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. */
105 regexp *rx; /* perl core regexp structure */
106 regexp_internal *rxi; /* internal data for regexp object pprivate field */
107 char *start; /* Start of input for compile */
108 char *end; /* End of input for compile */
109 char *parse; /* Input-scan pointer. */
110 I32 whilem_seen; /* number of WHILEM in this expr */
111 regnode *emit_start; /* Start of emitted-code area */
112 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
113 I32 naughty; /* How bad is this pattern? */
114 I32 sawback; /* Did we see \1, ...? */
116 I32 size; /* Code size. */
117 I32 npar; /* Capture buffer count, (OPEN). */
118 I32 cpar; /* Capture buffer count, (CLOSE). */
119 I32 nestroot; /* root parens we are in - used by accept */
123 regnode **open_parens; /* pointers to open parens */
124 regnode **close_parens; /* pointers to close parens */
125 regnode *opend; /* END node in program */
127 HV *charnames; /* cache of named sequences */
128 HV *paren_names; /* Paren names */
129 regnode **recurse; /* Recurse regops */
130 I32 recurse_count; /* Number of recurse regops */
132 char *starttry; /* -Dr: where regtry was called. */
133 #define RExC_starttry (pRExC_state->starttry)
136 const char *lastparse;
138 #define RExC_lastparse (pRExC_state->lastparse)
139 #define RExC_lastnum (pRExC_state->lastnum)
143 #define RExC_flags (pRExC_state->flags)
144 #define RExC_precomp (pRExC_state->precomp)
145 #define RExC_rx (pRExC_state->rx)
146 #define RExC_rxi (pRExC_state->rxi)
147 #define RExC_start (pRExC_state->start)
148 #define RExC_end (pRExC_state->end)
149 #define RExC_parse (pRExC_state->parse)
150 #define RExC_whilem_seen (pRExC_state->whilem_seen)
151 #define RExC_offsets (pRExC_state->rxi->offsets) /* I am not like the others */
152 #define RExC_emit (pRExC_state->emit)
153 #define RExC_emit_start (pRExC_state->emit_start)
154 #define RExC_naughty (pRExC_state->naughty)
155 #define RExC_sawback (pRExC_state->sawback)
156 #define RExC_seen (pRExC_state->seen)
157 #define RExC_size (pRExC_state->size)
158 #define RExC_npar (pRExC_state->npar)
159 #define RExC_nestroot (pRExC_state->nestroot)
160 #define RExC_extralen (pRExC_state->extralen)
161 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
162 #define RExC_seen_evals (pRExC_state->seen_evals)
163 #define RExC_utf8 (pRExC_state->utf8)
164 #define RExC_charnames (pRExC_state->charnames)
165 #define RExC_open_parens (pRExC_state->open_parens)
166 #define RExC_close_parens (pRExC_state->close_parens)
167 #define RExC_opend (pRExC_state->opend)
168 #define RExC_paren_names (pRExC_state->paren_names)
169 #define RExC_recurse (pRExC_state->recurse)
170 #define RExC_recurse_count (pRExC_state->recurse_count)
172 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
173 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
174 ((*s) == '{' && regcurly(s)))
177 #undef SPSTART /* dratted cpp namespace... */
180 * Flags to be passed up and down.
182 #define WORST 0 /* Worst case. */
183 #define HASWIDTH 0x1 /* Known to match non-null strings. */
184 #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */
185 #define SPSTART 0x4 /* Starts with * or +. */
186 #define TRYAGAIN 0x8 /* Weeded out a declaration. */
188 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
190 /* whether trie related optimizations are enabled */
191 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
192 #define TRIE_STUDY_OPT
193 #define FULL_TRIE_STUDY
199 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
200 #define PBITVAL(paren) (1 << ((paren) & 7))
201 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
202 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
203 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
206 /* About scan_data_t.
208 During optimisation we recurse through the regexp program performing
209 various inplace (keyhole style) optimisations. In addition study_chunk
210 and scan_commit populate this data structure with information about
211 what strings MUST appear in the pattern. We look for the longest
212 string that must appear for at a fixed location, and we look for the
213 longest string that may appear at a floating location. So for instance
218 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
219 strings (because they follow a .* construct). study_chunk will identify
220 both FOO and BAR as being the longest fixed and floating strings respectively.
222 The strings can be composites, for instance
226 will result in a composite fixed substring 'foo'.
228 For each string some basic information is maintained:
230 - offset or min_offset
231 This is the position the string must appear at, or not before.
232 It also implicitly (when combined with minlenp) tells us how many
233 character must match before the string we are searching.
234 Likewise when combined with minlenp and the length of the string
235 tells us how many characters must appear after the string we have
239 Only used for floating strings. This is the rightmost point that
240 the string can appear at. Ifset to I32 max it indicates that the
241 string can occur infinitely far to the right.
244 A pointer to the minimum length of the pattern that the string
245 was found inside. This is important as in the case of positive
246 lookahead or positive lookbehind we can have multiple patterns
251 The minimum length of the pattern overall is 3, the minimum length
252 of the lookahead part is 3, but the minimum length of the part that
253 will actually match is 1. So 'FOO's minimum length is 3, but the
254 minimum length for the F is 1. This is important as the minimum length
255 is used to determine offsets in front of and behind the string being
256 looked for. Since strings can be composites this is the length of the
257 pattern at the time it was commited with a scan_commit. Note that
258 the length is calculated by study_chunk, so that the minimum lengths
259 are not known until the full pattern has been compiled, thus the
260 pointer to the value.
264 In the case of lookbehind the string being searched for can be
265 offset past the start point of the final matching string.
266 If this value was just blithely removed from the min_offset it would
267 invalidate some of the calculations for how many chars must match
268 before or after (as they are derived from min_offset and minlen and
269 the length of the string being searched for).
270 When the final pattern is compiled and the data is moved from the
271 scan_data_t structure into the regexp structure the information
272 about lookbehind is factored in, with the information that would
273 have been lost precalculated in the end_shift field for the
276 The fields pos_min and pos_delta are used to store the minimum offset
277 and the delta to the maximum offset at the current point in the pattern.
281 typedef struct scan_data_t {
282 /*I32 len_min; unused */
283 /*I32 len_delta; unused */
287 I32 last_end; /* min value, <0 unless valid. */
290 SV **longest; /* Either &l_fixed, or &l_float. */
291 SV *longest_fixed; /* longest fixed string found in pattern */
292 I32 offset_fixed; /* offset where it starts */
293 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
294 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
295 SV *longest_float; /* longest floating string found in pattern */
296 I32 offset_float_min; /* earliest point in string it can appear */
297 I32 offset_float_max; /* latest point in string it can appear */
298 I32 *minlen_float; /* pointer to the minlen relevent to the string */
299 I32 lookbehind_float; /* is the position of the string modified by LB */
303 struct regnode_charclass_class *start_class;
307 * Forward declarations for pregcomp()'s friends.
310 static const scan_data_t zero_scan_data =
311 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
313 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
314 #define SF_BEFORE_SEOL 0x0001
315 #define SF_BEFORE_MEOL 0x0002
316 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
317 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
320 # define SF_FIX_SHIFT_EOL (0+2)
321 # define SF_FL_SHIFT_EOL (0+4)
323 # define SF_FIX_SHIFT_EOL (+2)
324 # define SF_FL_SHIFT_EOL (+4)
327 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
328 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
330 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
331 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
332 #define SF_IS_INF 0x0040
333 #define SF_HAS_PAR 0x0080
334 #define SF_IN_PAR 0x0100
335 #define SF_HAS_EVAL 0x0200
336 #define SCF_DO_SUBSTR 0x0400
337 #define SCF_DO_STCLASS_AND 0x0800
338 #define SCF_DO_STCLASS_OR 0x1000
339 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
340 #define SCF_WHILEM_VISITED_POS 0x2000
342 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
343 #define SCF_SEEN_ACCEPT 0x8000
345 #define UTF (RExC_utf8 != 0)
346 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
347 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
349 #define OOB_UNICODE 12345678
350 #define OOB_NAMEDCLASS -1
352 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
353 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
356 /* length of regex to show in messages that don't mark a position within */
357 #define RegexLengthToShowInErrorMessages 127
360 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
361 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
362 * op/pragma/warn/regcomp.
364 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
365 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
367 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
370 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
371 * arg. Show regex, up to a maximum length. If it's too long, chop and add
374 #define _FAIL(code) STMT_START { \
375 const char *ellipses = ""; \
376 IV len = RExC_end - RExC_precomp; \
379 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
380 if (len > RegexLengthToShowInErrorMessages) { \
381 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
382 len = RegexLengthToShowInErrorMessages - 10; \
388 #define FAIL(msg) _FAIL( \
389 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
390 msg, (int)len, RExC_precomp, ellipses))
392 #define FAIL2(msg,arg) _FAIL( \
393 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
394 arg, (int)len, RExC_precomp, ellipses))
397 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
399 #define Simple_vFAIL(m) STMT_START { \
400 const IV offset = RExC_parse - RExC_precomp; \
401 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
402 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
406 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
408 #define vFAIL(m) STMT_START { \
410 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
415 * Like Simple_vFAIL(), but accepts two arguments.
417 #define Simple_vFAIL2(m,a1) STMT_START { \
418 const IV offset = RExC_parse - RExC_precomp; \
419 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
420 (int)offset, RExC_precomp, RExC_precomp + offset); \
424 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
426 #define vFAIL2(m,a1) STMT_START { \
428 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
429 Simple_vFAIL2(m, a1); \
434 * Like Simple_vFAIL(), but accepts three arguments.
436 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
437 const IV offset = RExC_parse - RExC_precomp; \
438 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
439 (int)offset, RExC_precomp, RExC_precomp + offset); \
443 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
445 #define vFAIL3(m,a1,a2) STMT_START { \
447 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
448 Simple_vFAIL3(m, a1, a2); \
452 * Like Simple_vFAIL(), but accepts four arguments.
454 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
455 const IV offset = RExC_parse - RExC_precomp; \
456 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
457 (int)offset, RExC_precomp, RExC_precomp + offset); \
460 #define vWARN(loc,m) STMT_START { \
461 const IV offset = loc - RExC_precomp; \
462 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
463 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
466 #define vWARNdep(loc,m) STMT_START { \
467 const IV offset = loc - RExC_precomp; \
468 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
469 "%s" REPORT_LOCATION, \
470 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
474 #define vWARN2(loc, m, a1) STMT_START { \
475 const IV offset = loc - RExC_precomp; \
476 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
477 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
480 #define vWARN3(loc, m, a1, a2) STMT_START { \
481 const IV offset = loc - RExC_precomp; \
482 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
483 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
486 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
487 const IV offset = loc - RExC_precomp; \
488 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
489 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
492 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
493 const IV offset = loc - RExC_precomp; \
494 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
495 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
499 /* Allow for side effects in s */
500 #define REGC(c,s) STMT_START { \
501 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
504 /* Macros for recording node offsets. 20001227 mjd@plover.com
505 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
506 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
507 * Element 0 holds the number n.
508 * Position is 1 indexed.
511 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
513 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
514 __LINE__, (int)(node), (int)(byte))); \
516 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
518 RExC_offsets[2*(node)-1] = (byte); \
523 #define Set_Node_Offset(node,byte) \
524 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
525 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
527 #define Set_Node_Length_To_R(node,len) STMT_START { \
529 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
530 __LINE__, (int)(node), (int)(len))); \
532 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
534 RExC_offsets[2*(node)] = (len); \
539 #define Set_Node_Length(node,len) \
540 Set_Node_Length_To_R((node)-RExC_emit_start, len)
541 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
542 #define Set_Node_Cur_Length(node) \
543 Set_Node_Length(node, RExC_parse - parse_start)
545 /* Get offsets and lengths */
546 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
547 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
549 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
550 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
551 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
555 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
556 #define EXPERIMENTAL_INPLACESCAN
559 #define DEBUG_STUDYDATA(data,depth) \
560 DEBUG_OPTIMISE_MORE_r(if(data){ \
561 PerlIO_printf(Perl_debug_log, \
562 "%*s"/* Len:%"IVdf"/%"IVdf" */"Pos:%"IVdf"/%"IVdf \
563 " Flags: %"IVdf" Whilem_c: %"IVdf" Lcp: %"IVdf" ", \
564 (int)(depth)*2, "", \
565 (IV)((data)->pos_min), \
566 (IV)((data)->pos_delta), \
567 (IV)((data)->flags), \
568 (IV)((data)->whilem_c), \
569 (IV)((data)->last_closep ? *((data)->last_closep) : -1) \
571 if ((data)->last_found) \
572 PerlIO_printf(Perl_debug_log, \
573 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
574 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
575 SvPVX_const((data)->last_found), \
576 (IV)((data)->last_end), \
577 (IV)((data)->last_start_min), \
578 (IV)((data)->last_start_max), \
579 ((data)->longest && \
580 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
581 SvPVX_const((data)->longest_fixed), \
582 (IV)((data)->offset_fixed), \
583 ((data)->longest && \
584 (data)->longest==&((data)->longest_float)) ? "*" : "", \
585 SvPVX_const((data)->longest_float), \
586 (IV)((data)->offset_float_min), \
587 (IV)((data)->offset_float_max) \
589 PerlIO_printf(Perl_debug_log,"\n"); \
592 static void clear_re(pTHX_ void *r);
594 /* Mark that we cannot extend a found fixed substring at this point.
595 Update the longest found anchored substring and the longest found
596 floating substrings if needed. */
599 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp)
601 const STRLEN l = CHR_SVLEN(data->last_found);
602 const STRLEN old_l = CHR_SVLEN(*data->longest);
603 GET_RE_DEBUG_FLAGS_DECL;
605 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
606 SvSetMagicSV(*data->longest, data->last_found);
607 if (*data->longest == data->longest_fixed) {
608 data->offset_fixed = l ? data->last_start_min : data->pos_min;
609 if (data->flags & SF_BEFORE_EOL)
611 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
613 data->flags &= ~SF_FIX_BEFORE_EOL;
614 data->minlen_fixed=minlenp;
615 data->lookbehind_fixed=0;
618 data->offset_float_min = l ? data->last_start_min : data->pos_min;
619 data->offset_float_max = (l
620 ? data->last_start_max
621 : data->pos_min + data->pos_delta);
622 if ((U32)data->offset_float_max > (U32)I32_MAX)
623 data->offset_float_max = I32_MAX;
624 if (data->flags & SF_BEFORE_EOL)
626 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
628 data->flags &= ~SF_FL_BEFORE_EOL;
629 data->minlen_float=minlenp;
630 data->lookbehind_float=0;
633 SvCUR_set(data->last_found, 0);
635 SV * const sv = data->last_found;
636 if (SvUTF8(sv) && SvMAGICAL(sv)) {
637 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
643 data->flags &= ~SF_BEFORE_EOL;
644 DEBUG_STUDYDATA(data,0);
647 /* Can match anything (initialization) */
649 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
651 ANYOF_CLASS_ZERO(cl);
652 ANYOF_BITMAP_SETALL(cl);
653 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
655 cl->flags |= ANYOF_LOCALE;
658 /* Can match anything (initialization) */
660 S_cl_is_anything(const struct regnode_charclass_class *cl)
664 for (value = 0; value <= ANYOF_MAX; value += 2)
665 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
667 if (!(cl->flags & ANYOF_UNICODE_ALL))
669 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
674 /* Can match anything (initialization) */
676 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
678 Zero(cl, 1, struct regnode_charclass_class);
680 cl_anything(pRExC_state, cl);
684 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
686 Zero(cl, 1, struct regnode_charclass_class);
688 cl_anything(pRExC_state, cl);
690 cl->flags |= ANYOF_LOCALE;
693 /* 'And' a given class with another one. Can create false positives */
694 /* We assume that cl is not inverted */
696 S_cl_and(struct regnode_charclass_class *cl,
697 const struct regnode_charclass_class *and_with)
700 assert(and_with->type == ANYOF);
701 if (!(and_with->flags & ANYOF_CLASS)
702 && !(cl->flags & ANYOF_CLASS)
703 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
704 && !(and_with->flags & ANYOF_FOLD)
705 && !(cl->flags & ANYOF_FOLD)) {
708 if (and_with->flags & ANYOF_INVERT)
709 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
710 cl->bitmap[i] &= ~and_with->bitmap[i];
712 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
713 cl->bitmap[i] &= and_with->bitmap[i];
714 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
715 if (!(and_with->flags & ANYOF_EOS))
716 cl->flags &= ~ANYOF_EOS;
718 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
719 !(and_with->flags & ANYOF_INVERT)) {
720 cl->flags &= ~ANYOF_UNICODE_ALL;
721 cl->flags |= ANYOF_UNICODE;
722 ARG_SET(cl, ARG(and_with));
724 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
725 !(and_with->flags & ANYOF_INVERT))
726 cl->flags &= ~ANYOF_UNICODE_ALL;
727 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
728 !(and_with->flags & ANYOF_INVERT))
729 cl->flags &= ~ANYOF_UNICODE;
732 /* 'OR' a given class with another one. Can create false positives */
733 /* We assume that cl is not inverted */
735 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
737 if (or_with->flags & ANYOF_INVERT) {
739 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
740 * <= (B1 | !B2) | (CL1 | !CL2)
741 * which is wasteful if CL2 is small, but we ignore CL2:
742 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
743 * XXXX Can we handle case-fold? Unclear:
744 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
745 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
747 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
748 && !(or_with->flags & ANYOF_FOLD)
749 && !(cl->flags & ANYOF_FOLD) ) {
752 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
753 cl->bitmap[i] |= ~or_with->bitmap[i];
754 } /* XXXX: logic is complicated otherwise */
756 cl_anything(pRExC_state, cl);
759 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
760 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
761 && (!(or_with->flags & ANYOF_FOLD)
762 || (cl->flags & ANYOF_FOLD)) ) {
765 /* OR char bitmap and class bitmap separately */
766 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
767 cl->bitmap[i] |= or_with->bitmap[i];
768 if (or_with->flags & ANYOF_CLASS) {
769 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
770 cl->classflags[i] |= or_with->classflags[i];
771 cl->flags |= ANYOF_CLASS;
774 else { /* XXXX: logic is complicated, leave it along for a moment. */
775 cl_anything(pRExC_state, cl);
778 if (or_with->flags & ANYOF_EOS)
779 cl->flags |= ANYOF_EOS;
781 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
782 ARG(cl) != ARG(or_with)) {
783 cl->flags |= ANYOF_UNICODE_ALL;
784 cl->flags &= ~ANYOF_UNICODE;
786 if (or_with->flags & ANYOF_UNICODE_ALL) {
787 cl->flags |= ANYOF_UNICODE_ALL;
788 cl->flags &= ~ANYOF_UNICODE;
792 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
793 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
794 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
795 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
800 dump_trie(trie,widecharmap,revcharmap)
801 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
802 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
804 These routines dump out a trie in a somewhat readable format.
805 The _interim_ variants are used for debugging the interim
806 tables that are used to generate the final compressed
807 representation which is what dump_trie expects.
809 Part of the reason for their existance is to provide a form
810 of documentation as to how the different representations function.
815 Dumps the final compressed table form of the trie to Perl_debug_log.
816 Used for debugging make_trie().
820 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
821 AV *revcharmap, U32 depth)
824 SV *sv=sv_newmortal();
825 int colwidth= widecharmap ? 6 : 4;
826 GET_RE_DEBUG_FLAGS_DECL;
829 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
830 (int)depth * 2 + 2,"",
831 "Match","Base","Ofs" );
833 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
834 SV ** const tmp = av_fetch( revcharmap, state, 0);
836 PerlIO_printf( Perl_debug_log, "%*s",
838 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
839 PL_colors[0], PL_colors[1],
840 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
841 PERL_PV_ESCAPE_FIRSTCHAR
846 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
847 (int)depth * 2 + 2,"");
849 for( state = 0 ; state < trie->uniquecharcount ; state++ )
850 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
851 PerlIO_printf( Perl_debug_log, "\n");
853 for( state = 1 ; state < trie->statecount ; state++ ) {
854 const U32 base = trie->states[ state ].trans.base;
856 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
858 if ( trie->states[ state ].wordnum ) {
859 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
861 PerlIO_printf( Perl_debug_log, "%6s", "" );
864 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
869 while( ( base + ofs < trie->uniquecharcount ) ||
870 ( base + ofs - trie->uniquecharcount < trie->lasttrans
871 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
874 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
876 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
877 if ( ( base + ofs >= trie->uniquecharcount ) &&
878 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
879 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
881 PerlIO_printf( Perl_debug_log, "%*"UVXf,
883 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
885 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
889 PerlIO_printf( Perl_debug_log, "]");
892 PerlIO_printf( Perl_debug_log, "\n" );
896 Dumps a fully constructed but uncompressed trie in list form.
897 List tries normally only are used for construction when the number of
898 possible chars (trie->uniquecharcount) is very high.
899 Used for debugging make_trie().
902 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
903 HV *widecharmap, AV *revcharmap, U32 next_alloc,
907 SV *sv=sv_newmortal();
908 int colwidth= widecharmap ? 6 : 4;
909 GET_RE_DEBUG_FLAGS_DECL;
910 /* print out the table precompression. */
911 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
912 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
913 "------:-----+-----------------\n" );
915 for( state=1 ; state < next_alloc ; state ++ ) {
918 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
919 (int)depth * 2 + 2,"", (UV)state );
920 if ( ! trie->states[ state ].wordnum ) {
921 PerlIO_printf( Perl_debug_log, "%5s| ","");
923 PerlIO_printf( Perl_debug_log, "W%4x| ",
924 trie->states[ state ].wordnum
927 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
928 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
930 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
932 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
933 PL_colors[0], PL_colors[1],
934 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
935 PERL_PV_ESCAPE_FIRSTCHAR
937 TRIE_LIST_ITEM(state,charid).forid,
938 (UV)TRIE_LIST_ITEM(state,charid).newstate
941 PerlIO_printf(Perl_debug_log, "\n%*s| ",
942 (int)((depth * 2) + 14), "");
945 PerlIO_printf( Perl_debug_log, "\n");
950 Dumps a fully constructed but uncompressed trie in table form.
951 This is the normal DFA style state transition table, with a few
952 twists to facilitate compression later.
953 Used for debugging make_trie().
956 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
957 HV *widecharmap, AV *revcharmap, U32 next_alloc,
962 SV *sv=sv_newmortal();
963 int colwidth= widecharmap ? 6 : 4;
964 GET_RE_DEBUG_FLAGS_DECL;
967 print out the table precompression so that we can do a visual check
968 that they are identical.
971 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
973 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
974 SV ** const tmp = av_fetch( revcharmap, charid, 0);
976 PerlIO_printf( Perl_debug_log, "%*s",
978 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
979 PL_colors[0], PL_colors[1],
980 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
981 PERL_PV_ESCAPE_FIRSTCHAR
987 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
989 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
990 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
993 PerlIO_printf( Perl_debug_log, "\n" );
995 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
997 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
998 (int)depth * 2 + 2,"",
999 (UV)TRIE_NODENUM( state ) );
1001 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1002 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1004 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1006 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1008 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1009 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1011 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1012 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1019 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1020 startbranch: the first branch in the whole branch sequence
1021 first : start branch of sequence of branch-exact nodes.
1022 May be the same as startbranch
1023 last : Thing following the last branch.
1024 May be the same as tail.
1025 tail : item following the branch sequence
1026 count : words in the sequence
1027 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1028 depth : indent depth
1030 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1032 A trie is an N'ary tree where the branches are determined by digital
1033 decomposition of the key. IE, at the root node you look up the 1st character and
1034 follow that branch repeat until you find the end of the branches. Nodes can be
1035 marked as "accepting" meaning they represent a complete word. Eg:
1039 would convert into the following structure. Numbers represent states, letters
1040 following numbers represent valid transitions on the letter from that state, if
1041 the number is in square brackets it represents an accepting state, otherwise it
1042 will be in parenthesis.
1044 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1048 (1) +-i->(6)-+-s->[7]
1050 +-s->(3)-+-h->(4)-+-e->[5]
1052 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1054 This shows that when matching against the string 'hers' we will begin at state 1
1055 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1056 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1057 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1058 single traverse. We store a mapping from accepting to state to which word was
1059 matched, and then when we have multiple possibilities we try to complete the
1060 rest of the regex in the order in which they occured in the alternation.
1062 The only prior NFA like behaviour that would be changed by the TRIE support is
1063 the silent ignoring of duplicate alternations which are of the form:
1065 / (DUPE|DUPE) X? (?{ ... }) Y /x
1067 Thus EVAL blocks follwing a trie may be called a different number of times with
1068 and without the optimisation. With the optimisations dupes will be silently
1069 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1070 the following demonstrates:
1072 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1074 which prints out 'word' three times, but
1076 'words'=~/(word|word|word)(?{ print $1 })S/
1078 which doesnt print it out at all. This is due to other optimisations kicking in.
