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. */
120 regnode **parens; /* offsets of each paren */
122 HV *charnames; /* cache of named sequences */
123 HV *paren_names; /* Paren names */
125 char *starttry; /* -Dr: where regtry was called. */
126 #define RExC_starttry (pRExC_state->starttry)
129 const char *lastparse;
131 #define RExC_lastparse (pRExC_state->lastparse)
132 #define RExC_lastnum (pRExC_state->lastnum)
136 #define RExC_flags (pRExC_state->flags)
137 #define RExC_precomp (pRExC_state->precomp)
138 #define RExC_rx (pRExC_state->rx)
139 #define RExC_start (pRExC_state->start)
140 #define RExC_end (pRExC_state->end)
141 #define RExC_parse (pRExC_state->parse)
142 #define RExC_whilem_seen (pRExC_state->whilem_seen)
143 #define RExC_offsets (pRExC_state->rx->offsets) /* I am not like the others */
144 #define RExC_emit (pRExC_state->emit)
145 #define RExC_emit_start (pRExC_state->emit_start)
146 #define RExC_naughty (pRExC_state->naughty)
147 #define RExC_sawback (pRExC_state->sawback)
148 #define RExC_seen (pRExC_state->seen)
149 #define RExC_size (pRExC_state->size)
150 #define RExC_npar (pRExC_state->npar)
151 #define RExC_extralen (pRExC_state->extralen)
152 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
153 #define RExC_seen_evals (pRExC_state->seen_evals)
154 #define RExC_utf8 (pRExC_state->utf8)
155 #define RExC_charnames (pRExC_state->charnames)
156 #define RExC_parens (pRExC_state->parens)
157 #define RExC_paren_names (pRExC_state->paren_names)
159 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
160 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
161 ((*s) == '{' && regcurly(s)))
164 #undef SPSTART /* dratted cpp namespace... */
167 * Flags to be passed up and down.
169 #define WORST 0 /* Worst case. */
170 #define HASWIDTH 0x1 /* Known to match non-null strings. */
171 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
172 #define SPSTART 0x4 /* Starts with * or +. */
173 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
175 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
177 /* whether trie related optimizations are enabled */
178 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
179 #define TRIE_STUDY_OPT
180 #define FULL_TRIE_STUDY
185 /* About scan_data_t.
187 During optimisation we recurse through the regexp program performing
188 various inplace (keyhole style) optimisations. In addition study_chunk
189 and scan_commit populate this data structure with information about
190 what strings MUST appear in the pattern. We look for the longest
191 string that must appear for at a fixed location, and we look for the
192 longest string that may appear at a floating location. So for instance
197 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
198 strings (because they follow a .* construct). study_chunk will identify
199 both FOO and BAR as being the longest fixed and floating strings respectively.
201 The strings can be composites, for instance
205 will result in a composite fixed substring 'foo'.
207 For each string some basic information is maintained:
209 - offset or min_offset
210 This is the position the string must appear at, or not before.
211 It also implicitly (when combined with minlenp) tells us how many
212 character must match before the string we are searching.
213 Likewise when combined with minlenp and the length of the string
214 tells us how many characters must appear after the string we have
218 Only used for floating strings. This is the rightmost point that
219 the string can appear at. Ifset to I32 max it indicates that the
220 string can occur infinitely far to the right.
223 A pointer to the minimum length of the pattern that the string
224 was found inside. This is important as in the case of positive
225 lookahead or positive lookbehind we can have multiple patterns
230 The minimum length of the pattern overall is 3, the minimum length
231 of the lookahead part is 3, but the minimum length of the part that
232 will actually match is 1. So 'FOO's minimum length is 3, but the
233 minimum length for the F is 1. This is important as the minimum length
234 is used to determine offsets in front of and behind the string being
235 looked for. Since strings can be composites this is the length of the
236 pattern at the time it was commited with a scan_commit. Note that
237 the length is calculated by study_chunk, so that the minimum lengths
238 are not known until the full pattern has been compiled, thus the
239 pointer to the value.
243 In the case of lookbehind the string being searched for can be
244 offset past the start point of the final matching string.
245 If this value was just blithely removed from the min_offset it would
246 invalidate some of the calculations for how many chars must match
247 before or after (as they are derived from min_offset and minlen and
248 the length of the string being searched for).
249 When the final pattern is compiled and the data is moved from the
250 scan_data_t structure into the regexp structure the information
251 about lookbehind is factored in, with the information that would
252 have been lost precalculated in the end_shift field for the
255 The fields pos_min and pos_delta are used to store the minimum offset
256 and the delta to the maximum offset at the current point in the pattern.
260 typedef struct scan_data_t {
261 /*I32 len_min; unused */
262 /*I32 len_delta; unused */
266 I32 last_end; /* min value, <0 unless valid. */
269 SV **longest; /* Either &l_fixed, or &l_float. */
270 SV *longest_fixed; /* longest fixed string found in pattern */
271 I32 offset_fixed; /* offset where it starts */
272 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
273 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
274 SV *longest_float; /* longest floating string found in pattern */
275 I32 offset_float_min; /* earliest point in string it can appear */
276 I32 offset_float_max; /* latest point in string it can appear */
277 I32 *minlen_float; /* pointer to the minlen relevent to the string */
278 I32 lookbehind_float; /* is the position of the string modified by LB */
282 struct regnode_charclass_class *start_class;
286 * Forward declarations for pregcomp()'s friends.
289 static const scan_data_t zero_scan_data =
290 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
292 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
293 #define SF_BEFORE_SEOL 0x0001
294 #define SF_BEFORE_MEOL 0x0002
295 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
296 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
299 # define SF_FIX_SHIFT_EOL (0+2)
300 # define SF_FL_SHIFT_EOL (0+4)
302 # define SF_FIX_SHIFT_EOL (+2)
303 # define SF_FL_SHIFT_EOL (+4)
306 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
307 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
309 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
310 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
311 #define SF_IS_INF 0x0040
312 #define SF_HAS_PAR 0x0080
313 #define SF_IN_PAR 0x0100
314 #define SF_HAS_EVAL 0x0200
315 #define SCF_DO_SUBSTR 0x0400
316 #define SCF_DO_STCLASS_AND 0x0800
317 #define SCF_DO_STCLASS_OR 0x1000
318 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
319 #define SCF_WHILEM_VISITED_POS 0x2000
321 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
324 #define UTF (RExC_utf8 != 0)
325 #define LOC ((RExC_flags & PMf_LOCALE) != 0)
326 #define FOLD ((RExC_flags & PMf_FOLD) != 0)
328 #define OOB_UNICODE 12345678
329 #define OOB_NAMEDCLASS -1
331 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
332 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
335 /* length of regex to show in messages that don't mark a position within */
336 #define RegexLengthToShowInErrorMessages 127
339 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
340 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
341 * op/pragma/warn/regcomp.
343 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
344 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
346 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
349 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
350 * arg. Show regex, up to a maximum length. If it's too long, chop and add
353 #define FAIL(msg) STMT_START { \
354 const char *ellipses = ""; \
355 IV len = RExC_end - RExC_precomp; \
358 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
359 if (len > RegexLengthToShowInErrorMessages) { \
360 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
361 len = RegexLengthToShowInErrorMessages - 10; \
364 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
365 msg, (int)len, RExC_precomp, ellipses); \
369 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
371 #define Simple_vFAIL(m) STMT_START { \
372 const IV offset = RExC_parse - RExC_precomp; \
373 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
374 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
378 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
380 #define vFAIL(m) STMT_START { \
382 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
387 * Like Simple_vFAIL(), but accepts two arguments.
389 #define Simple_vFAIL2(m,a1) STMT_START { \
390 const IV offset = RExC_parse - RExC_precomp; \
391 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
392 (int)offset, RExC_precomp, RExC_precomp + offset); \
396 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
398 #define vFAIL2(m,a1) STMT_START { \
400 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
401 Simple_vFAIL2(m, a1); \
406 * Like Simple_vFAIL(), but accepts three arguments.
408 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
409 const IV offset = RExC_parse - RExC_precomp; \
410 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
411 (int)offset, RExC_precomp, RExC_precomp + offset); \
415 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
417 #define vFAIL3(m,a1,a2) STMT_START { \
419 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
420 Simple_vFAIL3(m, a1, a2); \
424 * Like Simple_vFAIL(), but accepts four arguments.
426 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
427 const IV offset = RExC_parse - RExC_precomp; \
428 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
429 (int)offset, RExC_precomp, RExC_precomp + offset); \
432 #define vWARN(loc,m) STMT_START { \
433 const IV offset = loc - RExC_precomp; \
434 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
435 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
438 #define vWARNdep(loc,m) STMT_START { \
439 const IV offset = loc - RExC_precomp; \
440 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
441 "%s" REPORT_LOCATION, \
442 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
446 #define vWARN2(loc, m, a1) STMT_START { \
447 const IV offset = loc - RExC_precomp; \
448 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
449 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
452 #define vWARN3(loc, m, a1, a2) STMT_START { \
453 const IV offset = loc - RExC_precomp; \
454 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
455 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
458 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
459 const IV offset = loc - RExC_precomp; \
460 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
461 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
464 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
465 const IV offset = loc - RExC_precomp; \
466 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
467 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
471 /* Allow for side effects in s */
472 #define REGC(c,s) STMT_START { \
473 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
476 /* Macros for recording node offsets. 20001227 mjd@plover.com
477 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
478 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
479 * Element 0 holds the number n.
480 * Position is 1 indexed.
483 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
485 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
486 __LINE__, (node), (int)(byte))); \
488 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
490 RExC_offsets[2*(node)-1] = (byte); \
495 #define Set_Node_Offset(node,byte) \
496 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
497 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
499 #define Set_Node_Length_To_R(node,len) STMT_START { \
501 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
502 __LINE__, (int)(node), (int)(len))); \
504 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
506 RExC_offsets[2*(node)] = (len); \
511 #define Set_Node_Length(node,len) \
512 Set_Node_Length_To_R((node)-RExC_emit_start, len)
513 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
514 #define Set_Node_Cur_Length(node) \
515 Set_Node_Length(node, RExC_parse - parse_start)
517 /* Get offsets and lengths */
518 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
519 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
521 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
522 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
523 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
527 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
528 #define EXPERIMENTAL_INPLACESCAN
531 #define DEBUG_STUDYDATA(data,depth) \
532 DEBUG_OPTIMISE_MORE_r(if(data){ \
533 PerlIO_printf(Perl_debug_log, \
534 "%*s"/* Len:%"IVdf"/%"IVdf" */" Pos:%"IVdf"/%"IVdf \
535 " Flags: %"IVdf" Whilem_c: %"IVdf" Lcp: %"IVdf" ", \
536 (int)(depth)*2, "", \
537 (IV)((data)->pos_min), \
538 (IV)((data)->pos_delta), \
539 (IV)((data)->flags), \
540 (IV)((data)->whilem_c), \
541 (IV)((data)->last_closep ? *((data)->last_closep) : -1) \
543 if ((data)->last_found) \
544 PerlIO_printf(Perl_debug_log, \
545 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
546 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
547 SvPVX_const((data)->last_found), \
548 (IV)((data)->last_end), \
549 (IV)((data)->last_start_min), \
550 (IV)((data)->last_start_max), \
551 ((data)->longest && \
552 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
553 SvPVX_const((data)->longest_fixed), \
554 (IV)((data)->offset_fixed), \
555 ((data)->longest && \
556 (data)->longest==&((data)->longest_float)) ? "*" : "", \
557 SvPVX_const((data)->longest_float), \
558 (IV)((data)->offset_float_min), \
559 (IV)((data)->offset_float_max) \
561 PerlIO_printf(Perl_debug_log,"\n"); \
564 static void clear_re(pTHX_ void *r);
566 /* Mark that we cannot extend a found fixed substring at this point.
567 Update the longest found anchored substring and the longest found
568 floating substrings if needed. */
571 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp)
573 const STRLEN l = CHR_SVLEN(data->last_found);
574 const STRLEN old_l = CHR_SVLEN(*data->longest);
575 GET_RE_DEBUG_FLAGS_DECL;
577 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
578 SvSetMagicSV(*data->longest, data->last_found);
579 if (*data->longest == data->longest_fixed) {
580 data->offset_fixed = l ? data->last_start_min : data->pos_min;
581 if (data->flags & SF_BEFORE_EOL)
583 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
585 data->flags &= ~SF_FIX_BEFORE_EOL;
586 data->minlen_fixed=minlenp;
587 data->lookbehind_fixed=0;
590 data->offset_float_min = l ? data->last_start_min : data->pos_min;
591 data->offset_float_max = (l
592 ? data->last_start_max
593 : data->pos_min + data->pos_delta);
594 if ((U32)data->offset_float_max > (U32)I32_MAX)
595 data->offset_float_max = I32_MAX;
596 if (data->flags & SF_BEFORE_EOL)
598 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
600 data->flags &= ~SF_FL_BEFORE_EOL;
601 data->minlen_float=minlenp;
602 data->lookbehind_float=0;
605 SvCUR_set(data->last_found, 0);
607 SV * const sv = data->last_found;
608 if (SvUTF8(sv) && SvMAGICAL(sv)) {
609 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
615 data->flags &= ~SF_BEFORE_EOL;
616 DEBUG_STUDYDATA(data,0);
619 /* Can match anything (initialization) */
621 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
623 ANYOF_CLASS_ZERO(cl);
624 ANYOF_BITMAP_SETALL(cl);
625 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
627 cl->flags |= ANYOF_LOCALE;
630 /* Can match anything (initialization) */
632 S_cl_is_anything(const struct regnode_charclass_class *cl)
636 for (value = 0; value <= ANYOF_MAX; value += 2)
637 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
639 if (!(cl->flags & ANYOF_UNICODE_ALL))
641 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
646 /* Can match anything (initialization) */
648 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
650 Zero(cl, 1, struct regnode_charclass_class);
652 cl_anything(pRExC_state, cl);
656 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
658 Zero(cl, 1, struct regnode_charclass_class);
660 cl_anything(pRExC_state, cl);
662 cl->flags |= ANYOF_LOCALE;
665 /* 'And' a given class with another one. Can create false positives */
666 /* We assume that cl is not inverted */
668 S_cl_and(struct regnode_charclass_class *cl,
669 const struct regnode_charclass_class *and_with)
671 if (!(and_with->flags & ANYOF_CLASS)
672 && !(cl->flags & ANYOF_CLASS)
673 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
674 && !(and_with->flags & ANYOF_FOLD)
675 && !(cl->flags & ANYOF_FOLD)) {
678 if (and_with->flags & ANYOF_INVERT)
679 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
680 cl->bitmap[i] &= ~and_with->bitmap[i];
682 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
683 cl->bitmap[i] &= and_with->bitmap[i];
684 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
685 if (!(and_with->flags & ANYOF_EOS))
686 cl->flags &= ~ANYOF_EOS;
688 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
689 !(and_with->flags & ANYOF_INVERT)) {
690 cl->flags &= ~ANYOF_UNICODE_ALL;
691 cl->flags |= ANYOF_UNICODE;
692 ARG_SET(cl, ARG(and_with));
694 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
695 !(and_with->flags & ANYOF_INVERT))
696 cl->flags &= ~ANYOF_UNICODE_ALL;
697 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
698 !(and_with->flags & ANYOF_INVERT))
699 cl->flags &= ~ANYOF_UNICODE;
702 /* 'OR' a given class with another one. Can create false positives */
703 /* We assume that cl is not inverted */
705 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
707 if (or_with->flags & ANYOF_INVERT) {
709 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
710 * <= (B1 | !B2) | (CL1 | !CL2)
711 * which is wasteful if CL2 is small, but we ignore CL2:
712 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
713 * XXXX Can we handle case-fold? Unclear:
714 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
715 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
717 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
718 && !(or_with->flags & ANYOF_FOLD)
719 && !(cl->flags & ANYOF_FOLD) ) {
722 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
723 cl->bitmap[i] |= ~or_with->bitmap[i];
724 } /* XXXX: logic is complicated otherwise */
726 cl_anything(pRExC_state, cl);
729 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
730 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
731 && (!(or_with->flags & ANYOF_FOLD)
732 || (cl->flags & ANYOF_FOLD)) ) {
735 /* OR char bitmap and class bitmap separately */
736 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
737 cl->bitmap[i] |= or_with->bitmap[i];
738 if (or_with->flags & ANYOF_CLASS) {
739 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
740 cl->classflags[i] |= or_with->classflags[i];
741 cl->flags |= ANYOF_CLASS;
744 else { /* XXXX: logic is complicated, leave it along for a moment. */
745 cl_anything(pRExC_state, cl);
748 if (or_with->flags & ANYOF_EOS)
749 cl->flags |= ANYOF_EOS;
751 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
752 ARG(cl) != ARG(or_with)) {
753 cl->flags |= ANYOF_UNICODE_ALL;
754 cl->flags &= ~ANYOF_UNICODE;
756 if (or_with->flags & ANYOF_UNICODE_ALL) {
757 cl->flags |= ANYOF_UNICODE_ALL;
758 cl->flags &= ~ANYOF_UNICODE;
762 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
763 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
764 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
765 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
771 dump_trie_interim_list(trie,next_alloc)
772 dump_trie_interim_table(trie,next_alloc)
774 These routines dump out a trie in a somewhat readable format.
775 The _interim_ variants are used for debugging the interim
776 tables that are used to generate the final compressed
777 representation which is what dump_trie expects.
779 Part of the reason for their existance is to provide a form
780 of documentation as to how the different representations function.
786 Dumps the final compressed table form of the trie to Perl_debug_log.
787 Used for debugging make_trie().
791 S_dump_trie(pTHX_ const struct _reg_trie_data *trie,U32 depth)
794 SV *sv=sv_newmortal();
795 int colwidth= trie->widecharmap ? 6 : 4;
796 GET_RE_DEBUG_FLAGS_DECL;
799 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
800 (int)depth * 2 + 2,"",
801 "Match","Base","Ofs" );
803 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
804 SV ** const tmp = av_fetch( trie->revcharmap, state, 0);
806 PerlIO_printf( Perl_debug_log, "%*s",
808 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
809 PL_colors[0], PL_colors[1],
810 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
811 PERL_PV_ESCAPE_FIRSTCHAR
816 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
817 (int)depth * 2 + 2,"");
819 for( state = 0 ; state < trie->uniquecharcount ; state++ )
820 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
821 PerlIO_printf( Perl_debug_log, "\n");
823 for( state = 1 ; state < trie->laststate ; state++ ) {
824 const U32 base = trie->states[ state ].trans.base;
826 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
828 if ( trie->states[ state ].wordnum ) {
829 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
831 PerlIO_printf( Perl_debug_log, "%6s", "" );
834 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
839 while( ( base + ofs < trie->uniquecharcount ) ||
840 ( base + ofs - trie->uniquecharcount < trie->lasttrans
841 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
844 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
846 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
847 if ( ( base + ofs >= trie->uniquecharcount ) &&
848 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
849 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
851 PerlIO_printf( Perl_debug_log, "%*"UVXf,
853 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
855 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
859 PerlIO_printf( Perl_debug_log, "]");
862 PerlIO_printf( Perl_debug_log, "\n" );
866 dump_trie_interim_list(trie,next_alloc)
867 Dumps a fully constructed but uncompressed trie in list form.
868 List tries normally only are used for construction when the number of
869 possible chars (trie->uniquecharcount) is very high.
870 Used for debugging make_trie().
873 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie, U32 next_alloc,U32 depth)
876 SV *sv=sv_newmortal();
877 int colwidth= trie->widecharmap ? 6 : 4;
878 GET_RE_DEBUG_FLAGS_DECL;
879 /* print out the table precompression. */
880 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
881 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
882 "------:-----+-----------------\n" );
884 for( state=1 ; state < next_alloc ; state ++ ) {
887 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
888 (int)depth * 2 + 2,"", (UV)state );
889 if ( ! trie->states[ state ].wordnum ) {
890 PerlIO_printf( Perl_debug_log, "%5s| ","");
892 PerlIO_printf( Perl_debug_log, "W%4x| ",
893 trie->states[ state ].wordnum
896 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
897 SV ** const tmp = av_fetch( trie->revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
899 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
901 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
902 PL_colors[0], PL_colors[1],
903 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
904 PERL_PV_ESCAPE_FIRSTCHAR
906 TRIE_LIST_ITEM(state,charid).forid,
907 (UV)TRIE_LIST_ITEM(state,charid).newstate
911 PerlIO_printf( Perl_debug_log, "\n");
916 dump_trie_interim_table(trie,next_alloc)
917 Dumps a fully constructed but uncompressed trie in table form.
918 This is the normal DFA style state transition table, with a few
919 twists to facilitate compression later.
920 Used for debugging make_trie().
923 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie, U32 next_alloc, U32 depth)
927 SV *sv=sv_newmortal();
928 int colwidth= trie->widecharmap ? 6 : 4;
929 GET_RE_DEBUG_FLAGS_DECL;
932 print out the table precompression so that we can do a visual check
933 that they are identical.
936 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
938 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
939 SV ** const tmp = av_fetch( trie->revcharmap, charid, 0);
941 PerlIO_printf( Perl_debug_log, "%*s",
943 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
944 PL_colors[0], PL_colors[1],
945 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
946 PERL_PV_ESCAPE_FIRSTCHAR
952 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
954 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
955 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
958 PerlIO_printf( Perl_debug_log, "\n" );
960 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
962 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
963 (int)depth * 2 + 2,"",
964 (UV)TRIE_NODENUM( state ) );
966 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
967 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
969 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
971 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
973 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
974 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
976 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
977 trie->states[ TRIE_NODENUM( state ) ].wordnum );
984 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
985 startbranch: the first branch in the whole branch sequence
986 first : start branch of sequence of branch-exact nodes.
987 May be the same as startbranch
988 last : Thing following the last branch.
989 May be the same as tail.
990 tail : item following the branch sequence
991 count : words in the sequence
992 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
995 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
997 A trie is an N'ary tree where the branches are determined by digital
998 decomposition of the key. IE, at the root node you look up the 1st character and
999 follow that branch repeat until you find the end of the branches. Nodes can be
1000 marked as "accepting" meaning they represent a complete word. Eg:
1004 would convert into the following structure. Numbers represent states, letters
1005 following numbers represent valid transitions on the letter from that state, if
1006 the number is in square brackets it represents an accepting state, otherwise it
1007 will be in parenthesis.
1009 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1013 (1) +-i->(6)-+-s->[7]
1015 +-s->(3)-+-h->(4)-+-e->[5]
1017 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1019 This shows that when matching against the string 'hers' we will begin at state 1
1020 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1021 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1022 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1023 single traverse. We store a mapping from accepting to state to which word was
1024 matched, and then when we have multiple possibilities we try to complete the
1025 rest of the regex in the order in which they occured in the alternation.
1027 The only prior NFA like behaviour that would be changed by the TRIE support is
1028 the silent ignoring of duplicate alternations which are of the form:
1030 / (DUPE|DUPE) X? (?{ ... }) Y /x
1032 Thus EVAL blocks follwing a trie may be called a different number of times with
1033 and without the optimisation. With the optimisations dupes will be silently
1034 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1035 the following demonstrates:
1037 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1039 which prints out 'word' three times, but
1041 'words'=~/(word|word|word)(?{ print $1 })S/
1043 which doesnt print it out at all. This is due to other optimisations kicking in.
1045 Example of what happens on a structural level:
1047 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1049 1: CURLYM[1] {1,32767}(18)
1060 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1061 and should turn into:
1063 1: CURLYM[1] {1,32767}(18)
1065 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1073 Cases where tail != last would be like /(?foo|bar)baz/:
1083 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1084 and would end up looking like:
1087 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1094 d = uvuni_to_utf8_flags(d, uv, 0);
1096 is the recommended Unicode-aware way of saying
1101 #define TRIE_STORE_REVCHAR \
1103 SV *tmp = Perl_newSVpvf_nocontext( "%c", (int)uvc ); \
1104 if (UTF) SvUTF8_on(tmp); \
1105 av_push( TRIE_REVCHARMAP(trie), tmp ); \
1108 #define TRIE_READ_CHAR STMT_START { \
1112 if ( foldlen > 0 ) { \
1113 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1118 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1119 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1120 foldlen -= UNISKIP( uvc ); \
1121 scan = foldbuf + UNISKIP( uvc ); \
1124 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1134 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1135 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1136 TRIE_LIST_LEN( state ) *= 2; \
1137 Renew( trie->states[ state ].trans.list, \
1138 TRIE_LIST_LEN( state ), reg_trie_trans_le ); \
1140 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1141 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1142 TRIE_LIST_CUR( state )++; \
1145 #define TRIE_LIST_NEW(state) STMT_START { \
1146 Newxz( trie->states[ state ].trans.list, \
1147 4, reg_trie_trans_le ); \
1148 TRIE_LIST_CUR( state ) = 1; \
1149 TRIE_LIST_LEN( state ) = 4; \
1152 #define TRIE_HANDLE_WORD(state) STMT_START { \
1153 U16 dupe= trie->states[ state ].wordnum; \
1154 regnode * const noper_next = regnext( noper ); \
1156 if (trie->wordlen) \
1157 trie->wordlen[ curword ] = wordlen; \
1159 /* store the word for dumping */ \
1161 if (OP(noper) != NOTHING) \
1162 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1164 tmp = newSVpvn( "", 0 ); \
1165 if ( UTF ) SvUTF8_on( tmp ); \
1166 av_push( trie->words, tmp ); \
1171 if ( noper_next < tail ) { \
1173 Newxz( trie->jump, word_count + 1, U16); \
1174 trie->jump[curword] = (U16)(tail - noper_next); \
1176 jumper = noper_next; \
1178 nextbranch= regnext(cur); \
1182 /* So it's a dupe. This means we need to maintain a */\
1183 /* linked-list from the first to the next. */\
1184 /* we only allocate the nextword buffer when there */\
1185 /* a dupe, so first time we have to do the allocation */\
1186 if (!trie->nextword) \
1187 Newxz( trie->nextword, word_count + 1, U16); \
1188 while ( trie->nextword[dupe] ) \
1189 dupe= trie->nextword[dupe]; \
1190 trie->nextword[dupe]= curword; \
1192 /* we haven't inserted this word yet. */ \
1193 trie->states[ state ].wordnum = curword; \
1198 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1199 ( ( base + charid >= ucharcount \
1200 && base + charid < ubound \
1201 && state == trie->trans[ base - ucharcount + charid ].check \
1202 && trie->trans[ base - ucharcount + charid ].next ) \
1203 ? trie->trans[ base - ucharcount + charid ].next \
1204 : ( state==1 ? special : 0 ) \
1208 #define MADE_JUMP_TRIE 2
1209 #define MADE_EXACT_TRIE 4
1212 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1215 /* first pass, loop through and scan words */
1216 reg_trie_data *trie;
1218 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1223 regnode *jumper = NULL;
1224 regnode *nextbranch = NULL;
1225 /* we just use folder as a flag in utf8 */
1226 const U8 * const folder = ( flags == EXACTF
1228 : ( flags == EXACTFL
1234 const U32 data_slot = add_data( pRExC_state, 1, "t" );
1235 SV *re_trie_maxbuff;
1237 /* these are only used during construction but are useful during
1238 * debugging so we store them in the struct when debugging.
1240 STRLEN trie_charcount=0;
1241 AV *trie_revcharmap;
1243 GET_RE_DEBUG_FLAGS_DECL;
1245 PERL_UNUSED_ARG(depth);
1248 Newxz( trie, 1, reg_trie_data );
1250 trie->startstate = 1;
1251 trie->wordcount = word_count;
1252 RExC_rx->data->data[ data_slot ] = (void*)trie;
1253 Newxz( trie->charmap, 256, U16 );
1254 if (!(UTF && folder))
1255 Newxz( trie->bitmap, ANYOF_BITMAP_SIZE, char );
1257 trie->words = newAV();
1259 TRIE_REVCHARMAP(trie) = newAV();
1261 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1262 if (!SvIOK(re_trie_maxbuff)) {
1263 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1266 PerlIO_printf( Perl_debug_log,
1267 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1268 (int)depth * 2 + 2, "",
1269 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1270 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1273 /* -- First loop and Setup --
1275 We first traverse the branches and scan each word to determine if it
1276 contains widechars, and how many unique chars there are, this is
1277 important as we have to build a table with at least as many columns as we
1280 We use an array of integers to represent the character codes 0..255
1281 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1282 native representation of the character value as the key and IV's for the
1285 *TODO* If we keep track of how many times each character is used we can
1286 remap the columns so that the table compression later on is more
1287 efficient in terms of memory by ensuring most common value is in the
1288 middle and the least common are on the outside. IMO this would be better
1289 than a most to least common mapping as theres a decent chance the most
1290 common letter will share a node with the least common, meaning the node
1291 will not be compressable. With a middle is most common approach the worst
1292 case is when we have the least common nodes twice.
1296 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1297 regnode * const noper = NEXTOPER( cur );
1298 const U8 *uc = (U8*)STRING( noper );
1299 const U8 * const e = uc + STR_LEN( noper );
1301 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1302 const U8 *scan = (U8*)NULL;
1303 U32 wordlen = 0; /* required init */
1306 if (OP(noper) == NOTHING) {
1311 TRIE_BITMAP_SET(trie,*uc);
1312 if ( folder ) TRIE_BITMAP_SET(trie,folder[ *uc ]);
1314 for ( ; uc < e ; uc += len ) {
1315 TRIE_CHARCOUNT(trie)++;
1319 if ( !trie->charmap[ uvc ] ) {
1320 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1322 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1327 if ( !trie->widecharmap )
1328 trie->widecharmap = newHV();
1330 svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1333 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1335 if ( !SvTRUE( *svpp ) ) {
1336 sv_setiv( *svpp, ++trie->uniquecharcount );
1341 if( cur == first ) {
1344 } else if (chars < trie->minlen) {
1346 } else if (chars > trie->maxlen) {
1350 } /* end first pass */
1351 DEBUG_TRIE_COMPILE_r(
1352 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1353 (int)depth * 2 + 2,"",
1354 ( trie->widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1355 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1356 (int)trie->minlen, (int)trie->maxlen )
1358 Newxz( trie->wordlen, word_count, U32 );
1361 We now know what we are dealing with in terms of unique chars and
1362 string sizes so we can calculate how much memory a naive
1363 representation using a flat table will take. If it's over a reasonable
1364 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1365 conservative but potentially much slower representation using an array
1368 At the end we convert both representations into the same compressed
1369 form that will be used in regexec.c for matching with. The latter
1370 is a form that cannot be used to construct with but has memory
1371 properties similar to the list form and access properties similar
1372 to the table form making it both suitable for fast searches and
1373 small enough that its feasable to store for the duration of a program.
1375 See the comment in the code where the compressed table is produced
1376 inplace from the flat tabe representation for an explanation of how
1377 the compression works.
1382 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1384 Second Pass -- Array Of Lists Representation
1386 Each state will be represented by a list of charid:state records
1387 (reg_trie_trans_le) the first such element holds the CUR and LEN
1388 points of the allocated array. (See defines above).
1390 We build the initial structure using the lists, and then convert
1391 it into the compressed table form which allows faster lookups
1392 (but cant be modified once converted).
1395 STRLEN transcount = 1;
1397 Newxz( trie->states, TRIE_CHARCOUNT(trie) + 2, reg_trie_state );
1401 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1403 regnode * const noper = NEXTOPER( cur );
1404 U8 *uc = (U8*)STRING( noper );
1405 const U8 * const e = uc + STR_LEN( noper );
1406 U32 state = 1; /* required init */
1407 U16 charid = 0; /* sanity init */
1408 U8 *scan = (U8*)NULL; /* sanity init */
1409 STRLEN foldlen = 0; /* required init */
1410 U32 wordlen = 0; /* required init */
1411 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1413 if (OP(noper) != NOTHING) {
1414 for ( ; uc < e ; uc += len ) {
1419 charid = trie->charmap[ uvc ];
1421 SV** const svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0);
1425 charid=(U16)SvIV( *svpp );
1428 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1435 if ( !trie->states[ state ].trans.list ) {
1436 TRIE_LIST_NEW( state );
1438 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1439 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1440 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1445 newstate = next_alloc++;
1446 TRIE_LIST_PUSH( state, charid, newstate );
1451 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1455 TRIE_HANDLE_WORD(state);
1457 } /* end second pass */
1459 trie->laststate = next_alloc;
1460 Renew( trie->states, next_alloc, reg_trie_state );
1462 /* and now dump it out before we compress it */
1463 DEBUG_TRIE_COMPILE_MORE_r(
1464 dump_trie_interim_list(trie,next_alloc,depth+1)
1467 Newxz( trie->trans, transcount ,reg_trie_trans );
1474 for( state=1 ; state < next_alloc ; state ++ ) {
1478 DEBUG_TRIE_COMPILE_MORE_r(
1479 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1483 if (trie->states[state].trans.list) {
1484 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1488 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1489 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1490 if ( forid < minid ) {
1492 } else if ( forid > maxid ) {
1496 if ( transcount < tp + maxid - minid + 1) {
1498 Renew( trie->trans, transcount, reg_trie_trans );
1499 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1501 base = trie->uniquecharcount + tp - minid;
1502 if ( maxid == minid ) {
1504 for ( ; zp < tp ; zp++ ) {
1505 if ( ! trie->trans[ zp ].next ) {
1506 base = trie->uniquecharcount + zp - minid;
1507 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1508 trie->trans[ zp ].check = state;
1514 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1515 trie->trans[ tp ].check = state;
1520 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1521 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1522 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1523 trie->trans[ tid ].check = state;
1525 tp += ( maxid - minid + 1 );
1527 Safefree(trie->states[ state ].trans.list);
1530 DEBUG_TRIE_COMPILE_MORE_r(
1531 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1534 trie->states[ state ].trans.base=base;
1536 trie->lasttrans = tp + 1;
1540 Second Pass -- Flat Table Representation.
1542 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1543 We know that we will need Charcount+1 trans at most to store the data
1544 (one row per char at worst case) So we preallocate both structures
1545 assuming worst case.
1547 We then construct the trie using only the .next slots of the entry
1550 We use the .check field of the first entry of the node temporarily to
1551 make compression both faster and easier by keeping track of how many non
1552 zero fields are in the node.
1554 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1557 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1558 number representing the first entry of the node, and state as a
1559 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1560 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1561 are 2 entrys per node. eg:
1569 The table is internally in the right hand, idx form. However as we also
1570 have to deal with the states array which is indexed by nodenum we have to
1571 use TRIE_NODENUM() to convert.
1576 Newxz( trie->trans, ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1,
1578 Newxz( trie->states, TRIE_CHARCOUNT(trie) + 2, reg_trie_state );
1579 next_alloc = trie->uniquecharcount + 1;
1582 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1584 regnode * const noper = NEXTOPER( cur );
1585 const U8 *uc = (U8*)STRING( noper );
1586 const U8 * const e = uc + STR_LEN( noper );
1588 U32 state = 1; /* required init */
1590 U16 charid = 0; /* sanity init */
1591 U32 accept_state = 0; /* sanity init */
1592 U8 *scan = (U8*)NULL; /* sanity init */
1594 STRLEN foldlen = 0; /* required init */
1595 U32 wordlen = 0; /* required init */
1596 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1598 if ( OP(noper) != NOTHING ) {
1599 for ( ; uc < e ; uc += len ) {
1604 charid = trie->charmap[ uvc ];
1606 SV* const * const svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0);
1607 charid = svpp ? (U16)SvIV(*svpp) : 0;
1611 if ( !trie->trans[ state + charid ].next ) {
1612 trie->trans[ state + charid ].next = next_alloc;
1613 trie->trans[ state ].check++;
1614 next_alloc += trie->uniquecharcount;
1616 state = trie->trans[ state + charid ].next;
1618 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1620 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1623 accept_state = TRIE_NODENUM( state );
1624 TRIE_HANDLE_WORD(accept_state);
1626 } /* end second pass */
1628 /* and now dump it out before we compress it */
1629 DEBUG_TRIE_COMPILE_MORE_r(
1630 dump_trie_interim_table(trie,next_alloc,depth+1)
1635 * Inplace compress the table.*
1637 For sparse data sets the table constructed by the trie algorithm will
1638 be mostly 0/FAIL transitions or to put it another way mostly empty.
1639 (Note that leaf nodes will not contain any transitions.)
1641 This algorithm compresses the tables by eliminating most such
1642 transitions, at the cost of a modest bit of extra work during lookup:
1644 - Each states[] entry contains a .base field which indicates the
1645 index in the state[] array wheres its transition data is stored.
1647 - If .base is 0 there are no valid transitions from that node.
1649 - If .base is nonzero then charid is added to it to find an entry in
1652 -If trans[states[state].base+charid].check!=state then the
1653 transition is taken to be a 0/Fail transition. Thus if there are fail
1654 transitions at the front of the node then the .base offset will point
1655 somewhere inside the previous nodes data (or maybe even into a node
1656 even earlier), but the .check field determines if the transition is
1660 The following process inplace converts the table to the compressed
1661 table: We first do not compress the root node 1,and mark its all its
1662 .check pointers as 1 and set its .base pointer as 1 as well. This
1663 allows to do a DFA construction from the compressed table later, and
1664 ensures that any .base pointers we calculate later are greater than
1667 - We set 'pos' to indicate the first entry of the second node.
1669 - We then iterate over the columns of the node, finding the first and
1670 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1671 and set the .check pointers accordingly, and advance pos
1672 appropriately and repreat for the next node. Note that when we copy
1673 the next pointers we have to convert them from the original
1674 NODEIDX form to NODENUM form as the former is not valid post
1677 - If a node has no transitions used we mark its base as 0 and do not
1678 advance the pos pointer.
1680 - If a node only has one transition we use a second pointer into the
1681 structure to fill in allocated fail transitions from other states.
1682 This pointer is independent of the main pointer and scans forward
1683 looking for null transitions that are allocated to a state. When it
1684 finds one it writes the single transition into the "hole". If the
1685 pointer doesnt find one the single transition is appended as normal.
1687 - Once compressed we can Renew/realloc the structures to release the
1690 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1691 specifically Fig 3.47 and the associated pseudocode.
1695 const U32 laststate = TRIE_NODENUM( next_alloc );
1698 trie->laststate = laststate;
1700 for ( state = 1 ; state < laststate ; state++ ) {
1702 const U32 stateidx = TRIE_NODEIDX( state );
1703 const U32 o_used = trie->trans[ stateidx ].check;
1704 U32 used = trie->trans[ stateidx ].check;
1705 trie->trans[ stateidx ].check = 0;
1707 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1708 if ( flag || trie->trans[ stateidx + charid ].next ) {
1709 if ( trie->trans[ stateidx + charid ].next ) {
1711 for ( ; zp < pos ; zp++ ) {
1712 if ( ! trie->trans[ zp ].next ) {
1716 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1717 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1718 trie->trans[ zp ].check = state;
1719 if ( ++zp > pos ) pos = zp;
1726 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1728 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1729 trie->trans[ pos ].check = state;
1734 trie->lasttrans = pos + 1;
1735 Renew( trie->states, laststate + 1, reg_trie_state);
1736 DEBUG_TRIE_COMPILE_MORE_r(
1737 PerlIO_printf( Perl_debug_log,
1738 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1739 (int)depth * 2 + 2,"",
1740 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1743 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1746 } /* end table compress */
1748 /* resize the trans array to remove unused space */
1749 Renew( trie->trans, trie->lasttrans, reg_trie_trans);
1751 /* and now dump out the compressed format */
1752 DEBUG_TRIE_COMPILE_r(
1753 dump_trie(trie,depth+1)
1756 { /* Modify the program and insert the new TRIE node*/
1758 U8 nodetype =(U8)(flags & 0xFF);
1764 U32 mjd_nodelen = 0;
1767 This means we convert either the first branch or the first Exact,
1768 depending on whether the thing following (in 'last') is a branch
1769 or not and whther first is the startbranch (ie is it a sub part of
1770 the alternation or is it the whole thing.)
1771 Assuming its a sub part we conver the EXACT otherwise we convert
1772 the whole branch sequence, including the first.
1774 /* Find the node we are going to overwrite */
1775 if ( first == startbranch && OP( last ) != BRANCH ) {
1776 /* whole branch chain */
1779 const regnode *nop = NEXTOPER( convert );
1780 mjd_offset= Node_Offset((nop));
1781 mjd_nodelen= Node_Length((nop));
1784 /* branch sub-chain */
1785 convert = NEXTOPER( first );
1786 NEXT_OFF( first ) = (U16)(last - first);
1788 mjd_offset= Node_Offset((convert));
1789 mjd_nodelen= Node_Length((convert));
1793 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1794 (int)depth * 2 + 2, "",
1795 (UV)mjd_offset, (UV)mjd_nodelen)
1798 /* But first we check to see if there is a common prefix we can
1799 split out as an EXACT and put in front of the TRIE node. */
1800 trie->startstate= 1;
1801 if ( trie->bitmap && !trie->widecharmap && !trie->jump ) {
1804 PerlIO_printf(Perl_debug_log, "%*sLaststate:%"UVuf"\n",
1805 (int)depth * 2 + 2, "",
1806 (UV)trie->laststate)
1808 for ( state = 1 ; state < trie->laststate-1 ; state++ ) {
1812 const U32 base = trie->states[ state ].trans.base;
1814 if ( trie->states[state].wordnum )
1817 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1818 if ( ( base + ofs >= trie->uniquecharcount ) &&
1819 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1820 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1822 if ( ++count > 1 ) {
1823 SV **tmp = av_fetch( TRIE_REVCHARMAP(trie), ofs, 0);
1824 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1825 if ( state == 1 ) break;
1827 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1829 PerlIO_printf(Perl_debug_log,
1830 "%*sNew Start State=%"UVuf" Class: [",
1831 (int)depth * 2 + 2, "",
1834 SV ** const tmp = av_fetch( TRIE_REVCHARMAP(trie), idx, 0);
1835 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1837 TRIE_BITMAP_SET(trie,*ch);
1839 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1841 PerlIO_printf(Perl_debug_log, (char*)ch)
1845 TRIE_BITMAP_SET(trie,*ch);
1847 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1848 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1854 SV **tmp = av_fetch( TRIE_REVCHARMAP(trie), idx, 0);
1855 const char *ch = SvPV_nolen_const( *tmp );
1857 PerlIO_printf( Perl_debug_log,
1858 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1859 (int)depth * 2 + 2, "",
1860 (UV)state, (UV)idx, ch)
1863 OP( convert ) = nodetype;
1864 str=STRING(convert);
1873 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
1879 regnode *n = convert+NODE_SZ_STR(convert);
1880 NEXT_OFF(convert) = NODE_SZ_STR(convert);
1881 trie->startstate = state;
1882 trie->minlen -= (state - 1);
1883 trie->maxlen -= (state - 1);
1885 regnode *fix = convert;
1887 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
1888 while( ++fix < n ) {
1889 Set_Node_Offset_Length(fix, 0, 0);
1895 NEXT_OFF(convert) = (U16)(tail - convert);
1896 DEBUG_r(optimize= n);
1902 if ( trie->maxlen ) {
1903 NEXT_OFF( convert ) = (U16)(tail - convert);
1904 ARG_SET( convert, data_slot );
1905 /* Store the offset to the first unabsorbed branch in
1906 jump[0], which is otherwise unused by the jump logic.
1907 We use this when dumping a trie and during optimisation. */
1909 trie->jump[0] = (U16)(tail - nextbranch);
1912 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
1913 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
1915 OP( convert ) = TRIEC;
1916 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
1917 Safefree(trie->bitmap);
1920 OP( convert ) = TRIE;
1922 /* store the type in the flags */
1923 convert->flags = nodetype;
1927 + regarglen[ OP( convert ) ];
1929 /* XXX We really should free up the resource in trie now,
1930 as we won't use them - (which resources?) dmq */
1932 /* needed for dumping*/
1934 regnode *opt = convert;
1935 while (++opt<optimize) {
1936 Set_Node_Offset_Length(opt,0,0);
1939 Try to clean up some of the debris left after the
1942 while( optimize < jumper ) {
1943 mjd_nodelen += Node_Length((optimize));
1944 OP( optimize ) = OPTIMIZED;
1945 Set_Node_Offset_Length(optimize,0,0);
1948 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
1950 } /* end node insert */
1952 SvREFCNT_dec(TRIE_REVCHARMAP(trie));
1956 : trie->startstate>1
1962 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
1964 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
1966 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
1967 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
1970 We find the fail state for each state in the trie, this state is the longest proper
1971 suffix of the current states 'word' that is also a proper prefix of another word in our
1972 trie. State 1 represents the word '' and is the thus the default fail state. This allows
1973 the DFA not to have to restart after its tried and failed a word at a given point, it
1974 simply continues as though it had been matching the other word in the first place.
1976 'abcdgu'=~/abcdefg|cdgu/
1977 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
1978 fail, which would bring use to the state representing 'd' in the second word where we would
1979 try 'g' and succeed, prodceding to match 'cdgu'.
1981 /* add a fail transition */
1982 reg_trie_data *trie=(reg_trie_data *)RExC_rx->data->data[ARG(source)];
1984 const U32 ucharcount = trie->uniquecharcount;
1985 const U32 numstates = trie->laststate;
1986 const U32 ubound = trie->lasttrans + ucharcount;
1990 U32 base = trie->states[ 1 ].trans.base;
1993 const U32 data_slot = add_data( pRExC_state, 1, "T" );
1994 GET_RE_DEBUG_FLAGS_DECL;
1996 PERL_UNUSED_ARG(depth);
2000 ARG_SET( stclass, data_slot );
2001 Newxz( aho, 1, reg_ac_data );
2002 RExC_rx->data->data[ data_slot ] = (void*)aho;
2004 aho->states=(reg_trie_state *)savepvn((const char*)trie->states,
2005 (trie->laststate+1)*sizeof(reg_trie_state));
2006 Newxz( q, numstates, U32);
2007 Newxz( aho->fail, numstates, U32 );
2010 /* initialize fail[0..1] to be 1 so that we always have
2011 a valid final fail state */
2012 fail[ 0 ] = fail[ 1 ] = 1;
2014 for ( charid = 0; charid < ucharcount ; charid++ ) {
2015 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2017 q[ q_write ] = newstate;
2018 /* set to point at the root */
2019 fail[ q[ q_write++ ] ]=1;
2022 while ( q_read < q_write) {
2023 const U32 cur = q[ q_read++ % numstates ];
2024 base = trie->states[ cur ].trans.base;
2026 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2027 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2029 U32 fail_state = cur;
2032 fail_state = fail[ fail_state ];
2033 fail_base = aho->states[ fail_state ].trans.base;
2034 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2036 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2037 fail[ ch_state ] = fail_state;
2038 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2040 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2042 q[ q_write++ % numstates] = ch_state;
2046 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2047 when we fail in state 1, this allows us to use the
2048 charclass scan to find a valid start char. This is based on the principle
2049 that theres a good chance the string being searched contains lots of stuff
2050 that cant be a start char.
2052 fail[ 0 ] = fail[ 1 ] = 0;
2053 DEBUG_TRIE_COMPILE_r({
2054 PerlIO_printf(Perl_debug_log, "%*sStclass Failtable: 0", (int)(depth * 2), "");
2055 for( q_read=1; q_read<numstates; q_read++ ) {
2056 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2058 PerlIO_printf(Perl_debug_log, "\n");
2061 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2066 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2067 * These need to be revisited when a newer toolchain becomes available.
2069 #if defined(__sparc64__) && defined(__GNUC__)
2070 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2071 # undef SPARC64_GCC_WORKAROUND
2072 # define SPARC64_GCC_WORKAROUND 1
2076 #define DEBUG_PEEP(str,scan,depth) \
2077 DEBUG_OPTIMISE_r({ \
2078 SV * const mysv=sv_newmortal(); \
2079 regnode *Next = regnext(scan); \
2080 regprop(RExC_rx, mysv, scan); \
2081 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s [%d]\n", \
2082 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2083 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2090 #define JOIN_EXACT(scan,min,flags) \
2091 if (PL_regkind[OP(scan)] == EXACT) \
2092 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2095 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2096 /* Merge several consecutive EXACTish nodes into one. */
2097 regnode *n = regnext(scan);
2099 regnode *next = scan + NODE_SZ_STR(scan);
2103 regnode *stop = scan;
2104 GET_RE_DEBUG_FLAGS_DECL;
2106 PERL_UNUSED_ARG(depth);
2108 #ifndef EXPERIMENTAL_INPLACESCAN
2109 PERL_UNUSED_ARG(flags);
2110 PERL_UNUSED_ARG(val);
2112 DEBUG_PEEP("join",scan,depth);
2114 /* Skip NOTHING, merge EXACT*. */
2116 ( PL_regkind[OP(n)] == NOTHING ||
2117 (stringok && (OP(n) == OP(scan))))
2119 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2121 if (OP(n) == TAIL || n > next)
2123 if (PL_regkind[OP(n)] == NOTHING) {
2124 DEBUG_PEEP("skip:",n,depth);
2125 NEXT_OFF(scan) += NEXT_OFF(n);
2126 next = n + NODE_STEP_REGNODE;
2133 else if (stringok) {
2134 const unsigned int oldl = STR_LEN(scan);
2135 regnode * const nnext = regnext(n);
2137 DEBUG_PEEP("merg",n,depth);
2140 if (oldl + STR_LEN(n) > U8_MAX)
2142 NEXT_OFF(scan) += NEXT_OFF(n);
2143 STR_LEN(scan) += STR_LEN(n);
2144 next = n + NODE_SZ_STR(n);
2145 /* Now we can overwrite *n : */
2146 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2154 #ifdef EXPERIMENTAL_INPLACESCAN
2155 if (flags && !NEXT_OFF(n)) {
2156 DEBUG_PEEP("atch", val, depth);
2157 if (reg_off_by_arg[OP(n)]) {
2158 ARG_SET(n, val - n);
2161 NEXT_OFF(n) = val - n;
2168 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2170 Two problematic code points in Unicode casefolding of EXACT nodes:
2172 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2173 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2179 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2180 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2182 This means that in case-insensitive matching (or "loose matching",
2183 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2184 length of the above casefolded versions) can match a target string
2185 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2186 This would rather mess up the minimum length computation.
2188 What we'll do is to look for the tail four bytes, and then peek
2189 at the preceding two bytes to see whether we need to decrease
2190 the minimum length by four (six minus two).
2192 Thanks to the design of UTF-8, there cannot be false matches:
2193 A sequence of valid UTF-8 bytes cannot be a subsequence of
2194 another valid sequence of UTF-8 bytes.
2197 char * const s0 = STRING(scan), *s, *t;
2198 char * const s1 = s0 + STR_LEN(scan) - 1;
2199 char * const s2 = s1 - 4;
2200 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2201 const char t0[] = "\xaf\x49\xaf\x42";
2203 const char t0[] = "\xcc\x88\xcc\x81";
2205 const char * const t1 = t0 + 3;
2208 s < s2 && (t = ninstr(s, s1, t0, t1));
2211 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2212 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2214 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2215 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2223 n = scan + NODE_SZ_STR(scan);
2225 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2232 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2236 /* REx optimizer. Converts nodes into quickier variants "in place".
2237 Finds fixed substrings. */
2239 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2240 to the position after last scanned or to NULL. */
2245 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2246 I32 *minlenp, I32 *deltap,
2247 regnode *last, scan_data_t *data, U32 flags, U32 depth)
2248 /* scanp: Start here (read-write). */
2249 /* deltap: Write maxlen-minlen here. */
2250 /* last: Stop before this one. */
2253 I32 min = 0, pars = 0, code;
2254 regnode *scan = *scanp, *next;
2256 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2257 int is_inf_internal = 0; /* The studied chunk is infinite */
2258 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2259 scan_data_t data_fake;
2260 struct regnode_charclass_class and_with; /* Valid if flags & SCF_DO_STCLASS_OR */
2261 SV *re_trie_maxbuff = NULL;
2262 regnode *first_non_open = scan;
2265 GET_RE_DEBUG_FLAGS_DECL;
2267 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2270 while (first_non_open && OP(first_non_open) == OPEN)
2271 first_non_open=regnext(first_non_open);
2275 while (scan && OP(scan) != END && scan < last) {
2276 /* Peephole optimizer: */
2277 DEBUG_STUDYDATA(data,depth);
2278 DEBUG_PEEP("Peep",scan,depth);
2279 JOIN_EXACT(scan,&min,0);
2281 /* Follow the next-chain of the current node and optimize
2282 away all the NOTHINGs from it. */
2283 if (OP(scan) != CURLYX) {
2284 const int max = (reg_off_by_arg[OP(scan)]
2286 /* I32 may be smaller than U16 on CRAYs! */
2287 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2288 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2292 /* Skip NOTHING and LONGJMP. */
2293 while ((n = regnext(n))
2294 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2295 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2296 && off + noff < max)
2298 if (reg_off_by_arg[OP(scan)])
2301 NEXT_OFF(scan) = off;
2306 /* The principal pseudo-switch. Cannot be a switch, since we
2307 look into several different things. */
2308 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2309 || OP(scan) == IFTHEN || OP(scan) == SUSPEND) {
2310 next = regnext(scan);
2312 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2314 if (OP(next) == code || code == IFTHEN || code == SUSPEND) {
2315 /* NOTE - There is similar code to this block below for handling
2316 TRIE nodes on a re-study. If you change stuff here check there
2318 I32 max1 = 0, min1 = I32_MAX, num = 0;
2319 struct regnode_charclass_class accum;
2320 regnode * const startbranch=scan;
2322 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
2323 scan_commit(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2324 if (flags & SCF_DO_STCLASS)
2325 cl_init_zero(pRExC_state, &accum);
2327 while (OP(scan) == code) {
2328 I32 deltanext, minnext, f = 0, fake;
2329 struct regnode_charclass_class this_class;
2332 data_fake.flags = 0;
2334 data_fake.whilem_c = data->whilem_c;
2335 data_fake.last_closep = data->last_closep;
2338 data_fake.last_closep = &fake;
2339 next = regnext(scan);
2340 scan = NEXTOPER(scan);
2342 scan = NEXTOPER(scan);
2343 if (flags & SCF_DO_STCLASS) {
2344 cl_init(pRExC_state, &this_class);
2345 data_fake.start_class = &this_class;
2346 f = SCF_DO_STCLASS_AND;
2348 if (flags & SCF_WHILEM_VISITED_POS)
2349 f |= SCF_WHILEM_VISITED_POS;
2351 /* we suppose the run is continuous, last=next...*/
2352 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2353 next, &data_fake, f,depth+1);
2356 if (max1 < minnext + deltanext)
2357 max1 = minnext + deltanext;
2358 if (deltanext == I32_MAX)
2359 is_inf = is_inf_internal = 1;
2361 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2364 if (data_fake.flags & SF_HAS_EVAL)
2365 data->flags |= SF_HAS_EVAL;
2366 data->whilem_c = data_fake.whilem_c;
2368 if (flags & SCF_DO_STCLASS)
2369 cl_or(pRExC_state, &accum, &this_class);
2370 if (code == SUSPEND)
2373 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2375 if (flags & SCF_DO_SUBSTR) {
2376 data->pos_min += min1;
2377 data->pos_delta += max1 - min1;
2378 if (max1 != min1 || is_inf)
2379 data->longest = &(data->longest_float);
2382 delta += max1 - min1;
2383 if (flags & SCF_DO_STCLASS_OR) {
2384 cl_or(pRExC_state, data->start_class, &accum);
2386 cl_and(data->start_class, &and_with);
2387 flags &= ~SCF_DO_STCLASS;
2390 else if (flags & SCF_DO_STCLASS_AND) {
2392 cl_and(data->start_class, &accum);
2393 flags &= ~SCF_DO_STCLASS;
2396 /* Switch to OR mode: cache the old value of
2397 * data->start_class */
2398 StructCopy(data->start_class, &and_with,
2399 struct regnode_charclass_class);
2400 flags &= ~SCF_DO_STCLASS_AND;
2401 StructCopy(&accum, data->start_class,
2402 struct regnode_charclass_class);
2403 flags |= SCF_DO_STCLASS_OR;
2404 data->start_class->flags |= ANYOF_EOS;
2408 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2411 Assuming this was/is a branch we are dealing with: 'scan' now
2412 points at the item that follows the branch sequence, whatever
2413 it is. We now start at the beginning of the sequence and look
2420 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2422 If we can find such a subseqence we need to turn the first
2423 element into a trie and then add the subsequent branch exact
2424 strings to the trie.
2428 1. patterns where the whole set of branch can be converted.
2430 2. patterns where only a subset can be converted.
2432 In case 1 we can replace the whole set with a single regop
2433 for the trie. In case 2 we need to keep the start and end
2436 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2437 becomes BRANCH TRIE; BRANCH X;
2439 There is an additional case, that being where there is a
2440 common prefix, which gets split out into an EXACT like node
2441 preceding the TRIE node.
2443 If x(1..n)==tail then we can do a simple trie, if not we make
2444 a "jump" trie, such that when we match the appropriate word
2445 we "jump" to the appopriate tail node. Essentailly we turn
2446 a nested if into a case structure of sorts.
2451 if (!re_trie_maxbuff) {
2452 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2453 if (!SvIOK(re_trie_maxbuff))
2454 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2456 if ( SvIV(re_trie_maxbuff)>=0 ) {
2458 regnode *first = (regnode *)NULL;
2459 regnode *last = (regnode *)NULL;
2460 regnode *tail = scan;
2465 SV * const mysv = sv_newmortal(); /* for dumping */
2467 /* var tail is used because there may be a TAIL
2468 regop in the way. Ie, the exacts will point to the
2469 thing following the TAIL, but the last branch will
2470 point at the TAIL. So we advance tail. If we
2471 have nested (?:) we may have to move through several
2475 while ( OP( tail ) == TAIL ) {
2476 /* this is the TAIL generated by (?:) */
2477 tail = regnext( tail );
2482 regprop(RExC_rx, mysv, tail );
2483 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2484 (int)depth * 2 + 2, "",
2485 "Looking for TRIE'able sequences. Tail node is: ",
2486 SvPV_nolen_const( mysv )
2492 step through the branches, cur represents each
2493 branch, noper is the first thing to be matched
2494 as part of that branch and noper_next is the
2495 regnext() of that node. if noper is an EXACT
2496 and noper_next is the same as scan (our current
2497 position in the regex) then the EXACT branch is
2498 a possible optimization target. Once we have
2499 two or more consequetive such branches we can
2500 create a trie of the EXACT's contents and stich
2501 it in place. If the sequence represents all of
2502 the branches we eliminate the whole thing and
2503 replace it with a single TRIE. If it is a
2504 subsequence then we need to stitch it in. This
2505 means the first branch has to remain, and needs
2506 to be repointed at the item on the branch chain
2507 following the last branch optimized. This could
2508 be either a BRANCH, in which case the
2509 subsequence is internal, or it could be the
2510 item following the branch sequence in which
2511 case the subsequence is at the end.
2515 /* dont use tail as the end marker for this traverse */
2516 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2517 regnode * const noper = NEXTOPER( cur );
2518 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2519 regnode * const noper_next = regnext( noper );
2523 regprop(RExC_rx, mysv, cur);
2524 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2525 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2527 regprop(RExC_rx, mysv, noper);
2528 PerlIO_printf( Perl_debug_log, " -> %s",
2529 SvPV_nolen_const(mysv));
2532 regprop(RExC_rx, mysv, noper_next );
2533 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2534 SvPV_nolen_const(mysv));
2536 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2537 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2539 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2540 : PL_regkind[ OP( noper ) ] == EXACT )
2541 || OP(noper) == NOTHING )
2543 && noper_next == tail
2548 if ( !first || optype == NOTHING ) {
2549 if (!first) first = cur;
2550 optype = OP( noper );
2556 make_trie( pRExC_state,
2557 startbranch, first, cur, tail, count,
2560 if ( PL_regkind[ OP( noper ) ] == EXACT
2562 && noper_next == tail
2567 optype = OP( noper );
2577 regprop(RExC_rx, mysv, cur);
2578 PerlIO_printf( Perl_debug_log,
2579 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2580 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2584 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2585 #ifdef TRIE_STUDY_OPT
2586 if ( ((made == MADE_EXACT_TRIE &&
2587 startbranch == first)
2588 || ( first_non_open == first )) &&
2590 flags |= SCF_TRIE_RESTUDY;
2598 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2599 scan = NEXTOPER(NEXTOPER(scan));
2600 } else /* single branch is optimized. */
2601 scan = NEXTOPER(scan);
2604 else if (OP(scan) == EXACT) {
2605 I32 l = STR_LEN(scan);
2608 const U8 * const s = (U8*)STRING(scan);
2609 l = utf8_length(s, s + l);
2610 uc = utf8_to_uvchr(s, NULL);
2612 uc = *((U8*)STRING(scan));
2615 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2616 /* The code below prefers earlier match for fixed
2617 offset, later match for variable offset. */
2618 if (data->last_end == -1) { /* Update the start info. */
2619 data->last_start_min = data->pos_min;
2620 data->last_start_max = is_inf
2621 ? I32_MAX : data->pos_min + data->pos_delta;
2623 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2625 SvUTF8_on(data->last_found);
2627 SV * const sv = data->last_found;
2628 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2629 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2630 if (mg && mg->mg_len >= 0)
2631 mg->mg_len += utf8_length((U8*)STRING(scan),
2632 (U8*)STRING(scan)+STR_LEN(scan));
2634 data->last_end = data->pos_min + l;
2635 data->pos_min += l; /* As in the first entry. */
2636 data->flags &= ~SF_BEFORE_EOL;
2638 if (flags & SCF_DO_STCLASS_AND) {
2639 /* Check whether it is compatible with what we know already! */
2643 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2644 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2645 && (!(data->start_class->flags & ANYOF_FOLD)
2646 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2649 ANYOF_CLASS_ZERO(data->start_class);
2650 ANYOF_BITMAP_ZERO(data->start_class);
2652 ANYOF_BITMAP_SET(data->start_class, uc);
2653 data->start_class->flags &= ~ANYOF_EOS;
2655 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2657 else if (flags & SCF_DO_STCLASS_OR) {
2658 /* false positive possible if the class is case-folded */
2660 ANYOF_BITMAP_SET(data->start_class, uc);
2662 data->start_class->flags |= ANYOF_UNICODE_ALL;
2663 data->start_class->flags &= ~ANYOF_EOS;
2664 cl_and(data->start_class, &and_with);
2666 flags &= ~SCF_DO_STCLASS;
2668 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2669 I32 l = STR_LEN(scan);
2670 UV uc = *((U8*)STRING(scan));
2672 /* Search for fixed substrings supports EXACT only. */
2673 if (flags & SCF_DO_SUBSTR) {
2675 scan_commit(pRExC_state, data, minlenp);
2678 const U8 * const s = (U8 *)STRING(scan);
2679 l = utf8_length(s, s + l);
2680 uc = utf8_to_uvchr(s, NULL);
2683 if (flags & SCF_DO_SUBSTR)
2685 if (flags & SCF_DO_STCLASS_AND) {
2686 /* Check whether it is compatible with what we know already! */
2690 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2691 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2692 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2694 ANYOF_CLASS_ZERO(data->start_class);
2695 ANYOF_BITMAP_ZERO(data->start_class);
2697 ANYOF_BITMAP_SET(data->start_class, uc);
2698 data->start_class->flags &= ~ANYOF_EOS;
2699 data->start_class->flags |= ANYOF_FOLD;
2700 if (OP(scan) == EXACTFL)
2701 data->start_class->flags |= ANYOF_LOCALE;
2704 else if (flags & SCF_DO_STCLASS_OR) {
2705 if (data->start_class->flags & ANYOF_FOLD) {
2706 /* false positive possible if the class is case-folded.
2707 Assume that the locale settings are the same... */
2709 ANYOF_BITMAP_SET(data->start_class, uc);
2710 data->start_class->flags &= ~ANYOF_EOS;
2712 cl_and(data->start_class, &and_with);
2714 flags &= ~SCF_DO_STCLASS;
2716 else if (OP(scan)==RECURSE) {
2717 ARG2L_SET( scan, RExC_parens[ARG(scan)-1] - scan );
2719 else if (strchr((const char*)PL_varies,OP(scan))) {
2720 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2721 I32 f = flags, pos_before = 0;
2722 regnode * const oscan = scan;
2723 struct regnode_charclass_class this_class;
2724 struct regnode_charclass_class *oclass = NULL;
2725 I32 next_is_eval = 0;
2727 switch (PL_regkind[OP(scan)]) {
2728 case WHILEM: /* End of (?:...)* . */
2729 scan = NEXTOPER(scan);
2732 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2733 next = NEXTOPER(scan);
2734 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2736 maxcount = REG_INFTY;
2737 next = regnext(scan);
2738 scan = NEXTOPER(scan);
2742 if (flags & SCF_DO_SUBSTR)
2747 if (flags & SCF_DO_STCLASS) {
2749 maxcount = REG_INFTY;
2750 next = regnext(scan);
2751 scan = NEXTOPER(scan);
2754 is_inf = is_inf_internal = 1;
2755 scan = regnext(scan);
2756 if (flags & SCF_DO_SUBSTR) {
2757 scan_commit(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2758 data->longest = &(data->longest_float);
2760 goto optimize_curly_tail;
2762 mincount = ARG1(scan);
2763 maxcount = ARG2(scan);
2764 next = regnext(scan);
2765 if (OP(scan) == CURLYX) {
2766 I32 lp = (data ? *(data->last_closep) : 0);
2767 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
2769 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
2770 next_is_eval = (OP(scan) == EVAL);
2772 if (flags & SCF_DO_SUBSTR) {
2773 if (mincount == 0) scan_commit(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
2774 pos_before = data->pos_min;
2778 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
2780 data->flags |= SF_IS_INF;
2782 if (flags & SCF_DO_STCLASS) {
2783 cl_init(pRExC_state, &this_class);
2784 oclass = data->start_class;
2785 data->start_class = &this_class;
2786 f |= SCF_DO_STCLASS_AND;
2787 f &= ~SCF_DO_STCLASS_OR;
2789 /* These are the cases when once a subexpression
2790 fails at a particular position, it cannot succeed
2791 even after backtracking at the enclosing scope.
2793 XXXX what if minimal match and we are at the
2794 initial run of {n,m}? */
2795 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
2796 f &= ~SCF_WHILEM_VISITED_POS;
2798 /* This will finish on WHILEM, setting scan, or on NULL: */
2799 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, last, data,
2801 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
2803 if (flags & SCF_DO_STCLASS)
2804 data->start_class = oclass;
2805 if (mincount == 0 || minnext == 0) {
2806 if (flags & SCF_DO_STCLASS_OR) {
2807 cl_or(pRExC_state, data->start_class, &this_class);
2809 else if (flags & SCF_DO_STCLASS_AND) {
2810 /* Switch to OR mode: cache the old value of
2811 * data->start_class */
2812 StructCopy(data->start_class, &and_with,
2813 struct regnode_charclass_class);
2814 flags &= ~SCF_DO_STCLASS_AND;
2815 StructCopy(&this_class, data->start_class,
2816 struct regnode_charclass_class);
2817 flags |= SCF_DO_STCLASS_OR;
2818 data->start_class->flags |= ANYOF_EOS;
2820 } else { /* Non-zero len */
2821 if (flags & SCF_DO_STCLASS_OR) {
2822 cl_or(pRExC_state, data->start_class, &this_class);
2823 cl_and(data->start_class, &and_with);
2825 else if (flags & SCF_DO_STCLASS_AND)
2826 cl_and(data->start_class, &this_class);
2827 flags &= ~SCF_DO_STCLASS;
2829 if (!scan) /* It was not CURLYX, but CURLY. */
2831 if ( /* ? quantifier ok, except for (?{ ... }) */
2832 (next_is_eval || !(mincount == 0 && maxcount == 1))
2833 && (minnext == 0) && (deltanext == 0)
2834 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
2835 && maxcount <= REG_INFTY/3 /* Complement check for big count */
2836 && ckWARN(WARN_REGEXP))
2839 "Quantifier unexpected on zero-length expression");
2842 min += minnext * mincount;
2843 is_inf_internal |= ((maxcount == REG_INFTY
2844 && (minnext + deltanext) > 0)
2845 || deltanext == I32_MAX);
2846 is_inf |= is_inf_internal;
2847 delta += (minnext + deltanext) * maxcount - minnext * mincount;
2849 /* Try powerful optimization CURLYX => CURLYN. */
2850 if ( OP(oscan) == CURLYX && data
2851 && data->flags & SF_IN_PAR
2852 && !(data->flags & SF_HAS_EVAL)
2853 && !deltanext && minnext == 1 ) {
2854 /* Try to optimize to CURLYN. */
2855 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
2856 regnode * const nxt1 = nxt;
2863 if (!strchr((const char*)PL_simple,OP(nxt))
2864 && !(PL_regkind[OP(nxt)] == EXACT
2865 && STR_LEN(nxt) == 1))
2871 if (OP(nxt) != CLOSE)
2873 /* Now we know that nxt2 is the only contents: */
2874 oscan->flags = (U8)ARG(nxt);
2876 OP(nxt1) = NOTHING; /* was OPEN. */
2878 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
2879 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
2880 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
2881 OP(nxt) = OPTIMIZED; /* was CLOSE. */
2882 OP(nxt + 1) = OPTIMIZED; /* was count. */
2883 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
2888 /* Try optimization CURLYX => CURLYM. */
2889 if ( OP(oscan) == CURLYX && data
2890 && !(data->flags & SF_HAS_PAR)
2891 && !(data->flags & SF_HAS_EVAL)
2892 && !deltanext /* atom is fixed width */
2893 && minnext != 0 /* CURLYM can't handle zero width */
2895 /* XXXX How to optimize if data == 0? */
2896 /* Optimize to a simpler form. */
2897 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
2901 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
2902 && (OP(nxt2) != WHILEM))
2904 OP(nxt2) = SUCCEED; /* Whas WHILEM */
2905 /* Need to optimize away parenths. */
2906 if (data->flags & SF_IN_PAR) {
2907 /* Set the parenth number. */
2908 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
2910 if (OP(nxt) != CLOSE)
2911 FAIL("Panic opt close");
2912 oscan->flags = (U8)ARG(nxt);
2913 OP(nxt1) = OPTIMIZED; /* was OPEN. */
2914 OP(nxt) = OPTIMIZED; /* was CLOSE. */
2916 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
2917 OP(nxt + 1) = OPTIMIZED; /* was count. */
2918 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
2919 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
2922 while ( nxt1 && (OP(nxt1) != WHILEM)) {
2923 regnode *nnxt = regnext(nxt1);
2926 if (reg_off_by_arg[OP(nxt1)])
2927 ARG_SET(nxt1, nxt2 - nxt1);
2928 else if (nxt2 - nxt1 < U16_MAX)
2929 NEXT_OFF(nxt1) = nxt2 - nxt1;
2931 OP(nxt) = NOTHING; /* Cannot beautify */
2936 /* Optimize again: */
2937 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
2943 else if ((OP(oscan) == CURLYX)
2944 && (flags & SCF_WHILEM_VISITED_POS)
2945 /* See the comment on a similar expression above.
2946 However, this time it not a subexpression
2947 we care about, but the expression itself. */
2948 && (maxcount == REG_INFTY)
2949 && data && ++data->whilem_c < 16) {
2950 /* This stays as CURLYX, we can put the count/of pair. */
2951 /* Find WHILEM (as in regexec.c) */
2952 regnode *nxt = oscan + NEXT_OFF(oscan);
2954 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
2956 PREVOPER(nxt)->flags = (U8)(data->whilem_c
2957 | (RExC_whilem_seen << 4)); /* On WHILEM */
2959 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
2961 if (flags & SCF_DO_SUBSTR) {
2962 SV *last_str = NULL;
2963 int counted = mincount != 0;
2965 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
2966 #if defined(SPARC64_GCC_WORKAROUND)
2969 const char *s = NULL;
2972 if (pos_before >= data->last_start_min)
2975 b = data->last_start_min;
2978 s = SvPV_const(data->last_found, l);
2979 old = b - data->last_start_min;
2982 I32 b = pos_before >= data->last_start_min
2983 ? pos_before : data->last_start_min;
2985 const char * const s = SvPV_const(data->last_found, l);
2986 I32 old = b - data->last_start_min;
2990 old = utf8_hop((U8*)s, old) - (U8*)s;
2993 /* Get the added string: */
2994 last_str = newSVpvn(s + old, l);
2996 SvUTF8_on(last_str);
2997 if (deltanext == 0 && pos_before == b) {
2998 /* What was added is a constant string */
3000 SvGROW(last_str, (mincount * l) + 1);
3001 repeatcpy(SvPVX(last_str) + l,
3002 SvPVX_const(last_str), l, mincount - 1);
3003 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3004 /* Add additional parts. */
3005 SvCUR_set(data->last_found,
3006 SvCUR(data->last_found) - l);
3007 sv_catsv(data->last_found, last_str);
3009 SV * sv = data->last_found;
3011 SvUTF8(sv) && SvMAGICAL(sv) ?
3012 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3013 if (mg && mg->mg_len >= 0)
3014 mg->mg_len += CHR_SVLEN(last_str);
3016 data->last_end += l * (mincount - 1);
3019 /* start offset must point into the last copy */
3020 data->last_start_min += minnext * (mincount - 1);
3021 data->last_start_max += is_inf ? I32_MAX
3022 : (maxcount - 1) * (minnext + data->pos_delta);
3025 /* It is counted once already... */
3026 data->pos_min += minnext * (mincount - counted);
3027 data->pos_delta += - counted * deltanext +
3028 (minnext + deltanext) * maxcount - minnext * mincount;
3029 if (mincount != maxcount) {
3030 /* Cannot extend fixed substrings found inside
3032 scan_commit(pRExC_state,data,minlenp);
3033 if (mincount && last_str) {
3034 SV * const sv = data->last_found;
3035 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3036 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3040 sv_setsv(sv, last_str);
3041 data->last_end = data->pos_min;
3042 data->last_start_min =
3043 data->pos_min - CHR_SVLEN(last_str);
3044 data->last_start_max = is_inf
3046 : data->pos_min + data->pos_delta
3047 - CHR_SVLEN(last_str);
3049 data->longest = &(data->longest_float);
3051 SvREFCNT_dec(last_str);
3053 if (data && (fl & SF_HAS_EVAL))
3054 data->flags |= SF_HAS_EVAL;
3055 optimize_curly_tail:
3056 if (OP(oscan) != CURLYX) {
3057 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3059 NEXT_OFF(oscan) += NEXT_OFF(next);
3062 default: /* REF and CLUMP only? */
3063 if (flags & SCF_DO_SUBSTR) {
3064 scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
3065 data->longest = &(data->longest_float);
3067 is_inf = is_inf_internal = 1;
3068 if (flags & SCF_DO_STCLASS_OR)
3069 cl_anything(pRExC_state, data->start_class);
3070 flags &= ~SCF_DO_STCLASS;
3074 else if (strchr((const char*)PL_simple,OP(scan))) {
3077 if (flags & SCF_DO_SUBSTR) {
3078 scan_commit(pRExC_state,data,minlenp);
3082 if (flags & SCF_DO_STCLASS) {
3083 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3085 /* Some of the logic below assumes that switching
3086 locale on will only add false positives. */
3087 switch (PL_regkind[OP(scan)]) {
3091 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3092 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3093 cl_anything(pRExC_state, data->start_class);
3096 if (OP(scan) == SANY)
3098 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3099 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3100 || (data->start_class->flags & ANYOF_CLASS));
3101 cl_anything(pRExC_state, data->start_class);
3103 if (flags & SCF_DO_STCLASS_AND || !value)
3104 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3107 if (flags & SCF_DO_STCLASS_AND)
3108 cl_and(data->start_class,
3109 (struct regnode_charclass_class*)scan);
3111 cl_or(pRExC_state, data->start_class,
3112 (struct regnode_charclass_class*)scan);
3115 if (flags & SCF_DO_STCLASS_AND) {
3116 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3117 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3118 for (value = 0; value < 256; value++)
3119 if (!isALNUM(value))
3120 ANYOF_BITMAP_CLEAR(data->start_class, value);
3124 if (data->start_class->flags & ANYOF_LOCALE)
3125 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3127 for (value = 0; value < 256; value++)
3129 ANYOF_BITMAP_SET(data->start_class, value);
3134 if (flags & SCF_DO_STCLASS_AND) {
3135 if (data->start_class->flags & ANYOF_LOCALE)
3136 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3139 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3140 data->start_class->flags |= ANYOF_LOCALE;
3144 if (flags & SCF_DO_STCLASS_AND) {
3145 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3146 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3147 for (value = 0; value < 256; value++)
3149 ANYOF_BITMAP_CLEAR(data->start_class, value);
3153 if (data->start_class->flags & ANYOF_LOCALE)
3154 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3156 for (value = 0; value < 256; value++)
3157 if (!isALNUM(value))
3158 ANYOF_BITMAP_SET(data->start_class, value);
3163 if (flags & SCF_DO_STCLASS_AND) {
3164 if (data->start_class->flags & ANYOF_LOCALE)
3165 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3168 data->start_class->flags |= ANYOF_LOCALE;
3169 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3173 if (flags & SCF_DO_STCLASS_AND) {
3174 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3175 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3176 for (value = 0; value < 256; value++)
3177 if (!isSPACE(value))
3178 ANYOF_BITMAP_CLEAR(data->start_class, value);
3182 if (data->start_class->flags & ANYOF_LOCALE)
3183 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3185 for (value = 0; value < 256; value++)
3187 ANYOF_BITMAP_SET(data->start_class, value);
3192 if (flags & SCF_DO_STCLASS_AND) {
3193 if (data->start_class->flags & ANYOF_LOCALE)
3194 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3197 data->start_class->flags |= ANYOF_LOCALE;
3198 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3202 if (flags & SCF_DO_STCLASS_AND) {
3203 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3204 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3205 for (value = 0; value < 256; value++)
3207 ANYOF_BITMAP_CLEAR(data->start_class, value);
3211 if (data->start_class->flags & ANYOF_LOCALE)
3212 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3214 for (value = 0; value < 256; value++)
3215 if (!isSPACE(value))
3216 ANYOF_BITMAP_SET(data->start_class, value);
3221 if (flags & SCF_DO_STCLASS_AND) {
3222 if (data->start_class->flags & ANYOF_LOCALE) {
3223 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3224 for (value = 0; value < 256; value++)
3225 if (!isSPACE(value))
3226 ANYOF_BITMAP_CLEAR(data->start_class, value);
3230 data->start_class->flags |= ANYOF_LOCALE;
3231 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3235 if (flags & SCF_DO_STCLASS_AND) {
3236 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3237 for (value = 0; value < 256; value++)
3238 if (!isDIGIT(value))
3239 ANYOF_BITMAP_CLEAR(data->start_class, value);
3242 if (data->start_class->flags & ANYOF_LOCALE)
3243 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3245 for (value = 0; value < 256; value++)
3247 ANYOF_BITMAP_SET(data->start_class, value);
3252 if (flags & SCF_DO_STCLASS_AND) {
3253 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3254 for (value = 0; value < 256; value++)
3256 ANYOF_BITMAP_CLEAR(data->start_class, value);
3259 if (data->start_class->flags & ANYOF_LOCALE)
3260 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3262 for (value = 0; value < 256; value++)
3263 if (!isDIGIT(value))
3264 ANYOF_BITMAP_SET(data->start_class, value);
3269 if (flags & SCF_DO_STCLASS_OR)
3270 cl_and(data->start_class, &and_with);
3271 flags &= ~SCF_DO_STCLASS;
3274 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3275 data->flags |= (OP(scan) == MEOL
3279 else if ( PL_regkind[OP(scan)] == BRANCHJ
3280 /* Lookbehind, or need to calculate parens/evals/stclass: */
3281 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3282 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3283 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3284 || OP(scan) == UNLESSM )
3286 /* Negative Lookahead/lookbehind
3287 In this case we can't do fixed string optimisation.
3290 I32 deltanext, minnext, fake = 0;
3292 struct regnode_charclass_class intrnl;
3295 data_fake.flags = 0;
3297 data_fake.whilem_c = data->whilem_c;
3298 data_fake.last_closep = data->last_closep;
3301 data_fake.last_closep = &fake;
3302 if ( flags & SCF_DO_STCLASS && !scan->flags
3303 && OP(scan) == IFMATCH ) { /* Lookahead */
3304 cl_init(pRExC_state, &intrnl);
3305 data_fake.start_class = &intrnl;
3306 f |= SCF_DO_STCLASS_AND;
3308 if (flags & SCF_WHILEM_VISITED_POS)
3309 f |= SCF_WHILEM_VISITED_POS;
3310 next = regnext(scan);
3311 nscan = NEXTOPER(NEXTOPER(scan));
3312 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, last, &data_fake, f,depth+1);
3315 vFAIL("Variable length lookbehind not implemented");
3317 else if (minnext > (I32)U8_MAX) {
3318 vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3320 scan->flags = (U8)minnext;
3323 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3325 if (data_fake.flags & SF_HAS_EVAL)
3326 data->flags |= SF_HAS_EVAL;
3327 data->whilem_c = data_fake.whilem_c;
3329 if (f & SCF_DO_STCLASS_AND) {
3330 const int was = (data->start_class->flags & ANYOF_EOS);
3332 cl_and(data->start_class, &intrnl);
3334 data->start_class->flags |= ANYOF_EOS;
3337 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3339 /* Positive Lookahead/lookbehind
3340 In this case we can do fixed string optimisation,
3341 but we must be careful about it. Note in the case of
3342 lookbehind the positions will be offset by the minimum
3343 length of the pattern, something we won't know about
3344 until after the recurse.
3346 I32 deltanext, fake = 0;
3348 struct regnode_charclass_class intrnl;
3350 /* We use SAVEFREEPV so that when the full compile
3351 is finished perl will clean up the allocated
3352 minlens when its all done. This was we don't
3353 have to worry about freeing them when we know
3354 they wont be used, which would be a pain.
3357 Newx( minnextp, 1, I32 );
3358 SAVEFREEPV(minnextp);
3361 StructCopy(data, &data_fake, scan_data_t);
3362 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3365 scan_commit(pRExC_state, &data_fake,minlenp);
3366 data_fake.last_found=newSVsv(data->last_found);
3370 data_fake.last_closep = &fake;
3371 data_fake.flags = 0;
3373 data_fake.flags |= SF_IS_INF;
3374 if ( flags & SCF_DO_STCLASS && !scan->flags
3375 && OP(scan) == IFMATCH ) { /* Lookahead */
3376 cl_init(pRExC_state, &intrnl);
3377 data_fake.start_class = &intrnl;
3378 f |= SCF_DO_STCLASS_AND;
3380 if (flags & SCF_WHILEM_VISITED_POS)
3381 f |= SCF_WHILEM_VISITED_POS;
3382 next = regnext(scan);
3383 nscan = NEXTOPER(NEXTOPER(scan));
3385 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext, last, &data_fake, f,depth+1);
3388 vFAIL("Variable length lookbehind not implemented");
3390 else if (*minnextp > (I32)U8_MAX) {
3391 vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3393 scan->flags = (U8)*minnextp;
3399 if (f & SCF_DO_STCLASS_AND) {
3400 const int was = (data->start_class->flags & ANYOF_EOS);
3402 cl_and(data->start_class, &intrnl);
3404 data->start_class->flags |= ANYOF_EOS;
3407 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3409 if (data_fake.flags & SF_HAS_EVAL)
3410 data->flags |= SF_HAS_EVAL;
3411 data->whilem_c = data_fake.whilem_c;
3412 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3413 if (RExC_rx->minlen<*minnextp)
3414 RExC_rx->minlen=*minnextp;
3415 scan_commit(pRExC_state, &data_fake, minnextp);
3416 SvREFCNT_dec(data_fake.last_found);
3418 if ( data_fake.minlen_fixed != minlenp )
3420 data->offset_fixed= data_fake.offset_fixed;
3421 data->minlen_fixed= data_fake.minlen_fixed;
3422 data->lookbehind_fixed+= scan->flags;
3424 if ( data_fake.minlen_float != minlenp )
3426 data->minlen_float= data_fake.minlen_float;
3427 data->offset_float_min=data_fake.offset_float_min;
3428 data->offset_float_max=data_fake.offset_float_max;
3429 data->lookbehind_float+= scan->flags;
3438 else if (OP(scan) == OPEN) {
3441 else if (OP(scan) == CLOSE) {
3442 if ((I32)ARG(scan) == is_par) {
3443 next = regnext(scan);
3445 if ( next && (OP(next) != WHILEM) && next < last)
3446 is_par = 0; /* Disable optimization */
3449 *(data->last_closep) = ARG(scan);
3451 else if (OP(scan) == EVAL) {
3453 data->flags |= SF_HAS_EVAL;
3455 else if (OP(scan) == LOGICAL && scan->flags == 2) { /* Embedded follows */
3456 if (flags & SCF_DO_SUBSTR) {
3457 scan_commit(pRExC_state,data,minlenp);
3458 data->longest = &(data->longest_float);
3460 is_inf = is_inf_internal = 1;
3461 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3462 cl_anything(pRExC_state, data->start_class);
3463 flags &= ~SCF_DO_STCLASS;
3465 #ifdef TRIE_STUDY_OPT
3466 #ifdef FULL_TRIE_STUDY
3467 else if (PL_regkind[OP(scan)] == TRIE) {
3468 /* NOTE - There is similar code to this block above for handling
3469 BRANCH nodes on the initial study. If you change stuff here
3471 regnode *tail= regnext(scan);
3472 reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
3473 I32 max1 = 0, min1 = I32_MAX;
3474 struct regnode_charclass_class accum;
3476 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3477 scan_commit(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3478 if (flags & SCF_DO_STCLASS)
3479 cl_init_zero(pRExC_state, &accum);
3485 const regnode *nextbranch= NULL;
3488 for ( word=1 ; word <= trie->wordcount ; word++)
3490 I32 deltanext=0, minnext=0, f = 0, fake;
3491 struct regnode_charclass_class this_class;
3493 data_fake.flags = 0;
3495 data_fake.whilem_c = data->whilem_c;
3496 data_fake.last_closep = data->last_closep;
3499 data_fake.last_closep = &fake;
3501 if (flags & SCF_DO_STCLASS) {
3502 cl_init(pRExC_state, &this_class);
3503 data_fake.start_class = &this_class;
3504 f = SCF_DO_STCLASS_AND;
3506 if (flags & SCF_WHILEM_VISITED_POS)
3507 f |= SCF_WHILEM_VISITED_POS;
3509 if (trie->jump[word]) {
3511 nextbranch = tail - trie->jump[0];
3512 scan= tail - trie->jump[word];
3513 /* We go from the jump point to the branch that follows
3514 it. Note this means we need the vestigal unused branches
3515 even though they arent otherwise used.
3517 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3518 (regnode *)nextbranch, &data_fake, f,depth+1);
3520 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3521 nextbranch= regnext((regnode*)nextbranch);
3523 if (min1 > (I32)(minnext + trie->minlen))
3524 min1 = minnext + trie->minlen;
3525 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3526 max1 = minnext + deltanext + trie->maxlen;
3527 if (deltanext == I32_MAX)
3528 is_inf = is_inf_internal = 1;
3530 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3534 if (data_fake.flags & SF_HAS_EVAL)
3535 data->flags |= SF_HAS_EVAL;
3536 data->whilem_c = data_fake.whilem_c;
3538 if (flags & SCF_DO_STCLASS)
3539 cl_or(pRExC_state, &accum, &this_class);
3542 if (flags & SCF_DO_SUBSTR) {
3543 data->pos_min += min1;
3544 data->pos_delta += max1 - min1;
3545 if (max1 != min1 || is_inf)
3546 data->longest = &(data->longest_float);
3549 delta += max1 - min1;
3550 if (flags & SCF_DO_STCLASS_OR) {
3551 cl_or(pRExC_state, data->start_class, &accum);
3553 cl_and(data->start_class, &and_with);
3554 flags &= ~SCF_DO_STCLASS;
3557 else if (flags & SCF_DO_STCLASS_AND) {
3559 cl_and(data->start_class, &accum);
3560 flags &= ~SCF_DO_STCLASS;
3563 /* Switch to OR mode: cache the old value of
3564 * data->start_class */
3565 StructCopy(data->start_class, &and_with,
3566 struct regnode_charclass_class);
3567 flags &= ~SCF_DO_STCLASS_AND;
3568 StructCopy(&accum, data->start_class,
3569 struct regnode_charclass_class);
3570 flags |= SCF_DO_STCLASS_OR;
3571 data->start_class->flags |= ANYOF_EOS;
3578 else if (PL_regkind[OP(scan)] == TRIE) {
3579 reg_trie_data *trie = (reg_trie_data*)RExC_rx->data->data[ ARG(scan) ];
3582 min += trie->minlen;
3583 delta += (trie->maxlen - trie->minlen);
3584 flags &= ~SCF_DO_STCLASS; /* xxx */
3585 if (flags & SCF_DO_SUBSTR) {
3586 scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
3587 data->pos_min += trie->minlen;
3588 data->pos_delta += (trie->maxlen - trie->minlen);
3589 if (trie->maxlen != trie->minlen)
3590 data->longest = &(data->longest_float);
3592 if (trie->jump) /* no more substrings -- for now /grr*/
3593 flags &= ~SCF_DO_SUBSTR;
3595 #endif /* old or new */
3596 #endif /* TRIE_STUDY_OPT */
3597 /* Else: zero-length, ignore. */
3598 scan = regnext(scan);
3603 *deltap = is_inf_internal ? I32_MAX : delta;
3604 if (flags & SCF_DO_SUBSTR && is_inf)
3605 data->pos_delta = I32_MAX - data->pos_min;
3606 if (is_par > (I32)U8_MAX)
3608 if (is_par && pars==1 && data) {
3609 data->flags |= SF_IN_PAR;
3610 data->flags &= ~SF_HAS_PAR;
3612 else if (pars && data) {
3613 data->flags |= SF_HAS_PAR;
3614 data->flags &= ~SF_IN_PAR;
3616 if (flags & SCF_DO_STCLASS_OR)
3617 cl_and(data->start_class, &and_with);
3618 if (flags & SCF_TRIE_RESTUDY)
3619 data->flags |= SCF_TRIE_RESTUDY;
3621 DEBUG_STUDYDATA(data,depth);
3627 S_add_data(RExC_state_t *pRExC_state, I32 n, const char *s)
3629 if (RExC_rx->data) {
3630 Renewc(RExC_rx->data,
3631 sizeof(*RExC_rx->data) + sizeof(void*) * (RExC_rx->data->count + n - 1),
3632 char, struct reg_data);
3633 Renew(RExC_rx->data->what, RExC_rx->data->count + n, U8);
3634 RExC_rx->data->count += n;
3637 Newxc(RExC_rx->data, sizeof(*RExC_rx->data) + sizeof(void*) * (n - 1),
3638 char, struct reg_data);
3639 Newx(RExC_rx->data->what, n, U8);
3640 RExC_rx->data->count = n;
3642 Copy(s, RExC_rx->data->what + RExC_rx->data->count - n, n, U8);
3643 return RExC_rx->data->count - n;
3646 #ifndef PERL_IN_XSUB_RE
3648 Perl_reginitcolors(pTHX)
3651 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3653 char *t = savepv(s);
3657 t = strchr(t, '\t');
3663 PL_colors[i] = t = (char *)"";
3668 PL_colors[i++] = (char *)"";
3675 #ifdef TRIE_STUDY_OPT
3676 #define CHECK_RESTUDY_GOTO \
3678 (data.flags & SCF_TRIE_RESTUDY) \
3682 #define CHECK_RESTUDY_GOTO
3686 - pregcomp - compile a regular expression into internal code
3688 * We can't allocate space until we know how big the compiled form will be,
3689 * but we can't compile it (and thus know how big it is) until we've got a
3690 * place to put the code. So we cheat: we compile it twice, once with code
3691 * generation turned off and size counting turned on, and once "for real".
3692 * This also means that we don't allocate space until we are sure that the
3693 * thing really will compile successfully, and we never have to move the
3694 * code and thus invalidate pointers into it. (Note that it has to be in
3695 * one piece because free() must be able to free it all.) [NB: not true in perl]
3697 * Beware that the optimization-preparation code in here knows about some
3698 * of the structure of the compiled regexp. [I'll say.]
3700 #ifndef PERL_IN_XSUB_RE
3701 #define CORE_ONLY_BLOCK(c) {c}{
3702 #define RE_ENGINE_PTR &PL_core_reg_engine
3704 #define CORE_ONLY_BLOCK(c) {
3705 extern const struct regexp_engine my_reg_engine;
3706 #define RE_ENGINE_PTR &my_reg_engine
3711 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
3714 GET_RE_DEBUG_FLAGS_DECL;
3715 DEBUG_r(if (!PL_colorset) reginitcolors());
3717 /* Dispatch a request to compile a regexp to correct
3719 HV * const table = GvHV(PL_hintgv);
3721 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
3722 if (ptr && SvIOK(*ptr)) {
3723 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
3725 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
3728 return CALLREGCOMP_ENG(eng, exp, xend, pm);
3739 RExC_state_t RExC_state;
3740 RExC_state_t * const pRExC_state = &RExC_state;
3741 #ifdef TRIE_STUDY_OPT
3743 RExC_state_t copyRExC_state;
3746 FAIL("NULL regexp argument");
3748 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
3752 SV *dsv= sv_newmortal();
3753 RE_PV_QUOTED_DECL(s, RExC_utf8,
3754 dsv, RExC_precomp, (xend - exp), 60);
3755 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
3756 PL_colors[4],PL_colors[5],s);
3758 RExC_flags = pm->op_pmflags;
3762 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
3763 RExC_seen_evals = 0;
3766 /* First pass: determine size, legality. */
3773 RExC_emit = &PL_regdummy;
3774 RExC_whilem_seen = 0;
3775 RExC_charnames = NULL;
3777 RExC_paren_names = NULL;
3779 #if 0 /* REGC() is (currently) a NOP at the first pass.
3780 * Clever compilers notice this and complain. --jhi */
3781 REGC((U8)REG_MAGIC, (char*)RExC_emit);
3783 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
3784 if (reg(pRExC_state, 0, &flags,1) == NULL) {
3785 RExC_precomp = NULL;
3789 PerlIO_printf(Perl_debug_log,
3790 "Required size %"IVdf" nodes\n"
3791 "Starting second pass (creation)\n",
3794 RExC_lastparse=NULL;
3796 /* Small enough for pointer-storage convention?
3797 If extralen==0, this means that we will not need long jumps. */
3798 if (RExC_size >= 0x10000L && RExC_extralen)
3799 RExC_size += RExC_extralen;
3802 if (RExC_whilem_seen > 15)
3803 RExC_whilem_seen = 15;
3805 /* Allocate space and zero-initialize. Note, the two step process
3806 of zeroing when in debug mode, thus anything assigned has to
3807 happen after that */
3808 Newxc(r, sizeof(regexp) + (unsigned)RExC_size * sizeof(regnode),
3811 FAIL("Regexp out of space");
3813 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
3814 Zero(r, sizeof(regexp) + (unsigned)RExC_size * sizeof(regnode), char);
3816 /* initialization begins here */
3817 r->engine= RE_ENGINE_PTR;
3819 r->prelen = xend - exp;
3820 r->precomp = savepvn(RExC_precomp, r->prelen);
3822 #ifdef PERL_OLD_COPY_ON_WRITE
3823 r->saved_copy = NULL;
3825 r->reganch = pm->op_pmflags & PMf_COMPILETIME;
3826 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
3827 r->lastparen = 0; /* mg.c reads this. */
3829 r->substrs = 0; /* Useful during FAIL. */
3830 r->startp = 0; /* Useful during FAIL. */
3834 if (RExC_seen & REG_SEEN_RECURSE) {
3835 Newx(RExC_parens, RExC_npar,regnode *);
3836 SAVEFREEPV(RExC_parens);
3839 /* Useful during FAIL. */
3840 Newxz(r->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
3842 r->offsets[0] = RExC_size;
3844 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
3845 "%s %"UVuf" bytes for offset annotations.\n",
3846 r->offsets ? "Got" : "Couldn't get",
3847 (UV)((2*RExC_size+1) * sizeof(U32))));
3851 /* Second pass: emit code. */
3852 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
3857 RExC_emit_start = r->program;
3858 RExC_emit = r->program;
3859 /* Store the count of eval-groups for security checks: */
3860 RExC_emit->next_off = (RExC_seen_evals > (I32)U16_MAX) ? U16_MAX : (U16)RExC_seen_evals;
3861 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
3863 if (reg(pRExC_state, 0, &flags,1) == NULL)
3866 /* XXXX To minimize changes to RE engine we always allocate
3867 3-units-long substrs field. */
3868 Newx(r->substrs, 1, struct reg_substr_data);
3871 r->minlen = minlen = sawplus = sawopen = 0;
3872 Zero(r->substrs, 1, struct reg_substr_data);
3873 StructCopy(&zero_scan_data, &data, scan_data_t);
3875 #ifdef TRIE_STUDY_OPT
3877 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
3878 RExC_state=copyRExC_state;
3879 if (data.last_found) {
3880 SvREFCNT_dec(data.longest_fixed);
3881 SvREFCNT_dec(data.longest_float);
3882 SvREFCNT_dec(data.last_found);
3885 copyRExC_state=RExC_state;
3889 /* Dig out information for optimizations. */
3890 r->reganch = pm->op_pmflags & PMf_COMPILETIME; /* Again? */
3891 pm->op_pmflags = RExC_flags;
3893 r->reganch |= ROPT_UTF8; /* Unicode in it? */
3894 r->regstclass = NULL;
3895 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
3896 r->reganch |= ROPT_NAUGHTY;
3897 scan = r->program + 1; /* First BRANCH. */
3899 /* testing for BRANCH here tells us whether there is "must appear"
3900 data in the pattern. If there is then we can use it for optimisations */
3901 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
3903 STRLEN longest_float_length, longest_fixed_length;
3904 struct regnode_charclass_class ch_class; /* pointed to by data */
3906 I32 last_close = 0; /* pointed to by data */
3909 /* Skip introductions and multiplicators >= 1. */
3910 while ((OP(first) == OPEN && (sawopen = 1)) ||
3911 /* An OR of *one* alternative - should not happen now. */
3912 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
3913 /* for now we can't handle lookbehind IFMATCH*/
3914 (OP(first) == IFMATCH && !first->flags) ||
3915 (OP(first) == PLUS) ||
3916 (OP(first) == MINMOD) ||
3917 /* An {n,m} with n>0 */
3918 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
3921 if (OP(first) == PLUS)
3924 first += regarglen[OP(first)];
3925 if (OP(first) == IFMATCH) {
3926 first = NEXTOPER(first);
3927 first += EXTRA_STEP_2ARGS;
3928 } else /* XXX possible optimisation for /(?=)/ */
3929 first = NEXTOPER(first);
3932 /* Starting-point info. */
3934 DEBUG_PEEP("first:",first,0);
3935 /* Ignore EXACT as we deal with it later. */
3936 if (PL_regkind[OP(first)] == EXACT) {
3937 if (OP(first) == EXACT)
3938 NOOP; /* Empty, get anchored substr later. */
3939 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
3940 r->regstclass = first;
3943 else if (PL_regkind[OP(first)] == TRIE &&
3944 ((reg_trie_data *)r->data->data[ ARG(first) ])->minlen>0)
3947 /* this can happen only on restudy */
3948 if ( OP(first) == TRIE ) {
3949 struct regnode_1 *trieop;
3950 Newxz(trieop,1,struct regnode_1);
3951 StructCopy(first,trieop,struct regnode_1);
3952 trie_op=(regnode *)trieop;
3954 struct regnode_charclass *trieop;
3955 Newxz(trieop,1,struct regnode_charclass);
3956 StructCopy(first,trieop,struct regnode_charclass);
3957 trie_op=(regnode *)trieop;
3960 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
3961 r->regstclass = trie_op;
3964 else if (strchr((const char*)PL_simple,OP(first)))
3965 r->regstclass = first;
3966 else if (PL_regkind[OP(first)] == BOUND ||
3967 PL_regkind[OP(first)] == NBOUND)
3968 r->regstclass = first;
3969 else if (PL_regkind[OP(first)] == BOL) {
3970 r->reganch |= (OP(first) == MBOL
3972 : (OP(first) == SBOL
3975 first = NEXTOPER(first);
3978 else if (OP(first) == GPOS) {
3979 r->reganch |= ROPT_ANCH_GPOS;
3980 first = NEXTOPER(first);
3983 else if (!sawopen && (OP(first) == STAR &&
3984 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
3985 !(r->reganch & ROPT_ANCH) )
3987 /* turn .* into ^.* with an implied $*=1 */
3989 (OP(NEXTOPER(first)) == REG_ANY)
3992 r->reganch |= type | ROPT_IMPLICIT;
3993 first = NEXTOPER(first);
3996 if (sawplus && (!sawopen || !RExC_sawback)
3997 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
3998 /* x+ must match at the 1st pos of run of x's */
3999 r->reganch |= ROPT_SKIP;
4001 /* Scan is after the zeroth branch, first is atomic matcher. */
4002 #ifdef TRIE_STUDY_OPT
4005 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4006 (IV)(first - scan + 1))
4010 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4011 (IV)(first - scan + 1))
4017 * If there's something expensive in the r.e., find the
4018 * longest literal string that must appear and make it the
4019 * regmust. Resolve ties in favor of later strings, since
4020 * the regstart check works with the beginning of the r.e.
4021 * and avoiding duplication strengthens checking. Not a
4022 * strong reason, but sufficient in the absence of others.
4023 * [Now we resolve ties in favor of the earlier string if
4024 * it happens that c_offset_min has been invalidated, since the
4025 * earlier string may buy us something the later one won't.]
4029 data.longest_fixed = newSVpvs("");
4030 data.longest_float = newSVpvs("");
4031 data.last_found = newSVpvs("");
4032 data.longest = &(data.longest_fixed);
4034 if (!r->regstclass) {
4035 cl_init(pRExC_state, &ch_class);
4036 data.start_class = &ch_class;
4037 stclass_flag = SCF_DO_STCLASS_AND;
4038 } else /* XXXX Check for BOUND? */
4040 data.last_closep = &last_close;
4042 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4043 &data, SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4049 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4050 && data.last_start_min == 0 && data.last_end > 0
4051 && !RExC_seen_zerolen
4052 && (!(RExC_seen & REG_SEEN_GPOS) || (r->reganch & ROPT_ANCH_GPOS)))
4053 r->reganch |= ROPT_CHECK_ALL;
4054 scan_commit(pRExC_state, &data,&minlen);
4055 SvREFCNT_dec(data.last_found);
4057 /* Note that code very similar to this but for anchored string
4058 follows immediately below, changes may need to be made to both.
4061 longest_float_length = CHR_SVLEN(data.longest_float);
4062 if (longest_float_length
4063 || (data.flags & SF_FL_BEFORE_EOL
4064 && (!(data.flags & SF_FL_BEFORE_MEOL)
4065 || (RExC_flags & PMf_MULTILINE))))
4069 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4070 && data.offset_fixed == data.offset_float_min
4071 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4072 goto remove_float; /* As in (a)+. */
4074 /* copy the information about the longest float from the reg_scan_data
4075 over to the program. */
4076 if (SvUTF8(data.longest_float)) {
4077 r->float_utf8 = data.longest_float;
4078 r->float_substr = NULL;
4080 r->float_substr = data.longest_float;
4081 r->float_utf8 = NULL;
4083 /* float_end_shift is how many chars that must be matched that
4084 follow this item. We calculate it ahead of time as once the
4085 lookbehind offset is added in we lose the ability to correctly
4087 ml = data.minlen_float ? *(data.minlen_float)
4088 : (I32)longest_float_length;
4089 r->float_end_shift = ml - data.offset_float_min
4090 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4091 + data.lookbehind_float;
4092 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4093 r->float_max_offset = data.offset_float_max;
4094 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4095 r->float_max_offset -= data.lookbehind_float;
4097 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4098 && (!(data.flags & SF_FL_BEFORE_MEOL)
4099 || (RExC_flags & PMf_MULTILINE)));
4100 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4104 r->float_substr = r->float_utf8 = NULL;
4105 SvREFCNT_dec(data.longest_float);
4106 longest_float_length = 0;
4109 /* Note that code very similar to this but for floating string
4110 is immediately above, changes may need to be made to both.
4113 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4114 if (longest_fixed_length
4115 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4116 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4117 || (RExC_flags & PMf_MULTILINE))))
4121 /* copy the information about the longest fixed
4122 from the reg_scan_data over to the program. */
4123 if (SvUTF8(data.longest_fixed)) {
4124 r->anchored_utf8 = data.longest_fixed;
4125 r->anchored_substr = NULL;
4127 r->anchored_substr = data.longest_fixed;
4128 r->anchored_utf8 = NULL;
4130 /* fixed_end_shift is how many chars that must be matched that
4131 follow this item. We calculate it ahead of time as once the
4132 lookbehind offset is added in we lose the ability to correctly
4134 ml = data.minlen_fixed ? *(data.minlen_fixed)
4135 : (I32)longest_fixed_length;
4136 r->anchored_end_shift = ml - data.offset_fixed
4137 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4138 + data.lookbehind_fixed;
4139 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4141 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4142 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4143 || (RExC_flags & PMf_MULTILINE)));
4144 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4147 r->anchored_substr = r->anchored_utf8 = NULL;
4148 SvREFCNT_dec(data.longest_fixed);
4149 longest_fixed_length = 0;
4152 && (OP(r->regstclass) == REG_ANY || OP(r->regstclass) == SANY))
4153 r->regstclass = NULL;
4154 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4156 && !(data.start_class->flags & ANYOF_EOS)
4157 && !cl_is_anything(data.start_class))
4159 const I32 n = add_data(pRExC_state, 1, "f");
4161 Newx(RExC_rx->data->data[n], 1,
4162 struct regnode_charclass_class);
4163 StructCopy(data.start_class,
4164 (struct regnode_charclass_class*)RExC_rx->data->data[n],
4165 struct regnode_charclass_class);
4166 r->regstclass = (regnode*)RExC_rx->data->data[n];
4167 r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
4168 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4169 regprop(r, sv, (regnode*)data.start_class);
4170 PerlIO_printf(Perl_debug_log,
4171 "synthetic stclass \"%s\".\n",
4172 SvPVX_const(sv));});
4175 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4176 if (longest_fixed_length > longest_float_length) {
4177 r->check_end_shift = r->anchored_end_shift;
4178 r->check_substr = r->anchored_substr;
4179 r->check_utf8 = r->anchored_utf8;
4180 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4181 if (r->reganch & ROPT_ANCH_SINGLE)
4182 r->reganch |= ROPT_NOSCAN;
4185 r->check_end_shift = r->float_end_shift;
4186 r->check_substr = r->float_substr;
4187 r->check_utf8 = r->float_utf8;
4188 r->check_offset_min = r->float_min_offset;
4189 r->check_offset_max = r->float_max_offset;
4191 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4192 This should be changed ASAP! */
4193 if ((r->check_substr || r->check_utf8) && !(r->reganch & ROPT_ANCH_GPOS)) {
4194 r->reganch |= RE_USE_INTUIT;
4195 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4196 r->reganch |= RE_INTUIT_TAIL;
4198 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4199 if ( (STRLEN)minlen < longest_float_length )
4200 minlen= longest_float_length;
4201 if ( (STRLEN)minlen < longest_fixed_length )
4202 minlen= longest_fixed_length;
4206 /* Several toplevels. Best we can is to set minlen. */
4208 struct regnode_charclass_class ch_class;
4211 DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "\n"));
4213 scan = r->program + 1;
4214 cl_init(pRExC_state, &ch_class);
4215 data.start_class = &ch_class;
4216 data.last_closep = &last_close;
4218 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4219 &data, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4223 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4224 = r->float_substr = r->float_utf8 = NULL;
4225 if (!(data.start_class->flags & ANYOF_EOS)
4226 && !cl_is_anything(data.start_class))
4228 const I32 n = add_data(pRExC_state, 1, "f");
4230 Newx(RExC_rx->data->data[n], 1,
4231 struct regnode_charclass_class);
4232 StructCopy(data.start_class,
4233 (struct regnode_charclass_class*)RExC_rx->data->data[n],
4234 struct regnode_charclass_class);
4235 r->regstclass = (regnode*)RExC_rx->data->data[n];
4236 r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
4237 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4238 regprop(r, sv, (regnode*)data.start_class);
4239 PerlIO_printf(Perl_debug_log,
4240 "synthetic stclass \"%s\".\n",
4241 SvPVX_const(sv));});
4245 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4246 the "real" pattern. */
4247 if (r->minlen < minlen)
4250 if (RExC_seen & REG_SEEN_GPOS)
4251 r->reganch |= ROPT_GPOS_SEEN;
4252 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4253 r->reganch |= ROPT_LOOKBEHIND_SEEN;
4254 if (RExC_seen & REG_SEEN_EVAL)
4255 r->reganch |= ROPT_EVAL_SEEN;
4256 if (RExC_seen & REG_SEEN_CANY)
4257 r->reganch |= ROPT_CANY_SEEN;
4258 if (RExC_paren_names)
4259 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4261 r->paren_names = NULL;
4263 Newxz(r->startp, RExC_npar, I32);
4264 Newxz(r->endp, RExC_npar, I32);
4266 DEBUG_r( RX_DEBUG_on(r) );
4268 PerlIO_printf(Perl_debug_log,"Final program:\n");
4271 DEBUG_OFFSETS_r(if (r->offsets) {
4272 const U32 len = r->offsets[0];
4274 GET_RE_DEBUG_FLAGS_DECL;
4275 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)r->offsets[0]);
4276 for (i = 1; i <= len; i++) {
4277 if (r->offsets[i*2-1] || r->offsets[i*2])
4278 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4279 (UV)i, (UV)r->offsets[i*2-1], (UV)r->offsets[i*2]);
4281 PerlIO_printf(Perl_debug_log, "\n");
4287 #undef CORE_ONLY_BLOCK
4289 #undef RE_ENGINE_PTR
4292 Perl_reg_named_buff_sv(pTHX_ SV* namesv)
4294 I32 parno = 0; /* no match */
4296 const REGEXP * const rx = PM_GETRE(PL_curpm);
4297 if (rx && rx->paren_names) {
4298 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4301 SV* sv_dat=HeVAL(he_str);
4302 I32 *nums=(I32*)SvPVX(sv_dat);
4303 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4304 if ((I32)(rx->lastcloseparen) >= nums[i] &&
4305 rx->startp[nums[i]] != -1 &&
4306 rx->endp[nums[i]] != -1)
4319 SV *sv= sv_newmortal();
4320 Perl_sv_setpvf(aTHX_ sv, "%"IVdf,(IV)parno);
4321 gv_paren= Perl_gv_fetchsv(aTHX_ sv, GV_ADD, SVt_PVGV);
4322 return GvSVn(gv_paren);
4326 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4327 int rem=(int)(RExC_end - RExC_parse); \
4336 if (RExC_lastparse!=RExC_parse) \
4337 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4340 iscut ? "..." : "<" \
4343 PerlIO_printf(Perl_debug_log,"%16s",""); \
4348 num=REG_NODE_NUM(RExC_emit); \
4349 if (RExC_lastnum!=num) \
4350 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4352 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4353 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4354 (int)((depth*2)), "", \
4358 RExC_lastparse=RExC_parse; \
4363 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4364 DEBUG_PARSE_MSG((funcname)); \
4365 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4367 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4368 DEBUG_PARSE_MSG((funcname)); \
4369 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4372 - reg - regular expression, i.e. main body or parenthesized thing
4374 * Caller must absorb opening parenthesis.
4376 * Combining parenthesis handling with the base level of regular expression
4377 * is a trifle forced, but the need to tie the tails of the branches to what
4378 * follows makes it hard to avoid.
4380 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4382 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4384 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4388 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4389 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4392 register regnode *ret; /* Will be the head of the group. */
4393 register regnode *br;
4394 register regnode *lastbr;
4395 register regnode *ender = NULL;
4396 register I32 parno = 0;
4398 const I32 oregflags = RExC_flags;
4399 bool have_branch = 0;
4402 /* for (?g), (?gc), and (?o) warnings; warning
4403 about (?c) will warn about (?g) -- japhy */
4405 #define WASTED_O 0x01
4406 #define WASTED_G 0x02
4407 #define WASTED_C 0x04
4408 #define WASTED_GC (0x02|0x04)
4409 I32 wastedflags = 0x00;
4411 char * parse_start = RExC_parse; /* MJD */
4412 char * const oregcomp_parse = RExC_parse;
4414 GET_RE_DEBUG_FLAGS_DECL;
4415 DEBUG_PARSE("reg ");
4418 *flagp = 0; /* Tentatively. */
4421 /* Make an OPEN node, if parenthesized. */
4423 if (*RExC_parse == '?') { /* (?...) */
4424 U32 posflags = 0, negflags = 0;
4425 U32 *flagsp = &posflags;
4426 bool is_logical = 0;
4427 const char * const seqstart = RExC_parse;
4430 paren = *RExC_parse++;
4431 ret = NULL; /* For look-ahead/behind. */
4434 case '<': /* (?<...) */
4435 if (*RExC_parse == '!')
4437 else if (*RExC_parse != '=')
4441 case '\'': /* (?'...') */
4442 name_start= RExC_parse;
4445 while(isIDFIRST_uni(utf8n_to_uvchr((U8*)RExC_parse,
4446 RExC_end - RExC_parse,
4447 &numlen, UTF8_ALLOW_DEFAULT)))
4448 RExC_parse += numlen;
4450 while(isIDFIRST(*RExC_parse))
4453 if (RExC_parse == name_start)
4455 if (*RExC_parse != paren)
4456 vFAIL2("Sequence (?%c... not terminated",
4457 paren=='>' ? '<' : paren);
4459 SV *svname= Perl_newSVpvf(aTHX_ "%.*s",
4460 (int)(RExC_parse - name_start), name_start);
4464 if (!RExC_paren_names) {
4465 RExC_paren_names= newHV();
4466 sv_2mortal((SV*)RExC_paren_names);
4468 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
4470 sv_dat = HeVAL(he_str);
4472 /* croak baby croak */
4474 if (SvPOK(sv_dat)) {
4475 IV count=SvIV(sv_dat);
4476 I32 *pv=(I32*)SvGROW(sv_dat,SvCUR(sv_dat)+sizeof(I32)+1);
4477 SvCUR_set(sv_dat,SvCUR(sv_dat)+sizeof(I32));
4478 pv[count]=RExC_npar;
4481 (void)SvUPGRADE(sv_dat,SVt_PVNV);
4482 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
4486 /*sv_dump(sv_dat);*/
4488 nextchar(pRExC_state);
4490 goto capturing_parens;
4492 RExC_seen |= REG_SEEN_LOOKBEHIND;
4494 case '=': /* (?=...) */
4495 case '!': /* (?!...) */
4496 RExC_seen_zerolen++;
4497 case ':': /* (?:...) */
4498 case '>': /* (?>...) */
4500 case '$': /* (?$...) */
4501 case '@': /* (?@...) */
4502 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
4504 case '#': /* (?#...) */
4505 while (*RExC_parse && *RExC_parse != ')')
4507 if (*RExC_parse != ')')
4508 FAIL("Sequence (?#... not terminated");
4509 nextchar(pRExC_state);
4514 if (*RExC_parse != ')')
4515 FAIL("Sequence (?R) not terminated");
4516 reg_node(pRExC_state, SRECURSE);
4518 case '1': case '2': case '3': case '4': /* (?1) */
4519 case '5': case '6': case '7': case '8': case '9':
4522 const I32 num = atoi(RExC_parse);
4523 char * const parse_start = RExC_parse - 1; /* MJD */
4524 while (isDIGIT(*RExC_parse))
4526 if (*RExC_parse!=')')
4527 vFAIL("Expecting close bracket");
4528 ret = reganode(pRExC_state, RECURSE, num);
4530 if (num > (I32)RExC_rx->nparens) {
4532 vFAIL("Reference to nonexistent group");
4536 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
4537 "Recurse #%"UVuf" to %"IVdf"\n", ARG(ret), ARG2L(ret)));
4540 RExC_seen|=REG_SEEN_RECURSE;
4542 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
4543 Set_Node_Offset(ret, parse_start); /* MJD */
4545 nextchar(pRExC_state);
4548 case 'p': /* (?p...) */
4549 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
4550 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
4552 case '?': /* (??...) */
4554 if (*RExC_parse != '{')
4556 paren = *RExC_parse++;
4558 case '{': /* (?{...}) */
4560 I32 count = 1, n = 0;
4562 char *s = RExC_parse;
4564 RExC_seen_zerolen++;
4565 RExC_seen |= REG_SEEN_EVAL;
4566 while (count && (c = *RExC_parse)) {
4577 if (*RExC_parse != ')') {
4579 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
4583 OP_4tree *sop, *rop;
4584 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
4587 Perl_save_re_context(aTHX);
4588 rop = sv_compile_2op(sv, &sop, "re", &pad);
4589 sop->op_private |= OPpREFCOUNTED;
4590 /* re_dup will OpREFCNT_inc */
4591 OpREFCNT_set(sop, 1);
4594 n = add_data(pRExC_state, 3, "nop");
4595 RExC_rx->data->data[n] = (void*)rop;
4596 RExC_rx->data->data[n+1] = (void*)sop;
4597 RExC_rx->data->data[n+2] = (void*)pad;
4600 else { /* First pass */
4601 if (PL_reginterp_cnt < ++RExC_seen_evals
4603 /* No compiled RE interpolated, has runtime
4604 components ===> unsafe. */
4605 FAIL("Eval-group not allowed at runtime, use re 'eval'");
4606 if (PL_tainting && PL_tainted)
4607 FAIL("Eval-group in insecure regular expression");
4608 #if PERL_VERSION > 8
4609 if (IN_PERL_COMPILETIME)
4614 nextchar(pRExC_state);
4616 ret = reg_node(pRExC_state, LOGICAL);
4619 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
4620 /* deal with the length of this later - MJD */
4623 ret = reganode(pRExC_state, EVAL, n);
4624 Set_Node_Length(ret, RExC_parse - parse_start + 1);
4625 Set_Node_Offset(ret, parse_start);
4628 case '(': /* (?(?{...})...) and (?(?=...)...) */
4630 if (RExC_parse[0] == '?') { /* (?(?...)) */
4631 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
4632 || RExC_parse[1] == '<'
4633 || RExC_parse[1] == '{') { /* Lookahead or eval. */
4636 ret = reg_node(pRExC_state, LOGICAL);
4639 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
4643 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
4646 parno = atoi(RExC_parse++);
4648 while (isDIGIT(*RExC_parse))
4650 ret = reganode(pRExC_state, GROUPP, parno);
4652 if ((c = *nextchar(pRExC_state)) != ')')
4653 vFAIL("Switch condition not recognized");
4655 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
4656 br = regbranch(pRExC_state, &flags, 1,depth+1);
4658 br = reganode(pRExC_state, LONGJMP, 0);
4660 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
4661 c = *nextchar(pRExC_state);
4665 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
4666 regbranch(pRExC_state, &flags, 1,depth+1);
4667 REGTAIL(pRExC_state, ret, lastbr);
4670 c = *nextchar(pRExC_state);
4675 vFAIL("Switch (?(condition)... contains too many branches");
4676 ender = reg_node(pRExC_state, TAIL);
4677 REGTAIL(pRExC_state, br, ender);
4679 REGTAIL(pRExC_state, lastbr, ender);
4680 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
4683 REGTAIL(pRExC_state, ret, ender);
4687 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
4691 RExC_parse--; /* for vFAIL to print correctly */
4692 vFAIL("Sequence (? incomplete");
4696 parse_flags: /* (?i) */
4697 while (*RExC_parse && strchr("iogcmsx", *RExC_parse)) {
4698 /* (?g), (?gc) and (?o) are useless here
4699 and must be globally applied -- japhy */
4701 if (*RExC_parse == 'o' || *RExC_parse == 'g') {
4702 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
4703 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
4704 if (! (wastedflags & wflagbit) ) {
4705 wastedflags |= wflagbit;
4708 "Useless (%s%c) - %suse /%c modifier",
4709 flagsp == &negflags ? "?-" : "?",
4711 flagsp == &negflags ? "don't " : "",
4717 else if (*RExC_parse == 'c') {
4718 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
4719 if (! (wastedflags & WASTED_C) ) {
4720 wastedflags |= WASTED_GC;
4723 "Useless (%sc) - %suse /gc modifier",
4724 flagsp == &negflags ? "?-" : "?",
4725 flagsp == &negflags ? "don't " : ""
4730 else { pmflag(flagsp, *RExC_parse); }
4734 if (*RExC_parse == '-') {
4736 wastedflags = 0; /* reset so (?g-c) warns twice */
4740 RExC_flags |= posflags;
4741 RExC_flags &= ~negflags;
4742 if (*RExC_parse == ':') {
4748 if (*RExC_parse != ')') {
4750 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
4752 nextchar(pRExC_state);
4761 ret = reganode(pRExC_state, OPEN, parno);
4762 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
4763 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
4764 "Setting paren #%"IVdf" to %d\n", parno, REG_NODE_NUM(ret)));
4765 RExC_parens[parno-1]= ret;
4768 Set_Node_Length(ret, 1); /* MJD */
4769 Set_Node_Offset(ret, RExC_parse); /* MJD */
4776 /* Pick up the branches, linking them together. */
4777 parse_start = RExC_parse; /* MJD */
4778 br = regbranch(pRExC_state, &flags, 1,depth+1);
4779 /* branch_len = (paren != 0); */
4783 if (*RExC_parse == '|') {
4784 if (!SIZE_ONLY && RExC_extralen) {
4785 reginsert(pRExC_state, BRANCHJ, br, depth+1);
4788 reginsert(pRExC_state, BRANCH, br, depth+1);
4789 Set_Node_Length(br, paren != 0);
4790 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
4794 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
4796 else if (paren == ':') {
4797 *flagp |= flags&SIMPLE;
4799 if (is_open) { /* Starts with OPEN. */
4800 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
4802 else if (paren != '?') /* Not Conditional */
4804 *flagp |= flags & (SPSTART | HASWIDTH);
4806 while (*RExC_parse == '|') {
4807 if (!SIZE_ONLY && RExC_extralen) {
4808 ender = reganode(pRExC_state, LONGJMP,0);
4809 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
4812 RExC_extralen += 2; /* Account for LONGJMP. */
4813 nextchar(pRExC_state);
4814 br = regbranch(pRExC_state, &flags, 0, depth+1);
4818 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
4822 *flagp |= flags&SPSTART;
4825 if (have_branch || paren != ':') {
4826 /* Make a closing node, and hook it on the end. */
4829 ender = reg_node(pRExC_state, TAIL);
4832 ender = reganode(pRExC_state, CLOSE, parno);
4833 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
4834 Set_Node_Length(ender,1); /* MJD */
4840 *flagp &= ~HASWIDTH;
4843 ender = reg_node(pRExC_state, SUCCEED);
4846 ender = reg_node(pRExC_state, END);
4849 REGTAIL(pRExC_state, lastbr, ender);
4851 if (have_branch && !SIZE_ONLY) {
4853 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4855 /* Hook the tails of the branches to the closing node. */
4856 for (br = ret; br; br = regnext(br)) {
4857 const U8 op = PL_regkind[OP(br)];
4859 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
4861 else if (op == BRANCHJ) {
4862 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
4870 static const char parens[] = "=!<,>";
4872 if (paren && (p = strchr(parens, paren))) {
4873 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
4874 int flag = (p - parens) > 1;
4877 node = SUSPEND, flag = 0;
4878 reginsert(pRExC_state, node,ret, depth+1);
4879 Set_Node_Cur_Length(ret);
4880 Set_Node_Offset(ret, parse_start + 1);
4882 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
4886 /* Check for proper termination. */
4888 RExC_flags = oregflags;
4889 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
4890 RExC_parse = oregcomp_parse;
4891 vFAIL("Unmatched (");
4894 else if (!paren && RExC_parse < RExC_end) {
4895 if (*RExC_parse == ')') {
4897 vFAIL("Unmatched )");
4900 FAIL("Junk on end of regexp"); /* "Can't happen". */
4908 - regbranch - one alternative of an | operator
4910 * Implements the concatenation operator.
4913 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
4916 register regnode *ret;
4917 register regnode *chain = NULL;
4918 register regnode *latest;
4919 I32 flags = 0, c = 0;
4920 GET_RE_DEBUG_FLAGS_DECL;
4921 DEBUG_PARSE("brnc");
4925 if (!SIZE_ONLY && RExC_extralen)
4926 ret = reganode(pRExC_state, BRANCHJ,0);
4928 ret = reg_node(pRExC_state, BRANCH);
4929 Set_Node_Length(ret, 1);
4933 if (!first && SIZE_ONLY)
4934 RExC_extralen += 1; /* BRANCHJ */
4936 *flagp = WORST; /* Tentatively. */
4939 nextchar(pRExC_state);
4940 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
4942 latest = regpiece(pRExC_state, &flags,depth+1);
4943 if (latest == NULL) {
4944 if (flags & TRYAGAIN)
4948 else if (ret == NULL)
4950 *flagp |= flags&HASWIDTH;
4951 if (chain == NULL) /* First piece. */
4952 *flagp |= flags&SPSTART;
4955 REGTAIL(pRExC_state, chain, latest);
4960 if (chain == NULL) { /* Loop ran zero times. */
4961 chain = reg_node(pRExC_state, NOTHING);
4966 *flagp |= flags&SIMPLE;
4973 - regpiece - something followed by possible [*+?]
4975 * Note that the branching code sequences used for ? and the general cases
4976 * of * and + are somewhat optimized: they use the same NOTHING node as
4977 * both the endmarker for their branch list and the body of the last branch.
4978 * It might seem that this node could be dispensed with entirely, but the
4979 * endmarker role is not redundant.
4982 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
4985 register regnode *ret;
4987 register char *next;
4989 const char * const origparse = RExC_parse;
4991 I32 max = REG_INFTY;
4993 const char *maxpos = NULL;
4994 GET_RE_DEBUG_FLAGS_DECL;
4995 DEBUG_PARSE("piec");
4997 ret = regatom(pRExC_state, &flags,depth+1);
4999 if (flags & TRYAGAIN)
5006 if (op == '{' && regcurly(RExC_parse)) {
5008 parse_start = RExC_parse; /* MJD */
5009 next = RExC_parse + 1;
5010 while (isDIGIT(*next) || *next == ',') {
5019 if (*next == '}') { /* got one */
5023 min = atoi(RExC_parse);
5027 maxpos = RExC_parse;
5029 if (!max && *maxpos != '0')
5030 max = REG_INFTY; /* meaning "infinity" */
5031 else if (max >= REG_INFTY)
5032 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5034 nextchar(pRExC_state);
5037 if ((flags&SIMPLE)) {
5038 RExC_naughty += 2 + RExC_naughty / 2;
5039 reginsert(pRExC_state, CURLY, ret, depth+1);
5040 Set_Node_Offset(ret, parse_start+1); /* MJD */
5041 Set_Node_Cur_Length(ret);
5044 regnode * const w = reg_node(pRExC_state, WHILEM);
5047 REGTAIL(pRExC_state, ret, w);
5048 if (!SIZE_ONLY && RExC_extralen) {
5049 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5050 reginsert(pRExC_state, NOTHING,ret, depth+1);
5051 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5053 reginsert(pRExC_state, CURLYX,ret, depth+1);
5055 Set_Node_Offset(ret, parse_start+1);
5056 Set_Node_Length(ret,
5057 op == '{' ? (RExC_parse - parse_start) : 1);
5059 if (!SIZE_ONLY && RExC_extralen)
5060 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5061 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5063 RExC_whilem_seen++, RExC_extralen += 3;
5064 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5072 if (max && max < min)
5073 vFAIL("Can't do {n,m} with n > m");
5075 ARG1_SET(ret, (U16)min);
5076 ARG2_SET(ret, (U16)max);
5087 /* else if (OP(ret)==RECURSE) {
5089 vFAIL("Illegal quantifier on recursion group");
5092 #if 0 /* Now runtime fix should be reliable. */
5094 /* if this is reinstated, don't forget to put this back into perldiag:
5096 =item Regexp *+ operand could be empty at {#} in regex m/%s/
5098 (F) The part of the regexp subject to either the * or + quantifier
5099 could match an empty string. The {#} shows in the regular
5100 expression about where the problem was discovered.
5104 if (!(flags&HASWIDTH) && op != '?')
5105 vFAIL("Regexp *+ operand could be empty");
5108 parse_start = RExC_parse;
5109 nextchar(pRExC_state);
5111 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
5113 if (op == '*' && (flags&SIMPLE)) {
5114 reginsert(pRExC_state, STAR, ret, depth+1);
5118 else if (op == '*') {
5122 else if (op == '+' && (flags&SIMPLE)) {
5123 reginsert(pRExC_state, PLUS, ret, depth+1);
5127 else if (op == '+') {
5131 else if (op == '?') {
5136 if (!SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
5138 "%.*s matches null string many times",
5139 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
5143 if (*RExC_parse == '?') {
5144 nextchar(pRExC_state);
5145 reginsert(pRExC_state, MINMOD, ret, depth+1);
5146 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
5148 if (ISMULT2(RExC_parse)) {
5150 vFAIL("Nested quantifiers");
5157 /* reg_namedseq(pRExC_state,UVp)
5159 This is expected to be called by a parser routine that has
5160 recognized'\N' and needs to handle the rest. RExC_parse is
5161 expected to point at the first char following the N at the time
5164 If valuep is non-null then it is assumed that we are parsing inside
5165 of a charclass definition and the first codepoint in the resolved
5166 string is returned via *valuep and the routine will return NULL.
5167 In this mode if a multichar string is returned from the charnames
5168 handler a warning will be issued, and only the first char in the
5169 sequence will be examined. If the string returned is zero length
5170 then the value of *valuep is undefined and NON-NULL will
5171 be returned to indicate failure. (This will NOT be a valid pointer
5174 If value is null then it is assumed that we are parsing normal text
5175 and inserts a new EXACT node into the program containing the resolved
5176 string and returns a pointer to the new node. If the string is
5177 zerolength a NOTHING node is emitted.
5179 On success RExC_parse is set to the char following the endbrace.
5180 Parsing failures will generate a fatal errorvia vFAIL(...)
5182 NOTE: We cache all results from the charnames handler locally in
5183 the RExC_charnames hash (created on first use) to prevent a charnames
5184 handler from playing silly-buggers and returning a short string and
5185 then a long string for a given pattern. Since the regexp program
5186 size is calculated during an initial parse this would result
5187 in a buffer overrun so we cache to prevent the charname result from
5188 changing during the course of the parse.
5192 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
5194 char * name; /* start of the content of the name */
5195 char * endbrace; /* endbrace following the name */
5198 STRLEN len; /* this has various purposes throughout the code */
5199 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
5200 regnode *ret = NULL;
5202 if (*RExC_parse != '{') {
5203 vFAIL("Missing braces on \\N{}");
5205 name = RExC_parse+1;
5206 endbrace = strchr(RExC_parse, '}');
5209 vFAIL("Missing right brace on \\N{}");
5211 RExC_parse = endbrace + 1;
5214 /* RExC_parse points at the beginning brace,
5215 endbrace points at the last */
5216 if ( name[0]=='U' && name[1]=='+' ) {
5217 /* its a "unicode hex" notation {U+89AB} */
5218 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
5219 | PERL_SCAN_DISALLOW_PREFIX
5220 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
5222 len = (STRLEN)(endbrace - name - 2);
5223 cp = grok_hex(name + 2, &len, &fl, NULL);
5224 if ( len != (STRLEN)(endbrace - name - 2) ) {
5233 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
5235 /* fetch the charnames handler for this scope */
5236 HV * const table = GvHV(PL_hintgv);
5238 hv_fetchs(table, "charnames", FALSE) :
5240 SV *cv= cvp ? *cvp : NULL;
5243 /* create an SV with the name as argument */
5244 sv_name = newSVpvn(name, endbrace - name);
5246 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
5247 vFAIL2("Constant(\\N{%s}) unknown: "
5248 "(possibly a missing \"use charnames ...\")",
5251 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
5252 vFAIL2("Constant(\\N{%s}): "
5253 "$^H{charnames} is not defined",SvPVX(sv_name));
5258 if (!RExC_charnames) {
5259 /* make sure our cache is allocated */
5260 RExC_charnames = newHV();
5261 sv_2mortal((SV*)RExC_charnames);
5263 /* see if we have looked this one up before */
5264 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
5266 sv_str = HeVAL(he_str);
5279 count= call_sv(cv, G_SCALAR);
5281 if (count == 1) { /* XXXX is this right? dmq */
5283 SvREFCNT_inc_simple_void(sv_str);
5291 if ( !sv_str || !SvOK(sv_str) ) {
5292 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
5293 "did not return a defined value",SvPVX(sv_name));
5295 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
5300 char *p = SvPV(sv_str, len);
5303 if ( SvUTF8(sv_str) ) {
5304 *valuep = utf8_to_uvchr((U8*)p, &numlen);
5308 We have to turn on utf8 for high bit chars otherwise
5309 we get failures with
5311 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5312 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5314 This is different from what \x{} would do with the same
5315 codepoint, where the condition is > 0xFF.
5322 /* warn if we havent used the whole string? */
5324 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5326 "Ignoring excess chars from \\N{%s} in character class",
5330 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5332 "Ignoring zero length \\N{%s} in character class",
5337 SvREFCNT_dec(sv_name);
5339 SvREFCNT_dec(sv_str);
5340 return len ? NULL : (regnode *)&len;
5341 } else if(SvCUR(sv_str)) {
5346 char * parse_start = name-3; /* needed for the offsets */
5347 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
5349 ret = reg_node(pRExC_state,
5350 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
5353 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
5354 sv_utf8_upgrade(sv_str);
5355 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
5359 p = SvPV(sv_str, len);
5361 /* len is the length written, charlen is the size the char read */
5362 for ( len = 0; p < pend; p += charlen ) {
5364 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
5366 STRLEN foldlen,numlen;
5367 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
5368 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
5369 /* Emit all the Unicode characters. */
5371 for (foldbuf = tmpbuf;
5375 uvc = utf8_to_uvchr(foldbuf, &numlen);
5377 const STRLEN unilen = reguni(pRExC_state, uvc, s);
5380 /* In EBCDIC the numlen
5381 * and unilen can differ. */
5383 if (numlen >= foldlen)
5387 break; /* "Can't happen." */
5390 const STRLEN unilen = reguni(pRExC_state, uvc, s);
5402 RExC_size += STR_SZ(len);
5405 RExC_emit += STR_SZ(len);
5407 Set_Node_Cur_Length(ret); /* MJD */
5409 nextchar(pRExC_state);
5411 ret = reg_node(pRExC_state,NOTHING);
5414 SvREFCNT_dec(sv_str);
5417 SvREFCNT_dec(sv_name);
5426 - regatom - the lowest level
5428 * Optimization: gobbles an entire sequence of ordinary characters so that
5429 * it can turn them into a single node, which is smaller to store and
5430 * faster to run. Backslashed characters are exceptions, each becoming a
5431 * separate node; the code is simpler that way and it's not worth fixing.
5433 * [Yes, it is worth fixing, some scripts can run twice the speed.]
5434 * [It looks like its ok, as in S_study_chunk we merge adjacent EXACT nodes]
5437 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5440 register regnode *ret = NULL;
5442 char *parse_start = RExC_parse;
5443 GET_RE_DEBUG_FLAGS_DECL;
5444 DEBUG_PARSE("atom");
5445 *flagp = WORST; /* Tentatively. */
5448 switch (*RExC_parse) {
5450 RExC_seen_zerolen++;
5451 nextchar(pRExC_state);
5452 if (RExC_flags & PMf_MULTILINE)
5453 ret = reg_node(pRExC_state, MBOL);
5454 else if (RExC_flags & PMf_SINGLELINE)
5455 ret = reg_node(pRExC_state, SBOL);
5457 ret = reg_node(pRExC_state, BOL);
5458 Set_Node_Length(ret, 1); /* MJD */
5461 nextchar(pRExC_state);
5463 RExC_seen_zerolen++;
5464 if (RExC_flags & PMf_MULTILINE)
5465 ret = reg_node(pRExC_state, MEOL);
5466 else if (RExC_flags & PMf_SINGLELINE)
5467 ret = reg_node(pRExC_state, SEOL);
5469 ret = reg_node(pRExC_state, EOL);
5470 Set_Node_Length(ret, 1); /* MJD */
5473 nextchar(pRExC_state);
5474 if (RExC_flags & PMf_SINGLELINE)
5475 ret = reg_node(pRExC_state, SANY);
5477 ret = reg_node(pRExC_state, REG_ANY);
5478 *flagp |= HASWIDTH|SIMPLE;
5480 Set_Node_Length(ret, 1); /* MJD */
5484 char * const oregcomp_parse = ++RExC_parse;
5485 ret = regclass(pRExC_state,depth+1);
5486 if (*RExC_parse != ']') {
5487 RExC_parse = oregcomp_parse;
5488 vFAIL("Unmatched [");
5490 nextchar(pRExC_state);
5491 *flagp |= HASWIDTH|SIMPLE;
5492 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
5496 nextchar(pRExC_state);
5497 ret = reg(pRExC_state, 1, &flags,depth+1);
5499 if (flags & TRYAGAIN) {
5500 if (RExC_parse == RExC_end) {
5501 /* Make parent create an empty node if needed. */
5509 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
5513 if (flags & TRYAGAIN) {
5517 vFAIL("Internal urp");
5518 /* Supposed to be caught earlier. */
5521 if (!regcurly(RExC_parse)) {
5530 vFAIL("Quantifier follows nothing");
5533 switch (*++RExC_parse) {
5535 RExC_seen_zerolen++;
5536 ret = reg_node(pRExC_state, SBOL);
5538 nextchar(pRExC_state);
5539 Set_Node_Length(ret, 2); /* MJD */
5542 ret = reg_node(pRExC_state, GPOS);
5543 RExC_seen |= REG_SEEN_GPOS;
5545 nextchar(pRExC_state);
5546 Set_Node_Length(ret, 2); /* MJD */
5549 ret = reg_node(pRExC_state, SEOL);
5551 RExC_seen_zerolen++; /* Do not optimize RE away */
5552 nextchar(pRExC_state);
5555 ret = reg_node(pRExC_state, EOS);
5557 RExC_seen_zerolen++; /* Do not optimize RE away */
5558 nextchar(pRExC_state);
5559 Set_Node_Length(ret, 2); /* MJD */
5562 ret = reg_node(pRExC_state, CANY);
5563 RExC_seen |= REG_SEEN_CANY;
5564 *flagp |= HASWIDTH|SIMPLE;
5565 nextchar(pRExC_state);
5566 Set_Node_Length(ret, 2); /* MJD */
5569 ret = reg_node(pRExC_state, CLUMP);
5571 nextchar(pRExC_state);
5572 Set_Node_Length(ret, 2); /* MJD */
5575 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
5576 *flagp |= HASWIDTH|SIMPLE;
5577 nextchar(pRExC_state);
5578 Set_Node_Length(ret, 2); /* MJD */
5581 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
5582 *flagp |= HASWIDTH|SIMPLE;
5583 nextchar(pRExC_state);
5584 Set_Node_Length(ret, 2); /* MJD */
5587 RExC_seen_zerolen++;
5588 RExC_seen |= REG_SEEN_LOOKBEHIND;
5589 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
5591 nextchar(pRExC_state);
5592 Set_Node_Length(ret, 2); /* MJD */
5595 RExC_seen_zerolen++;
5596 RExC_seen |= REG_SEEN_LOOKBEHIND;
5597 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
5599 nextchar(pRExC_state);
5600 Set_Node_Length(ret, 2); /* MJD */
5603 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
5604 *flagp |= HASWIDTH|SIMPLE;
5605 nextchar(pRExC_state);
5606 Set_Node_Length(ret, 2); /* MJD */
5609 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
5610 *flagp |= HASWIDTH|SIMPLE;
5611 nextchar(pRExC_state);
5612 Set_Node_Length(ret, 2); /* MJD */
5615 ret = reg_node(pRExC_state, DIGIT);
5616 *flagp |= HASWIDTH|SIMPLE;
5617 nextchar(pRExC_state);
5618 Set_Node_Length(ret, 2); /* MJD */
5621 ret = reg_node(pRExC_state, NDIGIT);
5622 *flagp |= HASWIDTH|SIMPLE;
5623 nextchar(pRExC_state);
5624 Set_Node_Length(ret, 2); /* MJD */
5629 char* const oldregxend = RExC_end;
5630 char* parse_start = RExC_parse - 2;
5632 if (RExC_parse[1] == '{') {
5633 /* a lovely hack--pretend we saw [\pX] instead */
5634 RExC_end = strchr(RExC_parse, '}');
5636 const U8 c = (U8)*RExC_parse;
5638 RExC_end = oldregxend;
5639 vFAIL2("Missing right brace on \\%c{}", c);
5644 RExC_end = RExC_parse + 2;
5645 if (RExC_end > oldregxend)
5646 RExC_end = oldregxend;
5650 ret = regclass(pRExC_state,depth+1);
5652 RExC_end = oldregxend;
5655 Set_Node_Offset(ret, parse_start + 2);
5656 Set_Node_Cur_Length(ret);
5657 nextchar(pRExC_state);
5658 *flagp |= HASWIDTH|SIMPLE;
5662 /* Handle \N{NAME} here and not below because it can be
5663 multicharacter. join_exact() will join them up later on.
5664 Also this makes sure that things like /\N{BLAH}+/ and
5665 \N{BLAH} being multi char Just Happen. dmq*/
5667 ret= reg_namedseq(pRExC_state, NULL);
5671 char ch= RExC_parse[1];
5672 if (ch != '<' && ch != '\'') {
5674 vWARN( RExC_parse + 1,
5675 "Possible broken named back reference treated as literal k");
5679 char* name_start = (RExC_parse += 2);
5681 ch= (ch == '<') ? '>' : '\'';
5685 while(isIDFIRST_uni(utf8n_to_uvchr((U8*)RExC_parse,
5686 RExC_end - RExC_parse,
5687 &numlen, UTF8_ALLOW_DEFAULT)))
5688 RExC_parse += numlen;
5690 while(isIDFIRST(*RExC_parse))
5693 if (RExC_parse == name_start || *RExC_parse != ch)
5694 vFAIL2("Sequence \\k%c... not terminated",
5695 (ch == '>' ? '<' : ch));
5698 ret = reganode(pRExC_state,
5699 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5705 SV *svname = Perl_newSVpvf(aTHX_ "%.*s",
5706 (int)(RExC_parse - name_start), name_start);
5711 he_str = hv_fetch_ent( RExC_paren_names, svname, 0, 0 );
5712 SvREFCNT_dec(svname);
5714 sv_dat = HeVAL(he_str);
5716 vFAIL("Reference to nonexistent group");
5718 num = add_data( pRExC_state, 1, "S" );
5720 RExC_rx->data->data[num]=(void*)sv_dat;
5721 SvREFCNT_inc(sv_dat);
5723 /* override incorrect value set in reganode MJD */
5724 Set_Node_Offset(ret, parse_start+1);
5725 Set_Node_Cur_Length(ret); /* MJD */
5726 nextchar(pRExC_state);
5741 case '1': case '2': case '3': case '4':
5742 case '5': case '6': case '7': case '8': case '9':
5744 const I32 num = atoi(RExC_parse);
5746 if (num > 9 && num >= RExC_npar)
5749 char * const parse_start = RExC_parse - 1; /* MJD */
5750 while (isDIGIT(*RExC_parse))
5753 if (!SIZE_ONLY && num > (I32)RExC_rx->nparens)
5754 vFAIL("Reference to nonexistent group");
5756 ret = reganode(pRExC_state,
5757 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
5761 /* override incorrect value set in reganode MJD */
5762 Set_Node_Offset(ret, parse_start+1);
5763 Set_Node_Cur_Length(ret); /* MJD */
5765 nextchar(pRExC_state);
5770 if (RExC_parse >= RExC_end)
5771 FAIL("Trailing \\");
5774 /* Do not generate "unrecognized" warnings here, we fall
5775 back into the quick-grab loop below */
5782 if (RExC_flags & PMf_EXTENDED) {
5783 while (RExC_parse < RExC_end && *RExC_parse != '\n')
5785 if (RExC_parse < RExC_end)
5791 register STRLEN len;
5796 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
5798 parse_start = RExC_parse - 1;
5804 ret = reg_node(pRExC_state,
5805 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
5807 for (len = 0, p = RExC_parse - 1;
5808 len < 127 && p < RExC_end;
5811 char * const oldp = p;
5813 if (RExC_flags & PMf_EXTENDED)
5814 p = regwhite(p, RExC_end);
5862 ender = ASCII_TO_NATIVE('\033');
5866 ender = ASCII_TO_NATIVE('\007');
5871 char* const e = strchr(p, '}');
5875 vFAIL("Missing right brace on \\x{}");
5878 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
5879 | PERL_SCAN_DISALLOW_PREFIX;
5880 STRLEN numlen = e - p - 1;
5881 ender = grok_hex(p + 1, &numlen, &flags, NULL);
5888 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
5890 ender = grok_hex(p, &numlen, &flags, NULL);
5896 ender = UCHARAT(p++);
5897 ender = toCTRL(ender);
5899 case '0': case '1': case '2': case '3':case '4':
5900 case '5': case '6': case '7': case '8':case '9':
5902 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
5905 ender = grok_oct(p, &numlen, &flags, NULL);
5915 FAIL("Trailing \\");
5918 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
5919 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
5920 goto normal_default;
5925 if (UTF8_IS_START(*p) && UTF) {
5927 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
5928 &numlen, UTF8_ALLOW_DEFAULT);
5935 if (RExC_flags & PMf_EXTENDED)
5936 p = regwhite(p, RExC_end);
5938 /* Prime the casefolded buffer. */
5939 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
5941 if (ISMULT2(p)) { /* Back off on ?+*. */
5946 /* Emit all the Unicode characters. */
5948 for (foldbuf = tmpbuf;
5950 foldlen -= numlen) {
5951 ender = utf8_to_uvchr(foldbuf, &numlen);
5953 const STRLEN unilen = reguni(pRExC_state, ender, s);
5956 /* In EBCDIC the numlen
5957 * and unilen can differ. */
5959 if (numlen >= foldlen)
5963 break; /* "Can't happen." */
5967 const STRLEN unilen = reguni(pRExC_state, ender, s);
5976 REGC((char)ender, s++);
5982 /* Emit all the Unicode characters. */
5984 for (foldbuf = tmpbuf;
5986 foldlen -= numlen) {
5987 ender = utf8_to_uvchr(foldbuf, &numlen);
5989 const STRLEN unilen = reguni(pRExC_state, ender, s);
5992 /* In EBCDIC the numlen
5993 * and unilen can differ. */
5995 if (numlen >= foldlen)
6003 const STRLEN unilen = reguni(pRExC_state, ender, s);
6012 REGC((char)ender, s++);
6016 Set_Node_Cur_Length(ret); /* MJD */
6017 nextchar(pRExC_state);
6019 /* len is STRLEN which is unsigned, need to copy to signed */
6022 vFAIL("Internal disaster");
6026 if (len == 1 && UNI_IS_INVARIANT(ender))
6030 RExC_size += STR_SZ(len);
6033 RExC_emit += STR_SZ(len);
6039 /* If the encoding pragma is in effect recode the text of
6040 * any EXACT-kind nodes. */
6041 if (ret && PL_encoding && PL_regkind[OP(ret)] == EXACT) {
6042 const STRLEN oldlen = STR_LEN(ret);
6043 SV * const sv = sv_2mortal(newSVpvn(STRING(ret), oldlen));
6047 if (sv_utf8_downgrade(sv, TRUE)) {
6048 const char * const s = sv_recode_to_utf8(sv, PL_encoding);
6049 const STRLEN newlen = SvCUR(sv);
6054 GET_RE_DEBUG_FLAGS_DECL;
6055 DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "recode %*s to %*s\n",
6056 (int)oldlen, STRING(ret),
6058 Copy(s, STRING(ret), newlen, char);
6059 STR_LEN(ret) += newlen - oldlen;
6060 RExC_emit += STR_SZ(newlen) - STR_SZ(oldlen);
6062 RExC_size += STR_SZ(newlen) - STR_SZ(oldlen);
6070 S_regwhite(char *p, const char *e)
6075 else if (*p == '#') {
6078 } while (p < e && *p != '\n');
6086 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
6087 Character classes ([:foo:]) can also be negated ([:^foo:]).
6088 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
6089 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
6090 but trigger failures because they are currently unimplemented. */
6092 #define POSIXCC_DONE(c) ((c) == ':')
6093 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
6094 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
6097 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
6100 I32 namedclass = OOB_NAMEDCLASS;
6102 if (value == '[' && RExC_parse + 1 < RExC_end &&
6103 /* I smell either [: or [= or [. -- POSIX has been here, right? */
6104 POSIXCC(UCHARAT(RExC_parse))) {
6105 const char c = UCHARAT(RExC_parse);
6106 char* const s = RExC_parse++;
6108 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
6110 if (RExC_parse == RExC_end)
6111 /* Grandfather lone [:, [=, [. */
6114 const char* const t = RExC_parse++; /* skip over the c */
6117 if (UCHARAT(RExC_parse) == ']') {
6118 const char *posixcc = s + 1;
6119 RExC_parse++; /* skip over the ending ] */
6122 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
6123 const I32 skip = t - posixcc;
6125 /* Initially switch on the length of the name. */
6128 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
6129 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
6132 /* Names all of length 5. */
6133 /* alnum alpha ascii blank cntrl digit graph lower
6134 print punct space upper */
6135 /* Offset 4 gives the best switch position. */
6136 switch (posixcc[4]) {
6138 if (memEQ(posixcc, "alph", 4)) /* alpha */
6139 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
6142 if (memEQ(posixcc, "spac", 4)) /* space */
6143 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
6146 if (memEQ(posixcc, "grap", 4)) /* graph */
6147 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
6150 if (memEQ(posixcc, "asci", 4)) /* ascii */
6151 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
6154 if (memEQ(posixcc, "blan", 4)) /* blank */
6155 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
6158 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
6159 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
6162 if (memEQ(posixcc, "alnu", 4)) /* alnum */
6163 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
6166 if (memEQ(posixcc, "lowe", 4)) /* lower */
6167 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
6168 else if (memEQ(posixcc, "uppe", 4)) /* upper */
6169 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
6172 if (memEQ(posixcc, "digi", 4)) /* digit */
6173 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
6174 else if (memEQ(posixcc, "prin", 4)) /* print */
6175 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
6176 else if (memEQ(posixcc, "punc", 4)) /* punct */
6177 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
6182 if (memEQ(posixcc, "xdigit", 6))
6183 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
6187 if (namedclass == OOB_NAMEDCLASS)
6188 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
6190 assert (posixcc[skip] == ':');
6191 assert (posixcc[skip+1] == ']');
6192 } else if (!SIZE_ONLY) {
6193 /* [[=foo=]] and [[.foo.]] are still future. */
6195 /* adjust RExC_parse so the warning shows after
6197 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
6199 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6202 /* Maternal grandfather:
6203 * "[:" ending in ":" but not in ":]" */
6213 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
6216 if (POSIXCC(UCHARAT(RExC_parse))) {
6217 const char *s = RExC_parse;
6218 const char c = *s++;
6222 if (*s && c == *s && s[1] == ']') {
6223 if (ckWARN(WARN_REGEXP))
6225 "POSIX syntax [%c %c] belongs inside character classes",
6228 /* [[=foo=]] and [[.foo.]] are still future. */
6229 if (POSIXCC_NOTYET(c)) {
6230 /* adjust RExC_parse so the error shows after
6232 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
6234 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6242 parse a class specification and produce either an ANYOF node that
6243 matches the pattern. If the pattern matches a single char only and
6244 that char is < 256 then we produce an EXACT node instead.
6247 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
6250 register UV value = 0;
6251 register UV nextvalue;
6252 register IV prevvalue = OOB_UNICODE;
6253 register IV range = 0;
6254 register regnode *ret;
6257 char *rangebegin = NULL;
6258 bool need_class = 0;
6261 bool optimize_invert = TRUE;
6262 AV* unicode_alternate = NULL;
6264 UV literal_endpoint = 0;
6266 UV stored = 0; /* number of chars stored in the class */
6268 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
6269 case we need to change the emitted regop to an EXACT. */
6270 const char * orig_parse = RExC_parse;
6271 GET_RE_DEBUG_FLAGS_DECL;
6273 PERL_UNUSED_ARG(depth);
6276 DEBUG_PARSE("clas");
6278 /* Assume we are going to generate an ANYOF node. */
6279 ret = reganode(pRExC_state, ANYOF, 0);
6282 ANYOF_FLAGS(ret) = 0;
6284 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
6288 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
6292 RExC_size += ANYOF_SKIP;
6293 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
6296 RExC_emit += ANYOF_SKIP;
6298 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
6300 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
6301 ANYOF_BITMAP_ZERO(ret);
6302 listsv = newSVpvs("# comment\n");
6305 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6307 if (!SIZE_ONLY && POSIXCC(nextvalue))
6308 checkposixcc(pRExC_state);
6310 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
6311 if (UCHARAT(RExC_parse) == ']')
6315 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
6319 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
6322 rangebegin = RExC_parse;
6324 value = utf8n_to_uvchr((U8*)RExC_parse,
6325 RExC_end - RExC_parse,
6326 &numlen, UTF8_ALLOW_DEFAULT);
6327 RExC_parse += numlen;
6330 value = UCHARAT(RExC_parse++);
6332 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6333 if (value == '[' && POSIXCC(nextvalue))
6334 namedclass = regpposixcc(pRExC_state, value);
6335 else if (value == '\\') {
6337 value = utf8n_to_uvchr((U8*)RExC_parse,
6338 RExC_end - RExC_parse,
6339 &numlen, UTF8_ALLOW_DEFAULT);
6340 RExC_parse += numlen;
6343 value = UCHARAT(RExC_parse++);
6344 /* Some compilers cannot handle switching on 64-bit integer
6345 * values, therefore value cannot be an UV. Yes, this will
6346 * be a problem later if we want switch on Unicode.
6347 * A similar issue a little bit later when switching on
6348 * namedclass. --jhi */
6349 switch ((I32)value) {
6350 case 'w': namedclass = ANYOF_ALNUM; break;
6351 case 'W': namedclass = ANYOF_NALNUM; break;
6352 case 's': namedclass = ANYOF_SPACE; break;
6353 case 'S': namedclass = ANYOF_NSPACE; break;
6354 case 'd': namedclass = ANYOF_DIGIT; break;
6355 case 'D': namedclass = ANYOF_NDIGIT; break;
6356 case 'N': /* Handle \N{NAME} in class */
6358 /* We only pay attention to the first char of
6359 multichar strings being returned. I kinda wonder
6360 if this makes sense as it does change the behaviour
6361 from earlier versions, OTOH that behaviour was broken
6363 UV v; /* value is register so we cant & it /grrr */
6364 if (reg_namedseq(pRExC_state, &v)) {
6374 if (RExC_parse >= RExC_end)
6375 vFAIL2("Empty \\%c{}", (U8)value);
6376 if (*RExC_parse == '{') {
6377 const U8 c = (U8)value;
6378 e = strchr(RExC_parse++, '}');
6380 vFAIL2("Missing right brace on \\%c{}", c);
6381 while (isSPACE(UCHARAT(RExC_parse)))
6383 if (e == RExC_parse)
6384 vFAIL2("Empty \\%c{}", c);
6386 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
6394 if (UCHARAT(RExC_parse) == '^') {
6397 value = value == 'p' ? 'P' : 'p'; /* toggle */
6398 while (isSPACE(UCHARAT(RExC_parse))) {
6403 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
6404 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
6407 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
6408 namedclass = ANYOF_MAX; /* no official name, but it's named */
6411 case 'n': value = '\n'; break;
6412 case 'r': value = '\r'; break;
6413 case 't': value = '\t'; break;
6414 case 'f': value = '\f'; break;
6415 case 'b': value = '\b'; break;
6416 case 'e': value = ASCII_TO_NATIVE('\033');break;
6417 case 'a': value = ASCII_TO_NATIVE('\007');break;
6419 if (*RExC_parse == '{') {
6420 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6421 | PERL_SCAN_DISALLOW_PREFIX;
6422 char * const e = strchr(RExC_parse++, '}');
6424 vFAIL("Missing right brace on \\x{}");
6426 numlen = e - RExC_parse;
6427 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
6431 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6433 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
6434 RExC_parse += numlen;
6438 value = UCHARAT(RExC_parse++);
6439 value = toCTRL(value);
6441 case '0': case '1': case '2': case '3': case '4':
6442 case '5': case '6': case '7': case '8': case '9':
6446 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
6447 RExC_parse += numlen;
6451 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
6453 "Unrecognized escape \\%c in character class passed through",
6457 } /* end of \blah */
6463 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
6465 if (!SIZE_ONLY && !need_class)
6466 ANYOF_CLASS_ZERO(ret);
6470 /* a bad range like a-\d, a-[:digit:] ? */
6473 if (ckWARN(WARN_REGEXP)) {
6475 RExC_parse >= rangebegin ?
6476 RExC_parse - rangebegin : 0;
6478 "False [] range \"%*.*s\"",
6481 if (prevvalue < 256) {
6482 ANYOF_BITMAP_SET(ret, prevvalue);
6483 ANYOF_BITMAP_SET(ret, '-');
6486 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
6487 Perl_sv_catpvf(aTHX_ listsv,
6488 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
6492 range = 0; /* this was not a true range */
6496 const char *what = NULL;
6499 if (namedclass > OOB_NAMEDCLASS)
6500 optimize_invert = FALSE;
6501 /* Possible truncation here but in some 64-bit environments
6502 * the compiler gets heartburn about switch on 64-bit values.
6503 * A similar issue a little earlier when switching on value.
6505 switch ((I32)namedclass) {
6508 ANYOF_CLASS_SET(ret, ANYOF_ALNUM);
6510 for (value = 0; value < 256; value++)
6512 ANYOF_BITMAP_SET(ret, value);
6519 ANYOF_CLASS_SET(ret, ANYOF_NALNUM);
6521 for (value = 0; value < 256; value++)
6522 if (!isALNUM(value))
6523 ANYOF_BITMAP_SET(ret, value);
6530 ANYOF_CLASS_SET(ret, ANYOF_ALNUMC);
6532 for (value = 0; value < 256; value++)
6533 if (isALNUMC(value))
6534 ANYOF_BITMAP_SET(ret, value);
6541 ANYOF_CLASS_SET(ret, ANYOF_NALNUMC);
6543 for (value = 0; value < 256; value++)
6544 if (!isALNUMC(value))
6545 ANYOF_BITMAP_SET(ret, value);
6552 ANYOF_CLASS_SET(ret, ANYOF_ALPHA);
6554 for (value = 0; value < 256; value++)
6556 ANYOF_BITMAP_SET(ret, value);
6563 ANYOF_CLASS_SET(ret, ANYOF_NALPHA);
6565 for (value = 0; value < 256; value++)
6566 if (!isALPHA(value))
6567 ANYOF_BITMAP_SET(ret, value);
6574 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
6577 for (value = 0; value < 128; value++)
6578 ANYOF_BITMAP_SET(ret, value);
6580 for (value = 0; value < 256; value++) {
6582 ANYOF_BITMAP_SET(ret, value);
6591 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
6594 for (value = 128; value < 256; value++)
6595 ANYOF_BITMAP_SET(ret, value);
6597 for (value = 0; value < 256; value++) {
6598 if (!isASCII(value))
6599 ANYOF_BITMAP_SET(ret, value);
6608 ANYOF_CLASS_SET(ret, ANYOF_BLANK);
6610 for (value = 0; value < 256; value++)
6612 ANYOF_BITMAP_SET(ret, value);
6619 ANYOF_CLASS_SET(ret, ANYOF_NBLANK);
6621 for (value = 0; value < 256; value++)
6622 if (!isBLANK(value))
6623 ANYOF_BITMAP_SET(ret, value);
6630 ANYOF_CLASS_SET(ret, ANYOF_CNTRL);
6632 for (value = 0; value < 256; value++)
6634 ANYOF_BITMAP_SET(ret, value);
6641 ANYOF_CLASS_SET(ret, ANYOF_NCNTRL);
6643 for (value = 0; value < 256; value++)
6644 if (!isCNTRL(value))
6645 ANYOF_BITMAP_SET(ret, value);
6652 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
6654 /* consecutive digits assumed */
6655 for (value = '0'; value <= '9'; value++)
6656 ANYOF_BITMAP_SET(ret, value);
6663 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
6665 /* consecutive digits assumed */
6666 for (value = 0; value < '0'; value++)
6667 ANYOF_BITMAP_SET(ret, value);
6668 for (value = '9' + 1; value < 256; value++)
6669 ANYOF_BITMAP_SET(ret, value);
6676 ANYOF_CLASS_SET(ret, ANYOF_GRAPH);
6678 for (value = 0; value < 256; value++)
6680 ANYOF_BITMAP_SET(ret, value);
6687 ANYOF_CLASS_SET(ret, ANYOF_NGRAPH);
6689 for (value = 0; value < 256; value++)
6690 if (!isGRAPH(value))
6691 ANYOF_BITMAP_SET(ret, value);
6698 ANYOF_CLASS_SET(ret, ANYOF_LOWER);
6700 for (value = 0; value < 256; value++)
6702 ANYOF_BITMAP_SET(ret, value);
6709 ANYOF_CLASS_SET(ret, ANYOF_NLOWER);
6711 for (value = 0; value < 256; value++)
6712 if (!isLOWER(value))
6713 ANYOF_BITMAP_SET(ret, value);
6720 ANYOF_CLASS_SET(ret, ANYOF_PRINT);
6722 for (value = 0; value < 256; value++)
6724 ANYOF_BITMAP_SET(ret, value);
6731 ANYOF_CLASS_SET(ret, ANYOF_NPRINT);
6733 for (value = 0; value < 256; value++)
6734 if (!isPRINT(value))
6735 ANYOF_BITMAP_SET(ret, value);
6742 ANYOF_CLASS_SET(ret, ANYOF_PSXSPC);
6744 for (value = 0; value < 256; value++)
6745 if (isPSXSPC(value))
6746 ANYOF_BITMAP_SET(ret, value);
6753 ANYOF_CLASS_SET(ret, ANYOF_NPSXSPC);
6755 for (value = 0; value < 256; value++)
6756 if (!isPSXSPC(value))
6757 ANYOF_BITMAP_SET(ret, value);
6764 ANYOF_CLASS_SET(ret, ANYOF_PUNCT);
6766 for (value = 0; value < 256; value++)
6768 ANYOF_BITMAP_SET(ret, value);
6775 ANYOF_CLASS_SET(ret, ANYOF_NPUNCT);
6777 for (value = 0; value < 256; value++)
6778 if (!isPUNCT(value))
6779 ANYOF_BITMAP_SET(ret, value);
6786 ANYOF_CLASS_SET(ret, ANYOF_SPACE);
6788 for (value = 0; value < 256; value++)
6790 ANYOF_BITMAP_SET(ret, value);
6797 ANYOF_CLASS_SET(ret, ANYOF_NSPACE);
6799 for (value = 0; value < 256; value++)
6800 if (!isSPACE(value))
6801 ANYOF_BITMAP_SET(ret, value);
6808 ANYOF_CLASS_SET(ret, ANYOF_UPPER);
6810 for (value = 0; value < 256; value++)
6812 ANYOF_BITMAP_SET(ret, value);
6819 ANYOF_CLASS_SET(ret, ANYOF_NUPPER);
6821 for (value = 0; value < 256; value++)
6822 if (!isUPPER(value))
6823 ANYOF_BITMAP_SET(ret, value);
6830 ANYOF_CLASS_SET(ret, ANYOF_XDIGIT);
6832 for (value = 0; value < 256; value++)
6833 if (isXDIGIT(value))
6834 ANYOF_BITMAP_SET(ret, value);
6841 ANYOF_CLASS_SET(ret, ANYOF_NXDIGIT);
6843 for (value = 0; value < 256; value++)
6844 if (!isXDIGIT(value))
6845 ANYOF_BITMAP_SET(ret, value);
6851 /* this is to handle \p and \P */
6854 vFAIL("Invalid [::] class");
6858 /* Strings such as "+utf8::isWord\n" */
6859 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
6862 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
6865 } /* end of namedclass \blah */
6868 if (prevvalue > (IV)value) /* b-a */ {
6869 const int w = RExC_parse - rangebegin;
6870 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
6871 range = 0; /* not a valid range */
6875 prevvalue = value; /* save the beginning of the range */
6876 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
6877 RExC_parse[1] != ']') {
6880 /* a bad range like \w-, [:word:]- ? */
6881 if (namedclass > OOB_NAMEDCLASS) {
6882 if (ckWARN(WARN_REGEXP)) {
6884 RExC_parse >= rangebegin ?
6885 RExC_parse - rangebegin : 0;
6887 "False [] range \"%*.*s\"",
6891 ANYOF_BITMAP_SET(ret, '-');
6893 range = 1; /* yeah, it's a range! */
6894 continue; /* but do it the next time */
6898 /* now is the next time */
6899 /*stored += (value - prevvalue + 1);*/
6901 if (prevvalue < 256) {
6902 const IV ceilvalue = value < 256 ? value : 255;
6905 /* In EBCDIC [\x89-\x91] should include
6906 * the \x8e but [i-j] should not. */
6907 if (literal_endpoint == 2 &&
6908 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
6909 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
6911 if (isLOWER(prevvalue)) {
6912 for (i = prevvalue; i <= ceilvalue; i++)
6914 ANYOF_BITMAP_SET(ret, i);
6916 for (i = prevvalue; i <= ceilvalue; i++)
6918 ANYOF_BITMAP_SET(ret, i);
6923 for (i = prevvalue; i <= ceilvalue; i++) {
6924 if (!ANYOF_BITMAP_TEST(ret,i)) {
6926 ANYOF_BITMAP_SET(ret, i);
6930 if (value > 255 || UTF) {
6931 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
6932 const UV natvalue = NATIVE_TO_UNI(value);
6933 stored+=2; /* can't optimize this class */
6934 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
6935 if (prevnatvalue < natvalue) { /* what about > ? */
6936 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
6937 prevnatvalue, natvalue);
6939 else if (prevnatvalue == natvalue) {
6940 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
6942 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
6944 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
6946 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
6947 if (RExC_precomp[0] == ':' &&
6948 RExC_precomp[1] == '[' &&
6949 (f == 0xDF || f == 0x92)) {
6950 f = NATIVE_TO_UNI(f);
6953 /* If folding and foldable and a single
6954 * character, insert also the folded version
6955 * to the charclass. */
6957 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
6958 if ((RExC_precomp[0] == ':' &&
6959 RExC_precomp[1] == '[' &&
6961 (value == 0xFB05 || value == 0xFB06))) ?
6962 foldlen == ((STRLEN)UNISKIP(f) - 1) :
6963 foldlen == (STRLEN)UNISKIP(f) )
6965 if (foldlen == (STRLEN)UNISKIP(f))
6967 Perl_sv_catpvf(aTHX_ listsv,
6970 /* Any multicharacter foldings
6971 * require the following transform:
6972 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
6973 * where E folds into "pq" and F folds
6974 * into "rst", all other characters
6975 * fold to single characters. We save
6976 * away these multicharacter foldings,
6977 * to be later saved as part of the
6978 * additional "s" data. */
6981 if (!unicode_alternate)
6982 unicode_alternate = newAV();
6983 sv = newSVpvn((char*)foldbuf, foldlen);
6985 av_push(unicode_alternate, sv);
6989 /* If folding and the value is one of the Greek
6990 * sigmas insert a few more sigmas to make the
6991 * folding rules of the sigmas to work right.
6992 * Note that not all the possible combinations
6993 * are handled here: some of them are handled
6994 * by the standard folding rules, and some of
6995 * them (literal or EXACTF cases) are handled
6996 * during runtime in regexec.c:S_find_byclass(). */
6997 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
6998 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
6999 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7000 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7001 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7003 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7004 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7005 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7010 literal_endpoint = 0;
7014 range = 0; /* this range (if it was one) is done now */
7018 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7020 RExC_size += ANYOF_CLASS_ADD_SKIP;
7022 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7028 /****** !SIZE_ONLY AFTER HERE *********/
7030 if( stored == 1 && value < 256
7031 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7033 /* optimize single char class to an EXACT node
7034 but *only* when its not a UTF/high char */
7035 const char * cur_parse= RExC_parse;
7036 RExC_emit = (regnode *)orig_emit;
7037 RExC_parse = (char *)orig_parse;
7038 ret = reg_node(pRExC_state,
7039 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7040 RExC_parse = (char *)cur_parse;
7041 *STRING(ret)= (char)value;
7043 RExC_emit += STR_SZ(1);
7046 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7047 if ( /* If the only flag is folding (plus possibly inversion). */
7048 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7050 for (value = 0; value < 256; ++value) {
7051 if (ANYOF_BITMAP_TEST(ret, value)) {
7052 UV fold = PL_fold[value];
7055 ANYOF_BITMAP_SET(ret, fold);
7058 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7061 /* optimize inverted simple patterns (e.g. [^a-z]) */
7062 if (optimize_invert &&
7063 /* If the only flag is inversion. */
7064 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7065 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7066 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7067 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7070 AV * const av = newAV();
7072 /* The 0th element stores the character class description
7073 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7074 * to initialize the appropriate swash (which gets stored in
7075 * the 1st element), and also useful for dumping the regnode.
7076 * The 2nd element stores the multicharacter foldings,
7077 * used later (regexec.c:S_reginclass()). */
7078 av_store(av, 0, listsv);
7079 av_store(av, 1, NULL);
7080 av_store(av, 2, (SV*)unicode_alternate);
7081 rv = newRV_noinc((SV*)av);
7082 n = add_data(pRExC_state, 1, "s");
7083 RExC_rx->data->data[n] = (void*)rv;
7090 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7092 char* const retval = RExC_parse++;
7095 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7096 RExC_parse[2] == '#') {
7097 while (*RExC_parse != ')') {
7098 if (RExC_parse == RExC_end)
7099 FAIL("Sequence (?#... not terminated");
7105 if (RExC_flags & PMf_EXTENDED) {
7106 if (isSPACE(*RExC_parse)) {
7110 else if (*RExC_parse == '#') {
7111 while (RExC_parse < RExC_end)
7112 if (*RExC_parse++ == '\n') break;
7121 - reg_node - emit a node
7123 STATIC regnode * /* Location. */
7124 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7127 register regnode *ptr;
7128 regnode * const ret = RExC_emit;
7129 GET_RE_DEBUG_FLAGS_DECL;
7132 SIZE_ALIGN(RExC_size);
7136 NODE_ALIGN_FILL(ret);
7138 FILL_ADVANCE_NODE(ptr, op);
7139 if (RExC_offsets) { /* MJD */
7140 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7141 "reg_node", __LINE__,
7143 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7144 ? "Overwriting end of array!\n" : "OK",
7145 (UV)(RExC_emit - RExC_emit_start),
7146 (UV)(RExC_parse - RExC_start),
7147 (UV)RExC_offsets[0]));
7148 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7157 - reganode - emit a node with an argument
7159 STATIC regnode * /* Location. */
7160 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7163 register regnode *ptr;
7164 regnode * const ret = RExC_emit;
7165 GET_RE_DEBUG_FLAGS_DECL;
7168 SIZE_ALIGN(RExC_size);
7173 assert(2==regarglen[op]+1);
7175 Anything larger than this has to allocate the extra amount.
7176 If we changed this to be:
7178 RExC_size += (1 + regarglen[op]);
7180 then it wouldn't matter. Its not clear what side effect
7181 might come from that so its not done so far.
7187 NODE_ALIGN_FILL(ret);
7189 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7190 if (RExC_offsets) { /* MJD */
7191 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7195 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7196 "Overwriting end of array!\n" : "OK",
7197 (UV)(RExC_emit - RExC_emit_start),
7198 (UV)(RExC_parse - RExC_start),
7199 (UV)RExC_offsets[0]));
7200 Set_Cur_Node_Offset;
7209 - reguni - emit (if appropriate) a Unicode character
7212 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
7215 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
7219 - reginsert - insert an operator in front of already-emitted operand
7221 * Means relocating the operand.
7224 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
7227 register regnode *src;
7228 register regnode *dst;
7229 register regnode *place;
7230 const int offset = regarglen[(U8)op];
7231 const int size = NODE_STEP_REGNODE + offset;
7232 GET_RE_DEBUG_FLAGS_DECL;
7233 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
7234 DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
7245 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
7246 if ( RExC_parens[paren] >= src )
7247 RExC_parens[paren] += size;
7251 while (src > opnd) {
7252 StructCopy(--src, --dst, regnode);
7253 if (RExC_offsets) { /* MJD 20010112 */
7254 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
7258 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
7259 ? "Overwriting end of array!\n" : "OK",
7260 (UV)(src - RExC_emit_start),
7261 (UV)(dst - RExC_emit_start),
7262 (UV)RExC_offsets[0]));
7263 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
7264 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
7269 place = opnd; /* Op node, where operand used to be. */
7270 if (RExC_offsets) { /* MJD */
7271 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7275 (UV)(place - RExC_emit_start) > RExC_offsets[0]
7276 ? "Overwriting end of array!\n" : "OK",
7277 (UV)(place - RExC_emit_start),
7278 (UV)(RExC_parse - RExC_start),
7279 (UV)RExC_offsets[0]));
7280 Set_Node_Offset(place, RExC_parse);
7281 Set_Node_Length(place, 1);
7283 src = NEXTOPER(place);
7284 FILL_ADVANCE_NODE(place, op);
7285 Zero(src, offset, regnode);
7289 - regtail - set the next-pointer at the end of a node chain of p to val.
7290 - SEE ALSO: regtail_study
7292 /* TODO: All three parms should be const */
7294 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7297 register regnode *scan;
7298 GET_RE_DEBUG_FLAGS_DECL;
7300 PERL_UNUSED_ARG(depth);
7306 /* Find last node. */
7309 regnode * const temp = regnext(scan);
7311 SV * const mysv=sv_newmortal();
7312 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
7313 regprop(RExC_rx, mysv, scan);
7314 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
7315 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
7316 (temp == NULL ? "->" : ""),
7317 (temp == NULL ? reg_name[OP(val)] : "")
7325 if (reg_off_by_arg[OP(scan)]) {
7326 ARG_SET(scan, val - scan);
7329 NEXT_OFF(scan) = val - scan;
7335 - regtail_study - set the next-pointer at the end of a node chain of p to val.
7336 - Look for optimizable sequences at the same time.
7337 - currently only looks for EXACT chains.
7339 This is expermental code. The idea is to use this routine to perform
7340 in place optimizations on branches and groups as they are constructed,
7341 with the long term intention of removing optimization from study_chunk so
7342 that it is purely analytical.
7344 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
7345 to control which is which.
7348 /* TODO: All four parms should be const */
7351 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7354 register regnode *scan;
7356 #ifdef EXPERIMENTAL_INPLACESCAN
7360 GET_RE_DEBUG_FLAGS_DECL;
7366 /* Find last node. */
7370 regnode * const temp = regnext(scan);
7371 #ifdef EXPERIMENTAL_INPLACESCAN
7372 if (PL_regkind[OP(scan)] == EXACT)
7373 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
7381 if( exact == PSEUDO )
7383 else if ( exact != OP(scan) )
7392 SV * const mysv=sv_newmortal();
7393 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
7394 regprop(RExC_rx, mysv, scan);
7395 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
7396 SvPV_nolen_const(mysv),
7405 SV * const mysv_val=sv_newmortal();
7406 DEBUG_PARSE_MSG("");
7407 regprop(RExC_rx, mysv_val, val);
7408 PerlIO_printf(Perl_debug_log, "~ attach to %s (%d) offset to %d\n",
7409 SvPV_nolen_const(mysv_val),
7414 if (reg_off_by_arg[OP(scan)]) {
7415 ARG_SET(scan, val - scan);
7418 NEXT_OFF(scan) = val - scan;
7426 - regcurly - a little FSA that accepts {\d+,?\d*}
7429 S_regcurly(register const char *s)
7448 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
7451 Perl_regdump(pTHX_ const regexp *r)
7455 SV * const sv = sv_newmortal();
7456 SV *dsv= sv_newmortal();
7458 (void)dumpuntil(r, r->program, r->program + 1, NULL, NULL, sv, 0, 0);
7460 /* Header fields of interest. */
7461 if (r->anchored_substr) {
7462 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
7463 RE_SV_DUMPLEN(r->anchored_substr), 30);
7464 PerlIO_printf(Perl_debug_log,
7465 "anchored %s%s at %"IVdf" ",
7466 s, RE_SV_TAIL(r->anchored_substr),
7467 (IV)r->anchored_offset);
7468 } else if (r->anchored_utf8) {
7469 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
7470 RE_SV_DUMPLEN(r->anchored_utf8), 30);
7471 PerlIO_printf(Perl_debug_log,
7472 "anchored utf8 %s%s at %"IVdf" ",
7473 s, RE_SV_TAIL(r->anchored_utf8),
7474 (IV)r->anchored_offset);
7476 if (r->float_substr) {
7477 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
7478 RE_SV_DUMPLEN(r->float_substr), 30);
7479 PerlIO_printf(Perl_debug_log,
7480 "floating %s%s at %"IVdf"..%"UVuf" ",
7481 s, RE_SV_TAIL(r->float_substr),
7482 (IV)r->float_min_offset, (UV)r->float_max_offset);
7483 } else if (r->float_utf8) {
7484 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
7485 RE_SV_DUMPLEN(r->float_utf8), 30);
7486 PerlIO_printf(Perl_debug_log,
7487 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
7488 s, RE_SV_TAIL(r->float_utf8),
7489 (IV)r->float_min_offset, (UV)r->float_max_offset);
7491 if (r->check_substr || r->check_utf8)
7492 PerlIO_printf(Perl_debug_log,
7494 (r->check_substr == r->float_substr
7495 && r->check_utf8 == r->float_utf8
7496 ? "(checking floating" : "(checking anchored"));
7497 if (r->reganch & ROPT_NOSCAN)
7498 PerlIO_printf(Perl_debug_log, " noscan");
7499 if (r->reganch & ROPT_CHECK_ALL)
7500 PerlIO_printf(Perl_debug_log, " isall");
7501 if (r->check_substr || r->check_utf8)
7502 PerlIO_printf(Perl_debug_log, ") ");
7504 if (r->regstclass) {
7505 regprop(r, sv, r->regstclass);
7506 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
7508 if (r->reganch & ROPT_ANCH) {
7509 PerlIO_printf(Perl_debug_log, "anchored");
7510 if (r->reganch & ROPT_ANCH_BOL)
7511 PerlIO_printf(Perl_debug_log, "(BOL)");
7512 if (r->reganch & ROPT_ANCH_MBOL)
7513 PerlIO_printf(Perl_debug_log, "(MBOL)");
7514 if (r->reganch & ROPT_ANCH_SBOL)
7515 PerlIO_printf(Perl_debug_log, "(SBOL)");
7516 if (r->reganch & ROPT_ANCH_GPOS)
7517 PerlIO_printf(Perl_debug_log, "(GPOS)");
7518 PerlIO_putc(Perl_debug_log, ' ');
7520 if (r->reganch & ROPT_GPOS_SEEN)
7521 PerlIO_printf(Perl_debug_log, "GPOS ");
7522 if (r->reganch & ROPT_SKIP)
7523 PerlIO_printf(Perl_debug_log, "plus ");
7524 if (r->reganch & ROPT_IMPLICIT)
7525 PerlIO_printf(Perl_debug_log, "implicit ");
7526 PerlIO_printf(Perl_debug_log, "minlen %ld ", (long) r->minlen);
7527 if (r->reganch & ROPT_EVAL_SEEN)
7528 PerlIO_printf(Perl_debug_log, "with eval ");
7529 PerlIO_printf(Perl_debug_log, "\n");
7531 PERL_UNUSED_CONTEXT;
7533 #endif /* DEBUGGING */
7537 - regprop - printable representation of opcode
7540 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
7545 GET_RE_DEBUG_FLAGS_DECL;
7547 sv_setpvn(sv, "", 0);
7548 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
7549 /* It would be nice to FAIL() here, but this may be called from
7550 regexec.c, and it would be hard to supply pRExC_state. */
7551 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
7552 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
7554 k = PL_regkind[OP(o)];
7557 SV * const dsv = sv_2mortal(newSVpvs(""));
7558 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
7559 * is a crude hack but it may be the best for now since
7560 * we have no flag "this EXACTish node was UTF-8"
7562 const char * const s =
7563 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
7564 PL_colors[0], PL_colors[1],
7565 PERL_PV_ESCAPE_UNI_DETECT |
7566 PERL_PV_PRETTY_ELIPSES |
7569 Perl_sv_catpvf(aTHX_ sv, " %s", s );
7570 } else if (k == TRIE) {
7571 /* print the details of the trie in dumpuntil instead, as
7572 * prog->data isn't available here */
7573 const char op = OP(o);
7574 const I32 n = ARG(o);
7575 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
7576 (reg_ac_data *)prog->data->data[n] :
7578 const reg_trie_data * const trie = !IS_TRIE_AC(op) ?
7579 (reg_trie_data*)prog->data->data[n] :
7582 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
7583 DEBUG_TRIE_COMPILE_r(
7584 Perl_sv_catpvf(aTHX_ sv,
7585 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
7586 (UV)trie->startstate,
7587 (IV)trie->laststate-1,
7588 (UV)trie->wordcount,
7591 (UV)TRIE_CHARCOUNT(trie),
7592 (UV)trie->uniquecharcount
7595 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
7597 int rangestart = -1;
7598 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
7599 Perl_sv_catpvf(aTHX_ sv, "[");
7600 for (i = 0; i <= 256; i++) {
7601 if (i < 256 && BITMAP_TEST(bitmap,i)) {
7602 if (rangestart == -1)
7604 } else if (rangestart != -1) {
7605 if (i <= rangestart + 3)
7606 for (; rangestart < i; rangestart++)
7607 put_byte(sv, rangestart);
7609 put_byte(sv, rangestart);
7611 put_byte(sv, i - 1);
7616 Perl_sv_catpvf(aTHX_ sv, "]");
7619 } else if (k == CURLY) {
7620 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
7621 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
7622 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
7624 else if (k == WHILEM && o->flags) /* Ordinal/of */
7625 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
7626 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP)
7627 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
7628 else if (k == RECURSE)
7629 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
7630 else if (k == LOGICAL)
7631 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
7632 else if (k == ANYOF) {
7633 int i, rangestart = -1;
7634 const U8 flags = ANYOF_FLAGS(o);
7636 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
7637 static const char * const anyofs[] = {
7670 if (flags & ANYOF_LOCALE)
7671 sv_catpvs(sv, "{loc}");
7672 if (flags & ANYOF_FOLD)
7673 sv_catpvs(sv, "{i}");
7674 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
7675 if (flags & ANYOF_INVERT)
7677 for (i = 0; i <= 256; i++) {
7678 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
7679 if (rangestart == -1)
7681 } else if (rangestart != -1) {
7682 if (i <= rangestart + 3)
7683 for (; rangestart < i; rangestart++)
7684 put_byte(sv, rangestart);
7686 put_byte(sv, rangestart);
7688 put_byte(sv, i - 1);
7694 if (o->flags & ANYOF_CLASS)
7695 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
7696 if (ANYOF_CLASS_TEST(o,i))
7697 sv_catpv(sv, anyofs[i]);
7699 if (flags & ANYOF_UNICODE)
7700 sv_catpvs(sv, "{unicode}");
7701 else if (flags & ANYOF_UNICODE_ALL)
7702 sv_catpvs(sv, "{unicode_all}");
7706 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
7710 U8 s[UTF8_MAXBYTES_CASE+1];
7712 for (i = 0; i <= 256; i++) { /* just the first 256 */
7713 uvchr_to_utf8(s, i);
7715 if (i < 256 && swash_fetch(sw, s, TRUE)) {
7716 if (rangestart == -1)
7718 } else if (rangestart != -1) {
7719 if (i <= rangestart + 3)
7720 for (; rangestart < i; rangestart++) {
7721 const U8 * const e = uvchr_to_utf8(s,rangestart);
7723 for(p = s; p < e; p++)
7727 const U8 *e = uvchr_to_utf8(s,rangestart);
7729 for (p = s; p < e; p++)
7732 e = uvchr_to_utf8(s, i-1);
7733 for (p = s; p < e; p++)
7740 sv_catpvs(sv, "..."); /* et cetera */
7744 char *s = savesvpv(lv);
7745 char * const origs = s;
7747 while (*s && *s != '\n')
7751 const char * const t = ++s;
7769 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
7771 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
7772 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
7774 PERL_UNUSED_CONTEXT;
7775 PERL_UNUSED_ARG(sv);
7777 PERL_UNUSED_ARG(prog);
7778 #endif /* DEBUGGING */
7782 Perl_re_intuit_string(pTHX_ regexp *prog)
7783 { /* Assume that RE_INTUIT is set */
7785 GET_RE_DEBUG_FLAGS_DECL;
7786 PERL_UNUSED_CONTEXT;
7790 const char * const s = SvPV_nolen_const(prog->check_substr
7791 ? prog->check_substr : prog->check_utf8);
7793 if (!PL_colorset) reginitcolors();
7794 PerlIO_printf(Perl_debug_log,
7795 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
7797 prog->check_substr ? "" : "utf8 ",
7798 PL_colors[5],PL_colors[0],
7801 (strlen(s) > 60 ? "..." : ""));
7804 return prog->check_substr ? prog->check_substr : prog->check_utf8;
7808 pregfree - free a regexp
7810 See regdupe below if you change anything here.
7814 Perl_pregfree(pTHX_ struct regexp *r)
7818 GET_RE_DEBUG_FLAGS_DECL;
7820 if (!r || (--r->refcnt > 0))
7826 SV *dsv= sv_newmortal();
7827 RE_PV_QUOTED_DECL(s, (r->reganch & ROPT_UTF8),
7828 dsv, r->precomp, r->prelen, 60);
7829 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
7830 PL_colors[4],PL_colors[5],s);
7834 /* gcov results gave these as non-null 100% of the time, so there's no
7835 optimisation in checking them before calling Safefree */
7836 Safefree(r->precomp);
7837 Safefree(r->offsets); /* 20010421 MJD */
7838 RX_MATCH_COPY_FREE(r);
7839 #ifdef PERL_OLD_COPY_ON_WRITE
7841 SvREFCNT_dec(r->saved_copy);
7844 if (r->anchored_substr)
7845 SvREFCNT_dec(r->anchored_substr);
7846 if (r->anchored_utf8)
7847 SvREFCNT_dec(r->anchored_utf8);
7848 if (r->float_substr)
7849 SvREFCNT_dec(r->float_substr);
7851 SvREFCNT_dec(r->float_utf8);
7852 Safefree(r->substrs);
7855 SvREFCNT_dec(r->paren_names);
7857 int n = r->data->count;
7858 PAD* new_comppad = NULL;
7863 /* If you add a ->what type here, update the comment in regcomp.h */
7864 switch (r->data->what[n]) {
7867 SvREFCNT_dec((SV*)r->data->data[n]);
7870 Safefree(r->data->data[n]);
7873 new_comppad = (AV*)r->data->data[n];
7876 if (new_comppad == NULL)
7877 Perl_croak(aTHX_ "panic: pregfree comppad");
7878 PAD_SAVE_LOCAL(old_comppad,
7879 /* Watch out for global destruction's random ordering. */
7880 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
7883 refcnt = OpREFCNT_dec((OP_4tree*)r->data->data[n]);
7886 op_free((OP_4tree*)r->data->data[n]);
7888 PAD_RESTORE_LOCAL(old_comppad);
7889 SvREFCNT_dec((SV*)new_comppad);
7895 { /* Aho Corasick add-on structure for a trie node.
7896 Used in stclass optimization only */
7898 reg_ac_data *aho=(reg_ac_data*)r->data->data[n];
7900 refcount = --aho->refcount;
7903 Safefree(aho->states);
7904 Safefree(aho->fail);
7905 aho->trie=NULL; /* not necessary to free this as it is
7906 handled by the 't' case */
7907 Safefree(r->data->data[n]); /* do this last!!!! */
7908 Safefree(r->regstclass);
7914 /* trie structure. */
7916 reg_trie_data *trie=(reg_trie_data*)r->data->data[n];
7918 refcount = --trie->refcount;
7921 Safefree(trie->charmap);
7922 if (trie->widecharmap)
7923 SvREFCNT_dec((SV*)trie->widecharmap);
7924 Safefree(trie->states);
7925 Safefree(trie->trans);
7927 Safefree(trie->bitmap);
7929 Safefree(trie->wordlen);
7931 Safefree(trie->jump);
7933 Safefree(trie->nextword);
7937 SvREFCNT_dec((SV*)trie->words);
7938 if (trie->revcharmap)
7939 SvREFCNT_dec((SV*)trie->revcharmap);
7942 Safefree(r->data->data[n]); /* do this last!!!! */
7947 Perl_croak(aTHX_ "panic: regfree data code '%c'", r->data->what[n]);
7950 Safefree(r->data->what);
7953 Safefree(r->startp);
7958 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
7959 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
7960 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
7961 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
7964 regdupe - duplicate a regexp.
7966 This routine is called by sv.c's re_dup and is expected to clone a
7967 given regexp structure. It is a no-op when not under USE_ITHREADS.
7968 (Originally this *was* re_dup() for change history see sv.c)
7970 See pregfree() above if you change anything here.
7972 #if defined(USE_ITHREADS)
7974 Perl_regdupe(pTHX_ const regexp *r, CLONE_PARAMS *param)
7979 struct reg_substr_datum *s;
7982 return (REGEXP *)NULL;
7984 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
7987 len = r->offsets[0];
7988 npar = r->nparens+1;
7990 Newxc(ret, sizeof(regexp) + (len+1)*sizeof(regnode), char, regexp);
7991 Copy(r->program, ret->program, len+1, regnode);
7993 Newx(ret->startp, npar, I32);
7994 Copy(r->startp, ret->startp, npar, I32);
7995 Newx(ret->endp, npar, I32);
7996 Copy(r->startp, ret->startp, npar, I32);
7998 Newx(ret->substrs, 1, struct reg_substr_data);
7999 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8000 s->min_offset = r->substrs->data[i].min_offset;
8001 s->max_offset = r->substrs->data[i].max_offset;
8002 s->end_shift = r->substrs->data[i].end_shift;
8003 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8004 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8007 ret->regstclass = NULL;
8010 const int count = r->data->count;
8013 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8014 char, struct reg_data);
8015 Newx(d->what, count, U8);
8018 for (i = 0; i < count; i++) {
8019 d->what[i] = r->data->what[i];
8020 switch (d->what[i]) {
8021 /* legal options are one of: sfpont
8022 see also regcomp.h and pregfree() */
8025 d->data[i] = sv_dup_inc((SV *)r->data->data[i], param);
8028 d->data[i] = av_dup_inc((AV *)r->data->data[i], param);
8031 /* This is cheating. */
8032 Newx(d->data[i], 1, struct regnode_charclass_class);
8033 StructCopy(r->data->data[i], d->data[i],
8034 struct regnode_charclass_class);
8035 ret->regstclass = (regnode*)d->data[i];
8038 /* Compiled op trees are readonly, and can thus be
8039 shared without duplication. */
8041 d->data[i] = (void*)OpREFCNT_inc((OP*)r->data->data[i]);
8045 d->data[i] = r->data->data[i];
8048 d->data[i] = r->data->data[i];
8050 ((reg_trie_data*)d->data[i])->refcount++;
8054 d->data[i] = r->data->data[i];
8056 ((reg_ac_data*)d->data[i])->refcount++;
8058 /* Trie stclasses are readonly and can thus be shared
8059 * without duplication. We free the stclass in pregfree
8060 * when the corresponding reg_ac_data struct is freed.
8062 ret->regstclass= r->regstclass;
8065 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", r->data->what[i]);
8074 Newx(ret->offsets, 2*len+1, U32);
8075 Copy(r->offsets, ret->offsets, 2*len+1, U32);
8077 ret->precomp = SAVEPVN(r->precomp, r->prelen);
8078 ret->refcnt = r->refcnt;
8079 ret->minlen = r->minlen;
8080 ret->prelen = r->prelen;
8081 ret->nparens = r->nparens;
8082 ret->lastparen = r->lastparen;
8083 ret->lastcloseparen = r->lastcloseparen;
8084 ret->reganch = r->reganch;
8086 ret->sublen = r->sublen;
8088 ret->engine = r->engine;
8090 ret->paren_names = hv_dup_inc(r->paren_names, param);
8092 if (RX_MATCH_COPIED(ret))
8093 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8096 #ifdef PERL_OLD_COPY_ON_WRITE
8097 ret->saved_copy = NULL;
8100 ptr_table_store(PL_ptr_table, r, ret);
8105 #ifndef PERL_IN_XSUB_RE
8107 - regnext - dig the "next" pointer out of a node
8110 Perl_regnext(pTHX_ register regnode *p)
8113 register I32 offset;
8115 if (p == &PL_regdummy)
8118 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
8127 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
8130 STRLEN l1 = strlen(pat1);
8131 STRLEN l2 = strlen(pat2);
8134 const char *message;
8140 Copy(pat1, buf, l1 , char);
8141 Copy(pat2, buf + l1, l2 , char);
8142 buf[l1 + l2] = '\n';
8143 buf[l1 + l2 + 1] = '\0';
8145 /* ANSI variant takes additional second argument */
8146 va_start(args, pat2);
8150 msv = vmess(buf, &args);
8152 message = SvPV_const(msv,l1);
8155 Copy(message, buf, l1 , char);
8156 buf[l1-1] = '\0'; /* Overwrite \n */
8157 Perl_croak(aTHX_ "%s", buf);
8160 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
8162 #ifndef PERL_IN_XSUB_RE
8164 Perl_save_re_context(pTHX)
8168 struct re_save_state *state;
8170 SAVEVPTR(PL_curcop);
8171 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
8173 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
8174 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
8175 SSPUSHINT(SAVEt_RE_STATE);
8177 Copy(&PL_reg_state, state, 1, struct re_save_state);
8179 PL_reg_start_tmp = 0;
8180 PL_reg_start_tmpl = 0;
8181 PL_reg_oldsaved = NULL;
8182 PL_reg_oldsavedlen = 0;
8184 PL_reg_leftiter = 0;
8185 PL_reg_poscache = NULL;
8186 PL_reg_poscache_size = 0;
8187 #ifdef PERL_OLD_COPY_ON_WRITE
8191 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
8193 const REGEXP * const rx = PM_GETRE(PL_curpm);
8196 for (i = 1; i <= rx->nparens; i++) {
8197 char digits[TYPE_CHARS(long)];
8198 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
8199 GV *const *const gvp
8200 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
8203 GV * const gv = *gvp;
8204 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
8214 clear_re(pTHX_ void *r)
8217 ReREFCNT_dec((regexp *)r);
8223 S_put_byte(pTHX_ SV *sv, int c)
8225 if (isCNTRL(c) || c == 255 || !isPRINT(c))
8226 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
8227 else if (c == '-' || c == ']' || c == '\\' || c == '^')
8228 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
8230 Perl_sv_catpvf(aTHX_ sv, "%c", c);
8234 #define CLEAR_OPTSTART \
8235 if (optstart) STMT_START { \
8236 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%d nodes)\n", node - optstart)); \
8240 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
8242 STATIC const regnode *
8243 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
8244 const regnode *last, const regnode *plast,
8245 SV* sv, I32 indent, U32 depth)
8248 register U8 op = PSEUDO; /* Arbitrary non-END op. */
8249 register const regnode *next;
8250 const regnode *optstart= NULL;
8251 GET_RE_DEBUG_FLAGS_DECL;
8253 #ifdef DEBUG_DUMPUNTIL
8254 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
8255 last ? last-start : 0,plast ? plast-start : 0);
8258 if (plast && plast < last)
8261 while (PL_regkind[op] != END && (!last || node < last)) {
8262 /* While that wasn't END last time... */
8268 next = regnext((regnode *)node);
8271 if (OP(node) == OPTIMIZED) {
8272 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
8279 regprop(r, sv, node);
8280 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
8281 (int)(2*indent + 1), "", SvPVX_const(sv));
8283 if (OP(node) != OPTIMIZED) {
8284 if (next == NULL) /* Next ptr. */
8285 PerlIO_printf(Perl_debug_log, "(0)");
8286 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
8287 PerlIO_printf(Perl_debug_log, "(FAIL)");
8289 PerlIO_printf(Perl_debug_log, "(%"IVdf")", (IV)(next - start));
8291 /*if (PL_regkind[(U8)op] != TRIE)*/
8292 (void)PerlIO_putc(Perl_debug_log, '\n');
8296 if (PL_regkind[(U8)op] == BRANCHJ) {
8299 register const regnode *nnode = (OP(next) == LONGJMP
8300 ? regnext((regnode *)next)
8302 if (last && nnode > last)
8304 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
8307 else if (PL_regkind[(U8)op] == BRANCH) {
8309 DUMPUNTIL(NEXTOPER(node), next);
8311 else if ( PL_regkind[(U8)op] == TRIE ) {
8312 const char op = OP(node);
8313 const I32 n = ARG(node);
8314 const reg_ac_data * const ac = op>=AHOCORASICK ?
8315 (reg_ac_data *)r->data->data[n] :
8317 const reg_trie_data * const trie = op<AHOCORASICK ?
8318 (reg_trie_data*)r->data->data[n] :
8320 const regnode *nextbranch= NULL;
8322 sv_setpvn(sv, "", 0);
8323 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
8324 SV ** const elem_ptr = av_fetch(trie->words,word_idx,0);
8326 PerlIO_printf(Perl_debug_log, "%*s%s ",
8327 (int)(2*(indent+3)), "",
8328 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
8329 PL_colors[0], PL_colors[1],
8330 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
8331 PERL_PV_PRETTY_ELIPSES |
8337 U16 dist= trie->jump[word_idx+1];
8338 PerlIO_printf(Perl_debug_log, "(%u)\n",(next - dist) - start);
8341 nextbranch= next - trie->jump[0];
8342 DUMPUNTIL(next - dist, nextbranch);
8344 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
8345 nextbranch= regnext((regnode *)nextbranch);
8347 PerlIO_printf(Perl_debug_log, "\n");
8350 if (last && next > last)
8355 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
8356 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
8357 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
8359 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
8361 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
8363 else if ( op == PLUS || op == STAR) {
8364 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
8366 else if (op == ANYOF) {
8367 /* arglen 1 + class block */
8368 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
8369 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
8370 node = NEXTOPER(node);
8372 else if (PL_regkind[(U8)op] == EXACT) {
8373 /* Literal string, where present. */
8374 node += NODE_SZ_STR(node) - 1;
8375 node = NEXTOPER(node);
8378 node = NEXTOPER(node);
8379 node += regarglen[(U8)op];
8381 if (op == CURLYX || op == OPEN)
8383 else if (op == WHILEM)
8387 #ifdef DEBUG_DUMPUNTIL
8388 PerlIO_printf(Perl_debug_log, "--- %d\n",indent);
8393 #endif /* DEBUGGING */
8397 * c-indentation-style: bsd
8399 * indent-tabs-mode: t
8402 * ex: set ts=8 sts=4 sw=4 noet: