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;</
https://perl5.git.perl.org / perl5.git / blob