1080 Example of what happens on a structural level:
1082 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1084 1: CURLYM[1] {1,32767}(18)
1095 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1096 and should turn into:
1098 1: CURLYM[1] {1,32767}(18)
1100 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1108 Cases where tail != last would be like /(?foo|bar)baz/:
1118 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1119 and would end up looking like:
1122 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1129 d = uvuni_to_utf8_flags(d, uv, 0);
1131 is the recommended Unicode-aware way of saying
1136 #define TRIE_STORE_REVCHAR \
1138 SV *tmp = newSVpvs(""); \
1139 if (UTF) SvUTF8_on(tmp); \
1140 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1141 av_push( revcharmap, tmp ); \
1144 #define TRIE_READ_CHAR STMT_START { \
1148 if ( foldlen > 0 ) { \
1149 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1154 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1155 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1156 foldlen -= UNISKIP( uvc ); \
1157 scan = foldbuf + UNISKIP( uvc ); \
1160 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1170 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1171 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1172 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1173 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1175 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1176 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1177 TRIE_LIST_CUR( state )++; \
1180 #define TRIE_LIST_NEW(state) STMT_START { \
1181 Newxz( trie->states[ state ].trans.list, \
1182 4, reg_trie_trans_le ); \
1183 TRIE_LIST_CUR( state ) = 1; \
1184 TRIE_LIST_LEN( state ) = 4; \
1187 #define TRIE_HANDLE_WORD(state) STMT_START { \
1188 U16 dupe= trie->states[ state ].wordnum; \
1189 regnode * const noper_next = regnext( noper ); \
1191 if (trie->wordlen) \
1192 trie->wordlen[ curword ] = wordlen; \
1194 /* store the word for dumping */ \
1196 if (OP(noper) != NOTHING) \
1197 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1199 tmp = newSVpvn( "", 0 ); \
1200 if ( UTF ) SvUTF8_on( tmp ); \
1201 av_push( trie_words, tmp ); \
1206 if ( noper_next < tail ) { \
1208 trie->jump = PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1209 trie->jump[curword] = (U16)(noper_next - convert); \
1211 jumper = noper_next; \
1213 nextbranch= regnext(cur); \
1217 /* So it's a dupe. This means we need to maintain a */\
1218 /* linked-list from the first to the next. */\
1219 /* we only allocate the nextword buffer when there */\
1220 /* a dupe, so first time we have to do the allocation */\
1221 if (!trie->nextword) \
1223 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1224 while ( trie->nextword[dupe] ) \
1225 dupe= trie->nextword[dupe]; \
1226 trie->nextword[dupe]= curword; \
1228 /* we haven't inserted this word yet. */ \
1229 trie->states[ state ].wordnum = curword; \
1234 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1235 ( ( base + charid >= ucharcount \
1236 && base + charid < ubound \
1237 && state == trie->trans[ base - ucharcount + charid ].check \
1238 && trie->trans[ base - ucharcount + charid ].next ) \
1239 ? trie->trans[ base - ucharcount + charid ].next \
1240 : ( state==1 ? special : 0 ) \
1244 #define MADE_JUMP_TRIE 2
1245 #define MADE_EXACT_TRIE 4
1248 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1251 /* first pass, loop through and scan words */
1252 reg_trie_data *trie;
1253 HV *widecharmap = NULL;
1254 AV *revcharmap = newAV();
1256 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1261 regnode *jumper = NULL;
1262 regnode *nextbranch = NULL;
1263 regnode *convert = NULL;
1264 /* we just use folder as a flag in utf8 */
1265 const U8 * const folder = ( flags == EXACTF
1267 : ( flags == EXACTFL
1274 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1275 AV *trie_words = NULL;
1276 /* along with revcharmap, this only used during construction but both are
1277 * useful during debugging so we store them in the struct when debugging.
1280 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1281 STRLEN trie_charcount=0;
1283 SV *re_trie_maxbuff;
1284 GET_RE_DEBUG_FLAGS_DECL;
1286 PERL_UNUSED_ARG(depth);
1289 trie = PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1291 trie->startstate = 1;
1292 trie->wordcount = word_count;
1293 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1294 trie->charmap = PerlMemShared_calloc( 256, sizeof(U16) );
1295 if (!(UTF && folder))
1296 trie->bitmap = PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1298 trie_words = newAV();
1301 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1302 if (!SvIOK(re_trie_maxbuff)) {
1303 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1306 PerlIO_printf( Perl_debug_log,
1307 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1308 (int)depth * 2 + 2, "",
1309 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1310 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1314 /* Find the node we are going to overwrite */
1315 if ( first == startbranch && OP( last ) != BRANCH ) {
1316 /* whole branch chain */
1319 /* branch sub-chain */
1320 convert = NEXTOPER( first );
1323 /* -- First loop and Setup --
1325 We first traverse the branches and scan each word to determine if it
1326 contains widechars, and how many unique chars there are, this is
1327 important as we have to build a table with at least as many columns as we
1330 We use an array of integers to represent the character codes 0..255
1331 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1332 native representation of the character value as the key and IV's for the
1335 *TODO* If we keep track of how many times each character is used we can
1336 remap the columns so that the table compression later on is more
1337 efficient in terms of memory by ensuring most common value is in the
1338 middle and the least common are on the outside. IMO this would be better
1339 than a most to least common mapping as theres a decent chance the most
1340 common letter will share a node with the least common, meaning the node
1341 will not be compressable. With a middle is most common approach the worst
1342 case is when we have the least common nodes twice.
1346 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1347 regnode * const noper = NEXTOPER( cur );
1348 const U8 *uc = (U8*)STRING( noper );
1349 const U8 * const e = uc + STR_LEN( noper );
1351 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1352 const U8 *scan = (U8*)NULL;
1353 U32 wordlen = 0; /* required init */
1356 if (OP(noper) == NOTHING) {
1361 TRIE_BITMAP_SET(trie,*uc);
1362 if ( folder ) TRIE_BITMAP_SET(trie,folder[ *uc ]);
1364 for ( ; uc < e ; uc += len ) {
1365 TRIE_CHARCOUNT(trie)++;
1369 if ( !trie->charmap[ uvc ] ) {
1370 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1372 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1378 widecharmap = newHV();
1380 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1383 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1385 if ( !SvTRUE( *svpp ) ) {
1386 sv_setiv( *svpp, ++trie->uniquecharcount );
1391 if( cur == first ) {
1394 } else if (chars < trie->minlen) {
1396 } else if (chars > trie->maxlen) {
1400 } /* end first pass */
1401 DEBUG_TRIE_COMPILE_r(
1402 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1403 (int)depth * 2 + 2,"",
1404 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1405 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1406 (int)trie->minlen, (int)trie->maxlen )
1408 trie->wordlen = PerlMemShared_calloc( word_count, sizeof(U32) );
1411 We now know what we are dealing with in terms of unique chars and
1412 string sizes so we can calculate how much memory a naive
1413 representation using a flat table will take. If it's over a reasonable
1414 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1415 conservative but potentially much slower representation using an array
1418 At the end we convert both representations into the same compressed
1419 form that will be used in regexec.c for matching with. The latter
1420 is a form that cannot be used to construct with but has memory
1421 properties similar to the list form and access properties similar
1422 to the table form making it both suitable for fast searches and
1423 small enough that its feasable to store for the duration of a program.
1425 See the comment in the code where the compressed table is produced
1426 inplace from the flat tabe representation for an explanation of how
1427 the compression works.
1432 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1434 Second Pass -- Array Of Lists Representation
1436 Each state will be represented by a list of charid:state records
1437 (reg_trie_trans_le) the first such element holds the CUR and LEN
1438 points of the allocated array. (See defines above).
1440 We build the initial structure using the lists, and then convert
1441 it into the compressed table form which allows faster lookups
1442 (but cant be modified once converted).
1445 STRLEN transcount = 1;
1447 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1448 "%*sCompiling trie using list compiler\n",
1449 (int)depth * 2 + 2, ""));
1451 trie->states = PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1452 sizeof(reg_trie_state) );
1456 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1458 regnode * const noper = NEXTOPER( cur );
1459 U8 *uc = (U8*)STRING( noper );
1460 const U8 * const e = uc + STR_LEN( noper );
1461 U32 state = 1; /* required init */
1462 U16 charid = 0; /* sanity init */
1463 U8 *scan = (U8*)NULL; /* sanity init */
1464 STRLEN foldlen = 0; /* required init */
1465 U32 wordlen = 0; /* required init */
1466 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1468 if (OP(noper) != NOTHING) {
1469 for ( ; uc < e ; uc += len ) {
1474 charid = trie->charmap[ uvc ];
1476 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1480 charid=(U16)SvIV( *svpp );
1483 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1490 if ( !trie->states[ state ].trans.list ) {
1491 TRIE_LIST_NEW( state );
1493 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1494 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1495 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1500 newstate = next_alloc++;
1501 TRIE_LIST_PUSH( state, charid, newstate );
1506 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1510 TRIE_HANDLE_WORD(state);
1512 } /* end second pass */
1514 /* next alloc is the NEXT state to be allocated */
1515 trie->statecount = next_alloc;
1516 trie->states = PerlMemShared_realloc( trie->states, next_alloc
1517 * sizeof(reg_trie_state) );
1519 /* and now dump it out before we compress it */
1520 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1521 revcharmap, next_alloc,
1526 = PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1533 for( state=1 ; state < next_alloc ; state ++ ) {
1537 DEBUG_TRIE_COMPILE_MORE_r(
1538 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1542 if (trie->states[state].trans.list) {
1543 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1547 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1548 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1549 if ( forid < minid ) {
1551 } else if ( forid > maxid ) {
1555 if ( transcount < tp + maxid - minid + 1) {
1558 = PerlMemShared_realloc( trie->trans,
1560 * sizeof(reg_trie_trans) );
1561 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1563 base = trie->uniquecharcount + tp - minid;
1564 if ( maxid == minid ) {
1566 for ( ; zp < tp ; zp++ ) {
1567 if ( ! trie->trans[ zp ].next ) {
1568 base = trie->uniquecharcount + zp - minid;
1569 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1570 trie->trans[ zp ].check = state;
1576 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1577 trie->trans[ tp ].check = state;
1582 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1583 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1584 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1585 trie->trans[ tid ].check = state;
1587 tp += ( maxid - minid + 1 );
1589 Safefree(trie->states[ state ].trans.list);
1592 DEBUG_TRIE_COMPILE_MORE_r(
1593 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1596 trie->states[ state ].trans.base=base;
1598 trie->lasttrans = tp + 1;
1602 Second Pass -- Flat Table Representation.
1604 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1605 We know that we will need Charcount+1 trans at most to store the data
1606 (one row per char at worst case) So we preallocate both structures
1607 assuming worst case.
1609 We then construct the trie using only the .next slots of the entry
1612 We use the .check field of the first entry of the node temporarily to
1613 make compression both faster and easier by keeping track of how many non
1614 zero fields are in the node.
1616 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1619 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1620 number representing the first entry of the node, and state as a
1621 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1622 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1623 are 2 entrys per node. eg:
1631 The table is internally in the right hand, idx form. However as we also
1632 have to deal with the states array which is indexed by nodenum we have to
1633 use TRIE_NODENUM() to convert.
1636 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1637 "%*sCompiling trie using table compiler\n",
1638 (int)depth * 2 + 2, ""));
1640 trie->trans = PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1641 * trie->uniquecharcount + 1,
1642 sizeof(reg_trie_trans) );
1643 trie->states = PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1644 sizeof(reg_trie_state) );
1645 next_alloc = trie->uniquecharcount + 1;
1648 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1650 regnode * const noper = NEXTOPER( cur );
1651 const U8 *uc = (U8*)STRING( noper );
1652 const U8 * const e = uc + STR_LEN( noper );
1654 U32 state = 1; /* required init */
1656 U16 charid = 0; /* sanity init */
1657 U32 accept_state = 0; /* sanity init */
1658 U8 *scan = (U8*)NULL; /* sanity init */
1660 STRLEN foldlen = 0; /* required init */
1661 U32 wordlen = 0; /* required init */
1662 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1664 if ( OP(noper) != NOTHING ) {
1665 for ( ; uc < e ; uc += len ) {
1670 charid = trie->charmap[ uvc ];
1672 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1673 charid = svpp ? (U16)SvIV(*svpp) : 0;
1677 if ( !trie->trans[ state + charid ].next ) {
1678 trie->trans[ state + charid ].next = next_alloc;
1679 trie->trans[ state ].check++;
1680 next_alloc += trie->uniquecharcount;
1682 state = trie->trans[ state + charid ].next;
1684 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1686 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1689 accept_state = TRIE_NODENUM( state );
1690 TRIE_HANDLE_WORD(accept_state);
1692 } /* end second pass */
1694 /* and now dump it out before we compress it */
1695 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1697 next_alloc, depth+1));
1701 * Inplace compress the table.*
1703 For sparse data sets the table constructed by the trie algorithm will
1704 be mostly 0/FAIL transitions or to put it another way mostly empty.
1705 (Note that leaf nodes will not contain any transitions.)
1707 This algorithm compresses the tables by eliminating most such
1708 transitions, at the cost of a modest bit of extra work during lookup:
1710 - Each states[] entry contains a .base field which indicates the
1711 index in the state[] array wheres its transition data is stored.
1713 - If .base is 0 there are no valid transitions from that node.
1715 - If .base is nonzero then charid is added to it to find an entry in
1718 -If trans[states[state].base+charid].check!=state then the
1719 transition is taken to be a 0/Fail transition. Thus if there are fail
1720 transitions at the front of the node then the .base offset will point
1721 somewhere inside the previous nodes data (or maybe even into a node
1722 even earlier), but the .check field determines if the transition is
1726 The following process inplace converts the table to the compressed
1727 table: We first do not compress the root node 1,and mark its all its
1728 .check pointers as 1 and set its .base pointer as 1 as well. This
1729 allows to do a DFA construction from the compressed table later, and
1730 ensures that any .base pointers we calculate later are greater than
1733 - We set 'pos' to indicate the first entry of the second node.
1735 - We then iterate over the columns of the node, finding the first and
1736 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1737 and set the .check pointers accordingly, and advance pos
1738 appropriately and repreat for the next node. Note that when we copy
1739 the next pointers we have to convert them from the original
1740 NODEIDX form to NODENUM form as the former is not valid post
1743 - If a node has no transitions used we mark its base as 0 and do not
1744 advance the pos pointer.
1746 - If a node only has one transition we use a second pointer into the
1747 structure to fill in allocated fail transitions from other states.
1748 This pointer is independent of the main pointer and scans forward
1749 looking for null transitions that are allocated to a state. When it
1750 finds one it writes the single transition into the "hole". If the
1751 pointer doesnt find one the single transition is appended as normal.
1753 - Once compressed we can Renew/realloc the structures to release the
1756 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1757 specifically Fig 3.47 and the associated pseudocode.
1761 const U32 laststate = TRIE_NODENUM( next_alloc );
1764 trie->statecount = laststate;
1766 for ( state = 1 ; state < laststate ; state++ ) {
1768 const U32 stateidx = TRIE_NODEIDX( state );
1769 const U32 o_used = trie->trans[ stateidx ].check;
1770 U32 used = trie->trans[ stateidx ].check;
1771 trie->trans[ stateidx ].check = 0;
1773 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1774 if ( flag || trie->trans[ stateidx + charid ].next ) {
1775 if ( trie->trans[ stateidx + charid ].next ) {
1777 for ( ; zp < pos ; zp++ ) {
1778 if ( ! trie->trans[ zp ].next ) {
1782 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1783 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1784 trie->trans[ zp ].check = state;
1785 if ( ++zp > pos ) pos = zp;
1792 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1794 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1795 trie->trans[ pos ].check = state;
1800 trie->lasttrans = pos + 1;
1801 trie->states = PerlMemShared_realloc( trie->states, laststate
1802 * sizeof(reg_trie_state) );
1803 DEBUG_TRIE_COMPILE_MORE_r(
1804 PerlIO_printf( Perl_debug_log,
1805 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1806 (int)depth * 2 + 2,"",
1807 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1810 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1813 } /* end table compress */
1815 DEBUG_TRIE_COMPILE_MORE_r(
1816 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1817 (int)depth * 2 + 2, "",
1818 (UV)trie->statecount,
1819 (UV)trie->lasttrans)
1821 /* resize the trans array to remove unused space */
1822 trie->trans = PerlMemShared_realloc( trie->trans, trie->lasttrans
1823 * sizeof(reg_trie_trans) );
1825 /* and now dump out the compressed format */
1826 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1828 { /* Modify the program and insert the new TRIE node*/
1829 U8 nodetype =(U8)(flags & 0xFF);
1833 regnode *optimize = NULL;
1835 U32 mjd_nodelen = 0;
1838 This means we convert either the first branch or the first Exact,
1839 depending on whether the thing following (in 'last') is a branch
1840 or not and whther first is the startbranch (ie is it a sub part of
1841 the alternation or is it the whole thing.)
1842 Assuming its a sub part we conver the EXACT otherwise we convert
1843 the whole branch sequence, including the first.
1845 /* Find the node we are going to overwrite */
1846 if ( first != startbranch || OP( last ) == BRANCH ) {
1847 /* branch sub-chain */
1848 NEXT_OFF( first ) = (U16)(last - first);
1850 mjd_offset= Node_Offset((convert));
1851 mjd_nodelen= Node_Length((convert));
1853 /* whole branch chain */
1856 const regnode *nop = NEXTOPER( convert );
1857 mjd_offset= Node_Offset((nop));
1858 mjd_nodelen= Node_Length((nop));
1863 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1864 (int)depth * 2 + 2, "",
1865 (UV)mjd_offset, (UV)mjd_nodelen)
1868 /* But first we check to see if there is a common prefix we can
1869 split out as an EXACT and put in front of the TRIE node. */
1870 trie->startstate= 1;
1871 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1873 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1877 const U32 base = trie->states[ state ].trans.base;
1879 if ( trie->states[state].wordnum )
1882 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1883 if ( ( base + ofs >= trie->uniquecharcount ) &&
1884 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1885 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1887 if ( ++count > 1 ) {
1888 SV **tmp = av_fetch( revcharmap, ofs, 0);
1889 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1890 if ( state == 1 ) break;
1892 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1894 PerlIO_printf(Perl_debug_log,
1895 "%*sNew Start State=%"UVuf" Class: [",
1896 (int)depth * 2 + 2, "",
1899 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1900 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1902 TRIE_BITMAP_SET(trie,*ch);
1904 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1906 PerlIO_printf(Perl_debug_log, (char*)ch)
1910 TRIE_BITMAP_SET(trie,*ch);
1912 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1913 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1919 SV **tmp = av_fetch( revcharmap, idx, 0);
1920 char *ch = SvPV_nolen( *tmp );
1922 SV *sv=sv_newmortal();
1923 PerlIO_printf( Perl_debug_log,
1924 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1925 (int)depth * 2 + 2, "",
1927 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1928 PL_colors[0], PL_colors[1],
1929 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1930 PERL_PV_ESCAPE_FIRSTCHAR
1935 OP( convert ) = nodetype;
1936 str=STRING(convert);
1947 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
1953 regnode *n = convert+NODE_SZ_STR(convert);
1954 NEXT_OFF(convert) = NODE_SZ_STR(convert);
1955 trie->startstate = state;
1956 trie->minlen -= (state - 1);
1957 trie->maxlen -= (state - 1);
1959 regnode *fix = convert;
1960 U32 word = trie->wordcount;
1962 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
1963 while( ++fix < n ) {
1964 Set_Node_Offset_Length(fix, 0, 0);
1967 SV ** const tmp = av_fetch( trie_words, word, 0 );
1969 if ( STR_LEN(convert) <= SvCUR(*tmp) )
1970 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
1972 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
1979 NEXT_OFF(convert) = (U16)(tail - convert);
1980 DEBUG_r(optimize= n);
1986 if ( trie->maxlen ) {
1987 NEXT_OFF( convert ) = (U16)(tail - convert);
1988 ARG_SET( convert, data_slot );
1989 /* Store the offset to the first unabsorbed branch in
1990 jump[0], which is otherwise unused by the jump logic.
1991 We use this when dumping a trie and during optimisation. */
1993 trie->jump[0] = (U16)(nextbranch - convert);
1996 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
1997 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
1999 OP( convert ) = TRIEC;
2000 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2001 PerlMemShared_free(trie->bitmap);
2004 OP( convert ) = TRIE;
2006 /* store the type in the flags */
2007 convert->flags = nodetype;
2011 + regarglen[ OP( convert ) ];
2013 /* XXX We really should free up the resource in trie now,
2014 as we won't use them - (which resources?) dmq */
2016 /* needed for dumping*/
2017 DEBUG_r(if (optimize) {
2018 regnode *opt = convert;
2019 while ( ++opt < optimize) {
2020 Set_Node_Offset_Length(opt,0,0);
2023 Try to clean up some of the debris left after the
2026 while( optimize < jumper ) {
2027 mjd_nodelen += Node_Length((optimize));
2028 OP( optimize ) = OPTIMIZED;
2029 Set_Node_Offset_Length(optimize,0,0);
2032 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2034 } /* end node insert */
2035 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2037 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2038 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2040 SvREFCNT_dec(revcharmap);
2044 : trie->startstate>1
2050 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2052 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2054 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2055 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2058 We find the fail state for each state in the trie, this state is the longest proper
2059 suffix of the current states 'word' that is also a proper prefix of another word in our
2060 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2061 the DFA not to have to restart after its tried and failed a word at a given point, it
2062 simply continues as though it had been matching the other word in the first place.
2064 'abcdgu'=~/abcdefg|cdgu/
2065 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2066 fail, which would bring use to the state representing 'd' in the second word where we would
2067 try 'g' and succeed, prodceding to match 'cdgu'.
2069 /* add a fail transition */
2070 const U32 trie_offset = ARG(source);
2071 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2073 const U32 ucharcount = trie->uniquecharcount;
2074 const U32 numstates = trie->statecount;
2075 const U32 ubound = trie->lasttrans + ucharcount;
2079 U32 base = trie->states[ 1 ].trans.base;
2082 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2083 GET_RE_DEBUG_FLAGS_DECL;
2085 PERL_UNUSED_ARG(depth);
2089 ARG_SET( stclass, data_slot );
2090 aho = PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2091 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2092 aho->trie=trie_offset;
2093 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2094 Copy( trie->states, aho->states, numstates, reg_trie_state );
2095 Newxz( q, numstates, U32);
2096 aho->fail = PerlMemShared_calloc( numstates, sizeof(U32) );
2099 /* initialize fail[0..1] to be 1 so that we always have
2100 a valid final fail state */
2101 fail[ 0 ] = fail[ 1 ] = 1;
2103 for ( charid = 0; charid < ucharcount ; charid++ ) {
2104 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2106 q[ q_write ] = newstate;
2107 /* set to point at the root */
2108 fail[ q[ q_write++ ] ]=1;
2111 while ( q_read < q_write) {
2112 const U32 cur = q[ q_read++ % numstates ];
2113 base = trie->states[ cur ].trans.base;
2115 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2116 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2118 U32 fail_state = cur;
2121 fail_state = fail[ fail_state ];
2122 fail_base = aho->states[ fail_state ].trans.base;
2123 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2125 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2126 fail[ ch_state ] = fail_state;
2127 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2129 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2131 q[ q_write++ % numstates] = ch_state;
2135 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2136 when we fail in state 1, this allows us to use the
2137 charclass scan to find a valid start char. This is based on the principle
2138 that theres a good chance the string being searched contains lots of stuff
2139 that cant be a start char.
2141 fail[ 0 ] = fail[ 1 ] = 0;
2142 DEBUG_TRIE_COMPILE_r({
2143 PerlIO_printf(Perl_debug_log,
2144 "%*sStclass Failtable (%"UVuf" states): 0",
2145 (int)(depth * 2), "", (UV)numstates
2147 for( q_read=1; q_read<numstates; q_read++ ) {
2148 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2150 PerlIO_printf(Perl_debug_log, "\n");
2153 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2158 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2159 * These need to be revisited when a newer toolchain becomes available.
2161 #if defined(__sparc64__) && defined(__GNUC__)
2162 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2163 # undef SPARC64_GCC_WORKAROUND
2164 # define SPARC64_GCC_WORKAROUND 1
2168 #define DEBUG_PEEP(str,scan,depth) \
2169 DEBUG_OPTIMISE_r({if (scan){ \
2170 SV * const mysv=sv_newmortal(); \
2171 regnode *Next = regnext(scan); \
2172 regprop(RExC_rx, mysv, scan); \
2173 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2174 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2175 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2182 #define JOIN_EXACT(scan,min,flags) \
2183 if (PL_regkind[OP(scan)] == EXACT) \
2184 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2187 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2188 /* Merge several consecutive EXACTish nodes into one. */
2189 regnode *n = regnext(scan);
2191 regnode *next = scan + NODE_SZ_STR(scan);
2195 regnode *stop = scan;
2196 GET_RE_DEBUG_FLAGS_DECL;
2198 PERL_UNUSED_ARG(depth);
2200 #ifndef EXPERIMENTAL_INPLACESCAN
2201 PERL_UNUSED_ARG(flags);
2202 PERL_UNUSED_ARG(val);
2204 DEBUG_PEEP("join",scan,depth);
2206 /* Skip NOTHING, merge EXACT*. */
2208 ( PL_regkind[OP(n)] == NOTHING ||
2209 (stringok && (OP(n) == OP(scan))))
2211 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2213 if (OP(n) == TAIL || n > next)
2215 if (PL_regkind[OP(n)] == NOTHING) {
2216 DEBUG_PEEP("skip:",n,depth);
2217 NEXT_OFF(scan) += NEXT_OFF(n);
2218 next = n + NODE_STEP_REGNODE;
2225 else if (stringok) {
2226 const unsigned int oldl = STR_LEN(scan);
2227 regnode * const nnext = regnext(n);
2229 DEBUG_PEEP("merg",n,depth);
2232 if (oldl + STR_LEN(n) > U8_MAX)
2234 NEXT_OFF(scan) += NEXT_OFF(n);
2235 STR_LEN(scan) += STR_LEN(n);
2236 next = n + NODE_SZ_STR(n);
2237 /* Now we can overwrite *n : */
2238 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2246 #ifdef EXPERIMENTAL_INPLACESCAN
2247 if (flags && !NEXT_OFF(n)) {
2248 DEBUG_PEEP("atch", val, depth);
2249 if (reg_off_by_arg[OP(n)]) {
2250 ARG_SET(n, val - n);
2253 NEXT_OFF(n) = val - n;
2260 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2262 Two problematic code points in Unicode casefolding of EXACT nodes:
2264 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2265 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2271 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2272 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2274 This means that in case-insensitive matching (or "loose matching",
2275 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2276 length of the above casefolded versions) can match a target string
2277 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2278 This would rather mess up the minimum length computation.
2280 What we'll do is to look for the tail four bytes, and then peek
2281 at the preceding two bytes to see whether we need to decrease
2282 the minimum length by four (six minus two).
2284 Thanks to the design of UTF-8, there cannot be false matches:
2285 A sequence of valid UTF-8 bytes cannot be a subsequence of
2286 another valid sequence of UTF-8 bytes.
2289 char * const s0 = STRING(scan), *s, *t;
2290 char * const s1 = s0 + STR_LEN(scan) - 1;
2291 char * const s2 = s1 - 4;
2292 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2293 const char t0[] = "\xaf\x49\xaf\x42";
2295 const char t0[] = "\xcc\x88\xcc\x81";
2297 const char * const t1 = t0 + 3;
2300 s < s2 && (t = ninstr(s, s1, t0, t1));
2303 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2304 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2306 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2307 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2315 n = scan + NODE_SZ_STR(scan);
2317 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2324 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2328 /* REx optimizer. Converts nodes into quickier variants "in place".
2329 Finds fixed substrings. */
2331 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2332 to the position after last scanned or to NULL. */
2334 #define INIT_AND_WITHP \
2335 assert(!and_withp); \
2336 Newx(and_withp,1,struct regnode_charclass_class); \
2337 SAVEFREEPV(and_withp)
2339 /* this is a chain of data about sub patterns we are processing that
2340 need to be handled seperately/specially in study_chunk. Its so
2341 we can simulate recursion without losing state. */
2343 typedef struct scan_frame {
2344 regnode *last; /* last node to process in this frame */
2345 regnode *next; /* next node to process when last is reached */
2346 struct scan_frame *prev; /*previous frame*/
2347 I32 stop; /* what stopparen do we use */
2351 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2352 I32 *minlenp, I32 *deltap,
2357 struct regnode_charclass_class *and_withp,
2358 U32 flags, U32 depth)
2359 /* scanp: Start here (read-write). */
2360 /* deltap: Write maxlen-minlen here. */
2361 /* last: Stop before this one. */
2362 /* data: string data about the pattern */
2363 /* stopparen: treat close N as END */
2364 /* recursed: which subroutines have we recursed into */
2365 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2368 I32 min = 0, pars = 0, code;
2369 regnode *scan = *scanp, *next;
2371 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2372 int is_inf_internal = 0; /* The studied chunk is infinite */
2373 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2374 scan_data_t data_fake;
2375 SV *re_trie_maxbuff = NULL;
2376 regnode *first_non_open = scan;
2377 I32 stopmin = I32_MAX;
2378 scan_frame *frame = NULL;
2380 GET_RE_DEBUG_FLAGS_DECL;
2383 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2387 while (first_non_open && OP(first_non_open) == OPEN)
2388 first_non_open=regnext(first_non_open);
2393 while ( scan && OP(scan) != END && scan < last ){
2394 /* Peephole optimizer: */
2395 DEBUG_STUDYDATA(data,depth);
2396 DEBUG_PEEP("Peep",scan,depth);
2397 JOIN_EXACT(scan,&min,0);
2399 /* Follow the next-chain of the current node and optimize
2400 away all the NOTHINGs from it. */
2401 if (OP(scan) != CURLYX) {
2402 const int max = (reg_off_by_arg[OP(scan)]
2404 /* I32 may be smaller than U16 on CRAYs! */
2405 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2406 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2410 /* Skip NOTHING and LONGJMP. */
2411 while ((n = regnext(n))
2412 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2413 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2414 && off + noff < max)
2416 if (reg_off_by_arg[OP(scan)])
2419 NEXT_OFF(scan) = off;
2424 /* The principal pseudo-switch. Cannot be a switch, since we
2425 look into several different things. */
2426 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2427 || OP(scan) == IFTHEN) {
2428 next = regnext(scan);
2430 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2432 if (OP(next) == code || code == IFTHEN) {
2433 /* NOTE - There is similar code to this block below for handling
2434 TRIE nodes on a re-study. If you change stuff here check there
2436 I32 max1 = 0, min1 = I32_MAX, num = 0;
2437 struct regnode_charclass_class accum;
2438 regnode * const startbranch=scan;
2440 if (flags & SCF_DO_SUBSTR)
2441 scan_commit(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2442 if (flags & SCF_DO_STCLASS)
2443 cl_init_zero(pRExC_state, &accum);
2445 while (OP(scan) == code) {
2446 I32 deltanext, minnext, f = 0, fake;
2447 struct regnode_charclass_class this_class;
2450 data_fake.flags = 0;
2452 data_fake.whilem_c = data->whilem_c;
2453 data_fake.last_closep = data->last_closep;
2456 data_fake.last_closep = &fake;
2458 data_fake.pos_delta = delta;
2459 next = regnext(scan);
2460 scan = NEXTOPER(scan);
2462 scan = NEXTOPER(scan);
2463 if (flags & SCF_DO_STCLASS) {
2464 cl_init(pRExC_state, &this_class);
2465 data_fake.start_class = &this_class;
2466 f = SCF_DO_STCLASS_AND;
2468 if (flags & SCF_WHILEM_VISITED_POS)
2469 f |= SCF_WHILEM_VISITED_POS;
2471 /* we suppose the run is continuous, last=next...*/
2472 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2474 stopparen, recursed, NULL, f,depth+1);
2477 if (max1 < minnext + deltanext)
2478 max1 = minnext + deltanext;
2479 if (deltanext == I32_MAX)
2480 is_inf = is_inf_internal = 1;
2482 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2484 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2485 if ( stopmin > minnext)
2486 stopmin = min + min1;
2487 flags &= ~SCF_DO_SUBSTR;
2489 data->flags |= SCF_SEEN_ACCEPT;
2492 if (data_fake.flags & SF_HAS_EVAL)
2493 data->flags |= SF_HAS_EVAL;
2494 data->whilem_c = data_fake.whilem_c;
2496 if (flags & SCF_DO_STCLASS)
2497 cl_or(pRExC_state, &accum, &this_class);
2499 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2501 if (flags & SCF_DO_SUBSTR) {
2502 data->pos_min += min1;
2503 data->pos_delta += max1 - min1;
2504 if (max1 != min1 || is_inf)
2505 data->longest = &(data->longest_float);
2508 delta += max1 - min1;
2509 if (flags & SCF_DO_STCLASS_OR) {
2510 cl_or(pRExC_state, data->start_class, &accum);
2512 cl_and(data->start_class, and_withp);
2513 flags &= ~SCF_DO_STCLASS;
2516 else if (flags & SCF_DO_STCLASS_AND) {
2518 cl_and(data->start_class, &accum);
2519 flags &= ~SCF_DO_STCLASS;
2522 /* Switch to OR mode: cache the old value of
2523 * data->start_class */
2525 StructCopy(data->start_class, and_withp,
2526 struct regnode_charclass_class);
2527 flags &= ~SCF_DO_STCLASS_AND;
2528 StructCopy(&accum, data->start_class,
2529 struct regnode_charclass_class);
2530 flags |= SCF_DO_STCLASS_OR;
2531 data->start_class->flags |= ANYOF_EOS;
2535 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2538 Assuming this was/is a branch we are dealing with: 'scan' now
2539 points at the item that follows the branch sequence, whatever
2540 it is. We now start at the beginning of the sequence and look
2547 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2549 If we can find such a subseqence we need to turn the first
2550 element into a trie and then add the subsequent branch exact
2551 strings to the trie.
2555 1. patterns where the whole set of branch can be converted.
2557 2. patterns where only a subset can be converted.
2559 In case 1 we can replace the whole set with a single regop
2560 for the trie. In case 2 we need to keep the start and end
2563 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2564 becomes BRANCH TRIE; BRANCH X;
2566 There is an additional case, that being where there is a
2567 common prefix, which gets split out into an EXACT like node
2568 preceding the TRIE node.
2570 If x(1..n)==tail then we can do a simple trie, if not we make
2571 a "jump" trie, such that when we match the appropriate word
2572 we "jump" to the appopriate tail node. Essentailly we turn
2573 a nested if into a case structure of sorts.
2578 if (!re_trie_maxbuff) {
2579 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2580 if (!SvIOK(re_trie_maxbuff))
2581 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2583 if ( SvIV(re_trie_maxbuff)>=0 ) {
2585 regnode *first = (regnode *)NULL;
2586 regnode *last = (regnode *)NULL;
2587 regnode *tail = scan;
2592 SV * const mysv = sv_newmortal(); /* for dumping */
2594 /* var tail is used because there may be a TAIL
2595 regop in the way. Ie, the exacts will point to the
2596 thing following the TAIL, but the last branch will
2597 point at the TAIL. So we advance tail. If we
2598 have nested (?:) we may have to move through several
2602 while ( OP( tail ) == TAIL ) {
2603 /* this is the TAIL generated by (?:) */
2604 tail = regnext( tail );
2609 regprop(RExC_rx, mysv, tail );
2610 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2611 (int)depth * 2 + 2, "",
2612 "Looking for TRIE'able sequences. Tail node is: ",
2613 SvPV_nolen_const( mysv )
2619 step through the branches, cur represents each
2620 branch, noper is the first thing to be matched
2621 as part of that branch and noper_next is the
2622 regnext() of that node. if noper is an EXACT
2623 and noper_next is the same as scan (our current
2624 position in the regex) then the EXACT branch is
2625 a possible optimization target. Once we have
2626 two or more consequetive such branches we can
2627 create a trie of the EXACT's contents and stich
2628 it in place. If the sequence represents all of
2629 the branches we eliminate the whole thing and
2630 replace it with a single TRIE. If it is a
2631 subsequence then we need to stitch it in. This
2632 means the first branch has to remain, and needs
2633 to be repointed at the item on the branch chain
2634 following the last branch optimized. This could
2635 be either a BRANCH, in which case the
2636 subsequence is internal, or it could be the
2637 item following the branch sequence in which
2638 case the subsequence is at the end.
2642 /* dont use tail as the end marker for this traverse */
2643 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2644 regnode * const noper = NEXTOPER( cur );
2645 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2646 regnode * const noper_next = regnext( noper );
2650 regprop(RExC_rx, mysv, cur);
2651 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2652 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2654 regprop(RExC_rx, mysv, noper);
2655 PerlIO_printf( Perl_debug_log, " -> %s",
2656 SvPV_nolen_const(mysv));
2659 regprop(RExC_rx, mysv, noper_next );
2660 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2661 SvPV_nolen_const(mysv));
2663 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2664 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2666 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2667 : PL_regkind[ OP( noper ) ] == EXACT )
2668 || OP(noper) == NOTHING )
2670 && noper_next == tail
2675 if ( !first || optype == NOTHING ) {
2676 if (!first) first = cur;
2677 optype = OP( noper );
2683 make_trie( pRExC_state,
2684 startbranch, first, cur, tail, count,
2687 if ( PL_regkind[ OP( noper ) ] == EXACT
2689 && noper_next == tail
2694 optype = OP( noper );
2704 regprop(RExC_rx, mysv, cur);
2705 PerlIO_printf( Perl_debug_log,
2706 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2707 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2711 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2712 #ifdef TRIE_STUDY_OPT
2713 if ( ((made == MADE_EXACT_TRIE &&
2714 startbranch == first)
2715 || ( first_non_open == first )) &&
2717 flags |= SCF_TRIE_RESTUDY;
2718 if ( startbranch == first
2721 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2731 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2732 scan = NEXTOPER(NEXTOPER(scan));
2733 } else /* single branch is optimized. */
2734 scan = NEXTOPER(scan);
2736 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2737 scan_frame *newframe = NULL;
2742 if (OP(scan) != SUSPEND) {
2743 /* set the pointer */
2744 if (OP(scan) == GOSUB) {
2746 RExC_recurse[ARG2L(scan)] = scan;
2747 start = RExC_open_parens[paren-1];
2748 end = RExC_close_parens[paren-1];
2751 start = RExC_rxi->program + 1;
2755 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2756 SAVEFREEPV(recursed);
2758 if (!PAREN_TEST(recursed,paren+1)) {
2759 PAREN_SET(recursed,paren+1);
2760 Newx(newframe,1,scan_frame);
2762 if (flags & SCF_DO_SUBSTR) {
2763 scan_commit(pRExC_state,data,minlenp);
2764 data->longest = &(data->longest_float);
2766 is_inf = is_inf_internal = 1;
2767 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2768 cl_anything(pRExC_state, data->start_class);
2769 flags &= ~SCF_DO_STCLASS;
2772 Newx(newframe,1,scan_frame);
2775 end = regnext(scan);
2780 SAVEFREEPV(newframe);
2781 newframe->next = regnext(scan);
2782 newframe->last = last;
2783 newframe->stop = stopparen;
2784 newframe->prev = frame;
2794 else if (OP(scan) == EXACT) {
2795 I32 l = STR_LEN(scan);
2798 const U8 * const s = (U8*)STRING(scan);
2799 l = utf8_length(s, s + l);
2800 uc = utf8_to_uvchr(s, NULL);
2802 uc = *((U8*)STRING(scan));
2805 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2806 /* The code below prefers earlier match for fixed
2807 offset, later match for variable offset. */
2808 if (data->last_end == -1) { /* Update the start info. */
2809 data->last_start_min = data->pos_min;
2810 data->last_start_max = is_inf
2811 ? I32_MAX : data->pos_min + data->pos_delta;
2813 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2815 SvUTF8_on(data->last_found);
2817 SV * const sv = data->last_found;
2818 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2819 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2820 if (mg && mg->mg_len >= 0)
2821 mg->mg_len += utf8_length((U8*)STRING(scan),
2822 (U8*)STRING(scan)+STR_LEN(scan));
2824 data->last_end = data->pos_min + l;
2825 data->pos_min += l; /* As in the first entry. */
2826 data->flags &= ~SF_BEFORE_EOL;
2828 if (flags & SCF_DO_STCLASS_AND) {
2829 /* Check whether it is compatible with what we know already! */
2833 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2834 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2835 && (!(data->start_class->flags & ANYOF_FOLD)
2836 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2839 ANYOF_CLASS_ZERO(data->start_class);
2840 ANYOF_BITMAP_ZERO(data->start_class);
2842 ANYOF_BITMAP_SET(data->start_class, uc);
2843 data->start_class->flags &= ~ANYOF_EOS;
2845 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2847 else if (flags & SCF_DO_STCLASS_OR) {
2848 /* false positive possible if the class is case-folded */
2850 ANYOF_BITMAP_SET(data->start_class, uc);
2852 data->start_class->flags |= ANYOF_UNICODE_ALL;
2853 data->start_class->flags &= ~ANYOF_EOS;
2854 cl_and(data->start_class, and_withp);
2856 flags &= ~SCF_DO_STCLASS;
2858 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2859 I32 l = STR_LEN(scan);
2860 UV uc = *((U8*)STRING(scan));
2862 /* Search for fixed substrings supports EXACT only. */
2863 if (flags & SCF_DO_SUBSTR) {
2865 scan_commit(pRExC_state, data, minlenp);
2868 const U8 * const s = (U8 *)STRING(scan);
2869 l = utf8_length(s, s + l);
2870 uc = utf8_to_uvchr(s, NULL);
2873 if (flags & SCF_DO_SUBSTR)
2875 if (flags & SCF_DO_STCLASS_AND) {
2876 /* Check whether it is compatible with what we know already! */
2880 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2881 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2882 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2884 ANYOF_CLASS_ZERO(data->start_class);
2885 ANYOF_BITMAP_ZERO(data->start_class);
2887 ANYOF_BITMAP_SET(data->start_class, uc);
2888 data->start_class->flags &= ~ANYOF_EOS;
2889 data->start_class->flags |= ANYOF_FOLD;
2890 if (OP(scan) == EXACTFL)
2891 data->start_class->flags |= ANYOF_LOCALE;
2894 else if (flags & SCF_DO_STCLASS_OR) {
2895 if (data->start_class->flags & ANYOF_FOLD) {
2896 /* false positive possible if the class is case-folded.
2897 Assume that the locale settings are the same... */
2899 ANYOF_BITMAP_SET(data->start_class, uc);
2900 data->start_class->flags &= ~ANYOF_EOS;
2902 cl_and(data->start_class, and_withp);
2904 flags &= ~SCF_DO_STCLASS;
2906 else if (strchr((const char*)PL_varies,OP(scan))) {
2907 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2908 I32 f = flags, pos_before = 0;
2909 regnode * const oscan = scan;
2910 struct regnode_charclass_class this_class;
2911 struct regnode_charclass_class *oclass = NULL;
2912 I32 next_is_eval = 0;
2914 switch (PL_regkind[OP(scan)]) {
2915 case WHILEM: /* End of (?:...)* . */
2916 scan = NEXTOPER(scan);
2919 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2920 next = NEXTOPER(scan);
2921 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2923 maxcount = REG_INFTY;
2924 next = regnext(scan);
2925 scan = NEXTOPER(scan);
2929 if (flags & SCF_DO_SUBSTR)
2934 if (flags & SCF_DO_STCLASS) {
2936 maxcount = REG_INFTY;
2937 next = regnext(scan);
2938 scan = NEXTOPER(scan);
2941 is_inf = is_inf_internal = 1;
2942 scan = regnext(scan);
2943 if (flags & SCF_DO_SUBSTR) {
2944 scan_commit(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2945 data->longest = &(data->longest_float);
2947 goto optimize_curly_tail;
2949 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
2950 && (scan->flags == stopparen))
2955 mincount = ARG1(scan);
2956 maxcount = ARG2(scan);
2958 next = regnext(scan);
2959 if (OP(scan) == CURLYX) {
2960 I32 lp = (data ? *(data->last_closep) : 0);
2961 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
2963 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
2964 next_is_eval = (OP(scan) == EVAL);
2966 if (flags & SCF_DO_SUBSTR) {
2967 if (mincount == 0) scan_commit(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
2968 pos_before = data->pos_min;
2972 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
2974 data->flags |= SF_IS_INF;
2976 if (flags & SCF_DO_STCLASS) {
2977 cl_init(pRExC_state, &this_class);
2978 oclass = data->start_class;
2979 data->start_class = &this_class;
2980 f |= SCF_DO_STCLASS_AND;
2981 f &= ~SCF_DO_STCLASS_OR;
2983 /* These are the cases when once a subexpression
2984 fails at a particular position, it cannot succeed
2985 even after backtracking at the enclosing scope.
2987 XXXX what if minimal match and we are at the
2988 initial run of {n,m}? */
2989 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
2990 f &= ~SCF_WHILEM_VISITED_POS;
2992 /* This will finish on WHILEM, setting scan, or on NULL: */
2993 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2994 last, data, stopparen, recursed, NULL,
2996 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
2998 if (flags & SCF_DO_STCLASS)
2999 data->start_class = oclass;
3000 if (mincount == 0 || minnext == 0) {
3001 if (flags & SCF_DO_STCLASS_OR) {
3002 cl_or(pRExC_state, data->start_class, &this_class);
3004 else if (flags & SCF_DO_STCLASS_AND) {
3005 /* Switch to OR mode: cache the old value of
3006 * data->start_class */
3008 StructCopy(data->start_class, and_withp,
3009 struct regnode_charclass_class);
3010 flags &= ~SCF_DO_STCLASS_AND;
3011 StructCopy(&this_class, data->start_class,
3012 struct regnode_charclass_class);
3013 flags |= SCF_DO_STCLASS_OR;
3014 data->start_class->flags |= ANYOF_EOS;
3016 } else { /* Non-zero len */
3017 if (flags & SCF_DO_STCLASS_OR) {
3018 cl_or(pRExC_state, data->start_class, &this_class);
3019 cl_and(data->start_class, and_withp);
3021 else if (flags & SCF_DO_STCLASS_AND)
3022 cl_and(data->start_class, &this_class);
3023 flags &= ~SCF_DO_STCLASS;
3025 if (!scan) /* It was not CURLYX, but CURLY. */
3027 if ( /* ? quantifier ok, except for (?{ ... }) */
3028 (next_is_eval || !(mincount == 0 && maxcount == 1))
3029 && (minnext == 0) && (deltanext == 0)
3030 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3031 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3032 && ckWARN(WARN_REGEXP))
3035 "Quantifier unexpected on zero-length expression");
3038 min += minnext * mincount;
3039 is_inf_internal |= ((maxcount == REG_INFTY
3040 && (minnext + deltanext) > 0)
3041 || deltanext == I32_MAX);
3042 is_inf |= is_inf_internal;
3043 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3045 /* Try powerful optimization CURLYX => CURLYN. */
3046 if ( OP(oscan) == CURLYX && data
3047 && data->flags & SF_IN_PAR
3048 && !(data->flags & SF_HAS_EVAL)
3049 && !deltanext && minnext == 1 ) {
3050 /* Try to optimize to CURLYN. */
3051 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3052 regnode * const nxt1 = nxt;
3059 if (!strchr((const char*)PL_simple,OP(nxt))
3060 && !(PL_regkind[OP(nxt)] == EXACT
3061 && STR_LEN(nxt) == 1))
3067 if (OP(nxt) != CLOSE)
3069 if (RExC_open_parens) {
3070 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3071 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3073 /* Now we know that nxt2 is the only contents: */
3074 oscan->flags = (U8)ARG(nxt);
3076 OP(nxt1) = NOTHING; /* was OPEN. */
3079 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3080 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3081 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3082 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3083 OP(nxt + 1) = OPTIMIZED; /* was count. */
3084 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3089 /* Try optimization CURLYX => CURLYM. */
3090 if ( OP(oscan) == CURLYX && data
3091 && !(data->flags & SF_HAS_PAR)
3092 && !(data->flags & SF_HAS_EVAL)
3093 && !deltanext /* atom is fixed width */
3094 && minnext != 0 /* CURLYM can't handle zero width */
3096 /* XXXX How to optimize if data == 0? */
3097 /* Optimize to a simpler form. */
3098 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3102 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3103 && (OP(nxt2) != WHILEM))
3105 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3106 /* Need to optimize away parenths. */
3107 if (data->flags & SF_IN_PAR) {
3108 /* Set the parenth number. */
3109 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3111 if (OP(nxt) != CLOSE)
3112 FAIL("Panic opt close");
3113 oscan->flags = (U8)ARG(nxt);
3114 if (RExC_open_parens) {
3115 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3116 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3118 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3119 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3122 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3123 OP(nxt + 1) = OPTIMIZED; /* was count. */
3124 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3125 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3128 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3129 regnode *nnxt = regnext(nxt1);
3132 if (reg_off_by_arg[OP(nxt1)])
3133 ARG_SET(nxt1, nxt2 - nxt1);
3134 else if (nxt2 - nxt1 < U16_MAX)
3135 NEXT_OFF(nxt1) = nxt2 - nxt1;
3137 OP(nxt) = NOTHING; /* Cannot beautify */
3142 /* Optimize again: */
3143 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3144 NULL, stopparen, recursed, NULL, 0,depth+1);
3149 else if ((OP(oscan) == CURLYX)
3150 && (flags & SCF_WHILEM_VISITED_POS)
3151 /* See the comment on a similar expression above.
3152 However, this time it not a subexpression
3153 we care about, but the expression itself. */
3154 && (maxcount == REG_INFTY)
3155 && data && ++data->whilem_c < 16) {
3156 /* This stays as CURLYX, we can put the count/of pair. */
3157 /* Find WHILEM (as in regexec.c) */
3158 regnode *nxt = oscan + NEXT_OFF(oscan);
3160 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3162 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3163 | (RExC_whilem_seen << 4)); /* On WHILEM */
3165 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3167 if (flags & SCF_DO_SUBSTR) {
3168 SV *last_str = NULL;
3169 int counted = mincount != 0;
3171 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3172 #if defined(SPARC64_GCC_WORKAROUND)
3175 const char *s = NULL;
3178 if (pos_before >= data->last_start_min)
3181 b = data->last_start_min;
3184 s = SvPV_const(data->last_found, l);
3185 old = b - data->last_start_min;
3188 I32 b = pos_before >= data->last_start_min
3189 ? pos_before : data->last_start_min;
3191 const char * const s = SvPV_const(data->last_found, l);
3192 I32 old = b - data->last_start_min;
3196 old = utf8_hop((U8*)s, old) - (U8*)s;
3199 /* Get the added string: */
3200 last_str = newSVpvn(s + old, l);
3202 SvUTF8_on(last_str);
3203 if (deltanext == 0 && pos_before == b) {
3204 /* What was added is a constant string */
3206 SvGROW(last_str, (mincount * l) + 1);
3207 repeatcpy(SvPVX(last_str) + l,
3208 SvPVX_const(last_str), l, mincount - 1);
3209 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3210 /* Add additional parts. */
3211 SvCUR_set(data->last_found,
3212 SvCUR(data->last_found) - l);
3213 sv_catsv(data->last_found, last_str);
3215 SV * sv = data->last_found;
3217 SvUTF8(sv) && SvMAGICAL(sv) ?
3218 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3219 if (mg && mg->mg_len >= 0)
3220 mg->mg_len += CHR_SVLEN(last_str);
3222 data->last_end += l * (mincount - 1);
3225 /* start offset must point into the last copy */
3226 data->last_start_min += minnext * (mincount - 1);
3227 data->last_start_max += is_inf ? I32_MAX
3228 : (maxcount - 1) * (minnext + data->pos_delta);
3231 /* It is counted once already... */
3232 data->pos_min += minnext * (mincount - counted);
3233 data->pos_delta += - counted * deltanext +
3234 (minnext + deltanext) * maxcount - minnext * mincount;
3235 if (mincount != maxcount) {
3236 /* Cannot extend fixed substrings found inside
3238 scan_commit(pRExC_state,data,minlenp);
3239 if (mincount && last_str) {
3240 SV * const sv = data->last_found;
3241 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3242 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3246 sv_setsv(sv, last_str);
3247 data->last_end = data->pos_min;
3248 data->last_start_min =
3249 data->pos_min - CHR_SVLEN(last_str);
3250 data->last_start_max = is_inf
3252 : data->pos_min + data->pos_delta
3253 - CHR_SVLEN(last_str);
3255 data->longest = &(data->longest_float);
3257 SvREFCNT_dec(last_str);
3259 if (data && (fl & SF_HAS_EVAL))
3260 data->flags |= SF_HAS_EVAL;
3261 optimize_curly_tail:
3262 if (OP(oscan) != CURLYX) {
3263 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3265 NEXT_OFF(oscan) += NEXT_OFF(next);
3268 default: /* REF and CLUMP only? */
3269 if (flags & SCF_DO_SUBSTR) {
3270 scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
3271 data->longest = &(data->longest_float);
3273 is_inf = is_inf_internal = 1;
3274 if (flags & SCF_DO_STCLASS_OR)
3275 cl_anything(pRExC_state, data->start_class);
3276 flags &= ~SCF_DO_STCLASS;
3280 else if (strchr((const char*)PL_simple,OP(scan))) {
3283 if (flags & SCF_DO_SUBSTR) {
3284 scan_commit(pRExC_state,data,minlenp);
3288 if (flags & SCF_DO_STCLASS) {
3289 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3291 /* Some of the logic below assumes that switching
3292 locale on will only add false positives. */
3293 switch (PL_regkind[OP(scan)]) {
3297 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3298 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3299 cl_anything(pRExC_state, data->start_class);
3302 if (OP(scan) == SANY)
3304 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3305 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3306 || (data->start_class->flags & ANYOF_CLASS));
3307 cl_anything(pRExC_state, data->start_class);
3309 if (flags & SCF_DO_STCLASS_AND || !value)
3310 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3313 if (flags & SCF_DO_STCLASS_AND)
3314 cl_and(data->start_class,
3315 (struct regnode_charclass_class*)scan);
3317 cl_or(pRExC_state, data->start_class,
3318 (struct regnode_charclass_class*)scan);
3321 if (flags & SCF_DO_STCLASS_AND) {
3322 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3323 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3324 for (value = 0; value < 256; value++)
3325 if (!isALNUM(value))
3326 ANYOF_BITMAP_CLEAR(data->start_class, value);
3330 if (data->start_class->flags & ANYOF_LOCALE)
3331 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3333 for (value = 0; value < 256; value++)
3335 ANYOF_BITMAP_SET(data->start_class, value);
3340 if (flags & SCF_DO_STCLASS_AND) {
3341 if (data->start_class->flags & ANYOF_LOCALE)
3342 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3345 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3346 data->start_class->flags |= ANYOF_LOCALE;
3350 if (flags & SCF_DO_STCLASS_AND) {
3351 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3352 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3353 for (value = 0; value < 256; value++)
3355 ANYOF_BITMAP_CLEAR(data->start_class, value);
3359 if (data->start_class->flags & ANYOF_LOCALE)
3360 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3362 for (value = 0; value < 256; value++)
3363 if (!isALNUM(value))
3364 ANYOF_BITMAP_SET(data->start_class, value);
3369 if (flags & SCF_DO_STCLASS_AND) {
3370 if (data->start_class->flags & ANYOF_LOCALE)
3371 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3374 data->start_class->flags |= ANYOF_LOCALE;
3375 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3379 if (flags & SCF_DO_STCLASS_AND) {
3380 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3381 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3382 for (value = 0; value < 256; value++)
3383 if (!isSPACE(value))
3384 ANYOF_BITMAP_CLEAR(data->start_class, value);
3388 if (data->start_class->flags & ANYOF_LOCALE)
3389 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3391 for (value = 0; value < 256; value++)
3393 ANYOF_BITMAP_SET(data->start_class, value);
3398 if (flags & SCF_DO_STCLASS_AND) {
3399 if (data->start_class->flags & ANYOF_LOCALE)
3400 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3403 data->start_class->flags |= ANYOF_LOCALE;
3404 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3408 if (flags & SCF_DO_STCLASS_AND) {
3409 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3410 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3411 for (value = 0; value < 256; value++)
3413 ANYOF_BITMAP_CLEAR(data->start_class, value);
3417 if (data->start_class->flags & ANYOF_LOCALE)
3418 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3420 for (value = 0; value < 256; value++)
3421 if (!isSPACE(value))
3422 ANYOF_BITMAP_SET(data->start_class, value);
3427 if (flags & SCF_DO_STCLASS_AND) {
3428 if (data->start_class->flags & ANYOF_LOCALE) {
3429 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3430 for (value = 0; value < 256; value++)
3431 if (!isSPACE(value))
3432 ANYOF_BITMAP_CLEAR(data->start_class, value);
3436 data->start_class->flags |= ANYOF_LOCALE;
3437 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3441 if (flags & SCF_DO_STCLASS_AND) {
3442 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3443 for (value = 0; value < 256; value++)
3444 if (!isDIGIT(value))
3445 ANYOF_BITMAP_CLEAR(data->start_class, value);
3448 if (data->start_class->flags & ANYOF_LOCALE)
3449 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3451 for (value = 0; value < 256; value++)
3453 ANYOF_BITMAP_SET(data->start_class, value);
3458 if (flags & SCF_DO_STCLASS_AND) {
3459 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3460 for (value = 0; value < 256; value++)
3462 ANYOF_BITMAP_CLEAR(data->start_class, value);
3465 if (data->start_class->flags & ANYOF_LOCALE)
3466 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3468 for (value = 0; value < 256; value++)
3469 if (!isDIGIT(value))
3470 ANYOF_BITMAP_SET(data->start_class, value);
3475 if (flags & SCF_DO_STCLASS_OR)
3476 cl_and(data->start_class, and_withp);
3477 flags &= ~SCF_DO_STCLASS;
3480 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3481 data->flags |= (OP(scan) == MEOL
3485 else if ( PL_regkind[OP(scan)] == BRANCHJ
3486 /* Lookbehind, or need to calculate parens/evals/stclass: */
3487 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3488 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3489 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3490 || OP(scan) == UNLESSM )
3492 /* Negative Lookahead/lookbehind
3493 In this case we can't do fixed string optimisation.
3496 I32 deltanext, minnext, fake = 0;
3498 struct regnode_charclass_class intrnl;
3501 data_fake.flags = 0;
3503 data_fake.whilem_c = data->whilem_c;
3504 data_fake.last_closep = data->last_closep;
3507 data_fake.last_closep = &fake;
3508 data_fake.pos_delta = delta;
3509 if ( flags & SCF_DO_STCLASS && !scan->flags
3510 && OP(scan) == IFMATCH ) { /* Lookahead */
3511 cl_init(pRExC_state, &intrnl);
3512 data_fake.start_class = &intrnl;
3513 f |= SCF_DO_STCLASS_AND;
3515 if (flags & SCF_WHILEM_VISITED_POS)
3516 f |= SCF_WHILEM_VISITED_POS;
3517 next = regnext(scan);
3518 nscan = NEXTOPER(NEXTOPER(scan));
3519 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3520 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3523 FAIL("Variable length lookbehind not implemented");
3525 else if (minnext > (I32)U8_MAX) {
3526 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3528 scan->flags = (U8)minnext;
3531 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3533 if (data_fake.flags & SF_HAS_EVAL)
3534 data->flags |= SF_HAS_EVAL;
3535 data->whilem_c = data_fake.whilem_c;
3537 if (f & SCF_DO_STCLASS_AND) {
3538 const int was = (data->start_class->flags & ANYOF_EOS);
3540 cl_and(data->start_class, &intrnl);
3542 data->start_class->flags |= ANYOF_EOS;
3545 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3547 /* Positive Lookahead/lookbehind
3548 In this case we can do fixed string optimisation,
3549 but we must be careful about it. Note in the case of
3550 lookbehind the positions will be offset by the minimum
3551 length of the pattern, something we won't know about
3552 until after the recurse.
3554 I32 deltanext, fake = 0;
3556 struct regnode_charclass_class intrnl;
3558 /* We use SAVEFREEPV so that when the full compile
3559 is finished perl will clean up the allocated
3560 minlens when its all done. This was we don't
3561 have to worry about freeing them when we know
3562 they wont be used, which would be a pain.
3565 Newx( minnextp, 1, I32 );
3566 SAVEFREEPV(minnextp);
3569 StructCopy(data, &data_fake, scan_data_t);
3570 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3573 scan_commit(pRExC_state, &data_fake,minlenp);
3574 data_fake.last_found=newSVsv(data->last_found);
3578 data_fake.last_closep = &fake;
3579 data_fake.flags = 0;
3580 data_fake.pos_delta = delta;
3582 data_fake.flags |= SF_IS_INF;
3583 if ( flags & SCF_DO_STCLASS && !scan->flags
3584 && OP(scan) == IFMATCH ) { /* Lookahead */
3585 cl_init(pRExC_state, &intrnl);
3586 data_fake.start_class = &intrnl;
3587 f |= SCF_DO_STCLASS_AND;
3589 if (flags & SCF_WHILEM_VISITED_POS)
3590 f |= SCF_WHILEM_VISITED_POS;
3591 next = regnext(scan);
3592 nscan = NEXTOPER(NEXTOPER(scan));
3594 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3595 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3598 FAIL("Variable length lookbehind not implemented");
3600 else if (*minnextp > (I32)U8_MAX) {
3601 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3603 scan->flags = (U8)*minnextp;
3608 if (f & SCF_DO_STCLASS_AND) {
3609 const int was = (data->start_class->flags & ANYOF_EOS);
3611 cl_and(data->start_class, &intrnl);
3613 data->start_class->flags |= ANYOF_EOS;
3616 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3618 if (data_fake.flags & SF_HAS_EVAL)
3619 data->flags |= SF_HAS_EVAL;
3620 data->whilem_c = data_fake.whilem_c;
3621 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3622 if (RExC_rx->minlen<*minnextp)
3623 RExC_rx->minlen=*minnextp;
3624 scan_commit(pRExC_state, &data_fake, minnextp);
3625 SvREFCNT_dec(data_fake.last_found);
3627 if ( data_fake.minlen_fixed != minlenp )
3629 data->offset_fixed= data_fake.offset_fixed;
3630 data->minlen_fixed= data_fake.minlen_fixed;
3631 data->lookbehind_fixed+= scan->flags;
3633 if ( data_fake.minlen_float != minlenp )
3635 data->minlen_float= data_fake.minlen_float;
3636 data->offset_float_min=data_fake.offset_float_min;
3637 data->offset_float_max=data_fake.offset_float_max;
3638 data->lookbehind_float+= scan->flags;
3647 else if (OP(scan) == OPEN) {
3648 if (stopparen != (I32)ARG(scan))
3651 else if (OP(scan) == CLOSE) {
3652 if (stopparen == (I32)ARG(scan)) {
3655 if ((I32)ARG(scan) == is_par) {
3656 next = regnext(scan);
3658 if ( next && (OP(next) != WHILEM) && next < last)
3659 is_par = 0; /* Disable optimization */
3662 *(data->last_closep) = ARG(scan);
3664 else if (OP(scan) == EVAL) {
3666 data->flags |= SF_HAS_EVAL;
3668 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3669 if (flags & SCF_DO_SUBSTR) {
3670 scan_commit(pRExC_state,data,minlenp);
3671 flags &= ~SCF_DO_SUBSTR;
3673 if (data && OP(scan)==ACCEPT) {
3674 data->flags |= SCF_SEEN_ACCEPT;
3679 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3681 if (flags & SCF_DO_SUBSTR) {
3682 scan_commit(pRExC_state,data,minlenp);
3683 data->longest = &(data->longest_float);
3685 is_inf = is_inf_internal = 1;
3686 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3687 cl_anything(pRExC_state, data->start_class);
3688 flags &= ~SCF_DO_STCLASS;
3690 else if (OP(scan) == GPOS) {
3691 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3692 !(delta || is_inf || (data && data->pos_delta)))
3694 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3695 RExC_rx->extflags |= RXf_ANCH_GPOS;
3696 if (RExC_rx->gofs < (U32)min)
3697 RExC_rx->gofs = min;
3699 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3703 #ifdef TRIE_STUDY_OPT
3704 #ifdef FULL_TRIE_STUDY
3705 else if (PL_regkind[OP(scan)] == TRIE) {
3706 /* NOTE - There is similar code to this block above for handling
3707 BRANCH nodes on the initial study. If you change stuff here
3709 regnode *trie_node= scan;
3710 regnode *tail= regnext(scan);
3711 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3712 I32 max1 = 0, min1 = I32_MAX;
3713 struct regnode_charclass_class accum;
3715 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3716 scan_commit(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3717 if (flags & SCF_DO_STCLASS)
3718 cl_init_zero(pRExC_state, &accum);
3724 const regnode *nextbranch= NULL;
3727 for ( word=1 ; word <= trie->wordcount ; word++)
3729 I32 deltanext=0, minnext=0, f = 0, fake;
3730 struct regnode_charclass_class this_class;
3732 data_fake.flags = 0;
3734 data_fake.whilem_c = data->whilem_c;
3735 data_fake.last_closep = data->last_closep;
3738 data_fake.last_closep = &fake;
3739 data_fake.pos_delta = delta;
3740 if (flags & SCF_DO_STCLASS) {
3741 cl_init(pRExC_state, &this_class);
3742 data_fake.start_class = &this_class;
3743 f = SCF_DO_STCLASS_AND;
3745 if (flags & SCF_WHILEM_VISITED_POS)
3746 f |= SCF_WHILEM_VISITED_POS;
3748 if (trie->jump[word]) {
3750 nextbranch = trie_node + trie->jump[0];
3751 scan= trie_node + trie->jump[word];
3752 /* We go from the jump point to the branch that follows
3753 it. Note this means we need the vestigal unused branches
3754 even though they arent otherwise used.
3756 minnext = study_chunk(pRExC_state, &scan, minlenp,
3757 &deltanext, (regnode *)nextbranch, &data_fake,
3758 stopparen, recursed, NULL, f,depth+1);
3760 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3761 nextbranch= regnext((regnode*)nextbranch);
3763 if (min1 > (I32)(minnext + trie->minlen))
3764 min1 = minnext + trie->minlen;
3765 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3766 max1 = minnext + deltanext + trie->maxlen;
3767 if (deltanext == I32_MAX)
3768 is_inf = is_inf_internal = 1;
3770 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3772 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3773 if ( stopmin > min + min1)
3774 stopmin = min + min1;
3775 flags &= ~SCF_DO_SUBSTR;
3777 data->flags |= SCF_SEEN_ACCEPT;
3780 if (data_fake.flags & SF_HAS_EVAL)
3781 data->flags |= SF_HAS_EVAL;
3782 data->whilem_c = data_fake.whilem_c;
3784 if (flags & SCF_DO_STCLASS)
3785 cl_or(pRExC_state, &accum, &this_class);
3788 if (flags & SCF_DO_SUBSTR) {
3789 data->pos_min += min1;
3790 data->pos_delta += max1 - min1;
3791 if (max1 != min1 || is_inf)
3792 data->longest = &(data->longest_float);
3795 delta += max1 - min1;
3796 if (flags & SCF_DO_STCLASS_OR) {
3797 cl_or(pRExC_state, data->start_class, &accum);
3799 cl_and(data->start_class, and_withp);
3800 flags &= ~SCF_DO_STCLASS;
3803 else if (flags & SCF_DO_STCLASS_AND) {
3805 cl_and(data->start_class, &accum);
3806 flags &= ~SCF_DO_STCLASS;
3809 /* Switch to OR mode: cache the old value of
3810 * data->start_class */
3812 StructCopy(data->start_class, and_withp,
3813 struct regnode_charclass_class);
3814 flags &= ~SCF_DO_STCLASS_AND;
3815 StructCopy(&accum, data->start_class,
3816 struct regnode_charclass_class);
3817 flags |= SCF_DO_STCLASS_OR;
3818 data->start_class->flags |= ANYOF_EOS;
3825 else if (PL_regkind[OP(scan)] == TRIE) {
3826 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3829 min += trie->minlen;
3830 delta += (trie->maxlen - trie->minlen);
3831 flags &= ~SCF_DO_STCLASS; /* xxx */
3832 if (flags & SCF_DO_SUBSTR) {
3833 scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
3834 data->pos_min += trie->minlen;
3835 data->pos_delta += (trie->maxlen - trie->minlen);
3836 if (trie->maxlen != trie->minlen)
3837 data->longest = &(data->longest_float);
3839 if (trie->jump) /* no more substrings -- for now /grr*/
3840 flags &= ~SCF_DO_SUBSTR;
3842 #endif /* old or new */
3843 #endif /* TRIE_STUDY_OPT */
3844 /* Else: zero-length, ignore. */
3845 scan = regnext(scan);
3850 stopparen = frame->stop;
3851 frame = frame->prev;
3852 goto fake_study_recurse;
3859 *deltap = is_inf_internal ? I32_MAX : delta;
3860 if (flags & SCF_DO_SUBSTR && is_inf)
3861 data->pos_delta = I32_MAX - data->pos_min;
3862 if (is_par > (I32)U8_MAX)
3864 if (is_par && pars==1 && data) {
3865 data->flags |= SF_IN_PAR;
3866 data->flags &= ~SF_HAS_PAR;
3868 else if (pars && data) {
3869 data->flags |= SF_HAS_PAR;
3870 data->flags &= ~SF_IN_PAR;
3872 if (flags & SCF_DO_STCLASS_OR)
3873 cl_and(data->start_class, and_withp);
3874 if (flags & SCF_TRIE_RESTUDY)
3875 data->flags |= SCF_TRIE_RESTUDY;
3877 DEBUG_STUDYDATA(data,depth);
3879 return min < stopmin ? min : stopmin;
3883 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3885 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3887 Renewc(RExC_rxi->data,
3888 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3889 char, struct reg_data);
3891 Renew(RExC_rxi->data->what, count + n, U8);
3893 Newx(RExC_rxi->data->what, n, U8);
3894 RExC_rxi->data->count = count + n;
3895 Copy(s, RExC_rxi->data->what + count, n, U8);
3899 #ifndef PERL_IN_XSUB_RE
3901 Perl_reginitcolors(pTHX)
3904 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3906 char *t = savepv(s);
3910 t = strchr(t, '\t');
3916 PL_colors[i] = t = (char *)"";
3921 PL_colors[i++] = (char *)"";
3928 #ifdef TRIE_STUDY_OPT
3929 #define CHECK_RESTUDY_GOTO \
3931 (data.flags & SCF_TRIE_RESTUDY) \
3935 #define CHECK_RESTUDY_GOTO
3939 - pregcomp - compile a regular expression into internal code
3941 * We can't allocate space until we know how big the compiled form will be,
3942 * but we can't compile it (and thus know how big it is) until we've got a
3943 * place to put the code. So we cheat: we compile it twice, once with code
3944 * generation turned off and size counting turned on, and once "for real".
3945 * This also means that we don't allocate space until we are sure that the
3946 * thing really will compile successfully, and we never have to move the
3947 * code and thus invalidate pointers into it. (Note that it has to be in
3948 * one piece because free() must be able to free it all.) [NB: not true in perl]
3950 * Beware that the optimization-preparation code in here knows about some
3951 * of the structure of the compiled regexp. [I'll say.]
3956 #ifndef PERL_IN_XSUB_RE
3957 #define RE_ENGINE_PTR &PL_core_reg_engine
3959 extern const struct regexp_engine my_reg_engine;
3960 #define RE_ENGINE_PTR &my_reg_engine
3962 /* these make a few things look better, to avoid indentation */
3963 #define BEGIN_BLOCK {
3967 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
3970 GET_RE_DEBUG_FLAGS_DECL;
3971 DEBUG_r(if (!PL_colorset) reginitcolors());
3972 #ifndef PERL_IN_XSUB_RE
3974 /* Dispatch a request to compile a regexp to correct
3976 HV * const table = GvHV(PL_hintgv);
3978 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
3979 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
3980 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
3982 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
3985 return CALLREGCOMP_ENG(eng, exp, xend, pm);
3992 register regexp_internal *ri;
4000 RExC_state_t RExC_state;
4001 RExC_state_t * const pRExC_state = &RExC_state;
4002 #ifdef TRIE_STUDY_OPT
4004 RExC_state_t copyRExC_state;
4007 FAIL("NULL regexp argument");
4009 RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4013 SV *dsv= sv_newmortal();
4014 RE_PV_QUOTED_DECL(s, RExC_utf8,
4015 dsv, RExC_precomp, (xend - exp), 60);
4016 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4017 PL_colors[4],PL_colors[5],s);
4019 RExC_flags = pm->op_pmflags;
4023 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4024 RExC_seen_evals = 0;
4027 /* First pass: determine size, legality. */
4035 RExC_emit = &PL_regdummy;
4036 RExC_whilem_seen = 0;
4037 RExC_charnames = NULL;
4038 RExC_open_parens = NULL;
4039 RExC_close_parens = NULL;
4041 RExC_paren_names = NULL;
4042 RExC_recurse = NULL;
4043 RExC_recurse_count = 0;
4045 #if 0 /* REGC() is (currently) a NOP at the first pass.
4046 * Clever compilers notice this and complain. --jhi */
4047 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4049 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4050 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4051 RExC_precomp = NULL;
4055 PerlIO_printf(Perl_debug_log,
4056 "Required size %"IVdf" nodes\n"
4057 "Starting second pass (creation)\n",
4060 RExC_lastparse=NULL;
4062 /* Small enough for pointer-storage convention?
4063 If extralen==0, this means that we will not need long jumps. */
4064 if (RExC_size >= 0x10000L && RExC_extralen)
4065 RExC_size += RExC_extralen;
4068 if (RExC_whilem_seen > 15)
4069 RExC_whilem_seen = 15;
4072 /* Make room for a sentinel value at the end of the program */
4076 /* Allocate space and zero-initialize. Note, the two step process
4077 of zeroing when in debug mode, thus anything assigned has to
4078 happen after that */
4079 Newxz(r, 1, regexp);
4080 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4081 char, regexp_internal);
4082 if ( r == NULL || ri == NULL )
4083 FAIL("Regexp out of space");
4085 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4086 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4088 /* bulk initialize base fields with 0. */
4089 Zero(ri, sizeof(regexp_internal), char);
4092 /* non-zero initialization begins here */
4094 r->engine= RE_ENGINE_PTR;
4096 r->prelen = xend - exp;
4097 r->precomp = savepvn(RExC_precomp, r->prelen);
4098 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4100 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4102 if (RExC_seen & REG_SEEN_RECURSE) {
4103 Newxz(RExC_open_parens, RExC_npar,regnode *);
4104 SAVEFREEPV(RExC_open_parens);
4105 Newxz(RExC_close_parens,RExC_npar,regnode *);
4106 SAVEFREEPV(RExC_close_parens);
4109 /* Useful during FAIL. */
4110 Newxz(ri->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4112 ri->offsets[0] = RExC_size;
4114 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4115 "%s %"UVuf" bytes for offset annotations.\n",
4116 ri->offsets ? "Got" : "Couldn't get",
4117 (UV)((2*RExC_size+1) * sizeof(U32))));
4122 /* Second pass: emit code. */
4123 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4128 RExC_emit_start = ri->program;
4129 RExC_emit = ri->program;
4131 /* put a sentinal on the end of the program so we can check for
4133 ri->program[RExC_size].type = 255;
4135 /* Store the count of eval-groups for security checks: */
4136 RExC_emit->next_off = (RExC_seen_evals > (I32)U16_MAX) ? U16_MAX : (U16)RExC_seen_evals;
4137 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4138 if (reg(pRExC_state, 0, &flags,1) == NULL)
4141 /* XXXX To minimize changes to RE engine we always allocate
4142 3-units-long substrs field. */
4143 Newx(r->substrs, 1, struct reg_substr_data);
4144 if (RExC_recurse_count) {
4145 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4146 SAVEFREEPV(RExC_recurse);
4150 r->minlen = minlen = sawplus = sawopen = 0;
4151 Zero(r->substrs, 1, struct reg_substr_data);
4153 #ifdef TRIE_STUDY_OPT
4156 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4158 RExC_state = copyRExC_state;
4159 if (seen & REG_TOP_LEVEL_BRANCHES)
4160 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4162 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4163 if (data.last_found) {
4164 SvREFCNT_dec(data.longest_fixed);
4165 SvREFCNT_dec(data.longest_float);
4166 SvREFCNT_dec(data.last_found);
4168 StructCopy(&zero_scan_data, &data, scan_data_t);
4170 StructCopy(&zero_scan_data, &data, scan_data_t);
4171 copyRExC_state = RExC_state;
4174 StructCopy(&zero_scan_data, &data, scan_data_t);
4177 /* Dig out information for optimizations. */
4178 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4179 pm->op_pmflags = RExC_flags;
4181 r->extflags |= RXf_UTF8; /* Unicode in it? */
4182 ri->regstclass = NULL;
4183 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4184 r->intflags |= PREGf_NAUGHTY;
4185 scan = ri->program + 1; /* First BRANCH. */
4187 /* testing for BRANCH here tells us whether there is "must appear"
4188 data in the pattern. If there is then we can use it for optimisations */
4189 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4191 STRLEN longest_float_length, longest_fixed_length;
4192 struct regnode_charclass_class ch_class; /* pointed to by data */
4194 I32 last_close = 0; /* pointed to by data */
4197 /* Skip introductions and multiplicators >= 1. */
4198 while ((OP(first) == OPEN && (sawopen = 1)) ||
4199 /* An OR of *one* alternative - should not happen now. */
4200 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4201 /* for now we can't handle lookbehind IFMATCH*/
4202 (OP(first) == IFMATCH && !first->flags) ||
4203 (OP(first) == PLUS) ||
4204 (OP(first) == MINMOD) ||
4205 /* An {n,m} with n>0 */
4206 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4209 if (OP(first) == PLUS)
4212 first += regarglen[OP(first)];
4213 if (OP(first) == IFMATCH) {
4214 first = NEXTOPER(first);
4215 first += EXTRA_STEP_2ARGS;
4216 } else /* XXX possible optimisation for /(?=)/ */
4217 first = NEXTOPER(first);
4220 /* Starting-point info. */
4222 DEBUG_PEEP("first:",first,0);
4223 /* Ignore EXACT as we deal with it later. */
4224 if (PL_regkind[OP(first)] == EXACT) {
4225 if (OP(first) == EXACT)
4226 NOOP; /* Empty, get anchored substr later. */
4227 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4228 ri->regstclass = first;
4231 else if (PL_regkind[OP(first)] == TRIE &&
4232 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4235 /* this can happen only on restudy */
4236 if ( OP(first) == TRIE ) {
4237 struct regnode_1 *trieop =
4238 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4239 StructCopy(first,trieop,struct regnode_1);
4240 trie_op=(regnode *)trieop;
4242 struct regnode_charclass *trieop =
4243 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4244 StructCopy(first,trieop,struct regnode_charclass);
4245 trie_op=(regnode *)trieop;
4248 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4249 ri->regstclass = trie_op;
4252 else if (strchr((const char*)PL_simple,OP(first)))
4253 ri->regstclass = first;
4254 else if (PL_regkind[OP(first)] == BOUND ||
4255 PL_regkind[OP(first)] == NBOUND)
4256 ri->regstclass = first;
4257 else if (PL_regkind[OP(first)] == BOL) {
4258 r->extflags |= (OP(first) == MBOL
4260 : (OP(first) == SBOL
4263 first = NEXTOPER(first);
4266 else if (OP(first) == GPOS) {
4267 r->extflags |= RXf_ANCH_GPOS;
4268 first = NEXTOPER(first);
4271 else if ((!sawopen || !RExC_sawback) &&
4272 (OP(first) == STAR &&
4273 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4274 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4276 /* turn .* into ^.* with an implied $*=1 */
4278 (OP(NEXTOPER(first)) == REG_ANY)
4281 r->extflags |= type;
4282 r->intflags |= PREGf_IMPLICIT;
4283 first = NEXTOPER(first);
4286 if (sawplus && (!sawopen || !RExC_sawback)
4287 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4288 /* x+ must match at the 1st pos of run of x's */
4289 r->intflags |= PREGf_SKIP;
4291 /* Scan is after the zeroth branch, first is atomic matcher. */
4292 #ifdef TRIE_STUDY_OPT
4295 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4296 (IV)(first - scan + 1))
4300 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4301 (IV)(first - scan + 1))
4307 * If there's something expensive in the r.e., find the
4308 * longest literal string that must appear and make it the
4309 * regmust. Resolve ties in favor of later strings, since
4310 * the regstart check works with the beginning of the r.e.
4311 * and avoiding duplication strengthens checking. Not a
4312 * strong reason, but sufficient in the absence of others.
4313 * [Now we resolve ties in favor of the earlier string if
4314 * it happens that c_offset_min has been invalidated, since the
4315 * earlier string may buy us something the later one won't.]
4318 data.longest_fixed = newSVpvs("");
4319 data.longest_float = newSVpvs("");
4320 data.last_found = newSVpvs("");
4321 data.longest = &(data.longest_fixed);
4323 if (!ri->regstclass) {
4324 cl_init(pRExC_state, &ch_class);
4325 data.start_class = &ch_class;
4326 stclass_flag = SCF_DO_STCLASS_AND;
4327 } else /* XXXX Check for BOUND? */
4329 data.last_closep = &last_close;
4331 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4332 &data, -1, NULL, NULL,
4333 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4339 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4340 && data.last_start_min == 0 && data.last_end > 0
4341 && !RExC_seen_zerolen
4342 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4343 r->extflags |= RXf_CHECK_ALL;
4344 scan_commit(pRExC_state, &data,&minlen);
4345 SvREFCNT_dec(data.last_found);
4347 /* Note that code very similar to this but for anchored string
4348 follows immediately below, changes may need to be made to both.
4351 longest_float_length = CHR_SVLEN(data.longest_float);
4352 if (longest_float_length
4353 || (data.flags & SF_FL_BEFORE_EOL
4354 && (!(data.flags & SF_FL_BEFORE_MEOL)
4355 || (RExC_flags & RXf_PMf_MULTILINE))))
4359 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4360 && data.offset_fixed == data.offset_float_min
4361 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4362 goto remove_float; /* As in (a)+. */
4364 /* copy the information about the longest float from the reg_scan_data
4365 over to the program. */
4366 if (SvUTF8(data.longest_float)) {
4367 r->float_utf8 = data.longest_float;
4368 r->float_substr = NULL;
4370 r->float_substr = data.longest_float;
4371 r->float_utf8 = NULL;
4373 /* float_end_shift is how many chars that must be matched that
4374 follow this item. We calculate it ahead of time as once the
4375 lookbehind offset is added in we lose the ability to correctly
4377 ml = data.minlen_float ? *(data.minlen_float)
4378 : (I32)longest_float_length;
4379 r->float_end_shift = ml - data.offset_float_min
4380 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4381 + data.lookbehind_float;
4382 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4383 r->float_max_offset = data.offset_float_max;
4384 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4385 r->float_max_offset -= data.lookbehind_float;
4387 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4388 && (!(data.flags & SF_FL_BEFORE_MEOL)
4389 || (RExC_flags & RXf_PMf_MULTILINE)));
4390 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4394 r->float_substr = r->float_utf8 = NULL;
4395 SvREFCNT_dec(data.longest_float);
4396 longest_float_length = 0;
4399 /* Note that code very similar to this but for floating string
4400 is immediately above, changes may need to be made to both.
4403 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4404 if (longest_fixed_length
4405 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4406 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4407 || (RExC_flags & RXf_PMf_MULTILINE))))
4411 /* copy the information about the longest fixed
4412 from the reg_scan_data over to the program. */
4413 if (SvUTF8(data.longest_fixed)) {
4414 r->anchored_utf8 = data.longest_fixed;
4415 r->anchored_substr = NULL;
4417 r->anchored_substr = data.longest_fixed;
4418 r->anchored_utf8 = NULL;
4420 /* fixed_end_shift is how many chars that must be matched that
4421 follow this item. We calculate it ahead of time as once the
4422 lookbehind offset is added in we lose the ability to correctly
4424 ml = data.minlen_fixed ? *(data.minlen_fixed)
4425 : (I32)longest_fixed_length;
4426 r->anchored_end_shift = ml - data.offset_fixed
4427 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4428 + data.lookbehind_fixed;
4429 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4431 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4432 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4433 || (RExC_flags & RXf_PMf_MULTILINE)));
4434 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4437 r->anchored_substr = r->anchored_utf8 = NULL;
4438 SvREFCNT_dec(data.longest_fixed);
4439 longest_fixed_length = 0;
4442 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4443 ri->regstclass = NULL;
4444 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4446 && !(data.start_class->flags & ANYOF_EOS)
4447 && !cl_is_anything(data.start_class))
4449 const U32 n = add_data(pRExC_state, 1, "f");
4451 Newx(RExC_rxi->data->data[n], 1,
4452 struct regnode_charclass_class);
4453 StructCopy(data.start_class,
4454 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4455 struct regnode_charclass_class);
4456 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4457 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4458 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4459 regprop(r, sv, (regnode*)data.start_class);
4460 PerlIO_printf(Perl_debug_log,
4461 "synthetic stclass \"%s\".\n",
4462 SvPVX_const(sv));});
4465 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4466 if (longest_fixed_length > longest_float_length) {
4467 r->check_end_shift = r->anchored_end_shift;
4468 r->check_substr = r->anchored_substr;
4469 r->check_utf8 = r->anchored_utf8;
4470 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4471 if (r->extflags & RXf_ANCH_SINGLE)
4472 r->extflags |= RXf_NOSCAN;
4475 r->check_end_shift = r->float_end_shift;
4476 r->check_substr = r->float_substr;
4477 r->check_utf8 = r->float_utf8;
4478 r->check_offset_min = r->float_min_offset;
4479 r->check_offset_max = r->float_max_offset;
4481 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4482 This should be changed ASAP! */
4483 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4484 r->extflags |= RXf_USE_INTUIT;
4485 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4486 r->extflags |= RXf_INTUIT_TAIL;
4488 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4489 if ( (STRLEN)minlen < longest_float_length )
4490 minlen= longest_float_length;
4491 if ( (STRLEN)minlen < longest_fixed_length )
4492 minlen= longest_fixed_length;
4496 /* Several toplevels. Best we can is to set minlen. */
4498 struct regnode_charclass_class ch_class;
4501 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4503 scan = ri->program + 1;
4504 cl_init(pRExC_state, &ch_class);
4505 data.start_class = &ch_class;
4506 data.last_closep = &last_close;
4509 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4510 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4514 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4515 = r->float_substr = r->float_utf8 = NULL;
4516 if (!(data.start_class->flags & ANYOF_EOS)
4517 && !cl_is_anything(data.start_class))
4519 const U32 n = add_data(pRExC_state, 1, "f");
4521 Newx(RExC_rxi->data->data[n], 1,
4522 struct regnode_charclass_class);
4523 StructCopy(data.start_class,
4524 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4525 struct regnode_charclass_class);
4526 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4527 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4528 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4529 regprop(r, sv, (regnode*)data.start_class);
4530 PerlIO_printf(Perl_debug_log,
4531 "synthetic stclass \"%s\".\n",
4532 SvPVX_const(sv));});
4536 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4537 the "real" pattern. */
4539 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4540 (IV)minlen, (IV)r->minlen);
4542 r->minlenret = minlen;
4543 if (r->minlen < minlen)
4546 if (RExC_seen & REG_SEEN_GPOS)
4547 r->extflags |= RXf_GPOS_SEEN;
4548 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4549 r->extflags |= RXf_LOOKBEHIND_SEEN;
4550 if (RExC_seen & REG_SEEN_EVAL)
4551 r->extflags |= RXf_EVAL_SEEN;
4552 if (RExC_seen & REG_SEEN_CANY)
4553 r->extflags |= RXf_CANY_SEEN;
4554 if (RExC_seen & REG_SEEN_VERBARG)
4555 r->intflags |= PREGf_VERBARG_SEEN;
4556 if (RExC_seen & REG_SEEN_CUTGROUP)
4557 r->intflags |= PREGf_CUTGROUP_SEEN;
4558 if (RExC_paren_names)
4559 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4561 r->paren_names = NULL;
4563 if (RExC_recurse_count) {
4564 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4565 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4566 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4569 Newxz(r->startp, RExC_npar, I32);
4570 Newxz(r->endp, RExC_npar, I32);
4571 /* assume we don't need to swap parens around before we match */
4574 PerlIO_printf(Perl_debug_log,"Final program:\n");
4577 DEBUG_OFFSETS_r(if (ri->offsets) {
4578 const U32 len = ri->offsets[0];
4580 GET_RE_DEBUG_FLAGS_DECL;
4581 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->offsets[0]);
4582 for (i = 1; i <= len; i++) {
4583 if (ri->offsets[i*2-1] || ri->offsets[i*2])
4584 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4585 (UV)i, (UV)ri->offsets[i*2-1], (UV)ri->offsets[i*2]);
4587 PerlIO_printf(Perl_debug_log, "\n");
4593 #undef CORE_ONLY_BLOCK
4595 #undef RE_ENGINE_PTR
4597 #ifndef PERL_IN_XSUB_RE
4599 Perl_reg_named_buff_sv(pTHX_ SV* namesv)
4601 I32 parno = 0; /* no match */
4603 const REGEXP * const rx = PM_GETRE(PL_curpm);
4604 if (rx && rx->paren_names) {
4605 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4608 SV* sv_dat=HeVAL(he_str);
4609 I32 *nums=(I32*)SvPVX(sv_dat);
4610 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4611 if ((I32)(rx->lastparen) >= nums[i] &&
4612 rx->endp[nums[i]] != -1)
4625 SV *sv= sv_newmortal();
4626 Perl_sv_setpvf(aTHX_ sv, "%"IVdf,(IV)parno);
4627 gv_paren= Perl_gv_fetchsv(aTHX_ sv, GV_ADD, SVt_PVGV);
4628 return GvSVn(gv_paren);
4633 /* Scans the name of a named buffer from the pattern.
4634 * If flags is REG_RSN_RETURN_NULL returns null.
4635 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4636 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4637 * to the parsed name as looked up in the RExC_paren_names hash.
4638 * If there is an error throws a vFAIL().. type exception.
4641 #define REG_RSN_RETURN_NULL 0
4642 #define REG_RSN_RETURN_NAME 1
4643 #define REG_RSN_RETURN_DATA 2
4646 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4647 char *name_start = RExC_parse;
4650 while( isIDFIRST_uni(utf8n_to_uvchr((U8*)RExC_parse,
4651 RExC_end - RExC_parse, &numlen, UTF8_ALLOW_DEFAULT)))
4653 RExC_parse += numlen;
4656 while( isIDFIRST(*RExC_parse) )
4660 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4661 (int)(RExC_parse - name_start)));
4664 if ( flags == REG_RSN_RETURN_NAME)
4666 else if (flags==REG_RSN_RETURN_DATA) {
4669 if ( ! sv_name ) /* should not happen*/
4670 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4671 if (RExC_paren_names)
4672 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4674 sv_dat = HeVAL(he_str);
4676 vFAIL("Reference to nonexistent named group");
4680 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4687 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4688 int rem=(int)(RExC_end - RExC_parse); \
4697 if (RExC_lastparse!=RExC_parse) \
4698 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4701 iscut ? "..." : "<" \
4704 PerlIO_printf(Perl_debug_log,"%16s",""); \
4709 num=REG_NODE_NUM(RExC_emit); \
4710 if (RExC_lastnum!=num) \
4711 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4713 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4714 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4715 (int)((depth*2)), "", \
4719 RExC_lastparse=RExC_parse; \
4724 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4725 DEBUG_PARSE_MSG((funcname)); \
4726 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4728 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4729 DEBUG_PARSE_MSG((funcname)); \
4730 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4733 - reg - regular expression, i.e. main body or parenthesized thing
4735 * Caller must absorb opening parenthesis.
4737 * Combining parenthesis handling with the base level of regular expression
4738 * is a trifle forced, but the need to tie the tails of the branches to what
4739 * follows makes it hard to avoid.
4741 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4743 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4745 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4748 /* this idea is borrowed from STR_WITH_LEN in handy.h */
4749 #define CHECK_WORD(s,v,l) \
4750 (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
4753 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4754 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4757 register regnode *ret; /* Will be the head of the group. */
4758 register regnode *br;
4759 register regnode *lastbr;
4760 register regnode *ender = NULL;
4761 register I32 parno = 0;
4763 const I32 oregflags = RExC_flags;
4764 bool have_branch = 0;
4767 /* for (?g), (?gc), and (?o) warnings; warning
4768 about (?c) will warn about (?g) -- japhy */
4770 #define WASTED_O 0x01
4771 #define WASTED_G 0x02
4772 #define WASTED_C 0x04
4773 #define WASTED_GC (0x02|0x04)
4774 I32 wastedflags = 0x00;
4776 char * parse_start = RExC_parse; /* MJD */
4777 char * const oregcomp_parse = RExC_parse;
4779 GET_RE_DEBUG_FLAGS_DECL;
4780 DEBUG_PARSE("reg ");
4783 *flagp = 0; /* Tentatively. */
4786 /* Make an OPEN node, if parenthesized. */
4788 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4789 char *start_verb = RExC_parse;
4790 STRLEN verb_len = 0;
4791 char *start_arg = NULL;
4792 unsigned char op = 0;
4794 int internal_argval = 0; /* internal_argval is only useful if !argok */
4795 while ( *RExC_parse && *RExC_parse != ')' ) {
4796 if ( *RExC_parse == ':' ) {
4797 start_arg = RExC_parse + 1;
4803 verb_len = RExC_parse - start_verb;
4806 while ( *RExC_parse && *RExC_parse != ')' )
4808 if ( *RExC_parse != ')' )
4809 vFAIL("Unterminated verb pattern argument");
4810 if ( RExC_parse == start_arg )
4813 if ( *RExC_parse != ')' )
4814 vFAIL("Unterminated verb pattern");
4817 switch ( *start_verb ) {
4818 case 'A': /* (*ACCEPT) */
4819 if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
4821 internal_argval = RExC_nestroot;
4824 case 'C': /* (*COMMIT) */
4825 if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
4828 case 'F': /* (*FAIL) */
4829 if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
4834 case ':': /* (*:NAME) */
4835 case 'M': /* (*MARK:NAME) */
4836 if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
4841 case 'P': /* (*PRUNE) */
4842 if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
4845 case 'S': /* (*SKIP) */
4846 if ( CHECK_WORD("SKIP",start_verb,verb_len) )
4849 case 'T': /* (*THEN) */
4850 /* [19:06] <TimToady> :: is then */
4851 if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
4853 RExC_seen |= REG_SEEN_CUTGROUP;
4859 vFAIL3("Unknown verb pattern '%.*s'",
4860 verb_len, start_verb);
4863 if ( start_arg && internal_argval ) {
4864 vFAIL3("Verb pattern '%.*s' may not have an argument",
4865 verb_len, start_verb);
4866 } else if ( argok < 0 && !start_arg ) {
4867 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
4868 verb_len, start_verb);
4870 ret = reganode(pRExC_state, op, internal_argval);
4871 if ( ! internal_argval && ! SIZE_ONLY ) {
4873 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
4874 ARG(ret) = add_data( pRExC_state, 1, "S" );
4875 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
4882 if (!internal_argval)
4883 RExC_seen |= REG_SEEN_VERBARG;
4884 } else if ( start_arg ) {
4885 vFAIL3("Verb pattern '%.*s' may not have an argument",
4886 verb_len, start_verb);
4888 ret = reg_node(pRExC_state, op);
4890 nextchar(pRExC_state);
4893 if (*RExC_parse == '?') { /* (?...) */
4894 U32 posflags = 0, negflags = 0;
4895 U32 *flagsp = &posflags;
4896 bool is_logical = 0;
4897 const char * const seqstart = RExC_parse;
4900 paren = *RExC_parse++;
4901 ret = NULL; /* For look-ahead/behind. */
4904 case '<': /* (?<...) */
4905 if (*RExC_parse == '!')
4907 else if (*RExC_parse != '=')
4912 case '\'': /* (?'...') */
4913 name_start= RExC_parse;
4914 svname = reg_scan_name(pRExC_state,
4915 SIZE_ONLY ? /* reverse test from the others */
4916 REG_RSN_RETURN_NAME :
4917 REG_RSN_RETURN_NULL);
4918 if (RExC_parse == name_start)
4920 if (*RExC_parse != paren)
4921 vFAIL2("Sequence (?%c... not terminated",
4922 paren=='>' ? '<' : paren);
4926 if (!svname) /* shouldnt happen */
4928 "panic: reg_scan_name returned NULL");
4929 if (!RExC_paren_names) {
4930 RExC_paren_names= newHV();
4931 sv_2mortal((SV*)RExC_paren_names);
4933 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
4935 sv_dat = HeVAL(he_str);
4937 /* croak baby croak */
4939 "panic: paren_name hash element allocation failed");
4940 } else if ( SvPOK(sv_dat) ) {
4941 IV count=SvIV(sv_dat);
4942 I32 *pv=(I32*)SvGROW(sv_dat,SvCUR(sv_dat)+sizeof(I32)+1);
4943 SvCUR_set(sv_dat,SvCUR(sv_dat)+sizeof(I32));
4944 pv[count]=RExC_npar;
4947 (void)SvUPGRADE(sv_dat,SVt_PVNV);
4948 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
4953 /*sv_dump(sv_dat);*/
4955 nextchar(pRExC_state);
4957 goto capturing_parens;
4959 RExC_seen |= REG_SEEN_LOOKBEHIND;
4961 case '=': /* (?=...) */
4962 case '!': /* (?!...) */
4963 RExC_seen_zerolen++;
4964 if (*RExC_parse == ')') {
4965 ret=reg_node(pRExC_state, OPFAIL);
4966 nextchar(pRExC_state);
4969 case ':': /* (?:...) */
4970 case '>': /* (?>...) */
4972 case '$': /* (?$...) */
4973 case '@': /* (?@...) */
4974 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
4976 case '#': /* (?#...) */
4977 while (*RExC_parse && *RExC_parse != ')')
4979 if (*RExC_parse != ')')
4980 FAIL("Sequence (?#... not terminated");
4981 nextchar(pRExC_state);
4984 case '0' : /* (?0) */
4985 case 'R' : /* (?R) */
4986 if (*RExC_parse != ')')
4987 FAIL("Sequence (?R) not terminated");
4988 ret = reg_node(pRExC_state, GOSTART);
4989 nextchar(pRExC_state);
4992 { /* named and numeric backreferences */
4995 case '&': /* (?&NAME) */
4996 parse_start = RExC_parse - 1;
4998 SV *sv_dat = reg_scan_name(pRExC_state,
4999 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5000 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5002 goto gen_recurse_regop;
5005 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5007 vFAIL("Illegal pattern");
5009 goto parse_recursion;
5011 case '-': /* (?-1) */
5012 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5013 RExC_parse--; /* rewind to let it be handled later */
5017 case '1': case '2': case '3': case '4': /* (?1) */
5018 case '5': case '6': case '7': case '8': case '9':
5021 num = atoi(RExC_parse);
5022 parse_start = RExC_parse - 1; /* MJD */
5023 if (*RExC_parse == '-')
5025 while (isDIGIT(*RExC_parse))
5027 if (*RExC_parse!=')')
5028 vFAIL("Expecting close bracket");
5031 if ( paren == '-' ) {
5033 Diagram of capture buffer numbering.
5034 Top line is the normal capture buffer numbers
5035 Botton line is the negative indexing as from
5039 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5043 num = RExC_npar + num;
5046 vFAIL("Reference to nonexistent group");
5048 } else if ( paren == '+' ) {
5049 num = RExC_npar + num - 1;
5052 ret = reganode(pRExC_state, GOSUB, num);
5054 if (num > (I32)RExC_rx->nparens) {
5056 vFAIL("Reference to nonexistent group");
5058 ARG2L_SET( ret, RExC_recurse_count++);
5060 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5061 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5065 RExC_seen |= REG_SEEN_RECURSE;
5066 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5067 Set_Node_Offset(ret, parse_start); /* MJD */
5069 nextchar(pRExC_state);
5071 } /* named and numeric backreferences */
5074 case 'p': /* (?p...) */
5075 if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
5076 vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
5078 case '?': /* (??...) */
5080 if (*RExC_parse != '{')
5082 paren = *RExC_parse++;
5084 case '{': /* (?{...}) */
5089 char *s = RExC_parse;
5091 RExC_seen_zerolen++;
5092 RExC_seen |= REG_SEEN_EVAL;
5093 while (count && (c = *RExC_parse)) {
5104 if (*RExC_parse != ')') {
5106 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5110 OP_4tree *sop, *rop;
5111 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5114 Perl_save_re_context(aTHX);
5115 rop = sv_compile_2op(sv, &sop, "re", &pad);
5116 sop->op_private |= OPpREFCOUNTED;
5117 /* re_dup will OpREFCNT_inc */
5118 OpREFCNT_set(sop, 1);
5121 n = add_data(pRExC_state, 3, "nop");
5122 RExC_rxi->data->data[n] = (void*)rop;
5123 RExC_rxi->data->data[n+1] = (void*)sop;
5124 RExC_rxi->data->data[n+2] = (void*)pad;
5127 else { /* First pass */
5128 if (PL_reginterp_cnt < ++RExC_seen_evals
5130 /* No compiled RE interpolated, has runtime
5131 components ===> unsafe. */
5132 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5133 if (PL_tainting && PL_tainted)
5134 FAIL("Eval-group in insecure regular expression");
5135 #if PERL_VERSION > 8
5136 if (IN_PERL_COMPILETIME)
5141 nextchar(pRExC_state);
5143 ret = reg_node(pRExC_state, LOGICAL);
5146 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5147 /* deal with the length of this later - MJD */
5150 ret = reganode(pRExC_state, EVAL, n);
5151 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5152 Set_Node_Offset(ret, parse_start);
5155 case '(': /* (?(?{...})...) and (?(?=...)...) */
5158 if (RExC_parse[0] == '?') { /* (?(?...)) */
5159 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5160 || RExC_parse[1] == '<'
5161 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5164 ret = reg_node(pRExC_state, LOGICAL);
5167 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5171 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5172 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5174 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5175 char *name_start= RExC_parse++;
5177 SV *sv_dat=reg_scan_name(pRExC_state,
5178 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5179 if (RExC_parse == name_start || *RExC_parse != ch)
5180 vFAIL2("Sequence (?(%c... not terminated",
5181 (ch == '>' ? '<' : ch));
5184 num = add_data( pRExC_state, 1, "S" );
5185 RExC_rxi->data->data[num]=(void*)sv_dat;
5186 SvREFCNT_inc(sv_dat);
5188 ret = reganode(pRExC_state,NGROUPP,num);
5189 goto insert_if_check_paren;
5191 else if (RExC_parse[0] == 'D' &&
5192 RExC_parse[1] == 'E' &&
5193 RExC_parse[2] == 'F' &&
5194 RExC_parse[3] == 'I' &&
5195 RExC_parse[4] == 'N' &&
5196 RExC_parse[5] == 'E')
5198 ret = reganode(pRExC_state,DEFINEP,0);
5201 goto insert_if_check_paren;
5203 else if (RExC_parse[0] == 'R') {
5206 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5207 parno = atoi(RExC_parse++);
5208 while (isDIGIT(*RExC_parse))
5210 } else if (RExC_parse[0] == '&') {
5213 sv_dat = reg_scan_name(pRExC_state,
5214 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5215 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5217 ret = reganode(pRExC_state,INSUBP,parno);
5218 goto insert_if_check_paren;
5220 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5223 parno = atoi(RExC_parse++);
5225 while (isDIGIT(*RExC_parse))
5227 ret = reganode(pRExC_state, GROUPP, parno);
5229 insert_if_check_paren:
5230 if ((c = *nextchar(pRExC_state)) != ')')
5231 vFAIL("Switch condition not recognized");
5233 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5234 br = regbranch(pRExC_state, &flags, 1,depth+1);
5236 br = reganode(pRExC_state, LONGJMP, 0);
5238 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5239 c = *nextchar(pRExC_state);
5244 vFAIL("(?(DEFINE)....) does not allow branches");
5245 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5246 regbranch(pRExC_state, &flags, 1,depth+1);
5247 REGTAIL(pRExC_state, ret, lastbr);
5250 c = *nextchar(pRExC_state);
5255 vFAIL("Switch (?(condition)... contains too many branches");
5256 ender = reg_node(pRExC_state, TAIL);
5257 REGTAIL(pRExC_state, br, ender);
5259 REGTAIL(pRExC_state, lastbr, ender);
5260 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5263 REGTAIL(pRExC_state, ret, ender);
5267 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5271 RExC_parse--; /* for vFAIL to print correctly */
5272 vFAIL("Sequence (? incomplete");
5276 parse_flags: /* (?i) */
5277 while (*RExC_parse && strchr("iogcmsx", *RExC_parse)) {
5278 /* (?g), (?gc) and (?o) are useless here
5279 and must be globally applied -- japhy */
5281 if (*RExC_parse == 'o' || *RExC_parse == 'g') {
5282 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5283 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5284 if (! (wastedflags & wflagbit) ) {
5285 wastedflags |= wflagbit;
5288 "Useless (%s%c) - %suse /%c modifier",
5289 flagsp == &negflags ? "?-" : "?",
5291 flagsp == &negflags ? "don't " : "",
5297 else if (*RExC_parse == 'c') {
5298 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5299 if (! (wastedflags & WASTED_C) ) {
5300 wastedflags |= WASTED_GC;
5303 "Useless (%sc) - %suse /gc modifier",
5304 flagsp == &negflags ? "?-" : "?",
5305 flagsp == &negflags ? "don't " : ""
5310 else { pmflag(flagsp, *RExC_parse); }
5314 if (*RExC_parse == '-') {
5316 wastedflags = 0; /* reset so (?g-c) warns twice */
5320 RExC_flags |= posflags;
5321 RExC_flags &= ~negflags;
5322 if (*RExC_parse == ':') {
5328 if (*RExC_parse != ')') {
5330 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5332 nextchar(pRExC_state);
5342 ret = reganode(pRExC_state, OPEN, parno);
5345 RExC_nestroot = parno;
5346 if (RExC_seen & REG_SEEN_RECURSE) {
5347 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5348 "Setting open paren #%"IVdf" to %d\n",
5349 (IV)parno, REG_NODE_NUM(ret)));
5350 RExC_open_parens[parno-1]= ret;
5353 Set_Node_Length(ret, 1); /* MJD */
5354 Set_Node_Offset(ret, RExC_parse); /* MJD */
5361 /* Pick up the branches, linking them together. */
5362 parse_start = RExC_parse; /* MJD */
5363 br = regbranch(pRExC_state, &flags, 1,depth+1);
5364 /* branch_len = (paren != 0); */
5368 if (*RExC_parse == '|') {
5369 if (!SIZE_ONLY && RExC_extralen) {
5370 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5373 reginsert(pRExC_state, BRANCH, br, depth+1);
5374 Set_Node_Length(br, paren != 0);
5375 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5379 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5381 else if (paren == ':') {
5382 *flagp |= flags&SIMPLE;
5384 if (is_open) { /* Starts with OPEN. */
5385 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5387 else if (paren != '?') /* Not Conditional */
5389 *flagp |= flags & (SPSTART | HASWIDTH);
5391 while (*RExC_parse == '|') {
5392 if (!SIZE_ONLY && RExC_extralen) {
5393 ender = reganode(pRExC_state, LONGJMP,0);
5394 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5397 RExC_extralen += 2; /* Account for LONGJMP. */
5398 nextchar(pRExC_state);
5399 br = regbranch(pRExC_state, &flags, 0, depth+1);
5403 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5407 *flagp |= flags&SPSTART;
5410 if (have_branch || paren != ':') {
5411 /* Make a closing node, and hook it on the end. */
5414 ender = reg_node(pRExC_state, TAIL);
5417 ender = reganode(pRExC_state, CLOSE, parno);
5418 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5419 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5420 "Setting close paren #%"IVdf" to %d\n",
5421 (IV)parno, REG_NODE_NUM(ender)));
5422 RExC_close_parens[parno-1]= ender;
5423 if (RExC_nestroot == parno)
5426 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5427 Set_Node_Length(ender,1); /* MJD */
5433 *flagp &= ~HASWIDTH;
5436 ender = reg_node(pRExC_state, SUCCEED);
5439 ender = reg_node(pRExC_state, END);
5441 assert(!RExC_opend); /* there can only be one! */
5446 REGTAIL(pRExC_state, lastbr, ender);
5448 if (have_branch && !SIZE_ONLY) {
5450 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5452 /* Hook the tails of the branches to the closing node. */
5453 for (br = ret; br; br = regnext(br)) {
5454 const U8 op = PL_regkind[OP(br)];
5456 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5458 else if (op == BRANCHJ) {
5459 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5467 static const char parens[] = "=!<,>";
5469 if (paren && (p = strchr(parens, paren))) {
5470 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5471 int flag = (p - parens) > 1;
5474 node = SUSPEND, flag = 0;
5475 reginsert(pRExC_state, node,ret, depth+1);
5476 Set_Node_Cur_Length(ret);
5477 Set_Node_Offset(ret, parse_start + 1);
5479 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5483 /* Check for proper termination. */
5485 RExC_flags = oregflags;
5486 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5487 RExC_parse = oregcomp_parse;
5488 vFAIL("Unmatched (");
5491 else if (!paren && RExC_parse < RExC_end) {
5492 if (*RExC_parse == ')') {
5494 vFAIL("Unmatched )");
5497 FAIL("Junk on end of regexp"); /* "Can't happen". */
5505 - regbranch - one alternative of an | operator
5507 * Implements the concatenation operator.
5510 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5513 register regnode *ret;
5514 register regnode *chain = NULL;
5515 register regnode *latest;
5516 I32 flags = 0, c = 0;
5517 GET_RE_DEBUG_FLAGS_DECL;
5518 DEBUG_PARSE("brnc");
5522 if (!SIZE_ONLY && RExC_extralen)
5523 ret = reganode(pRExC_state, BRANCHJ,0);
5525 ret = reg_node(pRExC_state, BRANCH);
5526 Set_Node_Length(ret, 1);
5530 if (!first && SIZE_ONLY)
5531 RExC_extralen += 1; /* BRANCHJ */
5533 *flagp = WORST; /* Tentatively. */
5536 nextchar(pRExC_state);
5537 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5539 latest = regpiece(pRExC_state, &flags,depth+1);
5540 if (latest == NULL) {
5541 if (flags & TRYAGAIN)
5545 else if (ret == NULL)
5547 *flagp |= flags&HASWIDTH;
5548 if (chain == NULL) /* First piece. */
5549 *flagp |= flags&SPSTART;
5552 REGTAIL(pRExC_state, chain, latest);
5557 if (chain == NULL) { /* Loop ran zero times. */
5558 chain = reg_node(pRExC_state, NOTHING);
5563 *flagp |= flags&SIMPLE;
5570 - regpiece - something followed by possible [*+?]
5572 * Note that the branching code sequences used for ? and the general cases
5573 * of * and + are somewhat optimized: they use the same NOTHING node as
5574 * both the endmarker for their branch list and the body of the last branch.
5575 * It might seem that this node could be dispensed with entirely, but the
5576 * endmarker role is not redundant.
5579 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5582 register regnode *ret;
5584 register char *next;
5586 const char * const origparse = RExC_parse;
5588 I32 max = REG_INFTY;
5590 const char *maxpos = NULL;
5591 GET_RE_DEBUG_FLAGS_DECL;
5592 DEBUG_PARSE("piec");
5594 ret = regatom(pRExC_state, &flags,depth+1);
5596 if (flags & TRYAGAIN)
5603 if (op == '{' && regcurly(RExC_parse)) {
5605 parse_start = RExC_parse; /* MJD */
5606 next = RExC_parse + 1;
5607 while (isDIGIT(*next) || *next == ',') {
5616 if (*next == '}') { /* got one */
5620 min = atoi(RExC_parse);
5624 maxpos = RExC_parse;
5626 if (!max && *maxpos != '0')
5627 max = REG_INFTY; /* meaning "infinity" */
5628 else if (max >= REG_INFTY)
5629 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5631 nextchar(pRExC_state);
5634 if ((flags&SIMPLE)) {
5635 RExC_naughty += 2 + RExC_naughty / 2;
5636 reginsert(pRExC_state, CURLY, ret, depth+1);
5637 Set_Node_Offset(ret, parse_start+1); /* MJD */
5638 Set_Node_Cur_Length(ret);
5641 regnode * const w = reg_node(pRExC_state, WHILEM);
5644 REGTAIL(pRExC_state, ret, w);
5645 if (!SIZE_ONLY && RExC_extralen) {
5646 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5647 reginsert(pRExC_state, NOTHING,ret, depth+1);
5648 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5650 reginsert(pRExC_state, CURLYX,ret, depth+1);
5652 Set_Node_Offset(ret, parse_start+1);
5653 Set_Node_Length(ret,
5654 op == '{' ? (RExC_parse - parse_start) : 1);
5656 if (!SIZE_ONLY && RExC_extralen)
5657 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5658 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5660 RExC_whilem_seen++, RExC_extralen += 3;
5661 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5669 if (max && max < min)
5670 vFAIL("Can't do {n,m} with n > m");
5672 ARG1_SET(ret, (U16)min);
5673 ARG2_SET(ret, (U16)max);
5685 #if 0 /* Now runtime fix should be reliable. */
5687 /* if this is reinstated, don't forget to put this back into perldiag:
5689 =item Regexp *+ operand could be empty at {#} in regex m/%s/
5691 (F) The part of the regexp subject to either the * or + quantifier
5692 could match an empty string. The {#} shows in the regular
5693 expression about where the problem was discovered.
5697 if (!(flags&HASWIDTH) && op != '?')
5698 vFAIL("Regexp *+ operand could be empty");
5701 parse_start = RExC_parse;
5702 nextchar(pRExC_state);
5704 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
5706 if (op == '*' && (flags&SIMPLE)) {
5707 reginsert(pRExC_state, STAR, ret, depth+1);
5711 else if (op == '*') {
5715 else if (op == '+' && (flags&SIMPLE)) {
5716 reginsert(pRExC_state, PLUS, ret, depth+1);
5720 else if (op == '+') {
5724 else if (op == '?') {
5729 if (!SIZE_ONLY && !(flags&HASWIDTH) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
5731 "%.*s matches null string many times",
5732 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
5736 if (RExC_parse < RExC_end && *RExC_parse == '?') {
5737 nextchar(pRExC_state);
5738 reginsert(pRExC_state, MINMOD, ret, depth+1);
5739 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
5741 #ifndef REG_ALLOW_MINMOD_SUSPEND
5744 if (RExC_parse < RExC_end && *RExC_parse == '+') {
5746 nextchar(pRExC_state);
5747 ender = reg_node(pRExC_state, SUCCEED);
5748 REGTAIL(pRExC_state, ret, ender);
5749 reginsert(pRExC_state, SUSPEND, ret, depth+1);
5751 ender = reg_node(pRExC_state, TAIL);
5752 REGTAIL(pRExC_state, ret, ender);
5756 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
5758 vFAIL("Nested quantifiers");
5765 /* reg_namedseq(pRExC_state,UVp)
5767 This is expected to be called by a parser routine that has
5768 recognized'\N' and needs to handle the rest. RExC_parse is
5769 expected to point at the first char following the N at the time
5772 If valuep is non-null then it is assumed that we are parsing inside
5773 of a charclass definition and the first codepoint in the resolved
5774 string is returned via *valuep and the routine will return NULL.
5775 In this mode if a multichar string is returned from the charnames
5776 handler a warning will be issued, and only the first char in the
5777 sequence will be examined. If the string returned is zero length
5778 then the value of *valuep is undefined and NON-NULL will
5779 be returned to indicate failure. (This will NOT be a valid pointer
5782 If value is null then it is assumed that we are parsing normal text
5783 and inserts a new EXACT node into the program containing the resolved
5784 string and returns a pointer to the new node. If the string is
5785 zerolength a NOTHING node is emitted.
5787 On success RExC_parse is set to the char following the endbrace.
5788 Parsing failures will generate a fatal errorvia vFAIL(...)
5790 NOTE: We cache all results from the charnames handler locally in
5791 the RExC_charnames hash (created on first use) to prevent a charnames
5792 handler from playing silly-buggers and returning a short string and
5793 then a long string for a given pattern. Since the regexp program
5794 size is calculated during an initial parse this would result
5795 in a buffer overrun so we cache to prevent the charname result from
5796 changing during the course of the parse.
5800 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
5802 char * name; /* start of the content of the name */
5803 char * endbrace; /* endbrace following the name */
5806 STRLEN len; /* this has various purposes throughout the code */
5807 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
5808 regnode *ret = NULL;
5810 if (*RExC_parse != '{') {
5811 vFAIL("Missing braces on \\N{}");
5813 name = RExC_parse+1;
5814 endbrace = strchr(RExC_parse, '}');
5817 vFAIL("Missing right brace on \\N{}");
5819 RExC_parse = endbrace + 1;
5822 /* RExC_parse points at the beginning brace,
5823 endbrace points at the last */
5824 if ( name[0]=='U' && name[1]=='+' ) {
5825 /* its a "unicode hex" notation {U+89AB} */
5826 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
5827 | PERL_SCAN_DISALLOW_PREFIX
5828 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
5830 len = (STRLEN)(endbrace - name - 2);
5831 cp = grok_hex(name + 2, &len, &fl, NULL);
5832 if ( len != (STRLEN)(endbrace - name - 2) ) {
5841 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
5843 /* fetch the charnames handler for this scope */
5844 HV * const table = GvHV(PL_hintgv);
5846 hv_fetchs(table, "charnames", FALSE) :
5848 SV *cv= cvp ? *cvp : NULL;
5851 /* create an SV with the name as argument */
5852 sv_name = newSVpvn(name, endbrace - name);
5854 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
5855 vFAIL2("Constant(\\N{%s}) unknown: "
5856 "(possibly a missing \"use charnames ...\")",
5859 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
5860 vFAIL2("Constant(\\N{%s}): "
5861 "$^H{charnames} is not defined",SvPVX(sv_name));
5866 if (!RExC_charnames) {
5867 /* make sure our cache is allocated */
5868 RExC_charnames = newHV();
5869 sv_2mortal((SV*)RExC_charnames);
5871 /* see if we have looked this one up before */
5872 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
5874 sv_str = HeVAL(he_str);
5887 count= call_sv(cv, G_SCALAR);
5889 if (count == 1) { /* XXXX is this right? dmq */
5891 SvREFCNT_inc_simple_void(sv_str);
5899 if ( !sv_str || !SvOK(sv_str) ) {
5900 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
5901 "did not return a defined value",SvPVX(sv_name));
5903 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
5908 char *p = SvPV(sv_str, len);
5911 if ( SvUTF8(sv_str) ) {
5912 *valuep = utf8_to_uvchr((U8*)p, &numlen);
5916 We have to turn on utf8 for high bit chars otherwise
5917 we get failures with
5919 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5920 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
5922 This is different from what \x{} would do with the same
5923 codepoint, where the condition is > 0xFF.
5930 /* warn if we havent used the whole string? */
5932 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5934 "Ignoring excess chars from \\N{%s} in character class",
5938 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5940 "Ignoring zero length \\N{%s} in character class",
5945 SvREFCNT_dec(sv_name);
5947 SvREFCNT_dec(sv_str);
5948 return len ? NULL : (regnode *)&len;
5949 } else if(SvCUR(sv_str)) {
5954 char * parse_start = name-3; /* needed for the offsets */
5955 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
5957 ret = reg_node(pRExC_state,
5958 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
5961 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
5962 sv_utf8_upgrade(sv_str);
5963 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
5967 p = SvPV(sv_str, len);
5969 /* len is the length written, charlen is the size the char read */
5970 for ( len = 0; p < pend; p += charlen ) {
5972 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
5974 STRLEN foldlen,numlen;
5975 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
5976 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
5977 /* Emit all the Unicode characters. */
5979 for (foldbuf = tmpbuf;
5983 uvc = utf8_to_uvchr(foldbuf, &numlen);
5985 const STRLEN unilen = reguni(pRExC_state, uvc, s);
5988 /* In EBCDIC the numlen
5989 * and unilen can differ. */
5991 if (numlen >= foldlen)
5995 break; /* "Can't happen." */
5998 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6010 RExC_size += STR_SZ(len);
6013 RExC_emit += STR_SZ(len);
6015 Set_Node_Cur_Length(ret); /* MJD */
6017 nextchar(pRExC_state);
6019 ret = reg_node(pRExC_state,NOTHING);
6022 SvREFCNT_dec(sv_str);
6025 SvREFCNT_dec(sv_name);
6035 * It returns the code point in utf8 for the value in *encp.
6036 * value: a code value in the source encoding
6037 * encp: a pointer to an Encode object
6039 * If the result from Encode is not a single character,
6040 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6043 S_reg_recode(pTHX_ const char value, SV **encp)
6046 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6047 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6049 const STRLEN newlen = SvCUR(sv);
6050 UV uv = UNICODE_REPLACEMENT;
6054 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6057 if (!newlen || numlen != newlen) {
6058 uv = UNICODE_REPLACEMENT;
6067 - regatom - the lowest level
6069 * Optimization: gobbles an entire sequence of ordinary characters so that
6070 * it can turn them into a single node, which is smaller to store and
6071 * faster to run. Backslashed characters are exceptions, each becoming a
6072 * separate node; the code is simpler that way and it's not worth fixing.
6074 * [Yes, it is worth fixing, some scripts can run twice the speed.]
6075 * [It looks like its ok, as in S_study_chunk we merge adjacent EXACT nodes]
6078 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6081 register regnode *ret = NULL;
6083 char *parse_start = RExC_parse;
6084 GET_RE_DEBUG_FLAGS_DECL;
6085 DEBUG_PARSE("atom");
6086 *flagp = WORST; /* Tentatively. */
6089 switch (*RExC_parse) {
6091 RExC_seen_zerolen++;
6092 nextchar(pRExC_state);
6093 if (RExC_flags & RXf_PMf_MULTILINE)
6094 ret = reg_node(pRExC_state, MBOL);
6095 else if (RExC_flags & RXf_PMf_SINGLELINE)
6096 ret = reg_node(pRExC_state, SBOL);
6098 ret = reg_node(pRExC_state, BOL);
6099 Set_Node_Length(ret, 1); /* MJD */
6102 nextchar(pRExC_state);
6104 RExC_seen_zerolen++;
6105 if (RExC_flags & RXf_PMf_MULTILINE)
6106 ret = reg_node(pRExC_state, MEOL);
6107 else if (RExC_flags & RXf_PMf_SINGLELINE)
6108 ret = reg_node(pRExC_state, SEOL);
6110 ret = reg_node(pRExC_state, EOL);
6111 Set_Node_Length(ret, 1); /* MJD */
6114 nextchar(pRExC_state);
6115 if (RExC_flags & RXf_PMf_SINGLELINE)
6116 ret = reg_node(pRExC_state, SANY);
6118 ret = reg_node(pRExC_state, REG_ANY);
6119 *flagp |= HASWIDTH|SIMPLE;
6121 Set_Node_Length(ret, 1); /* MJD */
6125 char * const oregcomp_parse = ++RExC_parse;
6126 ret = regclass(pRExC_state,depth+1);
6127 if (*RExC_parse != ']') {
6128 RExC_parse = oregcomp_parse;
6129 vFAIL("Unmatched [");
6131 nextchar(pRExC_state);
6132 *flagp |= HASWIDTH|SIMPLE;
6133 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6137 nextchar(pRExC_state);
6138 ret = reg(pRExC_state, 1, &flags,depth+1);
6140 if (flags & TRYAGAIN) {
6141 if (RExC_parse == RExC_end) {
6142 /* Make parent create an empty node if needed. */
6150 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE);
6154 if (flags & TRYAGAIN) {
6158 vFAIL("Internal urp");
6159 /* Supposed to be caught earlier. */
6162 if (!regcurly(RExC_parse)) {
6171 vFAIL("Quantifier follows nothing");
6174 switch (*++RExC_parse) {
6176 RExC_seen_zerolen++;
6177 ret = reg_node(pRExC_state, SBOL);
6179 nextchar(pRExC_state);
6180 Set_Node_Length(ret, 2); /* MJD */
6183 ret = reg_node(pRExC_state, GPOS);
6184 RExC_seen |= REG_SEEN_GPOS;
6186 nextchar(pRExC_state);
6187 Set_Node_Length(ret, 2); /* MJD */
6190 ret = reg_node(pRExC_state, SEOL);
6192 RExC_seen_zerolen++; /* Do not optimize RE away */
6193 nextchar(pRExC_state);
6196 ret = reg_node(pRExC_state, EOS);
6198 RExC_seen_zerolen++; /* Do not optimize RE away */
6199 nextchar(pRExC_state);
6200 Set_Node_Length(ret, 2); /* MJD */
6203 ret = reg_node(pRExC_state, CANY);
6204 RExC_seen |= REG_SEEN_CANY;
6205 *flagp |= HASWIDTH|SIMPLE;
6206 nextchar(pRExC_state);
6207 Set_Node_Length(ret, 2); /* MJD */
6210 ret = reg_node(pRExC_state, CLUMP);
6212 nextchar(pRExC_state);
6213 Set_Node_Length(ret, 2); /* MJD */
6216 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6217 *flagp |= HASWIDTH|SIMPLE;
6218 nextchar(pRExC_state);
6219 Set_Node_Length(ret, 2); /* MJD */
6222 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6223 *flagp |= HASWIDTH|SIMPLE;
6224 nextchar(pRExC_state);
6225 Set_Node_Length(ret, 2); /* MJD */
6228 RExC_seen_zerolen++;
6229 RExC_seen |= REG_SEEN_LOOKBEHIND;
6230 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6232 nextchar(pRExC_state);
6233 Set_Node_Length(ret, 2); /* MJD */
6236 RExC_seen_zerolen++;
6237 RExC_seen |= REG_SEEN_LOOKBEHIND;
6238 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6240 nextchar(pRExC_state);
6241 Set_Node_Length(ret, 2); /* MJD */
6244 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6245 *flagp |= HASWIDTH|SIMPLE;
6246 nextchar(pRExC_state);
6247 Set_Node_Length(ret, 2); /* MJD */
6250 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6251 *flagp |= HASWIDTH|SIMPLE;
6252 nextchar(pRExC_state);
6253 Set_Node_Length(ret, 2); /* MJD */
6256 ret = reg_node(pRExC_state, DIGIT);
6257 *flagp |= HASWIDTH|SIMPLE;
6258 nextchar(pRExC_state);
6259 Set_Node_Length(ret, 2); /* MJD */
6262 ret = reg_node(pRExC_state, NDIGIT);
6263 *flagp |= HASWIDTH|SIMPLE;
6264 nextchar(pRExC_state);
6265 Set_Node_Length(ret, 2); /* MJD */
6270 char* const oldregxend = RExC_end;
6271 char* parse_start = RExC_parse - 2;
6273 if (RExC_parse[1] == '{') {
6274 /* a lovely hack--pretend we saw [\pX] instead */
6275 RExC_end = strchr(RExC_parse, '}');
6277 const U8 c = (U8)*RExC_parse;
6279 RExC_end = oldregxend;
6280 vFAIL2("Missing right brace on \\%c{}", c);
6285 RExC_end = RExC_parse + 2;
6286 if (RExC_end > oldregxend)
6287 RExC_end = oldregxend;
6291 ret = regclass(pRExC_state,depth+1);
6293 RExC_end = oldregxend;
6296 Set_Node_Offset(ret, parse_start + 2);
6297 Set_Node_Cur_Length(ret);
6298 nextchar(pRExC_state);
6299 *flagp |= HASWIDTH|SIMPLE;
6303 /* Handle \N{NAME} here and not below because it can be
6304 multicharacter. join_exact() will join them up later on.
6305 Also this makes sure that things like /\N{BLAH}+/ and
6306 \N{BLAH} being multi char Just Happen. dmq*/
6308 ret= reg_namedseq(pRExC_state, NULL);
6310 case 'k': /* Handle \k<NAME> and \k'NAME' */
6312 char ch= RExC_parse[1];
6313 if (ch != '<' && ch != '\'') {
6315 vWARN( RExC_parse + 1,
6316 "Possible broken named back reference treated as literal k");
6320 char* name_start = (RExC_parse += 2);
6322 SV *sv_dat = reg_scan_name(pRExC_state,
6323 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6324 ch= (ch == '<') ? '>' : '\'';
6326 if (RExC_parse == name_start || *RExC_parse != ch)
6327 vFAIL2("Sequence \\k%c... not terminated",
6328 (ch == '>' ? '<' : ch));
6331 ret = reganode(pRExC_state,
6332 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6338 num = add_data( pRExC_state, 1, "S" );
6340 RExC_rxi->data->data[num]=(void*)sv_dat;
6341 SvREFCNT_inc(sv_dat);
6343 /* override incorrect value set in reganode MJD */
6344 Set_Node_Offset(ret, parse_start+1);
6345 Set_Node_Cur_Length(ret); /* MJD */
6346 nextchar(pRExC_state);
6362 case '1': case '2': case '3': case '4':
6363 case '5': case '6': case '7': case '8': case '9':
6366 bool isrel=(*RExC_parse=='R');
6369 num = atoi(RExC_parse);
6371 num = RExC_npar - num;
6373 vFAIL("Reference to nonexistent or unclosed group");
6375 if (num > 9 && num >= RExC_npar)
6378 char * const parse_start = RExC_parse - 1; /* MJD */
6379 while (isDIGIT(*RExC_parse))
6383 if (num > (I32)RExC_rx->nparens)
6384 vFAIL("Reference to nonexistent group");
6387 ret = reganode(pRExC_state,
6388 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6392 /* override incorrect value set in reganode MJD */
6393 Set_Node_Offset(ret, parse_start+1);
6394 Set_Node_Cur_Length(ret); /* MJD */
6396 nextchar(pRExC_state);
6401 if (RExC_parse >= RExC_end)
6402 FAIL("Trailing \\");
6405 /* Do not generate "unrecognized" warnings here, we fall
6406 back into the quick-grab loop below */
6413 if (RExC_flags & RXf_PMf_EXTENDED) {
6414 while (RExC_parse < RExC_end && *RExC_parse != '\n')
6416 if (RExC_parse < RExC_end)
6422 register STRLEN len;
6427 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6429 parse_start = RExC_parse - 1;
6435 ret = reg_node(pRExC_state,
6436 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6438 for (len = 0, p = RExC_parse - 1;
6439 len < 127 && p < RExC_end;
6442 char * const oldp = p;
6444 if (RExC_flags & RXf_PMf_EXTENDED)
6445 p = regwhite(p, RExC_end);
6495 ender = ASCII_TO_NATIVE('\033');
6499 ender = ASCII_TO_NATIVE('\007');
6504 char* const e = strchr(p, '}');
6508 vFAIL("Missing right brace on \\x{}");
6511 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6512 | PERL_SCAN_DISALLOW_PREFIX;
6513 STRLEN numlen = e - p - 1;
6514 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6521 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6523 ender = grok_hex(p, &numlen, &flags, NULL);
6526 if (PL_encoding && ender < 0x100)
6527 goto recode_encoding;
6531 ender = UCHARAT(p++);
6532 ender = toCTRL(ender);
6534 case '0': case '1': case '2': case '3':case '4':
6535 case '5': case '6': case '7': case '8':case '9':
6537 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6540 ender = grok_oct(p, &numlen, &flags, NULL);
6547 if (PL_encoding && ender < 0x100)
6548 goto recode_encoding;
6552 SV* enc = PL_encoding;
6553 ender = reg_recode((const char)(U8)ender, &enc);
6554 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6555 vWARN(p, "Invalid escape in the specified encoding");
6561 FAIL("Trailing \\");
6564 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6565 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6566 goto normal_default;
6571 if (UTF8_IS_START(*p) && UTF) {
6573 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6574 &numlen, UTF8_ALLOW_DEFAULT);
6581 if (RExC_flags & RXf_PMf_EXTENDED)
6582 p = regwhite(p, RExC_end);
6584 /* Prime the casefolded buffer. */
6585 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6587 if (ISMULT2(p)) { /* Back off on ?+*. */
6592 /* Emit all the Unicode characters. */
6594 for (foldbuf = tmpbuf;
6596 foldlen -= numlen) {
6597 ender = utf8_to_uvchr(foldbuf, &numlen);
6599 const STRLEN unilen = reguni(pRExC_state, ender, s);
6602 /* In EBCDIC the numlen
6603 * and unilen can differ. */
6605 if (numlen >= foldlen)
6609 break; /* "Can't happen." */
6613 const STRLEN unilen = reguni(pRExC_state, ender, s);
6622 REGC((char)ender, s++);
6628 /* Emit all the Unicode characters. */
6630 for (foldbuf = tmpbuf;
6632 foldlen -= numlen) {
6633 ender = utf8_to_uvchr(foldbuf, &numlen);
6635 const STRLEN unilen = reguni(pRExC_state, ender, s);
6638 /* In EBCDIC the numlen
6639 * and unilen can differ. */
6641 if (numlen >= foldlen)
6649 const STRLEN unilen = reguni(pRExC_state, ender, s);
6658 REGC((char)ender, s++);
6662 Set_Node_Cur_Length(ret); /* MJD */
6663 nextchar(pRExC_state);
6665 /* len is STRLEN which is unsigned, need to copy to signed */
6668 vFAIL("Internal disaster");
6672 if (len == 1 && UNI_IS_INVARIANT(ender))
6676 RExC_size += STR_SZ(len);
6679 RExC_emit += STR_SZ(len);
6689 S_regwhite(char *p, const char *e)
6694 else if (*p == '#') {
6697 } while (p < e && *p != '\n');
6705 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
6706 Character classes ([:foo:]) can also be negated ([:^foo:]).
6707 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
6708 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
6709 but trigger failures because they are currently unimplemented. */
6711 #define POSIXCC_DONE(c) ((c) == ':')
6712 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
6713 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
6716 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
6719 I32 namedclass = OOB_NAMEDCLASS;
6721 if (value == '[' && RExC_parse + 1 < RExC_end &&
6722 /* I smell either [: or [= or [. -- POSIX has been here, right? */
6723 POSIXCC(UCHARAT(RExC_parse))) {
6724 const char c = UCHARAT(RExC_parse);
6725 char* const s = RExC_parse++;
6727 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
6729 if (RExC_parse == RExC_end)
6730 /* Grandfather lone [:, [=, [. */
6733 const char* const t = RExC_parse++; /* skip over the c */
6736 if (UCHARAT(RExC_parse) == ']') {
6737 const char *posixcc = s + 1;
6738 RExC_parse++; /* skip over the ending ] */
6741 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
6742 const I32 skip = t - posixcc;
6744 /* Initially switch on the length of the name. */
6747 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
6748 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
6751 /* Names all of length 5. */
6752 /* alnum alpha ascii blank cntrl digit graph lower
6753 print punct space upper */
6754 /* Offset 4 gives the best switch position. */
6755 switch (posixcc[4]) {
6757 if (memEQ(posixcc, "alph", 4)) /* alpha */
6758 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
6761 if (memEQ(posixcc, "spac", 4)) /* space */
6762 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
6765 if (memEQ(posixcc, "grap", 4)) /* graph */
6766 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
6769 if (memEQ(posixcc, "asci", 4)) /* ascii */
6770 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
6773 if (memEQ(posixcc, "blan", 4)) /* blank */
6774 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
6777 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
6778 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
6781 if (memEQ(posixcc, "alnu", 4)) /* alnum */
6782 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
6785 if (memEQ(posixcc, "lowe", 4)) /* lower */
6786 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
6787 else if (memEQ(posixcc, "uppe", 4)) /* upper */
6788 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
6791 if (memEQ(posixcc, "digi", 4)) /* digit */
6792 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
6793 else if (memEQ(posixcc, "prin", 4)) /* print */
6794 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
6795 else if (memEQ(posixcc, "punc", 4)) /* punct */
6796 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
6801 if (memEQ(posixcc, "xdigit", 6))
6802 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
6806 if (namedclass == OOB_NAMEDCLASS)
6807 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
6809 assert (posixcc[skip] == ':');
6810 assert (posixcc[skip+1] == ']');
6811 } else if (!SIZE_ONLY) {
6812 /* [[=foo=]] and [[.foo.]] are still future. */
6814 /* adjust RExC_parse so the warning shows after
6816 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
6818 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6821 /* Maternal grandfather:
6822 * "[:" ending in ":" but not in ":]" */
6832 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
6835 if (POSIXCC(UCHARAT(RExC_parse))) {
6836 const char *s = RExC_parse;
6837 const char c = *s++;
6841 if (*s && c == *s && s[1] == ']') {
6842 if (ckWARN(WARN_REGEXP))
6844 "POSIX syntax [%c %c] belongs inside character classes",
6847 /* [[=foo=]] and [[.foo.]] are still future. */
6848 if (POSIXCC_NOTYET(c)) {
6849 /* adjust RExC_parse so the error shows after
6851 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
6853 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
6861 parse a class specification and produce either an ANYOF node that
6862 matches the pattern. If the pattern matches a single char only and
6863 that char is < 256 then we produce an EXACT node instead.
6866 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
6869 register UV value = 0;
6870 register UV nextvalue;
6871 register IV prevvalue = OOB_UNICODE;
6872 register IV range = 0;
6873 register regnode *ret;
6876 char *rangebegin = NULL;
6877 bool need_class = 0;
6880 bool optimize_invert = TRUE;
6881 AV* unicode_alternate = NULL;
6883 UV literal_endpoint = 0;
6885 UV stored = 0; /* number of chars stored in the class */
6887 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
6888 case we need to change the emitted regop to an EXACT. */
6889 const char * orig_parse = RExC_parse;
6890 GET_RE_DEBUG_FLAGS_DECL;
6892 PERL_UNUSED_ARG(depth);
6895 DEBUG_PARSE("clas");
6897 /* Assume we are going to generate an ANYOF node. */
6898 ret = reganode(pRExC_state, ANYOF, 0);
6901 ANYOF_FLAGS(ret) = 0;
6903 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
6907 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
6911 RExC_size += ANYOF_SKIP;
6912 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
6915 RExC_emit += ANYOF_SKIP;
6917 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
6919 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
6920 ANYOF_BITMAP_ZERO(ret);
6921 listsv = newSVpvs("# comment\n");
6924 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6926 if (!SIZE_ONLY && POSIXCC(nextvalue))
6927 checkposixcc(pRExC_state);
6929 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
6930 if (UCHARAT(RExC_parse) == ']')
6934 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
6938 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
6941 rangebegin = RExC_parse;
6943 value = utf8n_to_uvchr((U8*)RExC_parse,
6944 RExC_end - RExC_parse,
6945 &numlen, UTF8_ALLOW_DEFAULT);
6946 RExC_parse += numlen;
6949 value = UCHARAT(RExC_parse++);
6951 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
6952 if (value == '[' && POSIXCC(nextvalue))
6953 namedclass = regpposixcc(pRExC_state, value);
6954 else if (value == '\\') {
6956 value = utf8n_to_uvchr((U8*)RExC_parse,
6957 RExC_end - RExC_parse,
6958 &numlen, UTF8_ALLOW_DEFAULT);
6959 RExC_parse += numlen;
6962 value = UCHARAT(RExC_parse++);
6963 /* Some compilers cannot handle switching on 64-bit integer
6964 * values, therefore value cannot be an UV. Yes, this will
6965 * be a problem later if we want switch on Unicode.
6966 * A similar issue a little bit later when switching on
6967 * namedclass. --jhi */
6968 switch ((I32)value) {
6969 case 'w': namedclass = ANYOF_ALNUM; break;
6970 case 'W': namedclass = ANYOF_NALNUM; break;
6971 case 's': namedclass = ANYOF_SPACE; break;
6972 case 'S': namedclass = ANYOF_NSPACE; break;
6973 case 'd': namedclass = ANYOF_DIGIT; break;
6974 case 'D': namedclass = ANYOF_NDIGIT; break;
6975 case 'N': /* Handle \N{NAME} in class */
6977 /* We only pay attention to the first char of
6978 multichar strings being returned. I kinda wonder
6979 if this makes sense as it does change the behaviour
6980 from earlier versions, OTOH that behaviour was broken
6982 UV v; /* value is register so we cant & it /grrr */
6983 if (reg_namedseq(pRExC_state, &v)) {
6993 if (RExC_parse >= RExC_end)
6994 vFAIL2("Empty \\%c{}", (U8)value);
6995 if (*RExC_parse == '{') {
6996 const U8 c = (U8)value;
6997 e = strchr(RExC_parse++, '}');
6999 vFAIL2("Missing right brace on \\%c{}", c);
7000 while (isSPACE(UCHARAT(RExC_parse)))
7002 if (e == RExC_parse)
7003 vFAIL2("Empty \\%c{}", c);
7005 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7013 if (UCHARAT(RExC_parse) == '^') {
7016 value = value == 'p' ? 'P' : 'p'; /* toggle */
7017 while (isSPACE(UCHARAT(RExC_parse))) {
7022 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7023 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7026 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7027 namedclass = ANYOF_MAX; /* no official name, but it's named */
7030 case 'n': value = '\n'; break;
7031 case 'r': value = '\r'; break;
7032 case 't': value = '\t'; break;
7033 case 'f': value = '\f'; break;
7034 case 'b': value = '\b'; break;
7035 case 'e': value = ASCII_TO_NATIVE('\033');break;
7036 case 'a': value = ASCII_TO_NATIVE('\007');break;
7038 if (*RExC_parse == '{') {
7039 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7040 | PERL_SCAN_DISALLOW_PREFIX;
7041 char * const e = strchr(RExC_parse++, '}');
7043 vFAIL("Missing right brace on \\x{}");
7045 numlen = e - RExC_parse;
7046 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7050 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7052 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7053 RExC_parse += numlen;
7055 if (PL_encoding && value < 0x100)
7056 goto recode_encoding;
7059 value = UCHARAT(RExC_parse++);
7060 value = toCTRL(value);
7062 case '0': case '1': case '2': case '3': case '4':
7063 case '5': case '6': case '7': case '8': case '9':
7067 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7068 RExC_parse += numlen;
7069 if (PL_encoding && value < 0x100)
7070 goto recode_encoding;
7075 SV* enc = PL_encoding;
7076 value = reg_recode((const char)(U8)value, &enc);
7077 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7079 "Invalid escape in the specified encoding");
7083 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7085 "Unrecognized escape \\%c in character class passed through",
7089 } /* end of \blah */
7095 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7097 if (!SIZE_ONLY && !need_class)
7098 ANYOF_CLASS_ZERO(ret);
7102 /* a bad range like a-\d, a-[:digit:] ? */
7105 if (ckWARN(WARN_REGEXP)) {
7107 RExC_parse >= rangebegin ?
7108 RExC_parse - rangebegin : 0;
7110 "False [] range \"%*.*s\"",
7113 if (prevvalue < 256) {
7114 ANYOF_BITMAP_SET(ret, prevvalue);
7115 ANYOF_BITMAP_SET(ret, '-');
7118 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7119 Perl_sv_catpvf(aTHX_ listsv,
7120 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7124 range = 0; /* this was not a true range */
7128 const char *what = NULL;
7131 if (namedclass > OOB_NAMEDCLASS)
7132 optimize_invert = FALSE;
7133 /* Possible truncation here but in some 64-bit environments
7134 * the compiler gets heartburn about switch on 64-bit values.
7135 * A similar issue a little earlier when switching on value.
7137 switch ((I32)namedclass) {
7140 ANYOF_CLASS_SET(ret, ANYOF_ALNUM);
7142 for (value = 0; value < 256; value++)
7144 ANYOF_BITMAP_SET(ret, value);
7151 ANYOF_CLASS_SET(ret, ANYOF_NALNUM);
7153 for (value = 0; value < 256; value++)
7154 if (!isALNUM(value))
7155 ANYOF_BITMAP_SET(ret, value);
7162 ANYOF_CLASS_SET(ret, ANYOF_ALNUMC);
7164 for (value = 0; value < 256; value++)
7165 if (isALNUMC(value))
7166 ANYOF_BITMAP_SET(ret, value);
7173 ANYOF_CLASS_SET(ret, ANYOF_NALNUMC);
7175 for (value = 0; value < 256; value++)
7176 if (!isALNUMC(value))
7177 ANYOF_BITMAP_SET(ret, value);
7184 ANYOF_CLASS_SET(ret, ANYOF_ALPHA);
7186 for (value = 0; value < 256; value++)
7188 ANYOF_BITMAP_SET(ret, value);
7195 ANYOF_CLASS_SET(ret, ANYOF_NALPHA);
7197 for (value = 0; value < 256; value++)
7198 if (!isALPHA(value))
7199 ANYOF_BITMAP_SET(ret, value);
7206 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7209 for (value = 0; value < 128; value++)
7210 ANYOF_BITMAP_SET(ret, value);
7212 for (value = 0; value < 256; value++) {
7214 ANYOF_BITMAP_SET(ret, value);
7223 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7226 for (value = 128; value < 256; value++)
7227 ANYOF_BITMAP_SET(ret, value);
7229 for (value = 0; value < 256; value++) {
7230 if (!isASCII(value))
7231 ANYOF_BITMAP_SET(ret, value);
7240 ANYOF_CLASS_SET(ret, ANYOF_BLANK);
7242 for (value = 0; value < 256; value++)
7244 ANYOF_BITMAP_SET(ret, value);
7251 ANYOF_CLASS_SET(ret, ANYOF_NBLANK);
7253 for (value = 0; value < 256; value++)
7254 if (!isBLANK(value))
7255 ANYOF_BITMAP_SET(ret, value);
7262 ANYOF_CLASS_SET(ret, ANYOF_CNTRL);
7264 for (value = 0; value < 256; value++)
7266 ANYOF_BITMAP_SET(ret, value);
7273 ANYOF_CLASS_SET(ret, ANYOF_NCNTRL);
7275 for (value = 0; value < 256; value++)
7276 if (!isCNTRL(value))
7277 ANYOF_BITMAP_SET(ret, value);
7284 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7286 /* consecutive digits assumed */
7287 for (value = '0'; value <= '9'; value++)
7288 ANYOF_BITMAP_SET(ret, value);
7295 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7297 /* consecutive digits assumed */
7298 for (value = 0; value < '0'; value++)
7299 ANYOF_BITMAP_SET(ret, value);
7300 for (value = '9' + 1; value < 256; value++)
7301 ANYOF_BITMAP_SET(ret, value);
7308 ANYOF_CLASS_SET(ret, ANYOF_GRAPH);
7310 for (value = 0; value < 256; value++)
7312 ANYOF_BITMAP_SET(ret, value);
7319 ANYOF_CLASS_SET(ret, ANYOF_NGRAPH);
7321 for (value = 0; value < 256; value++)
7322 if (!isGRAPH(value))
7323 ANYOF_BITMAP_SET(ret, value);
7330 ANYOF_CLASS_SET(ret, ANYOF_LOWER);
7332 for (value = 0; value < 256; value++)
7334 ANYOF_BITMAP_SET(ret, value);
7341 ANYOF_CLASS_SET(ret, ANYOF_NLOWER);
7343 for (value = 0; value < 256; value++)
7344 if (!isLOWER(value))
7345 ANYOF_BITMAP_SET(ret, value);
7352 ANYOF_CLASS_SET(ret, ANYOF_PRINT);
7354 for (value = 0; value < 256; value++)
7356 ANYOF_BITMAP_SET(ret, value);
7363 ANYOF_CLASS_SET(ret, ANYOF_NPRINT);
7365 for (value = 0; value < 256; value++)
7366 if (!isPRINT(value))
7367 ANYOF_BITMAP_SET(ret, value);
7374 ANYOF_CLASS_SET(ret, ANYOF_PSXSPC);
7376 for (value = 0; value < 256; value++)
7377 if (isPSXSPC(value))
7378 ANYOF_BITMAP_SET(ret, value);
7385 ANYOF_CLASS_SET(ret, ANYOF_NPSXSPC);
7387 for (value = 0; value < 256; value++)
7388 if (!isPSXSPC(value))
7389 ANYOF_BITMAP_SET(ret, value);
7396 ANYOF_CLASS_SET(ret, ANYOF_PUNCT);
7398 for (value = 0; value < 256; value++)
7400 ANYOF_BITMAP_SET(ret, value);
7407 ANYOF_CLASS_SET(ret, ANYOF_NPUNCT);
7409 for (value = 0; value < 256; value++)
7410 if (!isPUNCT(value))
7411 ANYOF_BITMAP_SET(ret, value);
7418 ANYOF_CLASS_SET(ret, ANYOF_SPACE);
7420 for (value = 0; value < 256; value++)
7422 ANYOF_BITMAP_SET(ret, value);
7429 ANYOF_CLASS_SET(ret, ANYOF_NSPACE);
7431 for (value = 0; value < 256; value++)
7432 if (!isSPACE(value))
7433 ANYOF_BITMAP_SET(ret, value);
7440 ANYOF_CLASS_SET(ret, ANYOF_UPPER);
7442 for (value = 0; value < 256; value++)
7444 ANYOF_BITMAP_SET(ret, value);
7451 ANYOF_CLASS_SET(ret, ANYOF_NUPPER);
7453 for (value = 0; value < 256; value++)
7454 if (!isUPPER(value))
7455 ANYOF_BITMAP_SET(ret, value);
7462 ANYOF_CLASS_SET(ret, ANYOF_XDIGIT);
7464 for (value = 0; value < 256; value++)
7465 if (isXDIGIT(value))
7466 ANYOF_BITMAP_SET(ret, value);
7473 ANYOF_CLASS_SET(ret, ANYOF_NXDIGIT);
7475 for (value = 0; value < 256; value++)
7476 if (!isXDIGIT(value))
7477 ANYOF_BITMAP_SET(ret, value);
7483 /* this is to handle \p and \P */
7486 vFAIL("Invalid [::] class");
7490 /* Strings such as "+utf8::isWord\n" */
7491 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7494 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7497 } /* end of namedclass \blah */
7500 if (prevvalue > (IV)value) /* b-a */ {
7501 const int w = RExC_parse - rangebegin;
7502 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7503 range = 0; /* not a valid range */
7507 prevvalue = value; /* save the beginning of the range */
7508 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7509 RExC_parse[1] != ']') {
7512 /* a bad range like \w-, [:word:]- ? */
7513 if (namedclass > OOB_NAMEDCLASS) {
7514 if (ckWARN(WARN_REGEXP)) {
7516 RExC_parse >= rangebegin ?
7517 RExC_parse - rangebegin : 0;
7519 "False [] range \"%*.*s\"",
7523 ANYOF_BITMAP_SET(ret, '-');
7525 range = 1; /* yeah, it's a range! */
7526 continue; /* but do it the next time */
7530 /* now is the next time */
7531 /*stored += (value - prevvalue + 1);*/
7533 if (prevvalue < 256) {
7534 const IV ceilvalue = value < 256 ? value : 255;
7537 /* In EBCDIC [\x89-\x91] should include
7538 * the \x8e but [i-j] should not. */
7539 if (literal_endpoint == 2 &&
7540 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7541 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7543 if (isLOWER(prevvalue)) {
7544 for (i = prevvalue; i <= ceilvalue; i++)
7546 ANYOF_BITMAP_SET(ret, i);
7548 for (i = prevvalue; i <= ceilvalue; i++)
7550 ANYOF_BITMAP_SET(ret, i);
7555 for (i = prevvalue; i <= ceilvalue; i++) {
7556 if (!ANYOF_BITMAP_TEST(ret,i)) {
7558 ANYOF_BITMAP_SET(ret, i);
7562 if (value > 255 || UTF) {
7563 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7564 const UV natvalue = NATIVE_TO_UNI(value);
7565 stored+=2; /* can't optimize this class */
7566 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7567 if (prevnatvalue < natvalue) { /* what about > ? */
7568 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7569 prevnatvalue, natvalue);
7571 else if (prevnatvalue == natvalue) {
7572 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7574 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7576 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7578 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7579 if (RExC_precomp[0] == ':' &&
7580 RExC_precomp[1] == '[' &&
7581 (f == 0xDF || f == 0x92)) {
7582 f = NATIVE_TO_UNI(f);
7585 /* If folding and foldable and a single
7586 * character, insert also the folded version
7587 * to the charclass. */
7589 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7590 if ((RExC_precomp[0] == ':' &&
7591 RExC_precomp[1] == '[' &&
7593 (value == 0xFB05 || value == 0xFB06))) ?
7594 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7595 foldlen == (STRLEN)UNISKIP(f) )
7597 if (foldlen == (STRLEN)UNISKIP(f))
7599 Perl_sv_catpvf(aTHX_ listsv,
7602 /* Any multicharacter foldings
7603 * require the following transform:
7604 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7605 * where E folds into "pq" and F folds
7606 * into "rst", all other characters
7607 * fold to single characters. We save
7608 * away these multicharacter foldings,
7609 * to be later saved as part of the
7610 * additional "s" data. */
7613 if (!unicode_alternate)
7614 unicode_alternate = newAV();
7615 sv = newSVpvn((char*)foldbuf, foldlen);
7617 av_push(unicode_alternate, sv);
7621 /* If folding and the value is one of the Greek
7622 * sigmas insert a few more sigmas to make the
7623 * folding rules of the sigmas to work right.
7624 * Note that not all the possible combinations
7625 * are handled here: some of them are handled
7626 * by the standard folding rules, and some of
7627 * them (literal or EXACTF cases) are handled
7628 * during runtime in regexec.c:S_find_byclass(). */
7629 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7630 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7631 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7632 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7633 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7635 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7636 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7637 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7642 literal_endpoint = 0;
7646 range = 0; /* this range (if it was one) is done now */
7650 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7652 RExC_size += ANYOF_CLASS_ADD_SKIP;
7654 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7660 /****** !SIZE_ONLY AFTER HERE *********/
7662 if( stored == 1 && value < 256
7663 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7665 /* optimize single char class to an EXACT node
7666 but *only* when its not a UTF/high char */
7667 const char * cur_parse= RExC_parse;
7668 RExC_emit = (regnode *)orig_emit;
7669 RExC_parse = (char *)orig_parse;
7670 ret = reg_node(pRExC_state,
7671 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7672 RExC_parse = (char *)cur_parse;
7673 *STRING(ret)= (char)value;
7675 RExC_emit += STR_SZ(1);
7678 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7679 if ( /* If the only flag is folding (plus possibly inversion). */
7680 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7682 for (value = 0; value < 256; ++value) {
7683 if (ANYOF_BITMAP_TEST(ret, value)) {
7684 UV fold = PL_fold[value];
7687 ANYOF_BITMAP_SET(ret, fold);
7690 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7693 /* optimize inverted simple patterns (e.g. [^a-z]) */
7694 if (optimize_invert &&
7695 /* If the only flag is inversion. */
7696 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7697 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7698 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7699 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7702 AV * const av = newAV();
7704 /* The 0th element stores the character class description
7705 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7706 * to initialize the appropriate swash (which gets stored in
7707 * the 1st element), and also useful for dumping the regnode.
7708 * The 2nd element stores the multicharacter foldings,
7709 * used later (regexec.c:S_reginclass()). */
7710 av_store(av, 0, listsv);
7711 av_store(av, 1, NULL);
7712 av_store(av, 2, (SV*)unicode_alternate);
7713 rv = newRV_noinc((SV*)av);
7714 n = add_data(pRExC_state, 1, "s");
7715 RExC_rxi->data->data[n] = (void*)rv;
7722 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7724 char* const retval = RExC_parse++;
7727 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7728 RExC_parse[2] == '#') {
7729 while (*RExC_parse != ')') {
7730 if (RExC_parse == RExC_end)
7731 FAIL("Sequence (?#... not terminated");
7737 if (RExC_flags & RXf_PMf_EXTENDED) {
7738 if (isSPACE(*RExC_parse)) {
7742 else if (*RExC_parse == '#') {
7743 while (RExC_parse < RExC_end)
7744 if (*RExC_parse++ == '\n') break;
7753 - reg_node - emit a node
7755 STATIC regnode * /* Location. */
7756 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7759 register regnode *ptr;
7760 regnode * const ret = RExC_emit;
7761 GET_RE_DEBUG_FLAGS_DECL;
7764 SIZE_ALIGN(RExC_size);
7769 if (OP(RExC_emit) == 255)
7770 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %s: %d ",
7771 reg_name[op], OP(RExC_emit));
7773 NODE_ALIGN_FILL(ret);
7775 FILL_ADVANCE_NODE(ptr, op);
7776 if (RExC_offsets) { /* MJD */
7777 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7778 "reg_node", __LINE__,
7780 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7781 ? "Overwriting end of array!\n" : "OK",
7782 (UV)(RExC_emit - RExC_emit_start),
7783 (UV)(RExC_parse - RExC_start),
7784 (UV)RExC_offsets[0]));
7785 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7793 - reganode - emit a node with an argument
7795 STATIC regnode * /* Location. */
7796 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7799 register regnode *ptr;
7800 regnode * const ret = RExC_emit;
7801 GET_RE_DEBUG_FLAGS_DECL;
7804 SIZE_ALIGN(RExC_size);
7809 assert(2==regarglen[op]+1);
7811 Anything larger than this has to allocate the extra amount.
7812 If we changed this to be:
7814 RExC_size += (1 + regarglen[op]);
7816 then it wouldn't matter. Its not clear what side effect
7817 might come from that so its not done so far.
7823 if (OP(RExC_emit) == 255)
7824 Perl_croak(aTHX_ "panic: reganode overwriting end of allocated program space");
7826 NODE_ALIGN_FILL(ret);
7828 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7829 if (RExC_offsets) { /* MJD */
7830 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7834 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7835 "Overwriting end of array!\n" : "OK",
7836 (UV)(RExC_emit - RExC_emit_start),
7837 (UV)(RExC_parse - RExC_start),
7838 (UV)RExC_offsets[0]));
7839 Set_Cur_Node_Offset;
7847 - reguni - emit (if appropriate) a Unicode character
7850 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
7853 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
7857 - reginsert - insert an operator in front of already-emitted operand
7859 * Means relocating the operand.
7862 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
7865 register regnode *src;
7866 register regnode *dst;
7867 register regnode *place;
7868 const int offset = regarglen[(U8)op];
7869 const int size = NODE_STEP_REGNODE + offset;
7870 GET_RE_DEBUG_FLAGS_DECL;
7871 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
7872 DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
7881 if (RExC_open_parens) {
7883 DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);
7884 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
7885 if ( RExC_open_parens[paren] >= opnd ) {
7886 DEBUG_PARSE_FMT("open"," - %d",size);
7887 RExC_open_parens[paren] += size;
7889 DEBUG_PARSE_FMT("open"," - %s","ok");
7891 if ( RExC_close_parens[paren] >= opnd ) {
7892 DEBUG_PARSE_FMT("close"," - %d",size);
7893 RExC_close_parens[paren] += size;
7895 DEBUG_PARSE_FMT("close"," - %s","ok");
7900 while (src > opnd) {
7901 StructCopy(--src, --dst, regnode);
7902 if (RExC_offsets) { /* MJD 20010112 */
7903 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
7907 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
7908 ? "Overwriting end of array!\n" : "OK",
7909 (UV)(src - RExC_emit_start),
7910 (UV)(dst - RExC_emit_start),
7911 (UV)RExC_offsets[0]));
7912 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
7913 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
7918 place = opnd; /* Op node, where operand used to be. */
7919 if (RExC_offsets) { /* MJD */
7920 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7924 (UV)(place - RExC_emit_start) > RExC_offsets[0]
7925 ? "Overwriting end of array!\n" : "OK",
7926 (UV)(place - RExC_emit_start),
7927 (UV)(RExC_parse - RExC_start),
7928 (UV)RExC_offsets[0]));
7929 Set_Node_Offset(place, RExC_parse);
7930 Set_Node_Length(place, 1);
7932 src = NEXTOPER(place);
7933 FILL_ADVANCE_NODE(place, op);
7934 Zero(src, offset, regnode);
7938 - regtail - set the next-pointer at the end of a node chain of p to val.
7939 - SEE ALSO: regtail_study
7941 /* TODO: All three parms should be const */
7943 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
7946 register regnode *scan;
7947 GET_RE_DEBUG_FLAGS_DECL;
7949 PERL_UNUSED_ARG(depth);
7955 /* Find last node. */
7958 regnode * const temp = regnext(scan);
7960 SV * const mysv=sv_newmortal();
7961 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
7962 regprop(RExC_rx, mysv, scan);
7963 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
7964 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
7965 (temp == NULL ? "->" : ""),
7966 (temp == NULL ? reg_name[OP(val)] : "")
7974 if (reg_off_by_arg[OP(scan)]) {
7975 ARG_SET(scan, val - scan);
7978 NEXT_OFF(scan) = val - scan;
7984 - regtail_study - set the next-pointer at the end of a node chain of p to val.
7985 - Look for optimizable sequences at the same time.
7986 - currently only looks for EXACT chains.
7988 This is expermental code. The idea is to use this routine to perform
7989 in place optimizations on branches and groups as they are constructed,
7990 with the long term intention of removing optimization from study_chunk so
7991 that it is purely analytical.
7993 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
7994 to control which is which.
7997 /* TODO: All four parms should be const */
8000 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8003 register regnode *scan;
8005 #ifdef EXPERIMENTAL_INPLACESCAN
8009 GET_RE_DEBUG_FLAGS_DECL;
8015 /* Find last node. */
8019 regnode * const temp = regnext(scan);
8020 #ifdef EXPERIMENTAL_INPLACESCAN
8021 if (PL_regkind[OP(scan)] == EXACT)
8022 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8030 if( exact == PSEUDO )
8032 else if ( exact != OP(scan) )
8041 SV * const mysv=sv_newmortal();
8042 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8043 regprop(RExC_rx, mysv, scan);
8044 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8045 SvPV_nolen_const(mysv),
8054 SV * const mysv_val=sv_newmortal();
8055 DEBUG_PARSE_MSG("");
8056 regprop(RExC_rx, mysv_val, val);
8057 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8058 SvPV_nolen_const(mysv_val),
8059 (IV)REG_NODE_NUM(val),
8063 if (reg_off_by_arg[OP(scan)]) {
8064 ARG_SET(scan, val - scan);
8067 NEXT_OFF(scan) = val - scan;
8075 - regcurly - a little FSA that accepts {\d+,?\d*}
8078 S_regcurly(register const char *s)
8097 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8100 Perl_regdump(pTHX_ const regexp *r)
8104 SV * const sv = sv_newmortal();
8105 SV *dsv= sv_newmortal();
8108 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8110 /* Header fields of interest. */
8111 if (r->anchored_substr) {
8112 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8113 RE_SV_DUMPLEN(r->anchored_substr), 30);
8114 PerlIO_printf(Perl_debug_log,
8115 "anchored %s%s at %"IVdf" ",
8116 s, RE_SV_TAIL(r->anchored_substr),
8117 (IV)r->anchored_offset);
8118 } else if (r->anchored_utf8) {
8119 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8120 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8121 PerlIO_printf(Perl_debug_log,
8122 "anchored utf8 %s%s at %"IVdf" ",
8123 s, RE_SV_TAIL(r->anchored_utf8),
8124 (IV)r->anchored_offset);
8126 if (r->float_substr) {
8127 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8128 RE_SV_DUMPLEN(r->float_substr), 30);
8129 PerlIO_printf(Perl_debug_log,
8130 "floating %s%s at %"IVdf"..%"UVuf" ",
8131 s, RE_SV_TAIL(r->float_substr),
8132 (IV)r->float_min_offset, (UV)r->float_max_offset);
8133 } else if (r->float_utf8) {
8134 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8135 RE_SV_DUMPLEN(r->float_utf8), 30);
8136 PerlIO_printf(Perl_debug_log,
8137 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8138 s, RE_SV_TAIL(r->float_utf8),
8139 (IV)r->float_min_offset, (UV)r->float_max_offset);
8141 if (r->check_substr || r->check_utf8)
8142 PerlIO_printf(Perl_debug_log,
8144 (r->check_substr == r->float_substr
8145 && r->check_utf8 == r->float_utf8
8146 ? "(checking floating" : "(checking anchored"));
8147 if (r->extflags & RXf_NOSCAN)
8148 PerlIO_printf(Perl_debug_log, " noscan");
8149 if (r->extflags & RXf_CHECK_ALL)
8150 PerlIO_printf(Perl_debug_log, " isall");
8151 if (r->check_substr || r->check_utf8)
8152 PerlIO_printf(Perl_debug_log, ") ");
8154 if (ri->regstclass) {
8155 regprop(r, sv, ri->regstclass);
8156 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8158 if (r->extflags & RXf_ANCH) {
8159 PerlIO_printf(Perl_debug_log, "anchored");
8160 if (r->extflags & RXf_ANCH_BOL)
8161 PerlIO_printf(Perl_debug_log, "(BOL)");
8162 if (r->extflags & RXf_ANCH_MBOL)
8163 PerlIO_printf(Perl_debug_log, "(MBOL)");
8164 if (r->extflags & RXf_ANCH_SBOL)
8165 PerlIO_printf(Perl_debug_log, "(SBOL)");
8166 if (r->extflags & RXf_ANCH_GPOS)
8167 PerlIO_printf(Perl_debug_log, "(GPOS)");
8168 PerlIO_putc(Perl_debug_log, ' ');
8170 if (r->extflags & RXf_GPOS_SEEN)
8171 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8172 if (r->intflags & PREGf_SKIP)
8173 PerlIO_printf(Perl_debug_log, "plus ");
8174 if (r->intflags & PREGf_IMPLICIT)
8175 PerlIO_printf(Perl_debug_log, "implicit ");
8176 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8177 if (r->extflags & RXf_EVAL_SEEN)
8178 PerlIO_printf(Perl_debug_log, "with eval ");
8179 PerlIO_printf(Perl_debug_log, "\n");
8181 PERL_UNUSED_CONTEXT;
8183 #endif /* DEBUGGING */
8187 - regprop - printable representation of opcode
8190 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8195 RXi_GET_DECL(prog,progi);
8196 GET_RE_DEBUG_FLAGS_DECL;
8199 sv_setpvn(sv, "", 0);
8201 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8202 /* It would be nice to FAIL() here, but this may be called from
8203 regexec.c, and it would be hard to supply pRExC_state. */
8204 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8205 sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
8207 k = PL_regkind[OP(o)];
8210 SV * const dsv = sv_2mortal(newSVpvs(""));
8211 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8212 * is a crude hack but it may be the best for now since
8213 * we have no flag "this EXACTish node was UTF-8"
8215 const char * const s =
8216 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8217 PL_colors[0], PL_colors[1],
8218 PERL_PV_ESCAPE_UNI_DETECT |
8219 PERL_PV_PRETTY_ELIPSES |
8222 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8223 } else if (k == TRIE) {
8224 /* print the details of the trie in dumpuntil instead, as
8225 * progi->data isn't available here */
8226 const char op = OP(o);
8227 const I32 n = ARG(o);
8228 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8229 (reg_ac_data *)progi->data->data[n] :
8231 const reg_trie_data * const trie
8232 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8234 Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
8235 DEBUG_TRIE_COMPILE_r(
8236 Perl_sv_catpvf(aTHX_ sv,
8237 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8238 (UV)trie->startstate,
8239 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8240 (UV)trie->wordcount,
8243 (UV)TRIE_CHARCOUNT(trie),
8244 (UV)trie->uniquecharcount
8247 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8249 int rangestart = -1;
8250 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8251 Perl_sv_catpvf(aTHX_ sv, "[");
8252 for (i = 0; i <= 256; i++) {
8253 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8254 if (rangestart == -1)
8256 } else if (rangestart != -1) {
8257 if (i <= rangestart + 3)
8258 for (; rangestart < i; rangestart++)
8259 put_byte(sv, rangestart);
8261 put_byte(sv, rangestart);
8263 put_byte(sv, i - 1);
8268 Perl_sv_catpvf(aTHX_ sv, "]");
8271 } else if (k == CURLY) {
8272 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8273 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8274 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8276 else if (k == WHILEM && o->flags) /* Ordinal/of */
8277 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8278 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT)
8279 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8280 else if (k == GOSUB)
8281 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8282 else if (k == VERB) {
8284 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8285 (SV*)progi->data->data[ ARG( o ) ]);
8286 } else if (k == LOGICAL)
8287 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8288 else if (k == ANYOF) {
8289 int i, rangestart = -1;
8290 const U8 flags = ANYOF_FLAGS(o);
8292 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8293 static const char * const anyofs[] = {
8326 if (flags & ANYOF_LOCALE)
8327 sv_catpvs(sv, "{loc}");
8328 if (flags & ANYOF_FOLD)
8329 sv_catpvs(sv, "{i}");
8330 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8331 if (flags & ANYOF_INVERT)
8333 for (i = 0; i <= 256; i++) {
8334 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8335 if (rangestart == -1)
8337 } else if (rangestart != -1) {
8338 if (i <= rangestart + 3)
8339 for (; rangestart < i; rangestart++)
8340 put_byte(sv, rangestart);
8342 put_byte(sv, rangestart);
8344 put_byte(sv, i - 1);
8350 if (o->flags & ANYOF_CLASS)
8351 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8352 if (ANYOF_CLASS_TEST(o,i))
8353 sv_catpv(sv, anyofs[i]);
8355 if (flags & ANYOF_UNICODE)
8356 sv_catpvs(sv, "{unicode}");
8357 else if (flags & ANYOF_UNICODE_ALL)
8358 sv_catpvs(sv, "{unicode_all}");
8362 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8366 U8 s[UTF8_MAXBYTES_CASE+1];
8368 for (i = 0; i <= 256; i++) { /* just the first 256 */
8369 uvchr_to_utf8(s, i);
8371 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8372 if (rangestart == -1)
8374 } else if (rangestart != -1) {
8375 if (i <= rangestart + 3)
8376 for (; rangestart < i; rangestart++) {
8377 const U8 * const e = uvchr_to_utf8(s,rangestart);
8379 for(p = s; p < e; p++)
8383 const U8 *e = uvchr_to_utf8(s,rangestart);
8385 for (p = s; p < e; p++)
8388 e = uvchr_to_utf8(s, i-1);
8389 for (p = s; p < e; p++)
8396 sv_catpvs(sv, "..."); /* et cetera */
8400 char *s = savesvpv(lv);
8401 char * const origs = s;
8403 while (*s && *s != '\n')
8407 const char * const t = ++s;
8425 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8427 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8428 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8430 PERL_UNUSED_CONTEXT;
8431 PERL_UNUSED_ARG(sv);
8433 PERL_UNUSED_ARG(prog);
8434 #endif /* DEBUGGING */
8438 Perl_re_intuit_string(pTHX_ regexp *prog)
8439 { /* Assume that RE_INTUIT is set */
8441 GET_RE_DEBUG_FLAGS_DECL;
8442 PERL_UNUSED_CONTEXT;
8446 const char * const s = SvPV_nolen_const(prog->check_substr
8447 ? prog->check_substr : prog->check_utf8);
8449 if (!PL_colorset) reginitcolors();
8450 PerlIO_printf(Perl_debug_log,
8451 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8453 prog->check_substr ? "" : "utf8 ",
8454 PL_colors[5],PL_colors[0],
8457 (strlen(s) > 60 ? "..." : ""));
8460 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8464 pregfree - free a regexp
8466 See regdupe below if you change anything here.
8470 Perl_pregfree(pTHX_ struct regexp *r)
8474 GET_RE_DEBUG_FLAGS_DECL;
8476 if (!r || (--r->refcnt > 0))
8482 SV *dsv= sv_newmortal();
8483 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8484 dsv, r->precomp, r->prelen, 60);
8485 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8486 PL_colors[4],PL_colors[5],s);
8490 /* gcov results gave these as non-null 100% of the time, so there's no
8491 optimisation in checking them before calling Safefree */
8492 Safefree(r->precomp);
8493 Safefree(ri->offsets); /* 20010421 MJD */
8494 RX_MATCH_COPY_FREE(r);
8495 #ifdef PERL_OLD_COPY_ON_WRITE
8497 SvREFCNT_dec(r->saved_copy);
8500 if (r->anchored_substr)
8501 SvREFCNT_dec(r->anchored_substr);
8502 if (r->anchored_utf8)
8503 SvREFCNT_dec(r->anchored_utf8);
8504 if (r->float_substr)
8505 SvREFCNT_dec(r->float_substr);
8507 SvREFCNT_dec(r->float_utf8);
8508 Safefree(r->substrs);
8511 SvREFCNT_dec(r->paren_names);
8513 int n = ri->data->count;
8514 PAD* new_comppad = NULL;
8519 /* If you add a ->what type here, update the comment in regcomp.h */
8520 switch (ri->data->what[n]) {
8524 SvREFCNT_dec((SV*)ri->data->data[n]);
8527 Safefree(ri->data->data[n]);
8530 new_comppad = (AV*)ri->data->data[n];
8533 if (new_comppad == NULL)
8534 Perl_croak(aTHX_ "panic: pregfree comppad");
8535 PAD_SAVE_LOCAL(old_comppad,
8536 /* Watch out for global destruction's random ordering. */
8537 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8540 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8543 op_free((OP_4tree*)ri->data->data[n]);
8545 PAD_RESTORE_LOCAL(old_comppad);
8546 SvREFCNT_dec((SV*)new_comppad);
8552 { /* Aho Corasick add-on structure for a trie node.
8553 Used in stclass optimization only */
8555 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8557 refcount = --aho->refcount;
8560 PerlMemShared_free(aho->states);
8561 PerlMemShared_free(aho->fail);
8562 /* do this last!!!! */
8563 PerlMemShared_free(ri->data->data[n]);
8564 PerlMemShared_free(ri->regstclass);
8570 /* trie structure. */
8572 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8574 refcount = --trie->refcount;
8577 PerlMemShared_free(trie->charmap);
8578 PerlMemShared_free(trie->states);
8579 PerlMemShared_free(trie->trans);
8581 PerlMemShared_free(trie->bitmap);
8583 PerlMemShared_free(trie->wordlen);
8585 PerlMemShared_free(trie->jump);
8587 PerlMemShared_free(trie->nextword);
8588 /* do this last!!!! */
8589 PerlMemShared_free(ri->data->data[n]);
8594 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8597 Safefree(ri->data->what);
8600 Safefree(r->startp);
8603 Safefree(ri->swap->startp);
8604 Safefree(ri->swap->endp);
8611 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8612 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8613 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8614 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8617 regdupe - duplicate a regexp.
8619 This routine is called by sv.c's re_dup and is expected to clone a
8620 given regexp structure. It is a no-op when not under USE_ITHREADS.
8621 (Originally this *was* re_dup() for change history see sv.c)
8623 See pregfree() above if you change anything here.
8625 #if defined(USE_ITHREADS)
8627 Perl_regdupe(pTHX_ const regexp *r, CLONE_PARAMS *param)
8631 regexp_internal *reti;
8633 struct reg_substr_datum *s;
8637 return (REGEXP *)NULL;
8639 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8642 len = ri->offsets[0];
8643 npar = r->nparens+1;
8645 Newxz(ret, 1, regexp);
8646 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8648 Copy(ri->program, reti->program, len+1, regnode);
8650 Newx(ret->startp, npar, I32);
8651 Copy(r->startp, ret->startp, npar, I32);
8652 Newx(ret->endp, npar, I32);
8653 Copy(r->startp, ret->startp, npar, I32);
8655 Newx(reti->swap, 1, regexp_paren_ofs);
8656 /* no need to copy these */
8657 Newx(reti->swap->startp, npar, I32);
8658 Newx(reti->swap->endp, npar, I32);
8663 Newx(ret->substrs, 1, struct reg_substr_data);
8664 for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
8665 s->min_offset = r->substrs->data[i].min_offset;
8666 s->max_offset = r->substrs->data[i].max_offset;
8667 s->end_shift = r->substrs->data[i].end_shift;
8668 s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
8669 s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
8672 reti->regstclass = NULL;
8675 const int count = ri->data->count;
8678 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
8679 char, struct reg_data);
8680 Newx(d->what, count, U8);
8683 for (i = 0; i < count; i++) {
8684 d->what[i] = ri->data->what[i];
8685 switch (d->what[i]) {
8686 /* legal options are one of: sSfpontTu
8687 see also regcomp.h and pregfree() */
8690 case 'p': /* actually an AV, but the dup function is identical. */
8691 case 'u': /* actually an HV, but the dup function is identical. */
8692 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
8695 /* This is cheating. */
8696 Newx(d->data[i], 1, struct regnode_charclass_class);
8697 StructCopy(ri->data->data[i], d->data[i],
8698 struct regnode_charclass_class);
8699 reti->regstclass = (regnode*)d->data[i];
8702 /* Compiled op trees are readonly and in shared memory,
8703 and can thus be shared without duplication. */
8705 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
8709 /* Trie stclasses are readonly and can thus be shared
8710 * without duplication. We free the stclass in pregfree
8711 * when the corresponding reg_ac_data struct is freed.
8713 reti->regstclass= ri->regstclass;
8717 ((reg_trie_data*)ri->data->data[i])->refcount++;
8721 d->data[i] = ri->data->data[i];
8724 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
8733 Newx(reti->offsets, 2*len+1, U32);
8734 Copy(ri->offsets, reti->offsets, 2*len+1, U32);
8736 ret->precomp = SAVEPVN(r->precomp, r->prelen);
8737 ret->refcnt = r->refcnt;
8738 ret->minlen = r->minlen;
8739 ret->minlenret = r->minlenret;
8740 ret->prelen = r->prelen;
8741 ret->nparens = r->nparens;
8742 ret->lastparen = r->lastparen;
8743 ret->lastcloseparen = r->lastcloseparen;
8744 ret->intflags = r->intflags;
8745 ret->extflags = r->extflags;
8747 ret->sublen = r->sublen;
8749 ret->engine = r->engine;
8751 ret->paren_names = hv_dup_inc(r->paren_names, param);
8753 if (RX_MATCH_COPIED(ret))
8754 ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
8757 #ifdef PERL_OLD_COPY_ON_WRITE
8758 ret->saved_copy = NULL;
8761 ptr_table_store(PL_ptr_table, r, ret);
8769 converts a regexp embedded in a MAGIC struct to its stringified form,
8770 caching the converted form in the struct and returns the cached
8773 If lp is nonnull then it is used to return the length of the
8776 If flags is nonnull and the returned string contains UTF8 then
8777 (flags & 1) will be true.
8779 If haseval is nonnull then it is used to return whether the pattern
8782 Normally called via macro:
8784 CALLREG_STRINGIFY(mg,0,0);
8788 CALLREG_AS_STR(mg,lp,flags,haseval)
8790 See sv_2pv_flags() in sv.c for an example of internal usage.
8795 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
8797 const regexp * const re = (regexp *)mg->mg_obj;
8798 RXi_GET_DECL(re,ri);
8801 const char *fptr = "msix";
8806 bool need_newline = 0;
8807 U16 reganch = (U16)((re->extflags & RXf_PMf_COMPILETIME) >> 12);
8809 while((ch = *fptr++)) {
8811 reflags[left++] = ch;
8814 reflags[right--] = ch;
8819 reflags[left] = '-';
8823 mg->mg_len = re->prelen + 4 + left;
8825 * If /x was used, we have to worry about a regex ending with a
8826 * comment later being embedded within another regex. If so, we don't
8827 * want this regex's "commentization" to leak out to the right part of
8828 * the enclosing regex, we must cap it with a newline.
8830 * So, if /x was used, we scan backwards from the end of the regex. If
8831 * we find a '#' before we find a newline, we need to add a newline
8832 * ourself. If we find a '\n' first (or if we don't find '#' or '\n'),
8833 * we don't need to add anything. -jfriedl
8835 if (PMf_EXTENDED & re->extflags) {
8836 const char *endptr = re->precomp + re->prelen;
8837 while (endptr >= re->precomp) {
8838 const char c = *(endptr--);
8840 break; /* don't need another */
8842 /* we end while in a comment, so we need a newline */
8843 mg->mg_len++; /* save space for it */
8844 need_newline = 1; /* note to add it */
8850 Newx(mg->mg_ptr, mg->mg_len + 1 + left, char);
8851 mg->mg_ptr[0] = '(';
8852 mg->mg_ptr[1] = '?';
8853 Copy(reflags, mg->mg_ptr+2, left, char);
8854 *(mg->mg_ptr+left+2) = ':';
8855 Copy(re->precomp, mg->mg_ptr+3+left, re->prelen, char);
8857 mg->mg_ptr[mg->mg_len - 2] = '\n';
8858 mg->mg_ptr[mg->mg_len - 1] = ')';
8859 mg->mg_ptr[mg->mg_len] = 0;
8862 *haseval = ri->program[0].next_off;
8864 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
8872 #ifndef PERL_IN_XSUB_RE
8874 - regnext - dig the "next" pointer out of a node
8877 Perl_regnext(pTHX_ register regnode *p)
8880 register I32 offset;
8885 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
8894 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
8897 STRLEN l1 = strlen(pat1);
8898 STRLEN l2 = strlen(pat2);
8901 const char *message;
8907 Copy(pat1, buf, l1 , char);
8908 Copy(pat2, buf + l1, l2 , char);
8909 buf[l1 + l2] = '\n';
8910 buf[l1 + l2 + 1] = '\0';
8912 /* ANSI variant takes additional second argument */
8913 va_start(args, pat2);
8917 msv = vmess(buf, &args);
8919 message = SvPV_const(msv,l1);
8922 Copy(message, buf, l1 , char);
8923 buf[l1-1] = '\0'; /* Overwrite \n */
8924 Perl_croak(aTHX_ "%s", buf);
8927 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
8929 #ifndef PERL_IN_XSUB_RE
8931 Perl_save_re_context(pTHX)
8935 struct re_save_state *state;
8937 SAVEVPTR(PL_curcop);
8938 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
8940 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
8941 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
8942 SSPUSHINT(SAVEt_RE_STATE);
8944 Copy(&PL_reg_state, state, 1, struct re_save_state);
8946 PL_reg_start_tmp = 0;
8947 PL_reg_start_tmpl = 0;
8948 PL_reg_oldsaved = NULL;
8949 PL_reg_oldsavedlen = 0;
8951 PL_reg_leftiter = 0;
8952 PL_reg_poscache = NULL;
8953 PL_reg_poscache_size = 0;
8954 #ifdef PERL_OLD_COPY_ON_WRITE
8958 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
8960 const REGEXP * const rx = PM_GETRE(PL_curpm);
8963 for (i = 1; i <= rx->nparens; i++) {
8964 char digits[TYPE_CHARS(long)];
8965 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
8966 GV *const *const gvp
8967 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
8970 GV * const gv = *gvp;
8971 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
8981 clear_re(pTHX_ void *r)
8984 ReREFCNT_dec((regexp *)r);
8990 S_put_byte(pTHX_ SV *sv, int c)
8992 if (isCNTRL(c) || c == 255 || !isPRINT(c))
8993 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
8994 else if (c == '-' || c == ']' || c == '\\' || c == '^')
8995 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
8997 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9001 #define CLEAR_OPTSTART \
9002 if (optstart) STMT_START { \
9003 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9007 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9009 STATIC const regnode *
9010 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9011 const regnode *last, const regnode *plast,
9012 SV* sv, I32 indent, U32 depth)
9015 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9016 register const regnode *next;
9017 const regnode *optstart= NULL;
9019 GET_RE_DEBUG_FLAGS_DECL;
9021 #ifdef DEBUG_DUMPUNTIL
9022 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9023 last ? last-start : 0,plast ? plast-start : 0);
9026 if (plast && plast < last)
9029 while (PL_regkind[op] != END && (!last || node < last)) {
9030 /* While that wasn't END last time... */
9034 if (op == CLOSE || op == WHILEM)
9036 next = regnext((regnode *)node);
9039 if (OP(node) == OPTIMIZED) {
9040 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9047 regprop(r, sv, node);
9048 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9049 (int)(2*indent + 1), "", SvPVX_const(sv));
9051 if (OP(node) != OPTIMIZED) {
9052 if (next == NULL) /* Next ptr. */
9053 PerlIO_printf(Perl_debug_log, "(0)");
9054 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9055 PerlIO_printf(Perl_debug_log, "(FAIL)");
9057 PerlIO_printf(Perl_debug_log, "(%"IVdf")", (IV)(next - start));
9059 /*if (PL_regkind[(U8)op] != TRIE)*/
9060 (void)PerlIO_putc(Perl_debug_log, '\n');
9064 if (PL_regkind[(U8)op] == BRANCHJ) {
9067 register const regnode *nnode = (OP(next) == LONGJMP
9068 ? regnext((regnode *)next)
9070 if (last && nnode > last)
9072 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9075 else if (PL_regkind[(U8)op] == BRANCH) {
9077 DUMPUNTIL(NEXTOPER(node), next);
9079 else if ( PL_regkind[(U8)op] == TRIE ) {
9080 const regnode *this_trie = node;
9081 const char op = OP(node);
9082 const I32 n = ARG(node);
9083 const reg_ac_data * const ac = op>=AHOCORASICK ?
9084 (reg_ac_data *)ri->data->data[n] :
9086 const reg_trie_data * const trie =
9087 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9089 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9091 const regnode *nextbranch= NULL;
9093 sv_setpvn(sv, "", 0);
9094 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9095 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9097 PerlIO_printf(Perl_debug_log, "%*s%s ",
9098 (int)(2*(indent+3)), "",
9099 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9100 PL_colors[0], PL_colors[1],
9101 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9102 PERL_PV_PRETTY_ELIPSES |
9108 U16 dist= trie->jump[word_idx+1];
9109 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9110 (UV)((dist ? this_trie + dist : next) - start));
9113 nextbranch= this_trie + trie->jump[0];
9114 DUMPUNTIL(this_trie + dist, nextbranch);
9116 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9117 nextbranch= regnext((regnode *)nextbranch);
9119 PerlIO_printf(Perl_debug_log, "\n");
9122 if (last && next > last)
9127 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9128 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9129 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9131 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9133 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9135 else if ( op == PLUS || op == STAR) {
9136 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9138 else if (op == ANYOF) {
9139 /* arglen 1 + class block */
9140 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9141 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9142 node = NEXTOPER(node);
9144 else if (PL_regkind[(U8)op] == EXACT) {
9145 /* Literal string, where present. */
9146 node += NODE_SZ_STR(node) - 1;
9147 node = NEXTOPER(node);
9150 node = NEXTOPER(node);
9151 node += regarglen[(U8)op];
9153 if (op == CURLYX || op == OPEN)
9157 #ifdef DEBUG_DUMPUNTIL
9158 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9163 #endif /* DEBUGGING */
9167 * c-indentation-style: bsd
9169 * indent-tabs-mode: t
9172 * ex: set ts=8 sts=4 sw=4 noet: