DO_STCLASS_AND;
3368 if (flags & SCF_WHILEM_VISITED_POS)
3369 f |= SCF_WHILEM_VISITED_POS;
3370 next = regnext(scan);
3371 nscan = NEXTOPER(NEXTOPER(scan));
3373 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext, last, &data_fake, f,depth+1);
3376 vFAIL("Variable length lookbehind not implemented");
3378 else if (*minnextp > (I32)U8_MAX) {
3379 vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3381 scan->flags = (U8)*minnextp;
3387 if (f & SCF_DO_STCLASS_AND) {
3388 const int was = (data->start_class->flags & ANYOF_EOS);
3390 cl_and(data->start_class, &intrnl);
3392 data->start_class->flags |= ANYOF_EOS;
3395 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3397 if (data_fake.flags & SF_HAS_EVAL)
3398 data->flags |= SF_HAS_EVAL;
3399 data->whilem_c = data_fake.whilem_c;
3400 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3401 if (RExC_rx->minlen<*minnextp)
3402 RExC_rx->minlen=*minnextp;
3403 scan_commit(pRExC_state, &data_fake, minnextp);
3404 SvREFCNT_dec(data_fake.last_found);
3406 if ( data_fake.minlen_fixed != minlenp )
3408 data->offset_fixed= data_fake.offset_fixed;
3409 data->minlen_fixed= data_fake.minlen_fixed;
3410 data->lookbehind_fixed+= scan->flags;
3412 if ( data_fake.minlen_float != minlenp )
3414 data->minlen_float= data_fake.minlen_float;
3415 data->offset_float_min=data_fake.offset_float_min;
3416 data->offset_float_max=data_fake.offset_float_max;
3417 data->lookbehind_float+= scan->flags;
3426 else if (OP(scan) == OPEN) {
3429 else if (OP(scan) == CLOSE) {
3430 if ((I32)ARG(scan) == is_par) {
3431 next = regnext(scan);
3433 if ( next && (OP(next) != WHILEM) && next < last)
3434 is_par = 0; /* Disable optimization */
3437 *(data->last_closep) = ARG(scan);
3439 else if (OP(scan) == EVAL) {
3441 data->flags |= SF_HAS_EVAL;
3443 else if (OP(scan) == LOGICAL && scan->flags == 2) { /* Embedded follows */
3444 if (flags & SCF_DO_SUBSTR) {
3445 scan_commit(pRExC_state,data,minlenp);
3446 data->longest = &(data->longest_float);
3448 is_inf = is_inf_internal = 1;
3449 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3450 cl_anything(pRExC_state, data->start_class);
3451 flags &= ~SCF_DO_STCLASS;
3453 #ifdef TRIE_STUDY_OPT
3454 #ifdef FULL_TRIE_STUDY
3455 else if (PL_regkind[OP(scan)] == TRIE) {
3456 /* NOTE - There is similar code to this block above for handling
3457 BRANCH nodes on the initial study. If you change stuff here
3459 regnode *tail= regnext(scan);
3460 reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
3461 I32 max1 = 0, min1 = I32_MAX;
3462 struct regnode_charclass_class accum;
3464 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3465 scan_commit(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3466 if (flags & SCF_DO_STCLASS)
3467 cl_init_zero(pRExC_state, &accum);
3473 const regnode *nextbranch= NULL;
3476 for ( word=1 ; word <= trie->wordcount ; word++)
3478 I32 deltanext=0, minnext=0, f = 0, fake;
3479 struct regnode_charclass_class this_class;
3481 data_fake.flags = 0;
3483 data_fake.whilem_c = data->whilem_c;
3484 data_fake.last_closep = data->last_closep;
3487 data_fake.last_closep = &fake;
3489 if (flags & SCF_DO_STCLASS) {
3490 cl_init(pRExC_state, &this_class);
3491 data_fake.start_class = &this_class;
3492 f = SCF_DO_STCLASS_AND;
3494 if (flags & SCF_WHILEM_VISITED_POS)
3495 f |= SCF_WHILEM_VISITED_POS;
3497 if (trie->jump[word]) {
3499 nextbranch = tail - trie->jump[0];
3500 scan= tail - trie->jump[word];
3501 /* We go from the jump point to the branch that follows
3502 it. Note this means we need the vestigal unused branches
3503 even though they arent otherwise used.
3505 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3506 (regnode *)nextbranch, &data_fake, f,depth+1);
3508 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3509 nextbranch= regnext((regnode*)nextbranch);
3511 if (min1 > (I32)(minnext + trie->minlen))
3512 min1 = minnext + trie->minlen;
3513 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3514 max1 = minnext + deltanext + trie->maxlen;
3515 if (deltanext == I32_MAX)
3516 is_inf = is_inf_internal = 1;
3518 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3522 if (data_fake.flags & SF_HAS_EVAL)
3523 data->flags |= SF_HAS_EVAL;
3524 data->whilem_c = data_fake.whilem_c;
3526 if (flags & SCF_DO_STCLASS)
3527 cl_or(pRExC_state, &accum, &this_class);
3530 if (flags & SCF_DO_SUBSTR) {
3531 data->pos_min += min1;
3532 data->pos_delta += max1 - min1;
3533 if (max1 != min1 || is_inf)
3534 data->longest = &(data->longest_float);
3537 delta += max1 - min1;
3538 if (flags & SCF_DO_STCLASS_OR) {
3539 cl_or(pRExC_state, data->start_class, &accum);
3541 cl_and(data->start_class, &and_with);
3542 flags &= ~SCF_DO_STCLASS;
3545 else if (flags & SCF_DO_STCLASS_AND) {
3547 cl_and(data->start_class, &accum);
3548 flags &= ~SCF_DO_STCLASS;
3551 /* Switch to OR mode: cache the old value of
3552 * data->start_class */
3553 StructCopy(data->start_class, &and_with,
3554 struct regnode_charclass_class);
3555 flags &= ~SCF_DO_STCLASS_AND;
3556 StructCopy(&accum, data->start_class,
3557 struct regnode_charclass_class);
3558 flags |= SCF_DO_STCLASS_OR;
3559 data->start_class->flags |= ANYOF_EOS;
3566 else if (PL_regkind[OP(scan)] == TRIE) {
3567 reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
3570 min += trie->minlen;
3571 delta += (trie->maxlen - trie->minlen);
3572 flags &= ~SCF_DO_STCLASS; /* xxx */
3573 if (flags & SCF_DO_SUBSTR) {
3574 scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
3575 data->pos_min += trie->minlen;
3576 data->pos_delta += (trie->maxlen - trie->minlen);
3577 if (trie->maxlen != trie->minlen)
3578 data->longest = &(data->longest_float);
3580 if (trie->jump) /* no more substrings -- for now /grr*/
3581 flags &= ~SCF_DO_SUBSTR;
3583 #endif /* old or new */
3584 #endif /* TRIE_STUDY_OPT */
3585 /* Else: zero-length, ignore. */
3586 scan = regnext(scan);
3591 *deltap = is_inf_internal ? I32_MAX : delta;
3592 if (flags & SCF_DO_SUBSTR && is_inf)
3593 data->pos_delta = I32_MAX - data->pos_min;
3594 if (is_par > (I32)U8_MAX)
3596 if (is_par && pars==1 && data) {
3597 data->flags |= SF_IN_PAR;
3598 data->flags &= ~SF_HAS_PAR;
3600 else if (pars && data) {
3601 data->flags |= SF_HAS_PAR;
3602 data->flags &= ~SF_IN_PAR;
3604 if (flags & SCF_DO_STCLASS_OR)
3605 cl_and(data->start_class, &and_with);
3606 if (flags & SCF_TRIE_RESTUDY)
3607 data->flags |= SCF_TRIE_RESTUDY;
3609 DEBUG_STUDYDATA(data,depth);
3615 S_add_data(RExC_state_t *pRExC_state, I32 n, const char *s)
3617 if (RExC_rx->data) {
3618 Renewc(RExC_rx->data,
3619 sizeof(*RExC_rx->data) + sizeof(void*) * (RExC_rx->data->count + n - 1),
3620 char, struct reg_data);
3621 Renew(RExC_rx->data->what, RExC_rx->data->count + n, U8);
3622 RExC_rx->data->count += n;
3625 Newxc(RExC_rx->data, sizeof(*RExC_rx->data) + sizeof(void*) * (n - 1),
3626 char, struct reg_data);
3627 Newx(RExC_rx->data->what, n, U8);
3628 RExC_rx->data->count = n;
3630 Copy(s, RExC_rx->data->what + RExC_rx->data->count - n, n, U8);
3631 return RExC_rx->data->count - n;
3634 #ifndef PERL_IN_XSUB_RE
3636 Perl_reginitcolors(pTHX)
3639 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3641 char *t = savepv(s);
3645 t = strchr(t, '\t');
3651 PL_colors[i] = t = (char *)"";
3656 PL_colors[i++] = (char *)"";
3663 #ifdef TRIE_STUDY_OPT
3664 #define CHECK_RESTUDY_GOTO \
3666 (data.flags & SCF_TRIE_RESTUDY) \
3670 #define CHECK_RESTUDY_GOTO
3674 - pregcomp - compile a regular expression into internal code
3676 * We can't allocate space until we know how big the compiled form will be,
3677 * but we can't compile it (and thus know how big it is) until we've got a
3678 * place to put the code. So we cheat: we compile it twice, once with code
3679 * generation turned off and size counting turned on, and once "for real".
3680 * This also means that we don't allocate space until we are sure that the
3681 * thing really will compile successfully, and we never have to move the
3682 * code and thus invalidate pointers into it. (Note that it has to be in
3683 * one piece because free() must be able to free it all.) [NB: not true in perl]
3685 * Beware that the optimization-preparation code in here knows about some
3686 * of the structure of the compiled regexp. [I'll say.]
3688 #ifndef PERL_IN_XSUB_RE
3689 #define CORE_ONLY_BLOCK(c) {c}{
3690 #define RE_ENGINE_PTR &PL_core_reg_engine
3692 #define CORE_ONLY_BLOCK(c) {
3693 extern const struct regexp_engine my_reg_engine;
3694 #define RE_ENGINE_PTR &my_reg_engine
3699 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
3702 GET_RE_DEBUG_FLAGS_DECL;
3703 DEBUG_r(if (!PL_colorset) reginitcolors());
3705 /* Dispatch a request to compile a regexp to correct
3707 HV * const table = GvHV(PL_hintgv);
3709 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
3710 if (ptr && SvIOK(*ptr)) {
3711 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
3713 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
3716 return CALLREGCOMP_ENG(eng, exp, xend, pm);
3727 RExC_state_t RExC_state;
3728 RExC_state_t * const pRExC_state = &RExC_state;
3729 #ifdef TRIE_STUDY_OPT
3731 RExC_state_t copyRExC_state;
3734 FAIL("NULL regexp argument");
3736 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
3740 SV *dsv= sv_newmortal();
3741 RE_PV_QUOTED_DECL(s, RExC_utf8,
3742 dsv, RExC_precomp, (xend - exp), 60);
3743 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
3744 PL_colors[4],PL_colors[5],s);
3746 RExC_flags = pm->op_pmflags;
3750 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
3751 RExC_seen_evals = 0;
3754 /* First pass: determine size, legality. */
3761 RExC_emit = &PL_regdummy;
3762 RExC_whilem_seen = 0;
3763 RExC_charnames = NULL;
3765 #if 0 /* REGC() is (currently) a NOP at the first pass.
3766 * Clever compilers notice this and complain. --jhi */
3767 REGC((U8)REG_MAGIC, (char*)RExC_emit);
3769 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
3770 if (reg(pRExC_state, 0, &flags,1) == NULL) {
3771 RExC_precomp = NULL;
3774 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Required "));
3775 DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "size %"IVdf" nodes ", (IV)RExC_size));
3776 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nStarting second pass (creation)\n"));
3779 RExC_lastparse=NULL;
3783 /* Small enough for pointer-storage convention?
3784 If extralen==0, this means that we will not need long jumps. */
3785 if (RExC_size >= 0x10000L && RExC_extralen)
3786 RExC_size += RExC_extralen;
3789 if (RExC_whilem_seen > 15)
3790 RExC_whilem_seen = 15;
3792 /* Allocate space and zero-initialize. Note, the two step process
3793 of zeroing when in debug mode, thus anything assigned has to
3794 happen after that */
3795 Newxc(r, sizeof(regexp) + (unsigned)RExC_size * sizeof(regnode),
3798 FAIL("Regexp out of space");
3800 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
3801 Zero(r, sizeof(regexp) + (unsigned)RExC_size * sizeof(regnode), char);
3803 /* initialization begins here */
3804 r->engine= RE_ENGINE_PTR;
3806 r->prelen = xend - exp;
3807 r->precomp = savepvn(RExC_precomp, r->prelen);
3809 #ifdef PERL_OLD_COPY_ON_WRITE
3810 r->saved_copy = NULL;
3812 r->reganch = pm->op_pmflags & PMf_COMPILETIME;
3813 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
3814 r->lastparen = 0; /* mg.c reads this. */
3816 r->substrs = 0; /* Useful during FAIL. */
3817 r->startp = 0; /* Useful during FAIL. */
3818 r->endp = 0; /* Useful during FAIL. */
3820 Newxz(r->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
3822 r->offsets[0] = RExC_size;
3824 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
3825 "%s %"UVuf" bytes for offset annotations.\n",
3826 r->offsets ? "Got" : "Couldn't get",
3827 (UV)((2*RExC_size+1) * sizeof(U32))));
3831 /* Second pass: emit code. */
3832 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
3837 RExC_emit_start = r->program;
3838 RExC_emit = r->program;
3839 /* Store the count of eval-groups for security checks: */
3840 RExC_emit->next_off = (RExC_seen_evals > (I32)U16_MAX) ? U16_MAX : (U16)RExC_seen_evals;
3841 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
3843 if (reg(pRExC_state, 0, &flags,1) == NULL)
3845 /* XXXX To minimize changes to RE engine we always allocate
3846 3-units-long substrs field. */
3847 Newx(r->substrs, 1, struct reg_substr_data);
3850 r->minlen = minlen = sawplus = sawopen = 0;
3851 Zero(r->substrs, 1, struct reg_substr_data);
3852 StructCopy(&zero_scan_data, &data, scan_data_t);
3854 #ifdef TRIE_STUDY_OPT
3856 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
3857 RExC_state=copyRExC_state;
3858 if (data.last_found) {
3859 SvREFCNT_dec(data.longest_fixed);
3860 SvREFCNT_dec(data.longest_float);
3861 SvREFCNT_dec(data.last_found);
3864 copyRExC_state=RExC_state;
3868 /* Dig out information for optimizations. */
3869 r->reganch = pm->op_pmflags & PMf_COMPILETIME; /* Again? */
3870 pm->op_pmflags = RExC_flags;
3872 r->reganch |= ROPT_UTF8; /* Unicode in it? */
3873 r->regstclass = NULL;
3874 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
3875 r->reganch |= ROPT_NAUGHTY;
3876 scan = r->program + 1; /* First BRANCH. */
3878 /* testing for BRANCH here tells us whether there is "must appear"
3879 data in the pattern. If there is then we can use it for optimisations */
3880 if (OP(scan) != BRANCH) { /* Only one top-level choice. */
3882 STRLEN longest_float_length, longest_fixed_length;
3883 struct regnode_charclass_class ch_class; /* pointed to by data */
3885 I32 last_close = 0; /* pointed to by data */
3888 /* Skip introductions and multiplicators >= 1. */
3889 while ((OP(first) == OPEN && (sawopen = 1)) ||
3890 /* An OR of *one* alternative - should not happen now. */
3891 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
3892 /* for now we can't handle lookbehind IFMATCH*/
3893 (OP(first) == IFMATCH && !first->flags) ||
3894 (OP(first) == PLUS) ||
3895 (OP(first) == MINMOD) ||
3896 /* An {n,m} with n>0 */
3897 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
3900 if (OP(first) == PLUS)
3903 first += regarglen[OP(first)];
3904 if (OP(first) == IFMATCH) {
3905 first = NEXTOPER(first);
3906 first += EXTRA_STEP_2ARGS;
3907 } else /* XXX possible optimisation for /(?=)/ */
3908 first = NEXTOPER(first);
3911 /* Starting-point info. */
3913 DEBUG_PEEP("first:",first,0);
3914 /* Ignore EXACT as we deal with it later. */
3915 if (PL_regkind[OP(first)] == EXACT) {
3916 if (OP(first) == EXACT)
3917 NOOP; /* Empty, get anchored substr later. */
3918 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
3919 r->regstclass = first;
3922 else if (PL_regkind[OP(first)] == TRIE &&
3923 ((reg_trie_data *)r->data->data[ ARG(first) ])->minlen>0)
3926 /* this can happen only on restudy */
3927 if ( OP(first) == TRIE ) {
3928 struct regnode_1 *trieop;
3929 Newxz(trieop,1,struct regnode_1);
3930 StructCopy(first,trieop,struct regnode_1);
3931 trie_op=(regnode *)trieop;
3933 struct regnode_charclass *trieop;
3934 Newxz(trieop,1,struct regnode_charclass);
3935 StructCopy(first,trieop,struct regnode_charclass);
3936 trie_op=(regnode *)trieop;
3939 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
3940 r->regstclass = trie_op;
3943 else if (strchr((const char*)PL_simple,OP(first)))
3944 r->regstclass = first;
3945 else if (PL_regkind[OP(first)] == BOUND ||
3946 PL_regkind[OP(first)] == NBOUND)
3947 r->regstclass = first;
3948 else if (PL_regkind[OP(first)] == BOL) {
3949 r->reganch |= (OP(first) == MBOL
3951 : (OP(first) == SBOL
3954 first = NEXTOPER(first);
3957 else if (OP(first) == GPOS) {
3958 r->reganch |= ROPT_ANCH_GPOS;
3959 first = NEXTOPER(first);
3962 else if (!sawopen && (OP(first) == STAR &&
3963 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
3964 !(r->reganch & ROPT_ANCH) )
3966 /* turn .* into ^.* with an implied $*=1 */
3968 (OP(NEXTOPER(first)) == REG_ANY)
3971 r->reganch |= type | ROPT_IMPLICIT;
3972 first = NEXTOPER(first);
3975 if (sawplus && (!sawopen || !RExC_sawback)
3976 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
3977 /* x+ must match at the 1st pos of run of x's */
3978 r->reganch |= ROPT_SKIP;
3980 /* Scan is after the zeroth branch, first is atomic matcher. */
3981 #ifdef TRIE_STUDY_OPT
3984 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
3985 (IV)(first - scan + 1))
3989 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
3990 (IV)(first - scan + 1))
3996 * If there's something expensive in the r.e., find the
3997 * longest literal string that must appear and make it the
3998 * regmust. Resolve ties in favor of later strings, since
3999 * the regstart check works with the beginning of the r.e.
4000 * and avoiding duplication strengthens checking. Not a
4001 * strong reason, but sufficient in the absence of others.
4002 * [Now we resolve ties in favor of the earlier string if
4003 * it happens that c_offset_min has been invalidated, since the
4004 * earlier string may buy us something the later one won't.]
4008 data.longest_fixed = newSVpvs("");
4009 data.longest_float = newSVpvs("");
4010 data.last_found = newSVpvs("");
4011 data.longest = &(data.longest_fixed);
4013 if (!r->regstclass) {
4014 cl_init(pRExC_state, &ch_class);
4015 data.start_class = &ch_class;
4016 stclass_flag = SCF_DO_STCLASS_AND;
4017 } else /* XXXX Check for BOUND? */
4019 data.last_closep = &last_close;
4021 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4022 &data, SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4028 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4029 && data.last_start_min == 0 && data.last_end > 0
4030 && !RExC_seen_zerolen
4031 && (!(RExC_seen & REG_SEEN_GPOS) || (r->reganch & ROPT_ANCH_GPOS)))
4032 r->reganch |= ROPT_CHECK_ALL;
4033 scan_commit(pRExC_state, &data,&minlen);
4034 SvREFCNT_dec(data.last_found);
4036 /* Note that code very similar to this but for anchored string
4037 follows immediately below, changes may need to be made to both.
4040 longest_float_length = CHR_SVLEN(data.longest_float);
4041 if (longest_float_length
4042 || (data.flags & SF_FL_BEFORE_EOL
4043 && (!(data.flags & SF_FL_BEFORE_MEOL)
4044 || (RExC_flags & PMf_MULTILINE))))
4048 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4049 && data.offset_fixed == data.offset_float_min
4050 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4051 goto remove_float; /* As in (a)+. */
4053 /* copy the information about the longest float from the reg_scan_data
4054 over to the program. */
4055 if (SvUTF8(data.longest_float)) {
4056 r->float_utf8 = data.longest_float;
4057 r->float_substr = NULL;
4059 r->float_substr = data.longest_float;
4060 r->float_utf8 = NULL;
4062 /* float_end_shift is how many chars that must be matched that
4063 follow this item. We calculate it ahead of time as once the
4064 lookbehind offset is added in we lose the ability to correctly
4066 ml = data.minlen_float ? *(data.minlen_float)
4067 : (I32)longest_float_length;
4068 r->float_end_shift = ml - data.offset_float_min
4069 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4070 + data.lookbehind_float;
4071 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4072 r->float_max_offset = data.offset_float_max;
4073 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4074 r->float_max_offset -= data.lookbehind_float;
4076 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4077 && (!(data.flags & SF_FL_BEFORE_MEOL)
4078 || (RExC_flags & PMf_MULTILINE)));
4079 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4083 r->float_substr = r->float_utf8 = NULL;
4084 SvREFCNT_dec(data.longest_float);
4085 longest_float_length = 0;
4088 /* Note that code very similar to this but for floating string
4089 is immediately above, changes may need to be made to both.
4092 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4093 if (longest_fixed_length
4094 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4095 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4096 || (RExC_flags & PMf_MULTILINE))))
4100 /* copy the information about the longest fixed
4101 from the reg_scan_data over to the program. */
4102 if (SvUTF8(data.longest_fixed)) {
4103 r->anchored_utf8 = data.longest_fixed;
4104 r->anchored_substr = NULL;
4106 r->anchored_substr = data.longest_fixed;
4107 r->anchored_utf8 = NULL;
4109 /* fixed_end_shift is how many chars that must be matched that
4110 follow this item. We calculate it ahead of time as once the
4111 lookbehind offset is added in we lose the ability to correctly
4113 ml = data.minlen_fixed ? *(data.minlen_fixed)
4114 : (I32)longest_fixed_length;
4115 r->anchored_end_shift = ml - data.offset_fixed
4116 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4117 + data.lookbehind_fixed;
4118 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4120 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4121 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4122 || (RExC_flags & PMf_MULTILINE)));
4123 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4126 r->anchored_substr = r->anchored_utf8 = NULL;
4127 SvREFCNT_dec(data.longest_fixed);
4128 longest_fixed_length = 0;
4131 && (OP(r->regstclass) == REG_ANY || OP(r->regstclass) == SANY))
4132 r->regstclass = NULL;
4133 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4135 && !(data.start_class->flags & ANYOF_EOS)
4136 && !cl_is_anything(data.start_class))
4138 const I32 n = add_data(pRExC_state, 1, "f");
4140 Newx(RExC_rx->data->data[n], 1,
4141 struct regnode_charclass_class);
4142 StructCopy(data.start_class,
4143 (struct regnode_charclass_class*)RExC_rx->data->data[n],
4144 struct regnode_charclass_class);
4145 r->regstclass = (regnode*)RExC_rx->data->data[n];
4146 r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
4147 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4148 regprop(r, sv, (regnode*)data.start_class);
4149 PerlIO_printf(Perl_debug_log,
4150 "synthetic stclass \"%s\".\n",
4151 SvPVX_const(sv));});
4154 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4155 if (longest_fixed_length > longest_float_length) {
4156 r->check_end_shift = r->anchored_end_shift;
4157 r->check_substr = r->anchored_substr;
4158 r->check_utf8 = r->anchored_utf8;
4159 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4160 if (r->reganch & ROPT_ANCH_SINGLE)
4161 r->reganch |= ROPT_NOSCAN;
4164 r->check_end_shift = r->float_end_shift;
4165 r->check_substr = r->float_substr;
4166 r->check_utf8 = r->float_utf8;
4167 r->check_offset_min = r->float_min_offset;
4168 r->check_offset_max = r->float_max_offset;
4170 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4171 This should be changed ASAP! */
4172 if ((r->check_substr || r->check_utf8) && !(r->reganch & ROPT_ANCH_GPOS)) {
4173 r->reganch |= RE_USE_INTUIT;
4174 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4175 r->reganch |= RE_INTUIT_TAIL;
4177 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4178 if ( (STRLEN)minlen < longest_float_length )
4179 minlen= longest_float_length;
4180 if ( (STRLEN)minlen < longest_fixed_length )
4181 minlen= longest_fixed_length;
4185 /* Several toplevels. Best we can is to set minlen. */
4187 struct regnode_charclass_class ch_class;
4190 DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "\n"));
4192 scan = r->program + 1;
4193 cl_init(pRExC_state, &ch_class);
4194 data.start_class = &ch_class;
4195 data.last_closep = &last_close;
4197 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4198 &data, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4202 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4203 = r->float_substr = r->float_utf8 = NULL;
4204 if (!(data.start_class->flags & ANYOF_EOS)
4205 && !cl_is_anything(data.start_class))
4207 const I32 n = add_data(pRExC_state, 1, "f");
4209 Newx(RExC_rx->data->data[n], 1,
4210 struct regnode_charclass_class);
4211 StructCopy(data.start_class,
4212 (struct regnode_charclass_class*)RExC_rx->data->data[n],
4213 struct regnode_charclass_class);
4214 r->regstclass = (regnode*)RExC_rx->data->data[n];
4215 r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
4216 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4217 regprop(r, sv, (regnode*)data.start_class);
4218 PerlIO_printf(Perl_debug_log,
4219 "synthetic stclass \"%s\".\n",
4220 SvPVX_const(sv));});
4224 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4225 the "real" pattern. */
4226 if (r->minlen < minlen)
4229 if (RExC_seen & REG_SEEN_GPOS)
4230 r->reganch |= ROPT_GPOS_SEEN;
4231 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4232 r->reganch |= ROPT_LOOKBEHIND_SEEN;
4233 if (RExC_seen & REG_SEEN_EVAL)
4234 r->reganch |= ROPT_EVAL_SEEN;
4235 if (RExC_seen & REG_SEEN_CANY)
4236 r->reganch |= ROPT_CANY_SEEN;
4237 Newxz(r->startp, RExC_npar, I32);
4238 Newxz(r->endp, RExC_npar, I32);
4242 SvREFCNT_dec((SV*)(RExC_charnames));
4244 DEBUG_r( RX_DEBUG_on(r) );
4246 PerlIO_printf(Perl_debug_log,"Final program:\n");
4249 DEBUG_OFFSETS_r(if (r->offsets) {
4250 const U32 len = r->offsets[0];
4252 GET_RE_DEBUG_FLAGS_DECL;
4253 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)r->offsets[0]);
4254 for (i = 1; i <= len; i++) {
4255 if (r->offsets[i*2-1] || r->offsets[i*2])
4256 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4257 (UV)i, (UV)r->offsets[i*2-1], (UV)r->offsets[i*2]);
4259 PerlIO_printf(Perl_debug_log, "\n");
4265 #undef CORE_ONLY_BLOCK
4267 #undef RE_ENGINE_PTR
4269 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4270 int rem=(int)(RExC_end - RExC_parse); \
4279 if (RExC_lastparse!=RExC_parse) \
4280 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4283 iscut ? "..." : "<" \
4286 PerlIO_printf(Perl_debug_log,"%16s",""); \
4291 num=REG_NODE_NUM(RExC_emit); \
4292 if (RExC_lastnum!=num) \
4293 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4295 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4296 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4297 (int)((depth*2)), "", \
4301 RExC_lastparse=RExC_parse; \
4306 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4307 DEBUG_PARSE_MSG((funcname)); \
4308 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4311 - reg - regular expression, i.e. main body or parenthesized thing
4313 * Caller must absorb opening parenthesis.
4315 * Combining parenthesis handling with the base level of regular expression
4316 * is a trifle forced, but the need to tie the tails of the branches to what
4317 * follows makes it hard to avoid.
4319 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4321 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4323 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4327 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4328 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4331 register regnode *ret; /* Will be the head of the group. */
4332 register regnode *br;
4333 register regnode *lastbr;
4334 register regnode *ender = NULL;
4335 register I32 parno = 0;
4337 const I32 oregflags = RExC_flags;
4338 bool have_branch = 0;
4341 /* for (?g), (?gc), and (?o) warnings; warning
4342 about (?c) will warn about (?g) -- japhy */
4344 #define WASTED_O 0x01
4345 #define WASTED_G 0x02
4346 #define WASTED_C 0x04
4347 #define WASTED_GC (0x02|0x04)
4348 I32 wastedflags = 0x00;
4350 char * parse_start = RExC_parse; /* MJD */
4351 char * const oregcomp_parse = RExC_parse;
4353 GET_RE_DEBUG_FLAGS_DECL;
4354 DEBUG_PARSE("reg ");
4357 *flagp = 0; /* Tentatively. */
4360 /* Make an OPEN node, if parenthesized. */
4362 if (*RExC_parse == '?') { /* (?...) */
4363 U32 posflags = 0, negflags = 0;
4364 U32 *flagsp = &posflags;
4365 bool is_logical = 0;
4366 const char * const seqstart = RExC_parse;
4369 paren = *RExC_parse++;
4370 ret = NULL; /* For look-ahead/behind. */
4372 case '<': /* (?<...) */
4373 RExC_seen |= REG_SEEN_LOOKBEHIND;
4374 if (*RExC_parse == '!')
4376 if (*RExC_parse != '=' && *RExC_parse != '!')
4379 case '=': /* (?=...) */
4380 case '!': /* (?!...) */
4381 RExC_seen_zerolen++;
4382 case ':': /* (?:...) */
4383 case '>': /* (?>...) */
4385 case '$': /* (?$...) */
4386 case '@': /* (?@...) */
4387 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
4389 case '#': /* (?#...) */
4390 while (*RExC_parse && *RExC_parse != ')')
4392 if (*RExC_parse != ')')
4393 FAIL("Sequence (?#... not terminated");
4394 nextchar(pRExC_state);
4397 case 'p': /* (?p...) */
4398 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
4399 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
4401 case '?': /* (??...) */
4403 if (*RExC_parse != '{')
4405 paren = *RExC_parse++;
4407 case '{': /* (?{...}) */
4409 I32 count = 1, n = 0;
4411 char *s = RExC_parse;
4413 RExC_seen_zerolen++;
4414 RExC_seen |= REG_SEEN_EVAL;
4415 while (count && (c = *RExC_parse)) {
4426 if (*RExC_parse != ')') {
4428 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
4432 OP_4tree *sop, *rop;
4433 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
4436 Perl_save_re_context(aTHX);
4437 rop = sv_compile_2op(sv, &sop, "re", &pad);
4438 sop->op_private |= OPpREFCOUNTED;
4439 /* re_dup will OpREFCNT_inc */
4440 OpREFCNT_set(sop, 1);
4443 n = add_data(pRExC_state, 3, "nop");
4444 RExC_rx->data->data[n] = (void*)rop;
4445 RExC_rx->data->data[n+1] = (void*)sop;
4446 RExC_rx->data->data[n+2] = (void*)pad;
4449 else { /* First pass */
4450 if (PL_reginterp_cnt < ++RExC_seen_evals
4452 /* No compiled RE interpolated, has runtime
4453 components ===> unsafe. */
4454 FAIL("Eval-group not allowed at runtime, use re 'eval'");
4455 if (PL_tainting && PL_tainted)
4456 FAIL("Eval-group in insecure regular expression");
4457 #if PERL_VERSION > 8
4458 if (IN_PERL_COMPILETIME)
4463 nextchar(pRExC_state);
4465 ret = reg_node(pRExC_state, LOGICAL);
4468 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
4469 /* deal with the length of this later - MJD */
4472 ret = reganode(pRExC_state, EVAL, n);
4473 Set_Node_Length(ret, RExC_parse - parse_start + 1);
4474 Set_Node_Offset(ret, parse_start);
4477 case '(': /* (?(?{...})...) and (?(?=...)...) */
4479 if (RExC_parse[0] == '?') { /* (?(?...)) */
4480 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
4481 || RExC_parse[1] == '<'
4482 || RExC_parse[1] == '{') { /* Lookahead or eval. */
4485 ret = reg_node(pRExC_state, LOGICAL);
4488 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
4492 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
4495 parno = atoi(RExC_parse++);
4497 while (isDIGIT(*RExC_parse))
4499 ret = reganode(pRExC_state, GROUPP, parno);
4501 if ((c = *nextchar(pRExC_state)) != ')')
4502 vFAIL("Switch condition not recognized");
4504 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
4505 br = regbranch(pRExC_state, &flags, 1,depth+1);
4507 br = reganode(pRExC_state, LONGJMP, 0);
4509 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
4510 c = *nextchar(pRExC_state);
4514 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
4515 regbranch(pRExC_state, &flags, 1,depth+1);
4516 REGTAIL(pRExC_state, ret, lastbr);
4519 c = *nextchar(pRExC_state);
4524 vFAIL("Switch (?(condition)... contains too many branches");
4525 ender = reg_node(pRExC_state, TAIL);
4526 REGTAIL(pRExC_state, br, ender);
4528 REGTAIL(pRExC_state, lastbr, ender);
4529 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
4532 REGTAIL(pRExC_state, ret, ender);
4536 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
4540 RExC_parse--; /* for vFAIL to print correctly */
4541 vFAIL("Sequence (? incomplete");
4545 parse_flags: /* (?i) */
4546 while (*RExC_parse && strchr("iogcmsx", *RExC_parse)) {
4547 /* (?g), (?gc) and (?o) are useless here
4548 and must be globally applied -- japhy */
4550 if (*RExC_parse == 'o' || *RExC_parse == 'g') {
4551 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
4552 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
4553 if (! (wastedflags & wflagbit) ) {
4554 wastedflags |= wflagbit;
4557 "Useless (%s%c) - %suse /%c modifier",
4558 flagsp == &negflags ? "?-" : "?",
4560 flagsp == &negflags ? "don't " : "",
4566 else if (*RExC_parse == 'c') {
4567 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
4568 if (! (wastedflags & WASTED_C) ) {
4569 wastedflags |= WASTED_GC;
4572 "Useless (%sc) - %suse /gc modifier",
4573 flagsp == &negflags ? "?-" : "?",
4574 flagsp == &negflags ? "don't " : ""
4579 else { pmflag(flagsp, *RExC_parse); }
4583 if (*RExC_parse == '-') {
4585 wastedflags = 0; /* reset so (?g-c) warns twice */
4589 RExC_flags |= posflags;
4590 RExC_flags &= ~negflags;
4591 if (*RExC_parse == ':') {
4597 if (*RExC_parse != ')') {
4599 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
4601 nextchar(pRExC_state);
4609 ret = reganode(pRExC_state, OPEN, parno);
4610 Set_Node_Length(ret, 1); /* MJD */
4611 Set_Node_Offset(ret, RExC_parse); /* MJD */
4618 /* Pick up the branches, linking them together. */
4619 parse_start = RExC_parse; /* MJD */
4620 br = regbranch(pRExC_state, &flags, 1,depth+1);
4621 /* branch_len = (paren != 0); */
4625 if (*RExC_parse == '|') {
4626 if (!SIZE_ONLY && RExC_extralen) {
4627 reginsert(pRExC_state, BRANCHJ, br);
4630 reginsert(pRExC_state, BRANCH, br);
4631 Set_Node_Length(br, paren != 0);
4632 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
4636 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
4638 else if (paren == ':') {
4639 *flagp |= flags&SIMPLE;
4641 if (is_open) { /* Starts with OPEN. */
4642 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
4644 else if (paren != '?') /* Not Conditional */
4646 *flagp |= flags & (SPSTART | HASWIDTH);
4648 while (*RExC_parse == '|') {
4649 if (!SIZE_ONLY && RExC_extralen) {
4650 ender = reganode(pRExC_state, LONGJMP,0);
4651 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
4654 RExC_extralen += 2; /* Account for LONGJMP. */
4655 nextchar(pRExC_state);
4656 br = regbranch(pRExC_state, &flags, 0, depth+1);
4660 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
4664 *flagp |= flags&SPSTART;
4667 if (have_branch || paren != ':') {
4668 /* Make a closing node, and hook it on the end. */
4671 ender = reg_node(pRExC_state, TAIL);
4674 ender = reganode(pRExC_state, CLOSE, parno);
4675 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
4676 Set_Node_Length(ender,1); /* MJD */
4682 *flagp &= ~HASWIDTH;
4685 ender = reg_node(pRExC_state, SUCCEED);
4688 ender = reg_node(pRExC_state, END);
4691 REGTAIL_STUDY(pRExC_state, lastbr, ender);
4693 if (have_branch && !SIZE_ONLY) {
4694 /* Hook the tails of the branches to the closing node. */
4695 for (br = ret; br; br = regnext(br)) {
4696 const U8 op = PL_regkind[OP(br)];
4698 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
4700 else if (op == BRANCHJ) {
4701 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
4709 static const char parens[] = "=!<,>";
4711 if (paren && (p = strchr(parens, paren))) {
4712 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
4713 int flag = (p - parens) > 1;
4716 node = SUSPEND, flag = 0;
4717 reginsert(pRExC_state, node,ret);
4718 Set_Node_Cur_Length(ret);
4719 Set_Node_Offset(ret, parse_start + 1);
4721 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
4725 /* Check for proper termination. */
4727 RExC_flags = oregflags;
4728 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
4729 RExC_parse = oregcomp_parse;
4730 vFAIL("Unmatched (");
4733 else if (!paren && RExC_parse < RExC_end) {
4734 if (*RExC_parse == ')') {
4736 vFAIL("Unmatched )");
4739 FAIL("Junk on end of regexp"); /* "Can't happen". */
4747 - regbranch - one alternative of an | operator
4749 * Implements the concatenation operator.
4752 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
4755 register regnode *ret;
4756 register regnode *chain = NULL;
4757 register regnode *latest;
4758 I32 flags = 0, c = 0;
4759 GET_RE_DEBUG_FLAGS_DECL;
4760 DEBUG_PARSE("brnc");
4764 if (!SIZE_ONLY && RExC_extralen)
4765 ret = reganode(pRExC_state, BRANCHJ,0);
4767 ret = reg_node(pRExC_state, BRANCH);
4768 Set_Node_Length(ret, 1);
4772 if (!first && SIZE_ONLY)
4773 RExC_extralen += 1; /* BRANCHJ */
4775 *flagp = WORST; /* Tentatively. */
4778 nextchar(pRExC_state);
4779 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
4781 latest = regpiece(pRExC_state, &flags,depth+1);
4782 if (latest == NULL) {
4783 if (flags & TRYAGAIN)
4787 else if (ret == NULL)
4789 *flagp |= flags&HASWIDTH;
4790 if (chain == NULL) /* First piece. */
4791 *flagp |= flags&SPSTART;
4794 REGTAIL(pRExC_state, chain, latest);
4799 if (chain == NULL) { /* Loop ran zero times. */
4800 chain = reg_node(pRExC_state, NOTHING);
4805 *flagp |= flags&SIMPLE;
4812 - regpiece - something followed by possible [*+?]
4814 * Note that the branching code sequences used for ? and the general cases
4815 * of * and + are somewhat optimized: they use the same NOTHING node as
4816 * both the endmarker for their branch list and the body of the last branch.
4817 * It might seem that this node could be dispensed with entirely, but the
4818 * endmarker role is not redundant.
4821 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
4824 register regnode *ret;
4826 register char *next;
4828 const char * const origparse = RExC_parse;
4830 I32 max = REG_INFTY;
4832 const char *maxpos = NULL;
4833 GET_RE_DEBUG_FLAGS_DECL;
4834 DEBUG_PARSE("piec");
4836 ret = regatom(pRExC_state, &flags,depth+1);
4838 if (flags & TRYAGAIN)
4845 if (op == '{' && regcurly(RExC_parse)) {
4847 parse_start = RExC_parse; /* MJD */
4848 next = RExC_parse + 1;
4849 while (isDIGIT(*next) || *next == ',') {
4858 if (*next == '}') { /* got one */
4862 min = atoi(RExC_parse);
4866 maxpos = RExC_parse;
4868 if (!max && *maxpos != '0')
4869 max = REG_INFTY; /* meaning "infinity" */
4870 else if (max >= REG_INFTY)
4871 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
4873 nextchar(pRExC_state);
4876 if ((flags&SIMPLE)) {
4877 RExC_naughty += 2 + RExC_naughty / 2;
4878 reginsert(pRExC_state, CURLY, ret);
4879 Set_Node_Offset(ret, parse_start+1); /* MJD */
4880 Set_Node_Cur_Length(ret);
4883 regnode * const w = reg_node(pRExC_state, WHILEM);
4886 REGTAIL(pRExC_state, ret, w);
4887 if (!SIZE_ONLY && RExC_extralen) {
4888 reginsert(pRExC_state, LONGJMP,ret);
4889 reginsert(pRExC_state, NOTHING,ret);
4890 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
4892 reginsert(pRExC_state, CURLYX,ret);
4894 Set_Node_Offset(ret, parse_start+1);
4895 Set_Node_Length(ret,
4896 op == '{' ? (RExC_parse - parse_start) : 1);
4898 if (!SIZE_ONLY && RExC_extralen)
4899 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
4900 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
4902 RExC_whilem_seen++, RExC_extralen += 3;
4903 RExC_naughty += 4 + RExC_naughty; /* compound interest */
4911 if (max && max < min)
4912 vFAIL("Can't do {n,m} with n > m");
4914 ARG1_SET(ret, (U16)min);
4915 ARG2_SET(ret, (U16)max);
4927 #if 0 /* Now runtime fix should be reliable. */
4929 /* if this is reinstated, don't forget to put this back into perldiag:
4931 =item Regexp *+ operand could be empty at {#} in regex m/%s/
4933 (F) The part of the regexp subject to either the * or + quantifier
4934 could match an empty string. The {#} shows in the regular
4935 expression about where the problem was discovered.
4939 if (!(flags&HASWIDTH) && op != '?')
4940 vFAIL("Regexp *+ operand could be empty");
4943 parse_start = RExC_parse;
4944 nextchar(pRExC_state);
4946 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
4948 if (op == '*' && (flags&SIMPLE)) {
4949 reginsert(pRExC_state, STAR, ret);
4953 else if (op == '*') {
4957 else if (op == '+' && (flags&SIMPLE)) {
4958 reginsert(pRExC_state, PLUS, ret);
4962 else if (op == '+') {
4966 else if (op == '?') {
4971 if (!SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
4973 "%.*s matches null string many times",
4974 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
4978 if (*RExC_parse == '?') {
4979 nextchar(pRExC_state);
4980 reginsert(pRExC_state, MINMOD, ret);
4981 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
4983 if (ISMULT2(RExC_parse)) {
4985 vFAIL("Nested quantifiers");
4992 /* reg_namedseq(pRExC_state,UVp)
4994 This is expected to be called by a parser routine that has
4995 recognized'\N' and needs to handle the rest. RExC_parse is
4996 expected to point at the first char following the N at the time
4999 If valuep is non-null then it is assumed that we are parsing inside
5000 of a charclass definition and the first codepoint in the resolved
5001 string is returned via *valuep and the routine will return NULL.
5002 In this mode if a multichar string is returned from the charnames
5003 handler a warning will be issued, and only the first char in the
5004 sequence will be examined. If the string returned is zero length
5005 then the value of *valuep is undefined and NON-NULL will
5006 be returned to indicate failure. (This will NOT be a valid pointer
5009 If value is null then it is assumed that we are parsing normal text
5010 and inserts a new EXACT node into the program containing the resolved
5011 string and returns a pointer to the new node. If the string is
5012 zerolength a NOTHING node is emitted.
5014 On success RExC_parse is set to the char following the endbrace.
5015 Parsing failures will generate a fatal errorvia vFAIL(...)
5017 NOTE: We cache all results from the charnames handler locally in
5018 the RExC_charnames hash (created on first use) to prevent a charnames
5019 handler from playing silly-buggers and returning a short string and
5020 then a long string for a given pattern. Since the regexp program
5021 size is calculated during an initial parse this would result
5022 in a buffer overrun so we cache to prevent the charname result from
5023 changing during the course of the parse.
5027 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
5029 char * name; /* start of the content of the name */
5030 char * endbrace; /* endbrace following the name */
5033 STRLEN len; /* this has various purposes throughout the code */
5034 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
5035 regnode *ret = NULL;
5037 if (*RExC_parse != '{') {
5038 vFAIL("Missing braces on \\N{}");
5040 name = RExC_parse+1;
5041 endbrace = strchr(RExC_parse, '}');
5044 vFAIL("Missing right brace on \\N{}");
5046 RExC_parse = endbrace + 1;
5049 /* RExC_parse points at the beginning brace,
5050 endbrace points at the last */
5051 if ( name[0]=='U' && name[1]=='+' ) {
5052 /* its a "unicode hex" notation {U+89AB} */
5053 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
5054 | PERL_SCAN_DISALLOW_PREFIX
5055 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
5057 len = (STRLEN)(endbrace - name - 2);
5058 cp = grok_hex(name + 2, &len, &fl, NULL);
5059 if ( len != (STRLEN)(endbrace - name - 2) ) {
5068 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
5070 /* fetch the charnames handler for this scope */
5071 HV * const table = GvHV(PL_hintgv);
5073 hv_fetchs(table, "charnames", FALSE) :
5075 SV *cv= cvp ? *cvp : NULL;
5078 /* create an SV with the name as argument */
5079 sv_name = newSVpvn(name, endbrace - name);
5081 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
5082 vFAIL2("Constant(\\N{%s}) unknown: "
5083 "(possibly a missing \"use charnames ...\")",
5086 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
5087 vFAIL2("Constant(\\N{%s}): "
5088 "$^H{charnames} is not defined",SvPVX(sv_name));
5093 if (!RExC_charnames) {
5094 /* make sure our cache is allocated */
5095 RExC_charnames = newHV();
5097 /* see if we have looked this one up before */
5098 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
5100 sv_str = HeVAL(he_str);
5113 count= call_sv(cv, G_SCALAR);
5115 if (count == 1) { /* XXXX is this right? dmq */
5117 SvREFCNT_inc_simple_void(sv_str);
5125 if ( !sv_str || !SvOK(sv_str) ) {
5126 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
5127 "did not return a defined value",SvPVX(sv_name));
5129 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
5134 char *p = SvPV(sv_str, len);
5137 if ( SvUTF8(sv_str) ) {
5138 *valuep = utf8_to_uvchr((U8*)p, &numlen);
5142 We have to turn on utf8 for high bit chars otherwise
5143 we get failures with
5145 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5146 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5148 This is different from what \x{} would do with the same
5149 codepoint, where the condition is > 0xFF.
5156 /* warn if we havent used the whole string? */
5158 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5160 "Ignoring excess chars from \\N{%s} in character class",
5164 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5166 "Ignoring zero length \\N{%s} in character class",
5171 SvREFCNT_dec(sv_name);
5173 SvREFCNT_dec(sv_str);
5174 return len ? NULL : (regnode *)&len;
5175 } else if(SvCUR(sv_str)) {
5180 char * parse_start = name-3; /* needed for the offsets */
5181 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
5183 ret = reg_node(pRExC_state,
5184 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
5187 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
5188 sv_utf8_upgrade(sv_str);
5189 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
5193 p = SvPV(sv_str, len);
5195 /* len is the length written, charlen is the size the char read */
5196 for ( len = 0; p < pend; p += charlen ) {
5198 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
5200 STRLEN foldlen,numlen;
5201 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
5202 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
5203 /* Emit all the Unicode characters. */
5205 for (foldbuf = tmpbuf;
5209 uvc = utf8_to_uvchr(foldbuf, &numlen);
5211 const STRLEN unilen = reguni(pRExC_state, uvc, s);
5214 /* In EBCDIC the numlen
5215 * and unilen can differ. */
5217 if (numlen >= foldlen)
5221 break; /* "Can't happen." */
5224 const STRLEN unilen = reguni(pRExC_state, uvc, s);
5236 RExC_size += STR_SZ(len);
5239 RExC_emit += STR_SZ(len);
5241 Set_Node_Cur_Length(ret); /* MJD */
5243 nextchar(pRExC_state);
5245 ret = reg_node(pRExC_state,NOTHING);
5248 SvREFCNT_dec(sv_str);
5251 SvREFCNT_dec(sv_name);
5260 - regatom - the lowest level
5262 * Optimization: gobbles an entire sequence of ordinary characters so that
5263 * it can turn them into a single node, which is smaller to store and
5264 * faster to run. Backslashed characters are exceptions, each becoming a
5265 * separate node; the code is simpler that way and it's not worth fixing.
5267 * [Yes, it is worth fixing, some scripts can run twice the speed.]
5268 * [It looks like its ok, as in S_study_chunk we merge adjacent EXACT nodes]
5271 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5274 register regnode *ret = NULL;
5276 char *parse_start = RExC_parse;
5277 GET_RE_DEBUG_FLAGS_DECL;
5278 DEBUG_PARSE("atom");
5279 *flagp = WORST; /* Tentatively. */
5282 switch (*RExC_parse) {
5284 RExC_seen_zerolen++;
5285 nextchar(pRExC_state);
5286 if (RExC_flags & PMf_MULTILINE)
5287 ret = reg_node(pRExC_state, MBOL);
5288 else if (RExC_flags & PMf_SINGLELINE)
5289 ret = reg_node(pRExC_state, SBOL);
5291 ret = reg_node(pRExC_state, BOL);
5292 Set_Node_Length(ret, 1); /* MJD */
5295 nextchar(pRExC_state);
5297 RExC_seen_zerolen++;
5298 if (RExC_flags & PMf_MULTILINE)
5299 ret = reg_node(pRExC_state, MEOL);
5300 else if (RExC_flags & PMf_SINGLELINE)
5301 ret = reg_node(pRExC_state, SEOL);
5303 ret = reg_node(pRExC_state, EOL);
5304 Set_Node_Length(ret, 1); /* MJD */
5307 nextchar(pRExC_state);
5308 if (RExC_flags & PMf_SINGLELINE)
5309 ret = reg_node(pRExC_state, SANY);
5311 ret = reg_node(pRExC_state, REG_ANY);
5312 *flagp |= HASWIDTH|SIMPLE;
5314 Set_Node_Length(ret, 1); /* MJD */
5318 char * const oregcomp_parse = ++RExC_parse;
5319 ret = regclass(pRExC_state,depth+1);
5320 if (*RExC_parse != ']') {
5321 RExC_parse = oregcomp_parse;
5322 vFAIL("Unmatched [");
5324 nextchar(pRExC_state);
5325 *flagp |= HASWIDTH|SIMPLE;
5326 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
5330 nextchar(pRExC_state);
5331 ret = reg(pRExC_state, 1, &flags,depth+1);
5333 if (flags & TRYAGAIN) {
5334 if (RExC_parse == RExC_end) {
5335 /* Make parent create an empty node if needed. */
5343 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
5347 if (flags & TRYAGAIN) {
5351 vFAIL("Internal urp");
5352 /* Supposed to be caught earlier. */
5355 if (!regcurly(RExC_parse)) {
5364 vFAIL("Quantifier follows nothing");
5367 switch (*++RExC_parse) {
5369 RExC_seen_zerolen++;
5370 ret = reg_node(pRExC_state, SBOL);
5372 nextchar(pRExC_state);
5373 Set_Node_Length(ret, 2); /* MJD */
5376 ret = reg_node(pRExC_state, GPOS);
5377 RExC_seen |= REG_SEEN_GPOS;
5379 nextchar(pRExC_state);
5380 Set_Node_Length(ret, 2); /* MJD */
5383 ret = reg_node(pRExC_state, SEOL);
5385 RExC_seen_zerolen++; /* Do not optimize RE away */
5386 nextchar(pRExC_state);
5389 ret = reg_node(pRExC_state, EOS);
5391 RExC_seen_zerolen++; /* Do not optimize RE away */
5392 nextchar(pRExC_state);
5393 Set_Node_Length(ret, 2); /* MJD */
5396 ret = reg_node(pRExC_state, CANY);
5397 RExC_seen |= REG_SEEN_CANY;
5398 *flagp |= HASWIDTH|SIMPLE;
5399 nextchar(pRExC_state);
5400 Set_Node_Length(ret, 2); /* MJD */
5403 ret = reg_node(pRExC_state, CLUMP);
5405 nextchar(pRExC_state);
5406 Set_Node_Length(ret, 2); /* MJD */
5409 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
5410 *flagp |= HASWIDTH|SIMPLE;
5411 nextchar(pRExC_state);
5412 Set_Node_Length(ret, 2); /* MJD */
5415 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
5416 *flagp |= HASWIDTH|SIMPLE;
5417 nextchar(pRExC_state);
5418 Set_Node_Length(ret, 2); /* MJD */
5421 RExC_seen_zerolen++;
5422 RExC_seen |= REG_SEEN_LOOKBEHIND;
5423 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
5425 nextchar(pRExC_state);
5426 Set_Node_Length(ret, 2); /* MJD */
5429 RExC_seen_zerolen++;
5430 RExC_seen |= REG_SEEN_LOOKBEHIND;
5431 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
5433 nextchar(pRExC_state);
5434 Set_Node_Length(ret, 2); /* MJD */
5437 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
5438 *flagp |= HASWIDTH|SIMPLE;
5439 nextchar(pRExC_state);
5440 Set_Node_Length(ret, 2); /* MJD */
5443 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
5444 *flagp |= HASWIDTH|SIMPLE;
5445 nextchar(pRExC_state);
5446 Set_Node_Length(ret, 2); /* MJD */
5449 ret = reg_node(pRExC_state, DIGIT);
5450 *flagp |= HASWIDTH|SIMPLE;
5451 nextchar(pRExC_state);
5452 Set_Node_Length(ret, 2); /* MJD */
5455 ret = reg_node(pRExC_state, NDIGIT);
5456 *flagp |= HASWIDTH|SIMPLE;
5457 nextchar(pRExC_state);
5458 Set_Node_Length(ret, 2); /* MJD */
5463 char* const oldregxend = RExC_end;
5464 char* parse_start = RExC_parse - 2;
5466 if (RExC_parse[1] == '{') {
5467 /* a lovely hack--pretend we saw [\pX] instead */
5468 RExC_end = strchr(RExC_parse, '}');
5470 const U8 c = (U8)*RExC_parse;
5472 RExC_end = oldregxend;
5473 vFAIL2("Missing right brace on \\%c{}", c);
5478 RExC_end = RExC_parse + 2;
5479 if (RExC_end > oldregxend)
5480 RExC_end = oldregxend;
5484 ret = regclass(pRExC_state,depth+1);
5486 RExC_end = oldregxend;
5489 Set_Node_Offset(ret, parse_start + 2);
5490 Set_Node_Cur_Length(ret);
5491 nextchar(pRExC_state);
5492 *flagp |= HASWIDTH|SIMPLE;
5496 /* Handle \N{NAME} here and not below because it can be
5497 multicharacter. join_exact() will join them up later on.
5498 Also this makes sure that things like /\N{BLAH}+/ and
5499 \N{BLAH} being multi char Just Happen. dmq*/
5501 ret= reg_namedseq(pRExC_state, NULL);
5513 case '1': case '2': case '3': case '4':
5514 case '5': case '6': case '7': case '8': case '9':
5516 const I32 num = atoi(RExC_parse);
5518 if (num > 9 && num >= RExC_npar)
5521 char * const parse_start = RExC_parse - 1; /* MJD */
5522 while (isDIGIT(*RExC_parse))
5525 if (!SIZE_ONLY && num > (I32)RExC_rx->nparens)
5526 vFAIL("Reference to nonexistent group");
5528 ret = reganode(pRExC_state,
5529 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
5533 /* override incorrect value set in reganode MJD */
5534 Set_Node_Offset(ret, parse_start+1);
5535 Set_Node_Cur_Length(ret); /* MJD */
5537 nextchar(pRExC_state);
5542 if (RExC_parse >= RExC_end)
5543 FAIL("Trailing \\");
5546 /* Do not generate "unrecognized" warnings here, we fall
5547 back into the quick-grab loop below */
5554 if (RExC_flags & PMf_EXTENDED) {
5555 while (RExC_parse < RExC_end && *RExC_parse != '\n')
5557 if (RExC_parse < RExC_end)
5563 register STRLEN len;
5568 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
5570 parse_start = RExC_parse - 1;
5576 ret = reg_node(pRExC_state,
5577 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
5579 for (len = 0, p = RExC_parse - 1;
5580 len < 127 && p < RExC_end;
5583 char * const oldp = p;
5585 if (RExC_flags & PMf_EXTENDED)
5586 p = regwhite(p, RExC_end);
5634 ender = ASCII_TO_NATIVE('\033');
5638 ender = ASCII_TO_NATIVE('\007');
5643 char* const e = strchr(p, '}');
5647 vFAIL("Missing right brace on \\x{}");
5650 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
5651 | PERL_SCAN_DISALLOW_PREFIX;
5652 STRLEN numlen = e - p - 1;
5653 ender = grok_hex(p + 1, &numlen, &flags, NULL);
5660 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
5662 ender = grok_hex(p, &numlen, &flags, NULL);
5668 ender = UCHARAT(p++);
5669 ender = toCTRL(ender);
5671 case '0': case '1': case '2': case '3':case '4':
5672 case '5': case '6': case '7': case '8':case '9':
5674 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
5677 ender = grok_oct(p, &numlen, &flags, NULL);
5687 FAIL("Trailing \\");
5690 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
5691 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
5692 goto normal_default;
5697 if (UTF8_IS_START(*p) && UTF) {
5699 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
5700 &numlen, UTF8_ALLOW_DEFAULT);
5707 if (RExC_flags & PMf_EXTENDED)
5708 p = regwhite(p, RExC_end);
5710 /* Prime the casefolded buffer. */
5711 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
5713 if (ISMULT2(p)) { /* Back off on ?+*. */
5718 /* Emit all the Unicode characters. */
5720 for (foldbuf = tmpbuf;
5722 foldlen -= numlen) {
5723 ender = utf8_to_uvchr(foldbuf, &numlen);
5725 const STRLEN unilen = reguni(pRExC_state, ender, s);
5728 /* In EBCDIC the numlen
5729 * and unilen can differ. */
5731 if (numlen >= foldlen)
5735 break; /* "Can't happen." */
5739 const STRLEN unilen = reguni(pRExC_state, ender, s);
5748 REGC((char)ender, s++);
5754 /* Emit all the Unicode characters. */
5756 for (foldbuf = tmpbuf;
5758 foldlen -= numlen) {
5759 ender = utf8_to_uvchr(foldbuf, &numlen);
5761 const STRLEN unilen = reguni(pRExC_state, ender, s);
5764 /* In EBCDIC the numlen
5765 * and unilen can differ. */
5767 if (numlen >= foldlen)
5775 const STRLEN unilen = reguni(pRExC_state, ender, s);
5784 REGC((char)ender, s++);
5788 Set_Node_Cur_Length(ret); /* MJD */
5789 nextchar(pRExC_state);
5791 /* len is STRLEN which is unsigned, need to copy to signed */
5794 vFAIL("Internal disaster");
5798 if (len == 1 && UNI_IS_INVARIANT(ender))
5802 RExC_size += STR_SZ(len);
5805 RExC_emit += STR_SZ(len);
5811 /* If the encoding pragma is in effect recode the text of
5812 * any EXACT-kind nodes. */
5813 if (ret && PL_encoding && PL_regkind[OP(ret)] == EXACT) {
5814 const STRLEN oldlen = STR_LEN(ret);
5815 SV * const sv = sv_2mortal(newSVpvn(STRING(ret), oldlen));
5819 if (sv_utf8_downgrade(sv, TRUE)) {
5820 const char * const s = sv_recode_to_utf8(sv, PL_encoding);
5821 const STRLEN newlen = SvCUR(sv);
5826 GET_RE_DEBUG_FLAGS_DECL;
5827 DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "recode %*s to %*s\n",
5828 (int)oldlen, STRING(ret),
5830 Copy(s, STRING(ret), newlen, char);
5831 STR_LEN(ret) += newlen - oldlen;
5832 RExC_emit += STR_SZ(newlen) - STR_SZ(oldlen);
5834 RExC_size += STR_SZ(newlen) - STR_SZ(oldlen);
5842 S_regwhite(char *p, const char *e)
5847 else if (*p == '#') {
5850 } while (p < e && *p != '\n');
5858 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
5859 Character classes ([:foo:]) can also be negated ([:^foo:]).
5860 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
5861 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
5862 but trigger failures because they are currently unimplemented. */
5864 #define POSIXCC_DONE(c) ((c) == ':')
5865 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
5866 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
5869 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
5872 I32 namedclass = OOB_NAMEDCLASS;
5874 if (value == '[' && RExC_parse + 1 < RExC_end &&
5875 /* I smell either [: or [= or [. -- POSIX has been here, right? */
5876 POSIXCC(UCHARAT(RExC_parse))) {
5877 const char c = UCHARAT(RExC_parse);
5878 char* const s = RExC_parse++;
5880 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
5882 if (RExC_parse == RExC_end)
5883 /* Grandfather lone [:, [=, [. */
5886 const char* const t = RExC_parse++; /* skip over the c */
5889 if (UCHARAT(RExC_parse) == ']') {
5890 const char *posixcc = s + 1;
5891 RExC_parse++; /* skip over the ending ] */
5894 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
5895 const I32 skip = t - posixcc;
5897 /* Initially switch on the length of the name. */
5900 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
5901 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
5904 /* Names all of length 5. */
5905 /* alnum alpha ascii blank cntrl digit graph lower
5906 print punct space upper */
5907 /* Offset 4 gives the best switch position. */
5908 switch (posixcc[4]) {
5910 if (memEQ(posixcc, "alph", 4)) /* alpha */
5911 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
5914 if (memEQ(posixcc, "spac", 4)) /* space */
5915 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
5918 if (memEQ(posixcc, "grap", 4)) /* graph */
5919 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
5922 if (memEQ(posixcc, "asci", 4)) /* ascii */
5923 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
5926 if (memEQ(posixcc, "blan", 4)) /* blank */
5927 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
5930 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
5931 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
5934 if (memEQ(posixcc, "alnu", 4)) /* alnum */
5935 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
5938 if (memEQ(posixcc, "lowe", 4)) /* lower */
5939 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
5940 else if (memEQ(posixcc, "uppe", 4)) /* upper */
5941 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
5944 if (memEQ(posixcc, "digi", 4)) /* digit */
5945 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
5946 else if (memEQ(posixcc, "prin", 4)) /* print */
5947 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
5948 else if (memEQ(posixcc, "punc", 4)) /* punct */
5949 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
5954 if (memEQ(posixcc, "xdigit", 6))
5955 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
5959 if (namedclass == OOB_NAMEDCLASS)
5960 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
5962 assert (posixcc[skip] == ':');
5963 assert (posixcc[skip+1] == ']');
5964 } else if (!SIZE_ONLY) {
5965 /* [[=foo=]] and [[.foo.]] are still future. */
5967 /* adjust RExC_parse so the warning shows after
5969 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
5971 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
5974 /* Maternal grandfather:
5975 * "[:" ending in ":" but not in ":]" */
5985 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
5988 if (POSIXCC(UCHARAT(RExC_parse))) {
5989 const char *s = RExC_parse;
5990 const char c = *s++;
5994 if (*s && c == *s && s[1] == ']') {
5995 if (ckWARN(WARN_REGEXP))
5997 "POSIX syntax [%c %c] belongs inside character classes",
6000 /* [[=foo=]] and [[.foo.]] are still future. */
6001 if (POSIXCC_NOTYET(c)) {
6002 /* adjust RExC_parse so the error shows after
6004 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
6006 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6014 parse a class specification and produce either an ANYOF node that
6015 matches the pattern. If the pattern matches a single char only and
6016 that char is < 256 then we produce an EXACT node instead.
6019 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
6022 register UV value = 0;
6023 register UV nextvalue;
6024 register IV prevvalue = OOB_UNICODE;
6025 register IV range = 0;
6026 register regnode *ret;
6029 char *rangebegin = NULL;
6030 bool need_class = 0;
6033 bool optimize_invert = TRUE;
6034 AV* unicode_alternate = NULL;
6036 UV literal_endpoint = 0;
6038 UV stored = 0; /* number of chars stored in the class */
6040 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
6041 case we need to change the emitted regop to an EXACT. */
6042 const char * orig_parse = RExC_parse;
6043 GET_RE_DEBUG_FLAGS_DECL;
6045 PERL_UNUSED_ARG(depth);
6048 DEBUG_PARSE("clas");
6050 /* Assume we are going to generate an ANYOF node. */
6051 ret = reganode(pRExC_state, ANYOF, 0);
6054 ANYOF_FLAGS(ret) = 0;
6056 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
6060 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
6064 RExC_size += ANYOF_SKIP;
6065 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
6068 RExC_emit += ANYOF_SKIP;
6070 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
6072 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
6073 ANYOF_BITMAP_ZERO(ret);
6074 listsv = newSVpvs("# comment\n");
6077 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6079 if (!SIZE_ONLY && POSIXCC(nextvalue))
6080 checkposixcc(pRExC_state);
6082 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
6083 if (UCHARAT(RExC_parse) == ']')
6087 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
6091 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
6094 rangebegin = RExC_parse;
6096 value = utf8n_to_uvchr((U8*)RExC_parse,
6097 RExC_end - RExC_parse,
6098 &numlen, UTF8_ALLOW_DEFAULT);
6099 RExC_parse += numlen;
6102 value = UCHARAT(RExC_parse++);
6104 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6105 if (value == '[' && POSIXCC(nextvalue))
6106 namedclass = regpposixcc(pRExC_state, value);
6107 else if (value == '\\') {
6109 value = utf8n_to_uvchr((U8*)RExC_parse,
6110 RExC_end - RExC_parse,
6111 &numlen, UTF8_ALLOW_DEFAULT);
6112 RExC_parse += numlen;
6115 value = UCHARAT(RExC_parse++);
6116 /* Some compilers cannot handle switching on 64-bit integer
6117 * values, therefore value cannot be an UV. Yes, this will
6118 * be a problem later if we want switch on Unicode.
6119 * A similar issue a little bit later when switching on
6120 * namedclass. --jhi */
6121 switch ((I32)value) {
6122 case 'w': namedclass = ANYOF_ALNUM; break;
6123 case 'W': namedclass = ANYOF_NALNUM; break;
6124 case 's': namedclass = ANYOF_SPACE; break;
6125 case 'S': namedclass = ANYOF_NSPACE; break;
6126 case 'd': namedclass = ANYOF_DIGIT; break;
6127 case 'D': namedclass = ANYOF_NDIGIT; break;
6128 case 'N': /* Handle \N{NAME} in class */
6130 /* We only pay attention to the first char of
6131 multichar strings being returned. I kinda wonder
6132 if this makes sense as it does change the behaviour
6133 from earlier versions, OTOH that behaviour was broken
6135 UV v; /* value is register so we cant & it /grrr */
6136 if (reg_namedseq(pRExC_state, &v)) {
6146 if (RExC_parse >= RExC_end)
6147 vFAIL2("Empty \\%c{}", (U8)value);
6148 if (*RExC_parse == '{') {
6149 const U8 c = (U8)value;
6150 e = strchr(RExC_parse++, '}');
6152 vFAIL2("Missing right brace on \\%c{}", c);
6153 while (isSPACE(UCHARAT(RExC_parse)))
6155 if (e == RExC_parse)
6156 vFAIL2("Empty \\%c{}", c);
6158 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
6166 if (UCHARAT(RExC_parse) == '^') {
6169 value = value == 'p' ? 'P' : 'p'; /* toggle */
6170 while (isSPACE(UCHARAT(RExC_parse))) {
6175 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
6176 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
6179 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
6180 namedclass = ANYOF_MAX; /* no official name, but it's named */
6183 case 'n': value = '\n'; break;
6184 case 'r': value = '\r'; break;
6185 case 't': value = '\t'; break;
6186 case 'f': value = '\f'; break;
6187 case 'b': value = '\b'; break;
6188 case 'e': value = ASCII_TO_NATIVE('\033');break;
6189 case 'a': value = ASCII_TO_NATIVE('\007');break;
6191 if (*RExC_parse == '{') {
6192 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6193 | PERL_SCAN_DISALLOW_PREFIX;
6194 char * const e = strchr(RExC_parse++, '}');
6196 vFAIL("Missing right brace on \\x{}");
6198 numlen = e - RExC_parse;
6199 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
6203 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6205 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
6206 RExC_parse += numlen;
6210 value = UCHARAT(RExC_parse++);
6211 value = toCTRL(value);
6213 case '0': case '1': case '2': case '3': case '4':
6214 case '5': case '6': case '7': case '8': case '9':
6218 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
6219 RExC_parse += numlen;
6223 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
6225 "Unrecognized escape \\%c in character class passed through",
6229 } /* end of \blah */
6235 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
6237 if (!SIZE_ONLY && !need_class)
6238 ANYOF_CLASS_ZERO(ret);
6242 /* a bad range like a-\d, a-[:digit:] ? */
6245 if (ckWARN(WARN_REGEXP)) {
6247 RExC_parse >= rangebegin ?
6248 RExC_parse - rangebegin : 0;
6250 "False [] range \"%*.*s\"",
6253 if (prevvalue < 256) {
6254 ANYOF_BITMAP_SET(ret, prevvalue);
6255 ANYOF_BITMAP_SET(ret, '-');
6258 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
6259 Perl_sv_catpvf(aTHX_ listsv,
6260 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
6264 range = 0; /* this was not a true range */
6268 const char *what = NULL;
6271 if (namedclass > OOB_NAMEDCLASS)
6272 optimize_invert = FALSE;
6273 /* Possible truncation here but in some 64-bit environments
6274 * the compiler gets heartburn about switch on 64-bit values.
6275 * A similar issue a little earlier when switching on value.
6277 switch ((I32)namedclass) {
6280 ANYOF_CLASS_SET(ret, ANYOF_ALNUM);
6282 for (value = 0; value < 256; value++)
6284 ANYOF_BITMAP_SET(ret, value);
6291 ANYOF_CLASS_SET(ret, ANYOF_NALNUM);
6293 for (value = 0; value < 256; value++)
6294 if (!isALNUM(value))
6295 ANYOF_BITMAP_SET(ret, value);
6302 ANYOF_CLASS_SET(ret, ANYOF_ALNUMC);
6304 for (value = 0; value < 256; value++)
6305 if (isALNUMC(value))
6306 ANYOF_BITMAP_SET(ret, value);
6313 ANYOF_CLASS_SET(ret, ANYOF_NALNUMC);
6315 for (value = 0; value < 256; value++)
6316 if (!isALNUMC(value))
6317 ANYOF_BITMAP_SET(ret, value);
6324 ANYOF_CLASS_SET(ret, ANYOF_ALPHA);
6326 for (value = 0; value < 256; value++)
6328 ANYOF_BITMAP_SET(ret, value);
6335 ANYOF_CLASS_SET(ret, ANYOF_NALPHA);
6337 for (value = 0; value < 256; value++)
6338 if (!isALPHA(value))
6339 ANYOF_BITMAP_SET(ret, value);
6346 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
6349 for (value = 0; value < 128; value++)
6350 ANYOF_BITMAP_SET(ret, value);
6352 for (value = 0; value < 256; value++) {
6354 ANYOF_BITMAP_SET(ret, value);
6363 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
6366 for (value = 128; value < 256; value++)
6367 ANYOF_BITMAP_SET(ret, value);
6369 for (value = 0; value < 256; value++) {
6370 if (!isASCII(value))
6371 ANYOF_BITMAP_SET(ret, value);
6380 ANYOF_CLASS_SET(ret, ANYOF_BLANK);
6382 for (value = 0; value < 256; value++)
6384 ANYOF_BITMAP_SET(ret, value);
6391 ANYOF_CLASS_SET(ret, ANYOF_NBLANK);
6393 for (value = 0; value < 256; value++)
6394 if (!isBLANK(value))
6395 ANYOF_BITMAP_SET(ret, value);
6402 ANYOF_CLASS_SET(ret, ANYOF_CNTRL);
6404 for (value = 0; value < 256; value++)
6406 ANYOF_BITMAP_SET(ret, value);
6413 ANYOF_CLASS_SET(ret, ANYOF_NCNTRL);
6415 for (value = 0; value < 256; value++)
6416 if (!isCNTRL(value))
6417 ANYOF_BITMAP_SET(ret, value);
6424 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
6426 /* consecutive digits assumed */
6427 for (value = '0'; value <= '9'; value++)
6428 ANYOF_BITMAP_SET(ret, value);
6435 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
6437 /* consecutive digits assumed */
6438 for (value = 0; value < '0'; value++)
6439 ANYOF_BITMAP_SET(ret, value);
6440 for (value = '9' + 1; value < 256; value++)
6441 ANYOF_BITMAP_SET(ret, value);
6448 ANYOF_CLASS_SET(ret, ANYOF_GRAPH);
6450 for (value = 0; value < 256; value++)
6452 ANYOF_BITMAP_SET(ret, value);
6459 ANYOF_CLASS_SET(ret, ANYOF_NGRAPH);
6461 for (value = 0; value < 256; value++)
6462 if (!isGRAPH(value))
6463 ANYOF_BITMAP_SET(ret, value);
6470 ANYOF_CLASS_SET(ret, ANYOF_LOWER);
6472 for (value = 0; value < 256; value++)
6474 ANYOF_BITMAP_SET(ret, value);
6481 ANYOF_CLASS_SET(ret, ANYOF_NLOWER);
6483 for (value = 0; value < 256; value++)
6484 if (!isLOWER(value))
6485 ANYOF_BITMAP_SET(ret, value);
6492 ANYOF_CLASS_SET(ret, ANYOF_PRINT);
6494 for (value = 0; value < 256; value++)
6496 ANYOF_BITMAP_SET(ret, value);
6503 ANYOF_CLASS_SET(ret, ANYOF_NPRINT);
6505 for (value = 0; value < 256; value++)
6506 if (!isPRINT(value))
6507 ANYOF_BITMAP_SET(ret, value);
6514 ANYOF_CLASS_SET(ret, ANYOF_PSXSPC);
6516 for (value = 0; value < 256; value++)
6517 if (isPSXSPC(value))
6518 ANYOF_BITMAP_SET(ret, value);
6525 ANYOF_CLASS_SET(ret, ANYOF_NPSXSPC);
6527 for (value = 0; value < 256; value++)
6528 if (!isPSXSPC(value))
6529 ANYOF_BITMAP_SET(ret, value);
6536 ANYOF_CLASS_SET(ret, ANYOF_PUNCT);
6538 for (value = 0; value < 256; value++)
6540 ANYOF_BITMAP_SET(ret, value);
6547 ANYOF_CLASS_SET(ret, ANYOF_NPUNCT);
6549 for (value = 0; value < 256; value++)
6550 if (!isPUNCT(value))
6551 ANYOF_BITMAP_SET(ret, value);
6558 ANYOF_CLASS_SET(ret, ANYOF_SPACE);
6560 for (value = 0; value < 256; value++)
6562 ANYOF_BITMAP_SET(ret, value);
6569 ANYOF_CLASS_SET(ret, ANYOF_NSPACE);
6571 for (value = 0; value < 256; value++)
6572 if (!isSPACE(value))
6573 ANYOF_BITMAP_SET(ret, value);
6580 ANYOF_CLASS_SET(ret, ANYOF_UPPER);
6582 for (value = 0; value < 256; value++)
6584 ANYOF_BITMAP_SET(ret, value);
6591 ANYOF_CLASS_SET(ret, ANYOF_NUPPER);
6593 for (value = 0; value < 256; value++)
6594 if (!isUPPER(value))
6595 ANYOF_BITMAP_SET(ret, value);
6602 ANYOF_CLASS_SET(ret, ANYOF_XDIGIT);
6604 for (value = 0; value < 256; value++)
6605 if (isXDIGIT(value))
6606 ANYOF_BITMAP_SET(ret, value);
6613 ANYOF_CLASS_SET(ret, ANYOF_NXDIGIT);
6615 for (value = 0; value < 256; value++)
6616 if (!isXDIGIT(value))
6617 ANYOF_BITMAP_SET(ret, value);
6623 /* this is to handle \p and \P */
6626 vFAIL("Invalid [::] class");
6630 /* Strings such as "+utf8::isWord\n" */
6631 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
6634 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
6637 } /* end of namedclass \blah */
6640 if (prevvalue > (IV)value) /* b-a */ {
6641 const int w = RExC_parse - rangebegin;
6642 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
6643 range = 0; /* not a valid range */
6647 prevvalue = value; /* save the beginning of the range */
6648 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
6649 RExC_parse[1] != ']') {
6652 /* a bad range like \w-, [:word:]- ? */
6653 if (namedclass > OOB_NAMEDCLASS) {
6654 if (ckWARN(WARN_REGEXP)) {
6656 RExC_parse >= rangebegin ?
6657 RExC_parse - rangebegin : 0;
6659 "False [] range \"%*.*s\"",
6663 ANYOF_BITMAP_SET(ret, '-');
6665 range = 1; /* yeah, it's a range! */
6666 continue; /* but do it the next time */
6670 /* now is the next time */
6671 /*stored += (value - prevvalue + 1);*/
6673 if (prevvalue < 256) {
6674 const IV ceilvalue = value < 256 ? value : 255;
6677 /* In EBCDIC [\x89-\x91] should include
6678 * the \x8e but [i-j] should not. */
6679 if (literal_endpoint == 2 &&
6680 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
6681 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
6683 if (isLOWER(prevvalue)) {
6684 for (i = prevvalue; i <= ceilvalue; i++)
6686 ANYOF_BITMAP_SET(ret, i);
6688 for (i = prevvalue; i <= ceilvalue; i++)
6690 ANYOF_BITMAP_SET(ret, i);
6695 for (i = prevvalue; i <= ceilvalue; i++) {
6696 if (!ANYOF_BITMAP_TEST(ret,i)) {
6698 ANYOF_BITMAP_SET(ret, i);
6702 if (value > 255 || UTF) {
6703 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
6704 const UV natvalue = NATIVE_TO_UNI(value);
6705 stored+=2; /* can't optimize this class */
6706 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
6707 if (prevnatvalue < natvalue) { /* what about > ? */
6708 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
6709 prevnatvalue, natvalue);
6711 else if (prevnatvalue == natvalue) {
6712 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
6714 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
6716 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
6718 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
6719 if (RExC_precomp[0] == ':' &&
6720 RExC_precomp[1] == '[' &&
6721 (f == 0xDF || f == 0x92)) {
6722 f = NATIVE_TO_UNI(f);
6725 /* If folding and foldable and a single
6726 * character, insert also the folded version
6727 * to the charclass. */
6729 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
6730 if ((RExC_precomp[0] == ':' &&
6731 RExC_precomp[1] == '[' &&
6733 (value == 0xFB05 || value == 0xFB06))) ?
6734 foldlen == ((STRLEN)UNISKIP(f) - 1) :
6735 foldlen == (STRLEN)UNISKIP(f) )
6737 if (foldlen == (STRLEN)UNISKIP(f))
6739 Perl_sv_catpvf(aTHX_ listsv,
6742 /* Any multicharacter foldings
6743 * require the following transform:
6744 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
6745 * where E folds into "pq" and F folds
6746 * into "rst", all other characters
6747 * fold to single characters. We save
6748 * away these multicharacter foldings,
6749 * to be later saved as part of the
6750 * additional "s" data. */
6753 if (!unicode_alternate)
6754 unicode_alternate = newAV();
6755 sv = newSVpvn((char*)foldbuf, foldlen);
6757 av_push(unicode_alternate, sv);
6761 /* If folding and the value is one of the Greek
6762 * sigmas insert a few more sigmas to make the
6763 * folding rules of the sigmas to work right.
6764 * Note that not all the possible combinations
6765 * are handled here: some of them are handled
6766 * by the standard folding rules, and some of
6767 * them (literal or EXACTF cases) are handled
6768 * during runtime in regexec.c:S_find_byclass(). */
6769 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
6770 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
6771 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
6772 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
6773 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
6775 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
6776 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
6777 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
6782 literal_endpoint = 0;
6786 range = 0; /* this range (if it was one) is done now */
6790 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
6792 RExC_size += ANYOF_CLASS_ADD_SKIP;
6794 RExC_emit += ANYOF_CLASS_ADD_SKIP;
6800 /****** !SIZE_ONLY AFTER HERE *********/
6802 if( stored == 1 && value < 256
6803 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
6805 /* optimize single char class to an EXACT node
6806 but *only* when its not a UTF/high char */
6807 const char * cur_parse= RExC_parse;
6808 RExC_emit = (regnode *)orig_emit;
6809 RExC_parse = (char *)orig_parse;
6810 ret = reg_node(pRExC_state,
6811 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
6812 RExC_parse = (char *)cur_parse;
6813 *STRING(ret)= (char)value;
6815 RExC_emit += STR_SZ(1);
6818 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
6819 if ( /* If the only flag is folding (plus possibly inversion). */
6820 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
6822 for (value = 0; value < 256; ++value) {
6823 if (ANYOF_BITMAP_TEST(ret, value)) {
6824 UV fold = PL_fold[value];
6827 ANYOF_BITMAP_SET(ret, fold);
6830 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
6833 /* optimize inverted simple patterns (e.g. [^a-z]) */
6834 if (optimize_invert &&
6835 /* If the only flag is inversion. */
6836 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
6837 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
6838 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
6839 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
6842 AV * const av = newAV();
6844 /* The 0th element stores the character class description
6845 * in its textual form: used later (regexec.c:Perl_regclass_swash())
6846 * to initialize the appropriate swash (which gets stored in
6847 * the 1st element), and also useful for dumping the regnode.
6848 * The 2nd element stores the multicharacter foldings,
6849 * used later (regexec.c:S_reginclass()). */
6850 av_store(av, 0, listsv);
6851 av_store(av, 1, NULL);
6852 av_store(av, 2, (SV*)unicode_alternate);
6853 rv = newRV_noinc((SV*)av);
6854 n = add_data(pRExC_state, 1, "s");
6855 RExC_rx->data->data[n] = (void*)rv;
6862 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
6864 char* const retval = RExC_parse++;
6867 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
6868 RExC_parse[2] == '#') {
6869 while (*RExC_parse != ')') {
6870 if (RExC_parse == RExC_end)
6871 FAIL("Sequence (?#... not terminated");
6877 if (RExC_flags & PMf_EXTENDED) {
6878 if (isSPACE(*RExC_parse)) {
6882 else if (*RExC_parse == '#') {
6883 while (RExC_parse < RExC_end)
6884 if (*RExC_parse++ == '\n') break;
6893 - reg_node - emit a node
6895 STATIC regnode * /* Location. */
6896 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
6899 register regnode *ptr;
6900 regnode * const ret = RExC_emit;
6901 GET_RE_DEBUG_FLAGS_DECL;
6904 SIZE_ALIGN(RExC_size);
6908 NODE_ALIGN_FILL(ret);
6910 FILL_ADVANCE_NODE(ptr, op);
6911 if (RExC_offsets) { /* MJD */
6912 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
6913 "reg_node", __LINE__,
6915 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
6916 ? "Overwriting end of array!\n" : "OK",
6917 (UV)(RExC_emit - RExC_emit_start),
6918 (UV)(RExC_parse - RExC_start),
6919 (UV)RExC_offsets[0]));
6920 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
6929 - reganode - emit a node with an argument
6931 STATIC regnode * /* Location. */
6932 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
6935 register regnode *ptr;
6936 regnode * const ret = RExC_emit;
6937 GET_RE_DEBUG_FLAGS_DECL;
6940 SIZE_ALIGN(RExC_size);
6945 NODE_ALIGN_FILL(ret);
6947 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
6948 if (RExC_offsets) { /* MJD */
6949 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
6953 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
6954 "Overwriting end of array!\n" : "OK",
6955 (UV)(RExC_emit - RExC_emit_start),
6956 (UV)(RExC_parse - RExC_start),
6957 (UV)RExC_offsets[0]));
6958 Set_Cur_Node_Offset;
6967 - reguni - emit (if appropriate) a Unicode character
6970 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
6973 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
6977 - reginsert - insert an operator in front of already-emitted operand
6979 * Means relocating the operand.
6982 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd)
6985 register regnode *src;
6986 register regnode *dst;
6987 register regnode *place;
6988 const int offset = regarglen[(U8)op];
6989 GET_RE_DEBUG_FLAGS_DECL;
6990 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
6993 RExC_size += NODE_STEP_REGNODE + offset;
6998 RExC_emit += NODE_STEP_REGNODE + offset;
7000 while (src > opnd) {
7001 StructCopy(--src, --dst, regnode);
7002 if (RExC_offsets) { /* MJD 20010112 */
7003 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
7007 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
7008 ? "Overwriting end of array!\n" : "OK",
7009 (UV)(src - RExC_emit_start),
7010 (UV)(dst - RExC_emit_start),
7011 (UV)RExC_offsets[0]));
7012 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
7013 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
7018 place = opnd; /* Op node, where operand used to be. */
7019 if (RExC_offsets) { /* MJD */
7020 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7024 (UV)(place - RExC_emit_start) > RExC_offsets[0]
7025 ? "Overwriting end of array!\n" : "OK",
7026 (UV)(place - RExC_emit_start),
7027 (UV)(RExC_parse - RExC_start),
7028 (UV)RExC_offsets[0]));
7029 Set_Node_Offset(place, RExC_parse);
7030 Set_Node_Length(place, 1);
7032 src = NEXTOPER(place);
7033 FILL_ADVANCE_NODE(place, op);
7034 Zero(src, offset, regnode);
7038 - regtail - set the next-pointer at the end of a node chain of p to val.
7039 - SEE ALSO: regtail_study
7041 /* TODO: All three parms should be const */
7043 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7046 register regnode *scan;
7047 GET_RE_DEBUG_FLAGS_DECL;
7049 PERL_UNUSED_ARG(depth);
7055 /* Find last node. */
7058 regnode * const temp = regnext(scan);
7060 SV * const mysv=sv_newmortal();
7061 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
7062 regprop(RExC_rx, mysv, scan);
7063 PerlIO_printf(Perl_debug_log, "~ %s (%d)\n",
7064 SvPV_nolen_const(mysv), REG_NODE_NUM(scan));
7071 if (reg_off_by_arg[OP(scan)]) {
7072 ARG_SET(scan, val - scan);
7075 NEXT_OFF(scan) = val - scan;
7081 - regtail_study - set the next-pointer at the end of a node chain of p to val.
7082 - Look for optimizable sequences at the same time.
7083 - currently only looks for EXACT chains.
7085 This is expermental code. The idea is to use this routine to perform
7086 in place optimizations on branches and groups as they are constructed,
7087 with the long term intention of removing optimization from study_chunk so
7088 that it is purely analytical.
7090 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
7091 to control which is which.
7094 /* TODO: All four parms should be const */
7097 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7100 register regnode *scan;
7102 #ifdef EXPERIMENTAL_INPLACESCAN
7106 GET_RE_DEBUG_FLAGS_DECL;
7112 /* Find last node. */
7116 regnode * const temp = regnext(scan);
7117 #ifdef EXPERIMENTAL_INPLACESCAN
7118 if (PL_regkind[OP(scan)] == EXACT)
7119 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
7127 if( exact == PSEUDO )
7129 else if ( exact != OP(scan) )
7138 SV * const mysv=sv_newmortal();
7139 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
7140 regprop(RExC_rx, mysv, scan);
7141 PerlIO_printf(Perl_debug_log, "~ %s (%s) (%d)\n",
7142 SvPV_nolen_const(mysv),
7144 REG_NODE_NUM(scan));
7151 SV * const mysv_val=sv_newmortal();
7152 DEBUG_PARSE_MSG("");
7153 regprop(RExC_rx, mysv_val, val);
7154 PerlIO_printf(Perl_debug_log, "~ attach to %s (%d) offset to %d\n",
7155 SvPV_nolen_const(mysv_val),
7160 if (reg_off_by_arg[OP(scan)]) {
7161 ARG_SET(scan, val - scan);
7164 NEXT_OFF(scan) = val - scan;
7172 - regcurly - a little FSA that accepts {\d+,?\d*}
7175 S_regcurly(register const char *s)
7194 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
7197 Perl_regdump(pTHX_ const regexp *r)
7201 SV * const sv = sv_newmortal();
7202 SV *dsv= sv_newmortal();
7204 (void)dumpuntil(r, r->program, r->program + 1, NULL, NULL, sv, 0, 0);
7206 /* Header fields of interest. */
7207 if (r->anchored_substr) {
7208 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
7209 RE_SV_DUMPLEN(r->anchored_substr), 30);
7210 PerlIO_printf(Perl_debug_log,
7211 "anchored %s%s at %"IVdf" ",
7212 s, RE_SV_TAIL(r->anchored_substr),
7213 (IV)r->anchored_offset);
7214 } else if (r->anchored_utf8) {
7215 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
7216 RE_SV_DUMPLEN(r->anchored_utf8), 30);
7217 PerlIO_printf(Perl_debug_log,
7218 "anchored utf8 %s%s at %"IVdf" ",
7219 s, RE_SV_TAIL(r->anchored_utf8),
7220 (IV)r->anchored_offset);
7222 if (r->float_substr) {
7223 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
7224 RE_SV_DUMPLEN(r->float_substr), 30);
7225 PerlIO_printf(Perl_debug_log,
7226 "floating %s%s at %"IVdf"..%"UVuf" ",
7227 s, RE_SV_TAIL(r->float_substr),
7228 (IV)r->float_min_offset, (UV)r->float_max_offset);
7229 } else if (r->float_utf8) {
7230 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
7231 RE_SV_DUMPLEN(r->float_utf8), 30);
7232 PerlIO_printf(Perl_debug_log,
7233 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
7234 s, RE_SV_TAIL(r->float_utf8),
7235 (IV)r->float_min_offset, (UV)r->float_max_offset);
7237 if (r->check_substr || r->check_utf8)
7238 PerlIO_printf(Perl_debug_log,
7240 (r->check_substr == r->float_substr
7241 && r->check_utf8 == r->float_utf8
7242 ? "(checking floating" : "(checking anchored"));
7243 if (r->reganch & ROPT_NOSCAN)
7244 PerlIO_printf(Perl_debug_log, " noscan");
7245 if (r->reganch & ROPT_CHECK_ALL)
7246 PerlIO_printf(Perl_debug_log, " isall");
7247 if (r->check_substr || r->check_utf8)
7248 PerlIO_printf(Perl_debug_log, ") ");
7250 if (r->regstclass) {
7251 regprop(r, sv, r->regstclass);
7252 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
7254 if (r->reganch & ROPT_ANCH) {
7255 PerlIO_printf(Perl_debug_log, "anchored");
7256 if (r->reganch & ROPT_ANCH_BOL)
7257 PerlIO_printf(Perl_debug_log, "(BOL)");
7258 if (r->reganch & ROPT_ANCH_MBOL)
7259 PerlIO_printf(Perl_debug_log, "(MBOL)");
7260 if (r->reganch & ROPT_ANCH_SBOL)
7261 PerlIO_printf(Perl_debug_log, "(SBOL)");
7262 if (r->reganch & ROPT_ANCH_GPOS)
7263 PerlIO_printf(Perl_debug_log, "(GPOS)");
7264 PerlIO_putc(Perl_debug_log, ' ');
7266 if (r->reganch & ROPT_GPOS_SEEN)
7267 PerlIO_printf(Perl_debug_log, "GPOS ");
7268 if (r->reganch & ROPT_SKIP)
7269 PerlIO_printf(Perl_debug_log, "plus ");
7270 if (r->reganch & ROPT_IMPLICIT)
7271 PerlIO_printf(Perl_debug_log, "implicit ");
7272 PerlIO_printf(Perl_debug_log, "minlen %ld ", (long) r->minlen);
7273 if (r->reganch & ROPT_EVAL_SEEN)
7274 PerlIO_printf(Perl_debug_log, "with eval ");
7275 PerlIO_printf(Perl_debug_log, "\n");
7277 PERL_UNUSED_CONTEXT;
7279 #endif /* DEBUGGING */
7283 - regprop - printable representation of opcode
7286 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
7291 GET_RE_DEBUG_FLAGS_DECL;
7293 sv_setpvn(sv, "", 0);
7294 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
7295 /* It would be nice to FAIL() here, but this may be called from
7296 regexec.c, and it would be hard to supply pRExC_state. */
7297 Perl_croak(aTHX_ "Corrupted regexp opcode");
7298 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
7300 k = PL_regkind[OP(o)];
7303 SV * const dsv = sv_2mortal(newSVpvs(""));
7304 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
7305 * is a crude hack but it may be the best for now since
7306 * we have no flag "this EXACTish node was UTF-8"
7308 const char * const s =
7309 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
7310 PL_colors[0], PL_colors[1],
7311 PERL_PV_ESCAPE_UNI_DETECT |
7312 PERL_PV_PRETTY_ELIPSES |
7315 Perl_sv_catpvf(aTHX_ sv, " %s", s );
7316 } else if (k == TRIE) {
7317 /* print the details of the trie in dumpuntil instead, as
7318 * prog->data isn't available here */
7319 const char op = OP(o);
7320 const I32 n = ARG(o);
7321 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
7322 (reg_ac_data *)prog->data->data[n] :
7324 const reg_trie_data * const trie = !IS_TRIE_AC(op) ?
7325 (reg_trie_data*)prog->data->data[n] :
7328 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
7329 DEBUG_TRIE_COMPILE_r(
7330 Perl_sv_catpvf(aTHX_ sv,
7331 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
7332 (UV)trie->startstate,
7333 (IV)trie->laststate-1,
7334 (UV)trie->wordcount,
7337 (UV)TRIE_CHARCOUNT(trie),
7338 (UV)trie->uniquecharcount
7341 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
7343 int rangestart = -1;
7344 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
7345 Perl_sv_catpvf(aTHX_ sv, "[");
7346 for (i = 0; i <= 256; i++) {
7347 if (i < 256 && BITMAP_TEST(bitmap,i)) {
7348 if (rangestart == -1)
7350 } else if (rangestart != -1) {
7351 if (i <= rangestart + 3)
7352 for (; rangestart < i; rangestart++)
7353 put_byte(sv, rangestart);
7355 put_byte(sv, rangestart);
7357 put_byte(sv, i - 1);
7362 Perl_sv_catpvf(aTHX_ sv, "]");
7365 } else if (k == CURLY) {
7366 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
7367 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
7368 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
7370 else if (k == WHILEM && o->flags) /* Ordinal/of */
7371 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
7372 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP )
7373 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
7374 else if (k == LOGICAL)
7375 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
7376 else if (k == ANYOF) {
7377 int i, rangestart = -1;
7378 const U8 flags = ANYOF_FLAGS(o);
7380 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
7381 static const char * const anyofs[] = {
7414 if (flags & ANYOF_LOCALE)
7415 sv_catpvs(sv, "{loc}");
7416 if (flags & ANYOF_FOLD)
7417 sv_catpvs(sv, "{i}");
7418 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
7419 if (flags & ANYOF_INVERT)
7421 for (i = 0; i <= 256; i++) {
7422 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
7423 if (rangestart == -1)
7425 } else if (rangestart != -1) {
7426 if (i <= rangestart + 3)
7427 for (; rangestart < i; rangestart++)
7428 put_byte(sv, rangestart);
7430 put_byte(sv, rangestart);
7432 put_byte(sv, i - 1);
7438 if (o->flags & ANYOF_CLASS)
7439 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
7440 if (ANYOF_CLASS_TEST(o,i))
7441 sv_catpv(sv, anyofs[i]);
7443 if (flags & ANYOF_UNICODE)
7444 sv_catpvs(sv, "{unicode}");
7445 else if (flags & ANYOF_UNICODE_ALL)
7446 sv_catpvs(sv, "{unicode_all}");
7450 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
7454 U8 s[UTF8_MAXBYTES_CASE+1];
7456 for (i = 0; i <= 256; i++) { /* just the first 256 */
7457 uvchr_to_utf8(s, i);
7459 if (i < 256 && swash_fetch(sw, s, TRUE)) {
7460 if (rangestart == -1)
7462 } else if (rangestart != -1) {
7463 if (i <= rangestart + 3)
7464 for (; rangestart < i; rangestart++) {
7465 const U8 * const e = uvchr_to_utf8(s,rangestart);
7467 for(p = s; p < e; p++)
7471 const U8 *e = uvchr_to_utf8(s,rangestart);
7473 for (p = s; p < e; p++)
7476 e = uvchr_to_utf8(s, i-1);
7477 for (p = s; p < e; p++)
7484 sv_catpvs(sv, "..."); /* et cetera */
7488 char *s = savesvpv(lv);
7489 char * const origs = s;
7491 while (*s && *s != '\n')
7495 const char * const t = ++s;
7513 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
7515 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
7516 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
7518 PERL_UNUSED_CONTEXT;
7519 PERL_UNUSED_ARG(sv);
7521 PERL_UNUSED_ARG(prog);
7522 #endif /* DEBUGGING */
7526 Perl_re_intuit_string(pTHX_ regexp *prog)
7527 { /* Assume that RE_INTUIT is set */
7529 GET_RE_DEBUG_FLAGS_DECL;
7530 PERL_UNUSED_CONTEXT;
7534 const char * const s = SvPV_nolen_const(prog->check_substr
7535 ? prog->check_substr : prog->check_utf8);
7537 if (!PL_colorset) reginitcolors();
7538 PerlIO_printf(Perl_debug_log,
7539 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
7541 prog->check_substr ? "" : "utf8 ",
7542 PL_colors[5],PL_colors[0],
7545 (strlen(s) > 60 ? "..." : ""));
7548 return prog->check_substr ? prog->check_substr : prog->check_utf8;
7552 pregfree - free a regexp
7554 See regdupe below if you change anything here.
7558 Perl_pregfree(pTHX_ struct regexp *r)
7562 GET_RE_DEBUG_FLAGS_DECL;
7564 if (!r || (--r->refcnt > 0))
7570 SV *dsv= sv_newmortal();
7571 RE_PV_QUOTED_DECL(s, (r->reganch & ROPT_UTF8),
7572 dsv, r->precomp, r->prelen, 60);
7573 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
7574 PL_colors[4],PL_colors[5],s);
7578 /* gcov results gave these as non-null 100% of the time, so there's no
7579 optimisation in checking them before calling Safefree */
7580 Safefree(r->precomp);
7581 Safefree(r->offsets); /* 20010421 MJD */
7582 RX_MATCH_COPY_FREE(r);
7583 #ifdef PERL_OLD_COPY_ON_WRITE
7585 SvREFCNT_dec(r->saved_copy);
7588 if (r->anchored_substr)
7589 SvREFCNT_dec(r->anchored_substr);
7590 if (r->anchored_utf8)
7591 SvREFCNT_dec(r->anchored_utf8);
7592 if (r->float_substr)
7593 SvREFCNT_dec(r->float_substr);
7595 SvREFCNT_dec(r->float_utf8);
7596 Safefree(r->substrs);
7599 int n = r->data->count;
7600 PAD* new_comppad = NULL;
7605 /* If you add a ->what type here, update the comment in regcomp.h */
7606 switch (r->data->what[n]) {
7608 SvREFCNT_dec((SV*)r->data->data[n]);
7611 Safefree(r->data->data[n]);
7614 new_comppad = (AV*)r->data->data[n];
7617 if (new_comppad == NULL)
7618 Perl_croak(aTHX_ "panic: pregfree comppad");
7619 PAD_SAVE_LOCAL(old_comppad,
7620 /* Watch out for global destruction's random ordering. */
7621 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
7624 refcnt = OpREFCNT_dec((OP_4tree*)r->data->data[n]);
7627 op_free((OP_4tree*)r->data->data[n]);
7629 PAD_RESTORE_LOCAL(old_comppad);
7630 SvREFCNT_dec((SV*)new_comppad);
7636 { /* Aho Corasick add-on structure for a trie node.
7637 Used in stclass optimization only */
7639 reg_ac_data *aho=(reg_ac_data*)r->data->data[n];
7641 refcount = --aho->refcount;
7644 Safefree(aho->states);
7645 Safefree(aho->fail);
7646 aho->trie=NULL; /* not necessary to free this as it is
7647 handled by the 't' case */
7648 Safefree(r->data->data[n]); /* do this last!!!! */
7649 Safefree(r->regstclass);
7655 /* trie structure. */
7657 reg_trie_data *trie=(reg_trie_data*)r->data->data[n];
7659 refcount = --trie->refcount;
7662 Safefree(trie->charmap);
7663 if (trie->widecharmap)
7664 SvREFCNT_dec((SV*)trie->widecharmap);
7665 Safefree(trie->states);
7666 Safefree(trie->trans);
7668 Safefree(trie->bitmap);
7670 Safefree(trie->wordlen);
7672 Safefree(trie->jump);
7674 Safefree(trie->nextword);
7678 SvREFCNT_dec((SV*)trie->words);
7679 if (trie->revcharmap)
7680 SvREFCNT_dec((SV*)trie->revcharmap);
7683 Safefree(r->data->data[n]); /* do this last!!!! */
7688 Perl_croak(aTHX_ "panic: regfree data code '%c'", r->data->what[n]);
7691 Safefree(r->data->what);
7694 Safefree(r->startp);
7699 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
7700 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
7701 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
7704 regdupe - duplicate a regexp.
7706 This routine is called by sv.c's re_dup and is expected to clone a
7707 given regexp structure. It is a no-op when not under USE_ITHREADS.
7708 (Originally this *was* re_dup() for change history see sv.c)
7710 See pregfree() above if you change anything here.
7712 #if defined(USE_ITHREADS)
7714 Perl_regdupe(pTHX_ const regexp *r, CLONE_PARAMS *param)
7719 struct reg_substr_datum *s;
7722 return (REGEXP *)NULL;
7724 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
7727 len = r->offsets[0];
7728 npar = r->nparens+1;
7730 Newxc(ret, sizeof(regexp) + (len+1)*sizeof(regnode), char, regexp);
7731 Copy(r->program, ret->program, len+1, regnode);
7733 Newx(ret->startp, npar, I32);
7734 Copy(r->startp, ret->startp, npar, I32);
7735 Newx(ret->endp, npar, I32);
7736 Copy(r->startp, ret->startp, npar, I32);
7738 Newx(ret->substrs, 1, struct reg_substr_data);
7739 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
7740 s->min_offset = r->substrs->data[i].min_offset;
7741 s->max_offset = r->substrs->data[i].max_offset;
7742 s->end_shift = r->substrs->data[i].end_shift;
7743 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
7744 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
7747 ret->regstclass = NULL;
7750 const int count = r->data->count;
7753 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
7754 char, struct reg_data);
7755 Newx(d->what, count, U8);
7758 for (i = 0; i < count; i++) {
7759 d->what[i] = r->data->what[i];
7760 switch (d->what[i]) {
7761 /* legal options are one of: sfpont
7762 see also regcomp.h and pregfree() */
7764 d->data[i] = sv_dup_inc((SV *)r->data->data[i], param);
7767 d->data[i] = av_dup_inc((AV *)r->data->data[i], param);
7770 /* This is cheating. */
7771 Newx(d->data[i], 1, struct regnode_charclass_class);
7772 StructCopy(r->data->data[i], d->data[i],
7773 struct regnode_charclass_class);
7774 ret->regstclass = (regnode*)d->data[i];
7777 /* Compiled op trees are readonly, and can thus be
7778 shared without duplication. */
7780 d->data[i] = (void*)OpREFCNT_inc((OP*)r->data->data[i]);
7784 d->data[i] = r->data->data[i];
7787 d->data[i] = r->data->data[i];
7789 ((reg_trie_data*)d->data[i])->refcount++;
7793 d->data[i] = r->data->data[i];
7795 ((reg_ac_data*)d->data[i])->refcount++;
7797 /* Trie stclasses are readonly and can thus be shared
7798 * without duplication. We free the stclass in pregfree
7799 * when the corresponding reg_ac_data struct is freed.
7801 ret->regstclass= r->regstclass;
7804 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", r->data->what[i]);
7813 Newx(ret->offsets, 2*len+1, U32);
7814 Copy(r->offsets, ret->offsets, 2*len+1, U32);
7816 ret->precomp = SAVEPVN(r->precomp, r->prelen);
7817 ret->refcnt = r->refcnt;
7818 ret->minlen = r->minlen;
7819 ret->prelen = r->prelen;
7820 ret->nparens = r->nparens;
7821 ret->lastparen = r->lastparen;
7822 ret->lastcloseparen = r->lastcloseparen;
7823 ret->reganch = r->reganch;
7825 ret->sublen = r->sublen;
7827 ret->engine = r->engine;
7829 if (RX_MATCH_COPIED(ret))
7830 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
7833 #ifdef PERL_OLD_COPY_ON_WRITE
7834 ret->saved_copy = NULL;
7837 ptr_table_store(PL_ptr_table, r, ret);
7842 #ifndef PERL_IN_XSUB_RE
7844 - regnext - dig the "next" pointer out of a node
7847 Perl_regnext(pTHX_ register regnode *p)
7850 register I32 offset;
7852 if (p == &PL_regdummy)
7855 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
7864 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
7867 STRLEN l1 = strlen(pat1);
7868 STRLEN l2 = strlen(pat2);
7871 const char *message;
7877 Copy(pat1, buf, l1 , char);
7878 Copy(pat2, buf + l1, l2 , char);
7879 buf[l1 + l2] = '\n';
7880 buf[l1 + l2 + 1] = '\0';
7882 /* ANSI variant takes additional second argument */
7883 va_start(args, pat2);
7887 msv = vmess(buf, &args);
7889 message = SvPV_const(msv,l1);
7892 Copy(message, buf, l1 , char);
7893 buf[l1-1] = '\0'; /* Overwrite \n */
7894 Perl_croak(aTHX_ "%s", buf);
7897 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
7899 #ifndef PERL_IN_XSUB_RE
7901 Perl_save_re_context(pTHX)
7905 struct re_save_state *state;
7907 SAVEVPTR(PL_curcop);
7908 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
7910 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
7911 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
7912 SSPUSHINT(SAVEt_RE_STATE);
7914 Copy(&PL_reg_state, state, 1, struct re_save_state);
7916 PL_reg_start_tmp = 0;
7917 PL_reg_start_tmpl = 0;
7918 PL_reg_oldsaved = NULL;
7919 PL_reg_oldsavedlen = 0;
7921 PL_reg_leftiter = 0;
7922 PL_reg_poscache = NULL;
7923 PL_reg_poscache_size = 0;
7924 #ifdef PERL_OLD_COPY_ON_WRITE
7928 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
7930 const REGEXP * const rx = PM_GETRE(PL_curpm);
7933 for (i = 1; i <= rx->nparens; i++) {
7934 char digits[TYPE_CHARS(long)];
7935 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
7936 GV *const *const gvp
7937 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
7940 GV * const gv = *gvp;
7941 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
7951 clear_re(pTHX_ void *r)
7954 ReREFCNT_dec((regexp *)r);
7960 S_put_byte(pTHX_ SV *sv, int c)
7962 if (isCNTRL(c) || c == 255 || !isPRINT(c))
7963 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
7964 else if (c == '-' || c == ']' || c == '\\' || c == '^')
7965 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
7967 Perl_sv_catpvf(aTHX_ sv, "%c", c);
7971 #define CLEAR_OPTSTART \
7972 if (optstart) STMT_START { \
7973 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%d nodes)\n", node - optstart)); \
7977 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
7979 STATIC const regnode *
7980 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
7981 const regnode *last, const regnode *plast,
7982 SV* sv, I32 indent, U32 depth)
7985 register U8 op = PSEUDO; /* Arbitrary non-END op. */
7986 register const regnode *next;
7987 const regnode *optstart= NULL;
7988 GET_RE_DEBUG_FLAGS_DECL;
7990 #ifdef DEBUG_DUMPUNTIL
7991 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
7992 last ? last-start : 0,plast ? plast-start : 0);
7995 if (plast && plast < last)
7998 while (PL_regkind[op] != END && (!last || node < last)) {
7999 /* While that wasn't END last time... */
8005 next = regnext((regnode *)node);
8008 if (OP(node) == OPTIMIZED) {
8009 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
8016 regprop(r, sv, node);
8017 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
8018 (int)(2*indent + 1), "", SvPVX_const(sv));
8020 if (OP(node) != OPTIMIZED) {
8021 if (next == NULL) /* Next ptr. */
8022 PerlIO_printf(Perl_debug_log, "(0)");
8023 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
8024 PerlIO_printf(Perl_debug_log, "(FAIL)");
8026 PerlIO_printf(Perl_debug_log, "(%"IVdf")", (IV)(next - start));
8028 /*if (PL_regkind[(U8)op] != TRIE)*/
8029 (void)PerlIO_putc(Perl_debug_log, '\n');
8033 if (PL_regkind[(U8)op] == BRANCHJ) {
8036 register const regnode *nnode = (OP(next) == LONGJMP
8037 ? regnext((regnode *)next)
8039 if (last && nnode > last)
8041 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
8044 else if (PL_regkind[(U8)op] == BRANCH) {
8046 DUMPUNTIL(NEXTOPER(node), next);
8048 else if ( PL_regkind[(U8)op] == TRIE ) {
8049 const char op = OP(node);
8050 const I32 n = ARG(node);
8051 const reg_ac_data * const ac = op>=AHOCORASICK ?
8052 (reg_ac_data *)r->data->data[n] :
8054 const reg_trie_data * const trie = op<AHOCORASICK ?
8055 (reg_trie_data*)r->data->data[n] :
8057 const regnode *nextbranch= NULL;
8059 sv_setpvn(sv, "", 0);
8060 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
8061 SV ** const elem_ptr = av_fetch(trie->words,word_idx,0);
8063 PerlIO_printf(Perl_debug_log, "%*s%s ",
8064 (int)(2*(indent+3)), "",
8065 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
8066 PL_colors[0], PL_colors[1],
8067 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
8068 PERL_PV_PRETTY_ELIPSES |
8074 U16 dist= trie->jump[word_idx+1];
8075 PerlIO_printf(Perl_debug_log, "(%u)\n",(next - dist) - start);
8078 nextbranch= next - trie->jump[0];
8079 DUMPUNTIL(next - dist, nextbranch);
8081 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
8082 nextbranch= regnext((regnode *)nextbranch);
8084 PerlIO_printf(Perl_debug_log, "\n");
8087 if (last && next > last)
8092 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
8093 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
8094 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
8096 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
8098 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
8100 else if ( op == PLUS || op == STAR) {
8101 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
8103 else if (op == ANYOF) {
8104 /* arglen 1 + class block */
8105 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
8106 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
8107 node = NEXTOPER(node);
8109 else if (PL_regkind[(U8)op] == EXACT) {
8110 /* Literal string, where present. */
8111 node += NODE_SZ_STR(node) - 1;
8112 node = NEXTOPER(node);
8115 node = NEXTOPER(node);
8116 node += regarglen[(U8)op];
8118 if (op == CURLYX || op == OPEN)
8120 else if (op == WHILEM)
8124 #ifdef DEBUG_DUMPUNTIL
8125 PerlIO_printf(Perl_debug_log, "--- %d\n",indent);
8130 #endif /* DEBUGGING */
8134 * c-indentation-style: bsd
8136 * indent-tabs-mode: t
8139 * ex: set ts=8 sts=4 sw=4 noet: