5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
8 /* This file contains functions for compiling a regular expression. See
9 * also regexec.c which funnily enough, contains functions for executing
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_comp.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
57 **** Alterations to Henry's code are...
59 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
60 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
61 **** by Larry Wall and others
63 **** You may distribute under the terms of either the GNU General Public
64 **** License or the Artistic License, as specified in the README file.
67 * Beware that some of this code is subtly aware of the way operator
68 * precedence is structured in regular expressions. Serious changes in
69 * regular-expression syntax might require a total rethink.
72 #define PERL_IN_REGCOMP_C
75 #ifndef PERL_IN_XSUB_RE
80 #ifdef PERL_IN_XSUB_RE
91 # if defined(BUGGY_MSC6)
92 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
93 # pragma optimize("a",off)
94 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
95 # pragma optimize("w",on )
96 # endif /* BUGGY_MSC6 */
100 #define STATIC static
103 typedef struct RExC_state_t {
104 U32 flags; /* are we folding, multilining? */
105 char *precomp; /* uncompiled string. */
106 REGEXP *rx_sv; /* The SV that is the regexp. */
107 regexp *rx; /* perl core regexp structure */
108 regexp_internal *rxi; /* internal data for regexp object pprivate field */
109 char *start; /* Start of input for compile */
110 char *end; /* End of input for compile */
111 char *parse; /* Input-scan pointer. */
112 I32 whilem_seen; /* number of WHILEM in this expr */
113 regnode *emit_start; /* Start of emitted-code area */
114 regnode *emit_bound; /* First regnode outside of the allocated space */
115 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
116 I32 naughty; /* How bad is this pattern? */
117 I32 sawback; /* Did we see \1, ...? */
119 I32 size; /* Code size. */
120 I32 npar; /* Capture buffer count, (OPEN). */
121 I32 cpar; /* Capture buffer count, (CLOSE). */
122 I32 nestroot; /* root parens we are in - used by accept */
126 regnode **open_parens; /* pointers to open parens */
127 regnode **close_parens; /* pointers to close parens */
128 regnode *opend; /* END node in program */
129 I32 utf8; /* whether the pattern is utf8 or not */
130 I32 orig_utf8; /* whether the pattern was originally in utf8 */
131 /* XXX use this for future optimisation of case
132 * where pattern must be upgraded to utf8. */
133 HV *charnames; /* cache of named sequences */
134 HV *paren_names; /* Paren names */
136 regnode **recurse; /* Recurse regops */
137 I32 recurse_count; /* Number of recurse regops */
139 char *starttry; /* -Dr: where regtry was called. */
140 #define RExC_starttry (pRExC_state->starttry)
143 const char *lastparse;
145 AV *paren_name_list; /* idx -> name */
146 #define RExC_lastparse (pRExC_state->lastparse)
147 #define RExC_lastnum (pRExC_state->lastnum)
148 #define RExC_paren_name_list (pRExC_state->paren_name_list)
152 #define RExC_flags (pRExC_state->flags)
153 #define RExC_precomp (pRExC_state->precomp)
154 #define RExC_rx_sv (pRExC_state->rx_sv)
155 #define RExC_rx (pRExC_state->rx)
156 #define RExC_rxi (pRExC_state->rxi)
157 #define RExC_start (pRExC_state->start)
158 #define RExC_end (pRExC_state->end)
159 #define RExC_parse (pRExC_state->parse)
160 #define RExC_whilem_seen (pRExC_state->whilem_seen)
161 #ifdef RE_TRACK_PATTERN_OFFSETS
162 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
164 #define RExC_emit (pRExC_state->emit)
165 #define RExC_emit_start (pRExC_state->emit_start)
166 #define RExC_emit_bound (pRExC_state->emit_bound)
167 #define RExC_naughty (pRExC_state->naughty)
168 #define RExC_sawback (pRExC_state->sawback)
169 #define RExC_seen (pRExC_state->seen)
170 #define RExC_size (pRExC_state->size)
171 #define RExC_npar (pRExC_state->npar)
172 #define RExC_nestroot (pRExC_state->nestroot)
173 #define RExC_extralen (pRExC_state->extralen)
174 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
175 #define RExC_seen_evals (pRExC_state->seen_evals)
176 #define RExC_utf8 (pRExC_state->utf8)
177 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
178 #define RExC_charnames (pRExC_state->charnames)
179 #define RExC_open_parens (pRExC_state->open_parens)
180 #define RExC_close_parens (pRExC_state->close_parens)
181 #define RExC_opend (pRExC_state->opend)
182 #define RExC_paren_names (pRExC_state->paren_names)
183 #define RExC_recurse (pRExC_state->recurse)
184 #define RExC_recurse_count (pRExC_state->recurse_count)
187 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
188 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
189 ((*s) == '{' && regcurly(s)))
192 #undef SPSTART /* dratted cpp namespace... */
195 * Flags to be passed up and down.
197 #define WORST 0 /* Worst case. */
198 #define HASWIDTH 0x01 /* Known to match non-null strings. */
199 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
200 #define SPSTART 0x04 /* Starts with * or +. */
201 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
202 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
204 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
206 /* whether trie related optimizations are enabled */
207 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
208 #define TRIE_STUDY_OPT
209 #define FULL_TRIE_STUDY
215 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
216 #define PBITVAL(paren) (1 << ((paren) & 7))
217 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
218 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
219 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
222 /* About scan_data_t.
224 During optimisation we recurse through the regexp program performing
225 various inplace (keyhole style) optimisations. In addition study_chunk
226 and scan_commit populate this data structure with information about
227 what strings MUST appear in the pattern. We look for the longest
228 string that must appear for at a fixed location, and we look for the
229 longest string that may appear at a floating location. So for instance
234 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
235 strings (because they follow a .* construct). study_chunk will identify
236 both FOO and BAR as being the longest fixed and floating strings respectively.
238 The strings can be composites, for instance
242 will result in a composite fixed substring 'foo'.
244 For each string some basic information is maintained:
246 - offset or min_offset
247 This is the position the string must appear at, or not before.
248 It also implicitly (when combined with minlenp) tells us how many
249 character must match before the string we are searching.
250 Likewise when combined with minlenp and the length of the string
251 tells us how many characters must appear after the string we have
255 Only used for floating strings. This is the rightmost point that
256 the string can appear at. Ifset to I32 max it indicates that the
257 string can occur infinitely far to the right.
260 A pointer to the minimum length of the pattern that the string
261 was found inside. This is important as in the case of positive
262 lookahead or positive lookbehind we can have multiple patterns
267 The minimum length of the pattern overall is 3, the minimum length
268 of the lookahead part is 3, but the minimum length of the part that
269 will actually match is 1. So 'FOO's minimum length is 3, but the
270 minimum length for the F is 1. This is important as the minimum length
271 is used to determine offsets in front of and behind the string being
272 looked for. Since strings can be composites this is the length of the
273 pattern at the time it was commited with a scan_commit. Note that
274 the length is calculated by study_chunk, so that the minimum lengths
275 are not known until the full pattern has been compiled, thus the
276 pointer to the value.
280 In the case of lookbehind the string being searched for can be
281 offset past the start point of the final matching string.
282 If this value was just blithely removed from the min_offset it would
283 invalidate some of the calculations for how many chars must match
284 before or after (as they are derived from min_offset and minlen and
285 the length of the string being searched for).
286 When the final pattern is compiled and the data is moved from the
287 scan_data_t structure into the regexp structure the information
288 about lookbehind is factored in, with the information that would
289 have been lost precalculated in the end_shift field for the
292 The fields pos_min and pos_delta are used to store the minimum offset
293 and the delta to the maximum offset at the current point in the pattern.
297 typedef struct scan_data_t {
298 /*I32 len_min; unused */
299 /*I32 len_delta; unused */
303 I32 last_end; /* min value, <0 unless valid. */
306 SV **longest; /* Either &l_fixed, or &l_float. */
307 SV *longest_fixed; /* longest fixed string found in pattern */
308 I32 offset_fixed; /* offset where it starts */
309 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
310 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
311 SV *longest_float; /* longest floating string found in pattern */
312 I32 offset_float_min; /* earliest point in string it can appear */
313 I32 offset_float_max; /* latest point in string it can appear */
314 I32 *minlen_float; /* pointer to the minlen relevent to the string */
315 I32 lookbehind_float; /* is the position of the string modified by LB */
319 struct regnode_charclass_class *start_class;
323 * Forward declarations for pregcomp()'s friends.
326 static const scan_data_t zero_scan_data =
327 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
329 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
330 #define SF_BEFORE_SEOL 0x0001
331 #define SF_BEFORE_MEOL 0x0002
332 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
333 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
336 # define SF_FIX_SHIFT_EOL (0+2)
337 # define SF_FL_SHIFT_EOL (0+4)
339 # define SF_FIX_SHIFT_EOL (+2)
340 # define SF_FL_SHIFT_EOL (+4)
343 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
344 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
346 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
347 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
348 #define SF_IS_INF 0x0040
349 #define SF_HAS_PAR 0x0080
350 #define SF_IN_PAR 0x0100
351 #define SF_HAS_EVAL 0x0200
352 #define SCF_DO_SUBSTR 0x0400
353 #define SCF_DO_STCLASS_AND 0x0800
354 #define SCF_DO_STCLASS_OR 0x1000
355 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
356 #define SCF_WHILEM_VISITED_POS 0x2000
358 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
359 #define SCF_SEEN_ACCEPT 0x8000
361 #define UTF (RExC_utf8 != 0)
362 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
363 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
365 #define OOB_UNICODE 12345678
366 #define OOB_NAMEDCLASS -1
368 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
369 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
372 /* length of regex to show in messages that don't mark a position within */
373 #define RegexLengthToShowInErrorMessages 127
376 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
377 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
378 * op/pragma/warn/regcomp.
380 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
381 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
383 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
386 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
387 * arg. Show regex, up to a maximum length. If it's too long, chop and add
390 #define _FAIL(code) STMT_START { \
391 const char *ellipses = ""; \
392 IV len = RExC_end - RExC_precomp; \
395 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
396 if (len > RegexLengthToShowInErrorMessages) { \
397 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
398 len = RegexLengthToShowInErrorMessages - 10; \
404 #define FAIL(msg) _FAIL( \
405 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
406 msg, (int)len, RExC_precomp, ellipses))
408 #define FAIL2(msg,arg) _FAIL( \
409 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
410 arg, (int)len, RExC_precomp, ellipses))
413 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
415 #define Simple_vFAIL(m) STMT_START { \
416 const IV offset = RExC_parse - RExC_precomp; \
417 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
418 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
422 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
424 #define vFAIL(m) STMT_START { \
426 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
431 * Like Simple_vFAIL(), but accepts two arguments.
433 #define Simple_vFAIL2(m,a1) STMT_START { \
434 const IV offset = RExC_parse - RExC_precomp; \
435 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
436 (int)offset, RExC_precomp, RExC_precomp + offset); \
440 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
442 #define vFAIL2(m,a1) STMT_START { \
444 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
445 Simple_vFAIL2(m, a1); \
450 * Like Simple_vFAIL(), but accepts three arguments.
452 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
453 const IV offset = RExC_parse - RExC_precomp; \
454 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
455 (int)offset, RExC_precomp, RExC_precomp + offset); \
459 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
461 #define vFAIL3(m,a1,a2) STMT_START { \
463 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
464 Simple_vFAIL3(m, a1, a2); \
468 * Like Simple_vFAIL(), but accepts four arguments.
470 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
471 const IV offset = RExC_parse - RExC_precomp; \
472 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
473 (int)offset, RExC_precomp, RExC_precomp + offset); \
476 #define vWARN(loc,m) STMT_START { \
477 const IV offset = loc - RExC_precomp; \
478 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
479 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
482 #define vWARNdep(loc,m) STMT_START { \
483 const IV offset = loc - RExC_precomp; \
484 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
485 "%s" REPORT_LOCATION, \
486 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
490 #define vWARN2(loc, m, a1) STMT_START { \
491 const IV offset = loc - RExC_precomp; \
492 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
493 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
496 #define vWARN3(loc, m, a1, a2) STMT_START { \
497 const IV offset = loc - RExC_precomp; \
498 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
499 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
502 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
503 const IV offset = loc - RExC_precomp; \
504 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
505 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
508 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
509 const IV offset = loc - RExC_precomp; \
510 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
511 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
515 /* Allow for side effects in s */
516 #define REGC(c,s) STMT_START { \
517 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
520 /* Macros for recording node offsets. 20001227 mjd@plover.com
521 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
522 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
523 * Element 0 holds the number n.
524 * Position is 1 indexed.
526 #ifndef RE_TRACK_PATTERN_OFFSETS
527 #define Set_Node_Offset_To_R(node,byte)
528 #define Set_Node_Offset(node,byte)
529 #define Set_Cur_Node_Offset
530 #define Set_Node_Length_To_R(node,len)
531 #define Set_Node_Length(node,len)
532 #define Set_Node_Cur_Length(node)
533 #define Node_Offset(n)
534 #define Node_Length(n)
535 #define Set_Node_Offset_Length(node,offset,len)
536 #define ProgLen(ri) ri->u.proglen
537 #define SetProgLen(ri,x) ri->u.proglen = x
539 #define ProgLen(ri) ri->u.offsets[0]
540 #define SetProgLen(ri,x) ri->u.offsets[0] = x
541 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
543 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
544 __LINE__, (int)(node), (int)(byte))); \
546 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
548 RExC_offsets[2*(node)-1] = (byte); \
553 #define Set_Node_Offset(node,byte) \
554 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
555 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
557 #define Set_Node_Length_To_R(node,len) STMT_START { \
559 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
560 __LINE__, (int)(node), (int)(len))); \
562 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
564 RExC_offsets[2*(node)] = (len); \
569 #define Set_Node_Length(node,len) \
570 Set_Node_Length_To_R((node)-RExC_emit_start, len)
571 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
572 #define Set_Node_Cur_Length(node) \
573 Set_Node_Length(node, RExC_parse - parse_start)
575 /* Get offsets and lengths */
576 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
577 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
579 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
580 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
581 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
585 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
586 #define EXPERIMENTAL_INPLACESCAN
587 #endif /*RE_TRACK_PATTERN_OFFSETS*/
589 #define DEBUG_STUDYDATA(str,data,depth) \
590 DEBUG_OPTIMISE_MORE_r(if(data){ \
591 PerlIO_printf(Perl_debug_log, \
592 "%*s" str "Pos:%"IVdf"/%"IVdf \
593 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
594 (int)(depth)*2, "", \
595 (IV)((data)->pos_min), \
596 (IV)((data)->pos_delta), \
597 (UV)((data)->flags), \
598 (IV)((data)->whilem_c), \
599 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
600 is_inf ? "INF " : "" \
602 if ((data)->last_found) \
603 PerlIO_printf(Perl_debug_log, \
604 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
605 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
606 SvPVX_const((data)->last_found), \
607 (IV)((data)->last_end), \
608 (IV)((data)->last_start_min), \
609 (IV)((data)->last_start_max), \
610 ((data)->longest && \
611 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
612 SvPVX_const((data)->longest_fixed), \
613 (IV)((data)->offset_fixed), \
614 ((data)->longest && \
615 (data)->longest==&((data)->longest_float)) ? "*" : "", \
616 SvPVX_const((data)->longest_float), \
617 (IV)((data)->offset_float_min), \
618 (IV)((data)->offset_float_max) \
620 PerlIO_printf(Perl_debug_log,"\n"); \
623 static void clear_re(pTHX_ void *r);
625 /* Mark that we cannot extend a found fixed substring at this point.
626 Update the longest found anchored substring and the longest found
627 floating substrings if needed. */
630 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
632 const STRLEN l = CHR_SVLEN(data->last_found);
633 const STRLEN old_l = CHR_SVLEN(*data->longest);
634 GET_RE_DEBUG_FLAGS_DECL;
636 PERL_ARGS_ASSERT_SCAN_COMMIT;
638 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
639 SvSetMagicSV(*data->longest, data->last_found);
640 if (*data->longest == data->longest_fixed) {
641 data->offset_fixed = l ? data->last_start_min : data->pos_min;
642 if (data->flags & SF_BEFORE_EOL)
644 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
646 data->flags &= ~SF_FIX_BEFORE_EOL;
647 data->minlen_fixed=minlenp;
648 data->lookbehind_fixed=0;
650 else { /* *data->longest == data->longest_float */
651 data->offset_float_min = l ? data->last_start_min : data->pos_min;
652 data->offset_float_max = (l
653 ? data->last_start_max
654 : data->pos_min + data->pos_delta);
655 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
656 data->offset_float_max = I32_MAX;
657 if (data->flags & SF_BEFORE_EOL)
659 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
661 data->flags &= ~SF_FL_BEFORE_EOL;
662 data->minlen_float=minlenp;
663 data->lookbehind_float=0;
666 SvCUR_set(data->last_found, 0);
668 SV * const sv = data->last_found;
669 if (SvUTF8(sv) && SvMAGICAL(sv)) {
670 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
676 data->flags &= ~SF_BEFORE_EOL;
677 DEBUG_STUDYDATA("commit: ",data,0);
680 /* Can match anything (initialization) */
682 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
684 PERL_ARGS_ASSERT_CL_ANYTHING;
686 ANYOF_CLASS_ZERO(cl);
687 ANYOF_BITMAP_SETALL(cl);
688 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
690 cl->flags |= ANYOF_LOCALE;
693 /* Can match anything (initialization) */
695 S_cl_is_anything(const struct regnode_charclass_class *cl)
699 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
701 for (value = 0; value <= ANYOF_MAX; value += 2)
702 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
704 if (!(cl->flags & ANYOF_UNICODE_ALL))
706 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
711 /* Can match anything (initialization) */
713 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
715 PERL_ARGS_ASSERT_CL_INIT;
717 Zero(cl, 1, struct regnode_charclass_class);
719 cl_anything(pRExC_state, cl);
723 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
725 PERL_ARGS_ASSERT_CL_INIT_ZERO;
727 Zero(cl, 1, struct regnode_charclass_class);
729 cl_anything(pRExC_state, cl);
731 cl->flags |= ANYOF_LOCALE;
734 /* 'And' a given class with another one. Can create false positives */
735 /* We assume that cl is not inverted */
737 S_cl_and(struct regnode_charclass_class *cl,
738 const struct regnode_charclass_class *and_with)
740 PERL_ARGS_ASSERT_CL_AND;
742 assert(and_with->type == ANYOF);
743 if (!(and_with->flags & ANYOF_CLASS)
744 && !(cl->flags & ANYOF_CLASS)
745 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
746 && !(and_with->flags & ANYOF_FOLD)
747 && !(cl->flags & ANYOF_FOLD)) {
750 if (and_with->flags & ANYOF_INVERT)
751 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
752 cl->bitmap[i] &= ~and_with->bitmap[i];
754 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
755 cl->bitmap[i] &= and_with->bitmap[i];
756 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
757 if (!(and_with->flags & ANYOF_EOS))
758 cl->flags &= ~ANYOF_EOS;
760 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
761 !(and_with->flags & ANYOF_INVERT)) {
762 cl->flags &= ~ANYOF_UNICODE_ALL;
763 cl->flags |= ANYOF_UNICODE;
764 ARG_SET(cl, ARG(and_with));
766 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
767 !(and_with->flags & ANYOF_INVERT))
768 cl->flags &= ~ANYOF_UNICODE_ALL;
769 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
770 !(and_with->flags & ANYOF_INVERT))
771 cl->flags &= ~ANYOF_UNICODE;
774 /* 'OR' a given class with another one. Can create false positives */
775 /* We assume that cl is not inverted */
777 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
779 PERL_ARGS_ASSERT_CL_OR;
781 if (or_with->flags & ANYOF_INVERT) {
783 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
784 * <= (B1 | !B2) | (CL1 | !CL2)
785 * which is wasteful if CL2 is small, but we ignore CL2:
786 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
787 * XXXX Can we handle case-fold? Unclear:
788 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
789 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
791 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
792 && !(or_with->flags & ANYOF_FOLD)
793 && !(cl->flags & ANYOF_FOLD) ) {
796 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
797 cl->bitmap[i] |= ~or_with->bitmap[i];
798 } /* XXXX: logic is complicated otherwise */
800 cl_anything(pRExC_state, cl);
803 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
804 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
805 && (!(or_with->flags & ANYOF_FOLD)
806 || (cl->flags & ANYOF_FOLD)) ) {
809 /* OR char bitmap and class bitmap separately */
810 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
811 cl->bitmap[i] |= or_with->bitmap[i];
812 if (or_with->flags & ANYOF_CLASS) {
813 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
814 cl->classflags[i] |= or_with->classflags[i];
815 cl->flags |= ANYOF_CLASS;
818 else { /* XXXX: logic is complicated, leave it along for a moment. */
819 cl_anything(pRExC_state, cl);
822 if (or_with->flags & ANYOF_EOS)
823 cl->flags |= ANYOF_EOS;
825 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
826 ARG(cl) != ARG(or_with)) {
827 cl->flags |= ANYOF_UNICODE_ALL;
828 cl->flags &= ~ANYOF_UNICODE;
830 if (or_with->flags & ANYOF_UNICODE_ALL) {
831 cl->flags |= ANYOF_UNICODE_ALL;
832 cl->flags &= ~ANYOF_UNICODE;
836 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
837 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
838 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
839 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
844 dump_trie(trie,widecharmap,revcharmap)
845 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
846 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
848 These routines dump out a trie in a somewhat readable format.
849 The _interim_ variants are used for debugging the interim
850 tables that are used to generate the final compressed
851 representation which is what dump_trie expects.
853 Part of the reason for their existance is to provide a form
854 of documentation as to how the different representations function.
859 Dumps the final compressed table form of the trie to Perl_debug_log.
860 Used for debugging make_trie().
864 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
865 AV *revcharmap, U32 depth)
868 SV *sv=sv_newmortal();
869 int colwidth= widecharmap ? 6 : 4;
870 GET_RE_DEBUG_FLAGS_DECL;
872 PERL_ARGS_ASSERT_DUMP_TRIE;
874 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
875 (int)depth * 2 + 2,"",
876 "Match","Base","Ofs" );
878 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
879 SV ** const tmp = av_fetch( revcharmap, state, 0);
881 PerlIO_printf( Perl_debug_log, "%*s",
883 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
884 PL_colors[0], PL_colors[1],
885 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
886 PERL_PV_ESCAPE_FIRSTCHAR
891 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
892 (int)depth * 2 + 2,"");
894 for( state = 0 ; state < trie->uniquecharcount ; state++ )
895 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
896 PerlIO_printf( Perl_debug_log, "\n");
898 for( state = 1 ; state < trie->statecount ; state++ ) {
899 const U32 base = trie->states[ state ].trans.base;
901 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
903 if ( trie->states[ state ].wordnum ) {
904 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
906 PerlIO_printf( Perl_debug_log, "%6s", "" );
909 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
914 while( ( base + ofs < trie->uniquecharcount ) ||
915 ( base + ofs - trie->uniquecharcount < trie->lasttrans
916 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
919 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
921 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
922 if ( ( base + ofs >= trie->uniquecharcount ) &&
923 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
924 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
926 PerlIO_printf( Perl_debug_log, "%*"UVXf,
928 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
930 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
934 PerlIO_printf( Perl_debug_log, "]");
937 PerlIO_printf( Perl_debug_log, "\n" );
941 Dumps a fully constructed but uncompressed trie in list form.
942 List tries normally only are used for construction when the number of
943 possible chars (trie->uniquecharcount) is very high.
944 Used for debugging make_trie().
947 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
948 HV *widecharmap, AV *revcharmap, U32 next_alloc,
952 SV *sv=sv_newmortal();
953 int colwidth= widecharmap ? 6 : 4;
954 GET_RE_DEBUG_FLAGS_DECL;
956 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
958 /* print out the table precompression. */
959 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
960 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
961 "------:-----+-----------------\n" );
963 for( state=1 ; state < next_alloc ; state ++ ) {
966 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
967 (int)depth * 2 + 2,"", (UV)state );
968 if ( ! trie->states[ state ].wordnum ) {
969 PerlIO_printf( Perl_debug_log, "%5s| ","");
971 PerlIO_printf( Perl_debug_log, "W%4x| ",
972 trie->states[ state ].wordnum
975 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
976 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
978 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
980 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
981 PL_colors[0], PL_colors[1],
982 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
983 PERL_PV_ESCAPE_FIRSTCHAR
985 TRIE_LIST_ITEM(state,charid).forid,
986 (UV)TRIE_LIST_ITEM(state,charid).newstate
989 PerlIO_printf(Perl_debug_log, "\n%*s| ",
990 (int)((depth * 2) + 14), "");
993 PerlIO_printf( Perl_debug_log, "\n");
998 Dumps a fully constructed but uncompressed trie in table form.
999 This is the normal DFA style state transition table, with a few
1000 twists to facilitate compression later.
1001 Used for debugging make_trie().
1004 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1005 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1010 SV *sv=sv_newmortal();
1011 int colwidth= widecharmap ? 6 : 4;
1012 GET_RE_DEBUG_FLAGS_DECL;
1014 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1017 print out the table precompression so that we can do a visual check
1018 that they are identical.
1021 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1023 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1024 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1026 PerlIO_printf( Perl_debug_log, "%*s",
1028 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1029 PL_colors[0], PL_colors[1],
1030 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1031 PERL_PV_ESCAPE_FIRSTCHAR
1037 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1039 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1040 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1043 PerlIO_printf( Perl_debug_log, "\n" );
1045 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1047 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1048 (int)depth * 2 + 2,"",
1049 (UV)TRIE_NODENUM( state ) );
1051 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1052 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1054 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1056 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1058 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1059 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1061 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1062 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1069 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1070 startbranch: the first branch in the whole branch sequence
1071 first : start branch of sequence of branch-exact nodes.
1072 May be the same as startbranch
1073 last : Thing following the last branch.
1074 May be the same as tail.
1075 tail : item following the branch sequence
1076 count : words in the sequence
1077 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1078 depth : indent depth
1080 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1082 A trie is an N'ary tree where the branches are determined by digital
1083 decomposition of the key. IE, at the root node you look up the 1st character and
1084 follow that branch repeat until you find the end of the branches. Nodes can be
1085 marked as "accepting" meaning they represent a complete word. Eg:
1089 would convert into the following structure. Numbers represent states, letters
1090 following numbers represent valid transitions on the letter from that state, if
1091 the number is in square brackets it represents an accepting state, otherwise it
1092 will be in parenthesis.
1094 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1098 (1) +-i->(6)-+-s->[7]
1100 +-s->(3)-+-h->(4)-+-e->[5]
1102 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1104 This shows that when matching against the string 'hers' we will begin at state 1
1105 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1106 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1107 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1108 single traverse. We store a mapping from accepting to state to which word was
1109 matched, and then when we have multiple possibilities we try to complete the
1110 rest of the regex in the order in which they occured in the alternation.
1112 The only prior NFA like behaviour that would be changed by the TRIE support is
1113 the silent ignoring of duplicate alternations which are of the form:
1115 / (DUPE|DUPE) X? (?{ ... }) Y /x
1117 Thus EVAL blocks follwing a trie may be called a different number of times with
1118 and without the optimisation. With the optimisations dupes will be silently
1119 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1120 the following demonstrates:
1122 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1124 which prints out 'word' three times, but
1126 'words'=~/(word|word|word)(?{ print $1 })S/
1128 which doesnt print it out at all. This is due to other optimisations kicking in.
1130 Example of what happens on a structural level:
1132 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1134 1: CURLYM[1] {1,32767}(18)
1145 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1146 and should turn into:
1148 1: CURLYM[1] {1,32767}(18)
1150 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1158 Cases where tail != last would be like /(?foo|bar)baz/:
1168 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1169 and would end up looking like:
1172 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1179 d = uvuni_to_utf8_flags(d, uv, 0);
1181 is the recommended Unicode-aware way of saying
1186 #define TRIE_STORE_REVCHAR \
1189 SV *zlopp = newSV(2); \
1190 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1191 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1192 SvCUR_set(zlopp, kapow - flrbbbbb); \
1195 av_push(revcharmap, zlopp); \
1197 char ooooff = (char)uvc; \
1198 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1202 #define TRIE_READ_CHAR STMT_START { \
1206 if ( foldlen > 0 ) { \
1207 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1212 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1213 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1214 foldlen -= UNISKIP( uvc ); \
1215 scan = foldbuf + UNISKIP( uvc ); \
1218 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1228 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1229 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1230 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1231 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1233 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1234 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1235 TRIE_LIST_CUR( state )++; \
1238 #define TRIE_LIST_NEW(state) STMT_START { \
1239 Newxz( trie->states[ state ].trans.list, \
1240 4, reg_trie_trans_le ); \
1241 TRIE_LIST_CUR( state ) = 1; \
1242 TRIE_LIST_LEN( state ) = 4; \
1245 #define TRIE_HANDLE_WORD(state) STMT_START { \
1246 U16 dupe= trie->states[ state ].wordnum; \
1247 regnode * const noper_next = regnext( noper ); \
1249 if (trie->wordlen) \
1250 trie->wordlen[ curword ] = wordlen; \
1252 /* store the word for dumping */ \
1254 if (OP(noper) != NOTHING) \
1255 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1257 tmp = newSVpvn_utf8( "", 0, UTF ); \
1258 av_push( trie_words, tmp ); \
1263 if ( noper_next < tail ) { \
1265 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1266 trie->jump[curword] = (U16)(noper_next - convert); \
1268 jumper = noper_next; \
1270 nextbranch= regnext(cur); \
1274 /* So it's a dupe. This means we need to maintain a */\
1275 /* linked-list from the first to the next. */\
1276 /* we only allocate the nextword buffer when there */\
1277 /* a dupe, so first time we have to do the allocation */\
1278 if (!trie->nextword) \
1279 trie->nextword = (U16 *) \
1280 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1281 while ( trie->nextword[dupe] ) \
1282 dupe= trie->nextword[dupe]; \
1283 trie->nextword[dupe]= curword; \
1285 /* we haven't inserted this word yet. */ \
1286 trie->states[ state ].wordnum = curword; \
1291 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1292 ( ( base + charid >= ucharcount \
1293 && base + charid < ubound \
1294 && state == trie->trans[ base - ucharcount + charid ].check \
1295 && trie->trans[ base - ucharcount + charid ].next ) \
1296 ? trie->trans[ base - ucharcount + charid ].next \
1297 : ( state==1 ? special : 0 ) \
1301 #define MADE_JUMP_TRIE 2
1302 #define MADE_EXACT_TRIE 4
1305 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1308 /* first pass, loop through and scan words */
1309 reg_trie_data *trie;
1310 HV *widecharmap = NULL;
1311 AV *revcharmap = newAV();
1313 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1318 regnode *jumper = NULL;
1319 regnode *nextbranch = NULL;
1320 regnode *convert = NULL;
1321 /* we just use folder as a flag in utf8 */
1322 const U8 * const folder = ( flags == EXACTF
1324 : ( flags == EXACTFL
1331 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1332 AV *trie_words = NULL;
1333 /* along with revcharmap, this only used during construction but both are
1334 * useful during debugging so we store them in the struct when debugging.
1337 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1338 STRLEN trie_charcount=0;
1340 SV *re_trie_maxbuff;
1341 GET_RE_DEBUG_FLAGS_DECL;
1343 PERL_ARGS_ASSERT_MAKE_TRIE;
1345 PERL_UNUSED_ARG(depth);
1348 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1350 trie->startstate = 1;
1351 trie->wordcount = word_count;
1352 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1353 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1354 if (!(UTF && folder))
1355 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1357 trie_words = newAV();
1360 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1361 if (!SvIOK(re_trie_maxbuff)) {
1362 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1365 PerlIO_printf( Perl_debug_log,
1366 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1367 (int)depth * 2 + 2, "",
1368 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1369 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1373 /* Find the node we are going to overwrite */
1374 if ( first == startbranch && OP( last ) != BRANCH ) {
1375 /* whole branch chain */
1378 /* branch sub-chain */
1379 convert = NEXTOPER( first );
1382 /* -- First loop and Setup --
1384 We first traverse the branches and scan each word to determine if it
1385 contains widechars, and how many unique chars there are, this is
1386 important as we have to build a table with at least as many columns as we
1389 We use an array of integers to represent the character codes 0..255
1390 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1391 native representation of the character value as the key and IV's for the
1394 *TODO* If we keep track of how many times each character is used we can
1395 remap the columns so that the table compression later on is more
1396 efficient in terms of memory by ensuring most common value is in the
1397 middle and the least common are on the outside. IMO this would be better
1398 than a most to least common mapping as theres a decent chance the most
1399 common letter will share a node with the least common, meaning the node
1400 will not be compressable. With a middle is most common approach the worst
1401 case is when we have the least common nodes twice.
1405 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1406 regnode * const noper = NEXTOPER( cur );
1407 const U8 *uc = (U8*)STRING( noper );
1408 const U8 * const e = uc + STR_LEN( noper );
1410 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1411 const U8 *scan = (U8*)NULL;
1412 U32 wordlen = 0; /* required init */
1414 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1416 if (OP(noper) == NOTHING) {
1420 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1421 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1422 regardless of encoding */
1424 for ( ; uc < e ; uc += len ) {
1425 TRIE_CHARCOUNT(trie)++;
1429 if ( !trie->charmap[ uvc ] ) {
1430 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1432 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1436 /* store the codepoint in the bitmap, and if its ascii
1437 also store its folded equivelent. */
1438 TRIE_BITMAP_SET(trie,uvc);
1440 /* store the folded codepoint */
1441 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1444 /* store first byte of utf8 representation of
1445 codepoints in the 127 < uvc < 256 range */
1446 if (127 < uvc && uvc < 192) {
1447 TRIE_BITMAP_SET(trie,194);
1448 } else if (191 < uvc ) {
1449 TRIE_BITMAP_SET(trie,195);
1450 /* && uvc < 256 -- we know uvc is < 256 already */
1453 set_bit = 0; /* We've done our bit :-) */
1458 widecharmap = newHV();
1460 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1463 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1465 if ( !SvTRUE( *svpp ) ) {
1466 sv_setiv( *svpp, ++trie->uniquecharcount );
1471 if( cur == first ) {
1474 } else if (chars < trie->minlen) {
1476 } else if (chars > trie->maxlen) {
1480 } /* end first pass */
1481 DEBUG_TRIE_COMPILE_r(
1482 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1483 (int)depth * 2 + 2,"",
1484 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1485 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1486 (int)trie->minlen, (int)trie->maxlen )
1488 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1491 We now know what we are dealing with in terms of unique chars and
1492 string sizes so we can calculate how much memory a naive
1493 representation using a flat table will take. If it's over a reasonable
1494 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1495 conservative but potentially much slower representation using an array
1498 At the end we convert both representations into the same compressed
1499 form that will be used in regexec.c for matching with. The latter
1500 is a form that cannot be used to construct with but has memory
1501 properties similar to the list form and access properties similar
1502 to the table form making it both suitable for fast searches and
1503 small enough that its feasable to store for the duration of a program.
1505 See the comment in the code where the compressed table is produced
1506 inplace from the flat tabe representation for an explanation of how
1507 the compression works.
1512 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1514 Second Pass -- Array Of Lists Representation
1516 Each state will be represented by a list of charid:state records
1517 (reg_trie_trans_le) the first such element holds the CUR and LEN
1518 points of the allocated array. (See defines above).
1520 We build the initial structure using the lists, and then convert
1521 it into the compressed table form which allows faster lookups
1522 (but cant be modified once converted).
1525 STRLEN transcount = 1;
1527 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1528 "%*sCompiling trie using list compiler\n",
1529 (int)depth * 2 + 2, ""));
1531 trie->states = (reg_trie_state *)
1532 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1533 sizeof(reg_trie_state) );
1537 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1539 regnode * const noper = NEXTOPER( cur );
1540 U8 *uc = (U8*)STRING( noper );
1541 const U8 * const e = uc + STR_LEN( noper );
1542 U32 state = 1; /* required init */
1543 U16 charid = 0; /* sanity init */
1544 U8 *scan = (U8*)NULL; /* sanity init */
1545 STRLEN foldlen = 0; /* required init */
1546 U32 wordlen = 0; /* required init */
1547 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1549 if (OP(noper) != NOTHING) {
1550 for ( ; uc < e ; uc += len ) {
1555 charid = trie->charmap[ uvc ];
1557 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1561 charid=(U16)SvIV( *svpp );
1564 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1571 if ( !trie->states[ state ].trans.list ) {
1572 TRIE_LIST_NEW( state );
1574 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1575 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1576 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1581 newstate = next_alloc++;
1582 TRIE_LIST_PUSH( state, charid, newstate );
1587 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1591 TRIE_HANDLE_WORD(state);
1593 } /* end second pass */
1595 /* next alloc is the NEXT state to be allocated */
1596 trie->statecount = next_alloc;
1597 trie->states = (reg_trie_state *)
1598 PerlMemShared_realloc( trie->states,
1600 * sizeof(reg_trie_state) );
1602 /* and now dump it out before we compress it */
1603 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1604 revcharmap, next_alloc,
1608 trie->trans = (reg_trie_trans *)
1609 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1616 for( state=1 ; state < next_alloc ; state ++ ) {
1620 DEBUG_TRIE_COMPILE_MORE_r(
1621 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1625 if (trie->states[state].trans.list) {
1626 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1630 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1631 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1632 if ( forid < minid ) {
1634 } else if ( forid > maxid ) {
1638 if ( transcount < tp + maxid - minid + 1) {
1640 trie->trans = (reg_trie_trans *)
1641 PerlMemShared_realloc( trie->trans,
1643 * sizeof(reg_trie_trans) );
1644 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1646 base = trie->uniquecharcount + tp - minid;
1647 if ( maxid == minid ) {
1649 for ( ; zp < tp ; zp++ ) {
1650 if ( ! trie->trans[ zp ].next ) {
1651 base = trie->uniquecharcount + zp - minid;
1652 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1653 trie->trans[ zp ].check = state;
1659 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1660 trie->trans[ tp ].check = state;
1665 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1666 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1667 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1668 trie->trans[ tid ].check = state;
1670 tp += ( maxid - minid + 1 );
1672 Safefree(trie->states[ state ].trans.list);
1675 DEBUG_TRIE_COMPILE_MORE_r(
1676 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1679 trie->states[ state ].trans.base=base;
1681 trie->lasttrans = tp + 1;
1685 Second Pass -- Flat Table Representation.
1687 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1688 We know that we will need Charcount+1 trans at most to store the data
1689 (one row per char at worst case) So we preallocate both structures
1690 assuming worst case.
1692 We then construct the trie using only the .next slots of the entry
1695 We use the .check field of the first entry of the node temporarily to
1696 make compression both faster and easier by keeping track of how many non
1697 zero fields are in the node.
1699 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1702 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1703 number representing the first entry of the node, and state as a
1704 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1705 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1706 are 2 entrys per node. eg:
1714 The table is internally in the right hand, idx form. However as we also
1715 have to deal with the states array which is indexed by nodenum we have to
1716 use TRIE_NODENUM() to convert.
1719 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1720 "%*sCompiling trie using table compiler\n",
1721 (int)depth * 2 + 2, ""));
1723 trie->trans = (reg_trie_trans *)
1724 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1725 * trie->uniquecharcount + 1,
1726 sizeof(reg_trie_trans) );
1727 trie->states = (reg_trie_state *)
1728 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1729 sizeof(reg_trie_state) );
1730 next_alloc = trie->uniquecharcount + 1;
1733 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1735 regnode * const noper = NEXTOPER( cur );
1736 const U8 *uc = (U8*)STRING( noper );
1737 const U8 * const e = uc + STR_LEN( noper );
1739 U32 state = 1; /* required init */
1741 U16 charid = 0; /* sanity init */
1742 U32 accept_state = 0; /* sanity init */
1743 U8 *scan = (U8*)NULL; /* sanity init */
1745 STRLEN foldlen = 0; /* required init */
1746 U32 wordlen = 0; /* required init */
1747 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1749 if ( OP(noper) != NOTHING ) {
1750 for ( ; uc < e ; uc += len ) {
1755 charid = trie->charmap[ uvc ];
1757 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1758 charid = svpp ? (U16)SvIV(*svpp) : 0;
1762 if ( !trie->trans[ state + charid ].next ) {
1763 trie->trans[ state + charid ].next = next_alloc;
1764 trie->trans[ state ].check++;
1765 next_alloc += trie->uniquecharcount;
1767 state = trie->trans[ state + charid ].next;
1769 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1771 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1774 accept_state = TRIE_NODENUM( state );
1775 TRIE_HANDLE_WORD(accept_state);
1777 } /* end second pass */
1779 /* and now dump it out before we compress it */
1780 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1782 next_alloc, depth+1));
1786 * Inplace compress the table.*
1788 For sparse data sets the table constructed by the trie algorithm will
1789 be mostly 0/FAIL transitions or to put it another way mostly empty.
1790 (Note that leaf nodes will not contain any transitions.)
1792 This algorithm compresses the tables by eliminating most such
1793 transitions, at the cost of a modest bit of extra work during lookup:
1795 - Each states[] entry contains a .base field which indicates the
1796 index in the state[] array wheres its transition data is stored.
1798 - If .base is 0 there are no valid transitions from that node.
1800 - If .base is nonzero then charid is added to it to find an entry in
1803 -If trans[states[state].base+charid].check!=state then the
1804 transition is taken to be a 0/Fail transition. Thus if there are fail
1805 transitions at the front of the node then the .base offset will point
1806 somewhere inside the previous nodes data (or maybe even into a node
1807 even earlier), but the .check field determines if the transition is
1811 The following process inplace converts the table to the compressed
1812 table: We first do not compress the root node 1,and mark its all its
1813 .check pointers as 1 and set its .base pointer as 1 as well. This
1814 allows to do a DFA construction from the compressed table later, and
1815 ensures that any .base pointers we calculate later are greater than
1818 - We set 'pos' to indicate the first entry of the second node.
1820 - We then iterate over the columns of the node, finding the first and
1821 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1822 and set the .check pointers accordingly, and advance pos
1823 appropriately and repreat for the next node. Note that when we copy
1824 the next pointers we have to convert them from the original
1825 NODEIDX form to NODENUM form as the former is not valid post
1828 - If a node has no transitions used we mark its base as 0 and do not
1829 advance the pos pointer.
1831 - If a node only has one transition we use a second pointer into the
1832 structure to fill in allocated fail transitions from other states.
1833 This pointer is independent of the main pointer and scans forward
1834 looking for null transitions that are allocated to a state. When it
1835 finds one it writes the single transition into the "hole". If the
1836 pointer doesnt find one the single transition is appended as normal.
1838 - Once compressed we can Renew/realloc the structures to release the
1841 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1842 specifically Fig 3.47 and the associated pseudocode.
1846 const U32 laststate = TRIE_NODENUM( next_alloc );
1849 trie->statecount = laststate;
1851 for ( state = 1 ; state < laststate ; state++ ) {
1853 const U32 stateidx = TRIE_NODEIDX( state );
1854 const U32 o_used = trie->trans[ stateidx ].check;
1855 U32 used = trie->trans[ stateidx ].check;
1856 trie->trans[ stateidx ].check = 0;
1858 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1859 if ( flag || trie->trans[ stateidx + charid ].next ) {
1860 if ( trie->trans[ stateidx + charid ].next ) {
1862 for ( ; zp < pos ; zp++ ) {
1863 if ( ! trie->trans[ zp ].next ) {
1867 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1868 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1869 trie->trans[ zp ].check = state;
1870 if ( ++zp > pos ) pos = zp;
1877 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1879 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1880 trie->trans[ pos ].check = state;
1885 trie->lasttrans = pos + 1;
1886 trie->states = (reg_trie_state *)
1887 PerlMemShared_realloc( trie->states, laststate
1888 * sizeof(reg_trie_state) );
1889 DEBUG_TRIE_COMPILE_MORE_r(
1890 PerlIO_printf( Perl_debug_log,
1891 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1892 (int)depth * 2 + 2,"",
1893 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1896 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1899 } /* end table compress */
1901 DEBUG_TRIE_COMPILE_MORE_r(
1902 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1903 (int)depth * 2 + 2, "",
1904 (UV)trie->statecount,
1905 (UV)trie->lasttrans)
1907 /* resize the trans array to remove unused space */
1908 trie->trans = (reg_trie_trans *)
1909 PerlMemShared_realloc( trie->trans, trie->lasttrans
1910 * sizeof(reg_trie_trans) );
1912 /* and now dump out the compressed format */
1913 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1915 { /* Modify the program and insert the new TRIE node*/
1916 U8 nodetype =(U8)(flags & 0xFF);
1920 regnode *optimize = NULL;
1921 #ifdef RE_TRACK_PATTERN_OFFSETS
1924 U32 mjd_nodelen = 0;
1925 #endif /* RE_TRACK_PATTERN_OFFSETS */
1926 #endif /* DEBUGGING */
1928 This means we convert either the first branch or the first Exact,
1929 depending on whether the thing following (in 'last') is a branch
1930 or not and whther first is the startbranch (ie is it a sub part of
1931 the alternation or is it the whole thing.)
1932 Assuming its a sub part we conver the EXACT otherwise we convert
1933 the whole branch sequence, including the first.
1935 /* Find the node we are going to overwrite */
1936 if ( first != startbranch || OP( last ) == BRANCH ) {
1937 /* branch sub-chain */
1938 NEXT_OFF( first ) = (U16)(last - first);
1939 #ifdef RE_TRACK_PATTERN_OFFSETS
1941 mjd_offset= Node_Offset((convert));
1942 mjd_nodelen= Node_Length((convert));
1945 /* whole branch chain */
1947 #ifdef RE_TRACK_PATTERN_OFFSETS
1950 const regnode *nop = NEXTOPER( convert );
1951 mjd_offset= Node_Offset((nop));
1952 mjd_nodelen= Node_Length((nop));
1956 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1957 (int)depth * 2 + 2, "",
1958 (UV)mjd_offset, (UV)mjd_nodelen)
1961 /* But first we check to see if there is a common prefix we can
1962 split out as an EXACT and put in front of the TRIE node. */
1963 trie->startstate= 1;
1964 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1966 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1970 const U32 base = trie->states[ state ].trans.base;
1972 if ( trie->states[state].wordnum )
1975 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1976 if ( ( base + ofs >= trie->uniquecharcount ) &&
1977 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1978 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1980 if ( ++count > 1 ) {
1981 SV **tmp = av_fetch( revcharmap, ofs, 0);
1982 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1983 if ( state == 1 ) break;
1985 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1987 PerlIO_printf(Perl_debug_log,
1988 "%*sNew Start State=%"UVuf" Class: [",
1989 (int)depth * 2 + 2, "",
1992 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1993 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1995 TRIE_BITMAP_SET(trie,*ch);
1997 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1999 PerlIO_printf(Perl_debug_log, (char*)ch)
2003 TRIE_BITMAP_SET(trie,*ch);
2005 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2006 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2012 SV **tmp = av_fetch( revcharmap, idx, 0);
2014 char *ch = SvPV( *tmp, len );
2016 SV *sv=sv_newmortal();
2017 PerlIO_printf( Perl_debug_log,
2018 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2019 (int)depth * 2 + 2, "",
2021 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2022 PL_colors[0], PL_colors[1],
2023 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2024 PERL_PV_ESCAPE_FIRSTCHAR
2029 OP( convert ) = nodetype;
2030 str=STRING(convert);
2033 STR_LEN(convert) += len;
2039 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2045 regnode *n = convert+NODE_SZ_STR(convert);
2046 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2047 trie->startstate = state;
2048 trie->minlen -= (state - 1);
2049 trie->maxlen -= (state - 1);
2051 /* At least the UNICOS C compiler choked on this
2052 * being argument to DEBUG_r(), so let's just have
2055 #ifdef PERL_EXT_RE_BUILD
2061 regnode *fix = convert;
2062 U32 word = trie->wordcount;
2064 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2065 while( ++fix < n ) {
2066 Set_Node_Offset_Length(fix, 0, 0);
2069 SV ** const tmp = av_fetch( trie_words, word, 0 );
2071 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2072 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2074 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2082 NEXT_OFF(convert) = (U16)(tail - convert);
2083 DEBUG_r(optimize= n);
2089 if ( trie->maxlen ) {
2090 NEXT_OFF( convert ) = (U16)(tail - convert);
2091 ARG_SET( convert, data_slot );
2092 /* Store the offset to the first unabsorbed branch in
2093 jump[0], which is otherwise unused by the jump logic.
2094 We use this when dumping a trie and during optimisation. */
2096 trie->jump[0] = (U16)(nextbranch - convert);
2099 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2100 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2102 OP( convert ) = TRIEC;
2103 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2104 PerlMemShared_free(trie->bitmap);
2107 OP( convert ) = TRIE;
2109 /* store the type in the flags */
2110 convert->flags = nodetype;
2114 + regarglen[ OP( convert ) ];
2116 /* XXX We really should free up the resource in trie now,
2117 as we won't use them - (which resources?) dmq */
2119 /* needed for dumping*/
2120 DEBUG_r(if (optimize) {
2121 regnode *opt = convert;
2123 while ( ++opt < optimize) {
2124 Set_Node_Offset_Length(opt,0,0);
2127 Try to clean up some of the debris left after the
2130 while( optimize < jumper ) {
2131 mjd_nodelen += Node_Length((optimize));
2132 OP( optimize ) = OPTIMIZED;
2133 Set_Node_Offset_Length(optimize,0,0);
2136 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2138 } /* end node insert */
2139 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2141 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2142 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2144 SvREFCNT_dec(revcharmap);
2148 : trie->startstate>1
2154 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2156 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2158 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2159 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2162 We find the fail state for each state in the trie, this state is the longest proper
2163 suffix of the current states 'word' that is also a proper prefix of another word in our
2164 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2165 the DFA not to have to restart after its tried and failed a word at a given point, it
2166 simply continues as though it had been matching the other word in the first place.
2168 'abcdgu'=~/abcdefg|cdgu/
2169 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2170 fail, which would bring use to the state representing 'd' in the second word where we would
2171 try 'g' and succeed, prodceding to match 'cdgu'.
2173 /* add a fail transition */
2174 const U32 trie_offset = ARG(source);
2175 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2177 const U32 ucharcount = trie->uniquecharcount;
2178 const U32 numstates = trie->statecount;
2179 const U32 ubound = trie->lasttrans + ucharcount;
2183 U32 base = trie->states[ 1 ].trans.base;
2186 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2187 GET_RE_DEBUG_FLAGS_DECL;
2189 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2191 PERL_UNUSED_ARG(depth);
2195 ARG_SET( stclass, data_slot );
2196 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2197 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2198 aho->trie=trie_offset;
2199 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2200 Copy( trie->states, aho->states, numstates, reg_trie_state );
2201 Newxz( q, numstates, U32);
2202 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2205 /* initialize fail[0..1] to be 1 so that we always have
2206 a valid final fail state */
2207 fail[ 0 ] = fail[ 1 ] = 1;
2209 for ( charid = 0; charid < ucharcount ; charid++ ) {
2210 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2212 q[ q_write ] = newstate;
2213 /* set to point at the root */
2214 fail[ q[ q_write++ ] ]=1;
2217 while ( q_read < q_write) {
2218 const U32 cur = q[ q_read++ % numstates ];
2219 base = trie->states[ cur ].trans.base;
2221 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2222 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2224 U32 fail_state = cur;
2227 fail_state = fail[ fail_state ];
2228 fail_base = aho->states[ fail_state ].trans.base;
2229 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2231 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2232 fail[ ch_state ] = fail_state;
2233 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2235 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2237 q[ q_write++ % numstates] = ch_state;
2241 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2242 when we fail in state 1, this allows us to use the
2243 charclass scan to find a valid start char. This is based on the principle
2244 that theres a good chance the string being searched contains lots of stuff
2245 that cant be a start char.
2247 fail[ 0 ] = fail[ 1 ] = 0;
2248 DEBUG_TRIE_COMPILE_r({
2249 PerlIO_printf(Perl_debug_log,
2250 "%*sStclass Failtable (%"UVuf" states): 0",
2251 (int)(depth * 2), "", (UV)numstates
2253 for( q_read=1; q_read<numstates; q_read++ ) {
2254 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2256 PerlIO_printf(Perl_debug_log, "\n");
2259 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2264 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2265 * These need to be revisited when a newer toolchain becomes available.
2267 #if defined(__sparc64__) && defined(__GNUC__)
2268 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2269 # undef SPARC64_GCC_WORKAROUND
2270 # define SPARC64_GCC_WORKAROUND 1
2274 #define DEBUG_PEEP(str,scan,depth) \
2275 DEBUG_OPTIMISE_r({if (scan){ \
2276 SV * const mysv=sv_newmortal(); \
2277 regnode *Next = regnext(scan); \
2278 regprop(RExC_rx, mysv, scan); \
2279 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2280 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2281 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2288 #define JOIN_EXACT(scan,min,flags) \
2289 if (PL_regkind[OP(scan)] == EXACT) \
2290 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2293 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2294 /* Merge several consecutive EXACTish nodes into one. */
2295 regnode *n = regnext(scan);
2297 regnode *next = scan + NODE_SZ_STR(scan);
2301 regnode *stop = scan;
2302 GET_RE_DEBUG_FLAGS_DECL;
2304 PERL_UNUSED_ARG(depth);
2307 PERL_ARGS_ASSERT_JOIN_EXACT;
2308 #ifndef EXPERIMENTAL_INPLACESCAN
2309 PERL_UNUSED_ARG(flags);
2310 PERL_UNUSED_ARG(val);
2312 DEBUG_PEEP("join",scan,depth);
2314 /* Skip NOTHING, merge EXACT*. */
2316 ( PL_regkind[OP(n)] == NOTHING ||
2317 (stringok && (OP(n) == OP(scan))))
2319 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2321 if (OP(n) == TAIL || n > next)
2323 if (PL_regkind[OP(n)] == NOTHING) {
2324 DEBUG_PEEP("skip:",n,depth);
2325 NEXT_OFF(scan) += NEXT_OFF(n);
2326 next = n + NODE_STEP_REGNODE;
2333 else if (stringok) {
2334 const unsigned int oldl = STR_LEN(scan);
2335 regnode * const nnext = regnext(n);
2337 DEBUG_PEEP("merg",n,depth);
2340 if (oldl + STR_LEN(n) > U8_MAX)
2342 NEXT_OFF(scan) += NEXT_OFF(n);
2343 STR_LEN(scan) += STR_LEN(n);
2344 next = n + NODE_SZ_STR(n);
2345 /* Now we can overwrite *n : */
2346 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2354 #ifdef EXPERIMENTAL_INPLACESCAN
2355 if (flags && !NEXT_OFF(n)) {
2356 DEBUG_PEEP("atch", val, depth);
2357 if (reg_off_by_arg[OP(n)]) {
2358 ARG_SET(n, val - n);
2361 NEXT_OFF(n) = val - n;
2368 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2370 Two problematic code points in Unicode casefolding of EXACT nodes:
2372 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2373 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2379 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2380 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2382 This means that in case-insensitive matching (or "loose matching",
2383 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2384 length of the above casefolded versions) can match a target string
2385 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2386 This would rather mess up the minimum length computation.
2388 What we'll do is to look for the tail four bytes, and then peek
2389 at the preceding two bytes to see whether we need to decrease
2390 the minimum length by four (six minus two).
2392 Thanks to the design of UTF-8, there cannot be false matches:
2393 A sequence of valid UTF-8 bytes cannot be a subsequence of
2394 another valid sequence of UTF-8 bytes.
2397 char * const s0 = STRING(scan), *s, *t;
2398 char * const s1 = s0 + STR_LEN(scan) - 1;
2399 char * const s2 = s1 - 4;
2400 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2401 const char t0[] = "\xaf\x49\xaf\x42";
2403 const char t0[] = "\xcc\x88\xcc\x81";
2405 const char * const t1 = t0 + 3;
2408 s < s2 && (t = ninstr(s, s1, t0, t1));
2411 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2412 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2414 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2415 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2423 n = scan + NODE_SZ_STR(scan);
2425 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2432 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2436 /* REx optimizer. Converts nodes into quickier variants "in place".
2437 Finds fixed substrings. */
2439 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2440 to the position after last scanned or to NULL. */
2442 #define INIT_AND_WITHP \
2443 assert(!and_withp); \
2444 Newx(and_withp,1,struct regnode_charclass_class); \
2445 SAVEFREEPV(and_withp)
2447 /* this is a chain of data about sub patterns we are processing that
2448 need to be handled seperately/specially in study_chunk. Its so
2449 we can simulate recursion without losing state. */
2451 typedef struct scan_frame {
2452 regnode *last; /* last node to process in this frame */
2453 regnode *next; /* next node to process when last is reached */
2454 struct scan_frame *prev; /*previous frame*/
2455 I32 stop; /* what stopparen do we use */
2459 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2461 #define CASE_SYNST_FNC(nAmE) \
2463 if (flags & SCF_DO_STCLASS_AND) { \
2464 for (value = 0; value < 256; value++) \
2465 if (!is_ ## nAmE ## _cp(value)) \
2466 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2469 for (value = 0; value < 256; value++) \
2470 if (is_ ## nAmE ## _cp(value)) \
2471 ANYOF_BITMAP_SET(data->start_class, value); \
2475 if (flags & SCF_DO_STCLASS_AND) { \
2476 for (value = 0; value < 256; value++) \
2477 if (is_ ## nAmE ## _cp(value)) \
2478 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2481 for (value = 0; value < 256; value++) \
2482 if (!is_ ## nAmE ## _cp(value)) \
2483 ANYOF_BITMAP_SET(data->start_class, value); \
2490 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2491 I32 *minlenp, I32 *deltap,
2496 struct regnode_charclass_class *and_withp,
2497 U32 flags, U32 depth)
2498 /* scanp: Start here (read-write). */
2499 /* deltap: Write maxlen-minlen here. */
2500 /* last: Stop before this one. */
2501 /* data: string data about the pattern */
2502 /* stopparen: treat close N as END */
2503 /* recursed: which subroutines have we recursed into */
2504 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2507 I32 min = 0, pars = 0, code;
2508 regnode *scan = *scanp, *next;
2510 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2511 int is_inf_internal = 0; /* The studied chunk is infinite */
2512 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2513 scan_data_t data_fake;
2514 SV *re_trie_maxbuff = NULL;
2515 regnode *first_non_open = scan;
2516 I32 stopmin = I32_MAX;
2517 scan_frame *frame = NULL;
2518 GET_RE_DEBUG_FLAGS_DECL;
2520 PERL_ARGS_ASSERT_STUDY_CHUNK;
2523 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2527 while (first_non_open && OP(first_non_open) == OPEN)
2528 first_non_open=regnext(first_non_open);
2533 while ( scan && OP(scan) != END && scan < last ){
2534 /* Peephole optimizer: */
2535 DEBUG_STUDYDATA("Peep:", data,depth);
2536 DEBUG_PEEP("Peep",scan,depth);
2537 JOIN_EXACT(scan,&min,0);
2539 /* Follow the next-chain of the current node and optimize
2540 away all the NOTHINGs from it. */
2541 if (OP(scan) != CURLYX) {
2542 const int max = (reg_off_by_arg[OP(scan)]
2544 /* I32 may be smaller than U16 on CRAYs! */
2545 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2546 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2550 /* Skip NOTHING and LONGJMP. */
2551 while ((n = regnext(n))
2552 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2553 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2554 && off + noff < max)
2556 if (reg_off_by_arg[OP(scan)])
2559 NEXT_OFF(scan) = off;
2564 /* The principal pseudo-switch. Cannot be a switch, since we
2565 look into several different things. */
2566 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2567 || OP(scan) == IFTHEN) {
2568 next = regnext(scan);
2570 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2572 if (OP(next) == code || code == IFTHEN) {
2573 /* NOTE - There is similar code to this block below for handling
2574 TRIE nodes on a re-study. If you change stuff here check there
2576 I32 max1 = 0, min1 = I32_MAX, num = 0;
2577 struct regnode_charclass_class accum;
2578 regnode * const startbranch=scan;
2580 if (flags & SCF_DO_SUBSTR)
2581 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2582 if (flags & SCF_DO_STCLASS)
2583 cl_init_zero(pRExC_state, &accum);
2585 while (OP(scan) == code) {
2586 I32 deltanext, minnext, f = 0, fake;
2587 struct regnode_charclass_class this_class;
2590 data_fake.flags = 0;
2592 data_fake.whilem_c = data->whilem_c;
2593 data_fake.last_closep = data->last_closep;
2596 data_fake.last_closep = &fake;
2598 data_fake.pos_delta = delta;
2599 next = regnext(scan);
2600 scan = NEXTOPER(scan);
2602 scan = NEXTOPER(scan);
2603 if (flags & SCF_DO_STCLASS) {
2604 cl_init(pRExC_state, &this_class);
2605 data_fake.start_class = &this_class;
2606 f = SCF_DO_STCLASS_AND;
2608 if (flags & SCF_WHILEM_VISITED_POS)
2609 f |= SCF_WHILEM_VISITED_POS;
2611 /* we suppose the run is continuous, last=next...*/
2612 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2614 stopparen, recursed, NULL, f,depth+1);
2617 if (max1 < minnext + deltanext)
2618 max1 = minnext + deltanext;
2619 if (deltanext == I32_MAX)
2620 is_inf = is_inf_internal = 1;
2622 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2624 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2625 if ( stopmin > minnext)
2626 stopmin = min + min1;
2627 flags &= ~SCF_DO_SUBSTR;
2629 data->flags |= SCF_SEEN_ACCEPT;
2632 if (data_fake.flags & SF_HAS_EVAL)
2633 data->flags |= SF_HAS_EVAL;
2634 data->whilem_c = data_fake.whilem_c;
2636 if (flags & SCF_DO_STCLASS)
2637 cl_or(pRExC_state, &accum, &this_class);
2639 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2641 if (flags & SCF_DO_SUBSTR) {
2642 data->pos_min += min1;
2643 data->pos_delta += max1 - min1;
2644 if (max1 != min1 || is_inf)
2645 data->longest = &(data->longest_float);
2648 delta += max1 - min1;
2649 if (flags & SCF_DO_STCLASS_OR) {
2650 cl_or(pRExC_state, data->start_class, &accum);
2652 cl_and(data->start_class, and_withp);
2653 flags &= ~SCF_DO_STCLASS;
2656 else if (flags & SCF_DO_STCLASS_AND) {
2658 cl_and(data->start_class, &accum);
2659 flags &= ~SCF_DO_STCLASS;
2662 /* Switch to OR mode: cache the old value of
2663 * data->start_class */
2665 StructCopy(data->start_class, and_withp,
2666 struct regnode_charclass_class);
2667 flags &= ~SCF_DO_STCLASS_AND;
2668 StructCopy(&accum, data->start_class,
2669 struct regnode_charclass_class);
2670 flags |= SCF_DO_STCLASS_OR;
2671 data->start_class->flags |= ANYOF_EOS;
2675 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2678 Assuming this was/is a branch we are dealing with: 'scan' now
2679 points at the item that follows the branch sequence, whatever
2680 it is. We now start at the beginning of the sequence and look
2687 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2689 If we can find such a subseqence we need to turn the first
2690 element into a trie and then add the subsequent branch exact
2691 strings to the trie.
2695 1. patterns where the whole set of branch can be converted.
2697 2. patterns where only a subset can be converted.
2699 In case 1 we can replace the whole set with a single regop
2700 for the trie. In case 2 we need to keep the start and end
2703 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2704 becomes BRANCH TRIE; BRANCH X;
2706 There is an additional case, that being where there is a
2707 common prefix, which gets split out into an EXACT like node
2708 preceding the TRIE node.
2710 If x(1..n)==tail then we can do a simple trie, if not we make
2711 a "jump" trie, such that when we match the appropriate word
2712 we "jump" to the appopriate tail node. Essentailly we turn
2713 a nested if into a case structure of sorts.
2718 if (!re_trie_maxbuff) {
2719 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2720 if (!SvIOK(re_trie_maxbuff))
2721 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2723 if ( SvIV(re_trie_maxbuff)>=0 ) {
2725 regnode *first = (regnode *)NULL;
2726 regnode *last = (regnode *)NULL;
2727 regnode *tail = scan;
2732 SV * const mysv = sv_newmortal(); /* for dumping */
2734 /* var tail is used because there may be a TAIL
2735 regop in the way. Ie, the exacts will point to the
2736 thing following the TAIL, but the last branch will
2737 point at the TAIL. So we advance tail. If we
2738 have nested (?:) we may have to move through several
2742 while ( OP( tail ) == TAIL ) {
2743 /* this is the TAIL generated by (?:) */
2744 tail = regnext( tail );
2749 regprop(RExC_rx, mysv, tail );
2750 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2751 (int)depth * 2 + 2, "",
2752 "Looking for TRIE'able sequences. Tail node is: ",
2753 SvPV_nolen_const( mysv )
2759 step through the branches, cur represents each
2760 branch, noper is the first thing to be matched
2761 as part of that branch and noper_next is the
2762 regnext() of that node. if noper is an EXACT
2763 and noper_next is the same as scan (our current
2764 position in the regex) then the EXACT branch is
2765 a possible optimization target. Once we have
2766 two or more consequetive such branches we can
2767 create a trie of the EXACT's contents and stich
2768 it in place. If the sequence represents all of
2769 the branches we eliminate the whole thing and
2770 replace it with a single TRIE. If it is a
2771 subsequence then we need to stitch it in. This
2772 means the first branch has to remain, and needs
2773 to be repointed at the item on the branch chain
2774 following the last branch optimized. This could
2775 be either a BRANCH, in which case the
2776 subsequence is internal, or it could be the
2777 item following the branch sequence in which
2778 case the subsequence is at the end.
2782 /* dont use tail as the end marker for this traverse */
2783 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2784 regnode * const noper = NEXTOPER( cur );
2785 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2786 regnode * const noper_next = regnext( noper );
2790 regprop(RExC_rx, mysv, cur);
2791 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2792 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2794 regprop(RExC_rx, mysv, noper);
2795 PerlIO_printf( Perl_debug_log, " -> %s",
2796 SvPV_nolen_const(mysv));
2799 regprop(RExC_rx, mysv, noper_next );
2800 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2801 SvPV_nolen_const(mysv));
2803 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2804 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2806 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2807 : PL_regkind[ OP( noper ) ] == EXACT )
2808 || OP(noper) == NOTHING )
2810 && noper_next == tail
2815 if ( !first || optype == NOTHING ) {
2816 if (!first) first = cur;
2817 optype = OP( noper );
2823 Currently we assume that the trie can handle unicode and ascii
2824 matches fold cased matches. If this proves true then the following
2825 define will prevent tries in this situation.
2827 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2829 #define TRIE_TYPE_IS_SAFE 1
2830 if ( last && TRIE_TYPE_IS_SAFE ) {
2831 make_trie( pRExC_state,
2832 startbranch, first, cur, tail, count,
2835 if ( PL_regkind[ OP( noper ) ] == EXACT
2837 && noper_next == tail
2842 optype = OP( noper );
2852 regprop(RExC_rx, mysv, cur);
2853 PerlIO_printf( Perl_debug_log,
2854 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2855 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2859 if ( last && TRIE_TYPE_IS_SAFE ) {
2860 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2861 #ifdef TRIE_STUDY_OPT
2862 if ( ((made == MADE_EXACT_TRIE &&
2863 startbranch == first)
2864 || ( first_non_open == first )) &&
2866 flags |= SCF_TRIE_RESTUDY;
2867 if ( startbranch == first
2870 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2880 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2881 scan = NEXTOPER(NEXTOPER(scan));
2882 } else /* single branch is optimized. */
2883 scan = NEXTOPER(scan);
2885 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2886 scan_frame *newframe = NULL;
2891 if (OP(scan) != SUSPEND) {
2892 /* set the pointer */
2893 if (OP(scan) == GOSUB) {
2895 RExC_recurse[ARG2L(scan)] = scan;
2896 start = RExC_open_parens[paren-1];
2897 end = RExC_close_parens[paren-1];
2900 start = RExC_rxi->program + 1;
2904 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2905 SAVEFREEPV(recursed);
2907 if (!PAREN_TEST(recursed,paren+1)) {
2908 PAREN_SET(recursed,paren+1);
2909 Newx(newframe,1,scan_frame);
2911 if (flags & SCF_DO_SUBSTR) {
2912 SCAN_COMMIT(pRExC_state,data,minlenp);
2913 data->longest = &(data->longest_float);
2915 is_inf = is_inf_internal = 1;
2916 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2917 cl_anything(pRExC_state, data->start_class);
2918 flags &= ~SCF_DO_STCLASS;
2921 Newx(newframe,1,scan_frame);
2924 end = regnext(scan);
2929 SAVEFREEPV(newframe);
2930 newframe->next = regnext(scan);
2931 newframe->last = last;
2932 newframe->stop = stopparen;
2933 newframe->prev = frame;
2943 else if (OP(scan) == EXACT) {
2944 I32 l = STR_LEN(scan);
2947 const U8 * const s = (U8*)STRING(scan);
2948 l = utf8_length(s, s + l);
2949 uc = utf8_to_uvchr(s, NULL);
2951 uc = *((U8*)STRING(scan));
2954 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2955 /* The code below prefers earlier match for fixed
2956 offset, later match for variable offset. */
2957 if (data->last_end == -1) { /* Update the start info. */
2958 data->last_start_min = data->pos_min;
2959 data->last_start_max = is_inf
2960 ? I32_MAX : data->pos_min + data->pos_delta;
2962 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2964 SvUTF8_on(data->last_found);
2966 SV * const sv = data->last_found;
2967 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2968 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2969 if (mg && mg->mg_len >= 0)
2970 mg->mg_len += utf8_length((U8*)STRING(scan),
2971 (U8*)STRING(scan)+STR_LEN(scan));
2973 data->last_end = data->pos_min + l;
2974 data->pos_min += l; /* As in the first entry. */
2975 data->flags &= ~SF_BEFORE_EOL;
2977 if (flags & SCF_DO_STCLASS_AND) {
2978 /* Check whether it is compatible with what we know already! */
2982 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2983 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2984 && (!(data->start_class->flags & ANYOF_FOLD)
2985 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2988 ANYOF_CLASS_ZERO(data->start_class);
2989 ANYOF_BITMAP_ZERO(data->start_class);
2991 ANYOF_BITMAP_SET(data->start_class, uc);
2992 data->start_class->flags &= ~ANYOF_EOS;
2994 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2996 else if (flags & SCF_DO_STCLASS_OR) {
2997 /* false positive possible if the class is case-folded */
2999 ANYOF_BITMAP_SET(data->start_class, uc);
3001 data->start_class->flags |= ANYOF_UNICODE_ALL;
3002 data->start_class->flags &= ~ANYOF_EOS;
3003 cl_and(data->start_class, and_withp);
3005 flags &= ~SCF_DO_STCLASS;
3007 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3008 I32 l = STR_LEN(scan);
3009 UV uc = *((U8*)STRING(scan));
3011 /* Search for fixed substrings supports EXACT only. */
3012 if (flags & SCF_DO_SUBSTR) {
3014 SCAN_COMMIT(pRExC_state, data, minlenp);
3017 const U8 * const s = (U8 *)STRING(scan);
3018 l = utf8_length(s, s + l);
3019 uc = utf8_to_uvchr(s, NULL);
3022 if (flags & SCF_DO_SUBSTR)
3024 if (flags & SCF_DO_STCLASS_AND) {
3025 /* Check whether it is compatible with what we know already! */
3029 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3030 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3031 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3033 ANYOF_CLASS_ZERO(data->start_class);
3034 ANYOF_BITMAP_ZERO(data->start_class);
3036 ANYOF_BITMAP_SET(data->start_class, uc);
3037 data->start_class->flags &= ~ANYOF_EOS;
3038 data->start_class->flags |= ANYOF_FOLD;
3039 if (OP(scan) == EXACTFL)
3040 data->start_class->flags |= ANYOF_LOCALE;
3043 else if (flags & SCF_DO_STCLASS_OR) {
3044 if (data->start_class->flags & ANYOF_FOLD) {
3045 /* false positive possible if the class is case-folded.
3046 Assume that the locale settings are the same... */
3048 ANYOF_BITMAP_SET(data->start_class, uc);
3049 data->start_class->flags &= ~ANYOF_EOS;
3051 cl_and(data->start_class, and_withp);
3053 flags &= ~SCF_DO_STCLASS;
3055 else if (strchr((const char*)PL_varies,OP(scan))) {
3056 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3057 I32 f = flags, pos_before = 0;
3058 regnode * const oscan = scan;
3059 struct regnode_charclass_class this_class;
3060 struct regnode_charclass_class *oclass = NULL;
3061 I32 next_is_eval = 0;
3063 switch (PL_regkind[OP(scan)]) {
3064 case WHILEM: /* End of (?:...)* . */
3065 scan = NEXTOPER(scan);
3068 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3069 next = NEXTOPER(scan);
3070 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3072 maxcount = REG_INFTY;
3073 next = regnext(scan);
3074 scan = NEXTOPER(scan);
3078 if (flags & SCF_DO_SUBSTR)
3083 if (flags & SCF_DO_STCLASS) {
3085 maxcount = REG_INFTY;
3086 next = regnext(scan);
3087 scan = NEXTOPER(scan);
3090 is_inf = is_inf_internal = 1;
3091 scan = regnext(scan);
3092 if (flags & SCF_DO_SUBSTR) {
3093 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3094 data->longest = &(data->longest_float);
3096 goto optimize_curly_tail;
3098 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3099 && (scan->flags == stopparen))
3104 mincount = ARG1(scan);
3105 maxcount = ARG2(scan);
3107 next = regnext(scan);
3108 if (OP(scan) == CURLYX) {
3109 I32 lp = (data ? *(data->last_closep) : 0);
3110 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3112 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3113 next_is_eval = (OP(scan) == EVAL);
3115 if (flags & SCF_DO_SUBSTR) {
3116 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3117 pos_before = data->pos_min;
3121 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3123 data->flags |= SF_IS_INF;
3125 if (flags & SCF_DO_STCLASS) {
3126 cl_init(pRExC_state, &this_class);
3127 oclass = data->start_class;
3128 data->start_class = &this_class;
3129 f |= SCF_DO_STCLASS_AND;
3130 f &= ~SCF_DO_STCLASS_OR;
3132 /* These are the cases when once a subexpression
3133 fails at a particular position, it cannot succeed
3134 even after backtracking at the enclosing scope.
3136 XXXX what if minimal match and we are at the
3137 initial run of {n,m}? */
3138 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3139 f &= ~SCF_WHILEM_VISITED_POS;
3141 /* This will finish on WHILEM, setting scan, or on NULL: */
3142 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3143 last, data, stopparen, recursed, NULL,
3145 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3147 if (flags & SCF_DO_STCLASS)
3148 data->start_class = oclass;
3149 if (mincount == 0 || minnext == 0) {
3150 if (flags & SCF_DO_STCLASS_OR) {
3151 cl_or(pRExC_state, data->start_class, &this_class);
3153 else if (flags & SCF_DO_STCLASS_AND) {
3154 /* Switch to OR mode: cache the old value of
3155 * data->start_class */
3157 StructCopy(data->start_class, and_withp,
3158 struct regnode_charclass_class);
3159 flags &= ~SCF_DO_STCLASS_AND;
3160 StructCopy(&this_class, data->start_class,
3161 struct regnode_charclass_class);
3162 flags |= SCF_DO_STCLASS_OR;
3163 data->start_class->flags |= ANYOF_EOS;
3165 } else { /* Non-zero len */
3166 if (flags & SCF_DO_STCLASS_OR) {
3167 cl_or(pRExC_state, data->start_class, &this_class);
3168 cl_and(data->start_class, and_withp);
3170 else if (flags & SCF_DO_STCLASS_AND)
3171 cl_and(data->start_class, &this_class);
3172 flags &= ~SCF_DO_STCLASS;
3174 if (!scan) /* It was not CURLYX, but CURLY. */
3176 if ( /* ? quantifier ok, except for (?{ ... }) */
3177 (next_is_eval || !(mincount == 0 && maxcount == 1))
3178 && (minnext == 0) && (deltanext == 0)
3179 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3180 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3181 && ckWARN(WARN_REGEXP))
3184 "Quantifier unexpected on zero-length expression");
3187 min += minnext * mincount;
3188 is_inf_internal |= ((maxcount == REG_INFTY
3189 && (minnext + deltanext) > 0)
3190 || deltanext == I32_MAX);
3191 is_inf |= is_inf_internal;
3192 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3194 /* Try powerful optimization CURLYX => CURLYN. */
3195 if ( OP(oscan) == CURLYX && data
3196 && data->flags & SF_IN_PAR
3197 && !(data->flags & SF_HAS_EVAL)
3198 && !deltanext && minnext == 1 ) {
3199 /* Try to optimize to CURLYN. */
3200 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3201 regnode * const nxt1 = nxt;
3208 if (!strchr((const char*)PL_simple,OP(nxt))
3209 && !(PL_regkind[OP(nxt)] == EXACT
3210 && STR_LEN(nxt) == 1))
3216 if (OP(nxt) != CLOSE)
3218 if (RExC_open_parens) {
3219 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3220 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3222 /* Now we know that nxt2 is the only contents: */
3223 oscan->flags = (U8)ARG(nxt);
3225 OP(nxt1) = NOTHING; /* was OPEN. */
3228 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3229 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3230 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3231 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3232 OP(nxt + 1) = OPTIMIZED; /* was count. */
3233 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3238 /* Try optimization CURLYX => CURLYM. */
3239 if ( OP(oscan) == CURLYX && data
3240 && !(data->flags & SF_HAS_PAR)
3241 && !(data->flags & SF_HAS_EVAL)
3242 && !deltanext /* atom is fixed width */
3243 && minnext != 0 /* CURLYM can't handle zero width */
3245 /* XXXX How to optimize if data == 0? */
3246 /* Optimize to a simpler form. */
3247 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3251 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3252 && (OP(nxt2) != WHILEM))
3254 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3255 /* Need to optimize away parenths. */
3256 if (data->flags & SF_IN_PAR) {
3257 /* Set the parenth number. */
3258 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3260 if (OP(nxt) != CLOSE)
3261 FAIL("Panic opt close");
3262 oscan->flags = (U8)ARG(nxt);
3263 if (RExC_open_parens) {
3264 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3265 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3267 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3268 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3271 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3272 OP(nxt + 1) = OPTIMIZED; /* was count. */
3273 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3274 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3277 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3278 regnode *nnxt = regnext(nxt1);
3281 if (reg_off_by_arg[OP(nxt1)])
3282 ARG_SET(nxt1, nxt2 - nxt1);
3283 else if (nxt2 - nxt1 < U16_MAX)
3284 NEXT_OFF(nxt1) = nxt2 - nxt1;
3286 OP(nxt) = NOTHING; /* Cannot beautify */
3291 /* Optimize again: */
3292 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3293 NULL, stopparen, recursed, NULL, 0,depth+1);
3298 else if ((OP(oscan) == CURLYX)
3299 && (flags & SCF_WHILEM_VISITED_POS)
3300 /* See the comment on a similar expression above.
3301 However, this time it not a subexpression
3302 we care about, but the expression itself. */
3303 && (maxcount == REG_INFTY)
3304 && data && ++data->whilem_c < 16) {
3305 /* This stays as CURLYX, we can put the count/of pair. */
3306 /* Find WHILEM (as in regexec.c) */
3307 regnode *nxt = oscan + NEXT_OFF(oscan);
3309 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3311 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3312 | (RExC_whilem_seen << 4)); /* On WHILEM */
3314 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3316 if (flags & SCF_DO_SUBSTR) {
3317 SV *last_str = NULL;
3318 int counted = mincount != 0;
3320 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3321 #if defined(SPARC64_GCC_WORKAROUND)
3324 const char *s = NULL;
3327 if (pos_before >= data->last_start_min)
3330 b = data->last_start_min;
3333 s = SvPV_const(data->last_found, l);
3334 old = b - data->last_start_min;
3337 I32 b = pos_before >= data->last_start_min
3338 ? pos_before : data->last_start_min;
3340 const char * const s = SvPV_const(data->last_found, l);
3341 I32 old = b - data->last_start_min;
3345 old = utf8_hop((U8*)s, old) - (U8*)s;
3348 /* Get the added string: */
3349 last_str = newSVpvn_utf8(s + old, l, UTF);
3350 if (deltanext == 0 && pos_before == b) {
3351 /* What was added is a constant string */
3353 SvGROW(last_str, (mincount * l) + 1);
3354 repeatcpy(SvPVX(last_str) + l,
3355 SvPVX_const(last_str), l, mincount - 1);
3356 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3357 /* Add additional parts. */
3358 SvCUR_set(data->last_found,
3359 SvCUR(data->last_found) - l);
3360 sv_catsv(data->last_found, last_str);
3362 SV * sv = data->last_found;
3364 SvUTF8(sv) && SvMAGICAL(sv) ?
3365 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3366 if (mg && mg->mg_len >= 0)
3367 mg->mg_len += CHR_SVLEN(last_str) - l;
3369 data->last_end += l * (mincount - 1);
3372 /* start offset must point into the last copy */
3373 data->last_start_min += minnext * (mincount - 1);
3374 data->last_start_max += is_inf ? I32_MAX
3375 : (maxcount - 1) * (minnext + data->pos_delta);
3378 /* It is counted once already... */
3379 data->pos_min += minnext * (mincount - counted);
3380 data->pos_delta += - counted * deltanext +
3381 (minnext + deltanext) * maxcount - minnext * mincount;
3382 if (mincount != maxcount) {
3383 /* Cannot extend fixed substrings found inside
3385 SCAN_COMMIT(pRExC_state,data,minlenp);
3386 if (mincount && last_str) {
3387 SV * const sv = data->last_found;
3388 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3389 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3393 sv_setsv(sv, last_str);
3394 data->last_end = data->pos_min;
3395 data->last_start_min =
3396 data->pos_min - CHR_SVLEN(last_str);
3397 data->last_start_max = is_inf
3399 : data->pos_min + data->pos_delta
3400 - CHR_SVLEN(last_str);
3402 data->longest = &(data->longest_float);
3404 SvREFCNT_dec(last_str);
3406 if (data && (fl & SF_HAS_EVAL))
3407 data->flags |= SF_HAS_EVAL;
3408 optimize_curly_tail:
3409 if (OP(oscan) != CURLYX) {
3410 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3412 NEXT_OFF(oscan) += NEXT_OFF(next);
3415 default: /* REF and CLUMP only? */
3416 if (flags & SCF_DO_SUBSTR) {
3417 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3418 data->longest = &(data->longest_float);
3420 is_inf = is_inf_internal = 1;
3421 if (flags & SCF_DO_STCLASS_OR)
3422 cl_anything(pRExC_state, data->start_class);
3423 flags &= ~SCF_DO_STCLASS;
3427 else if (OP(scan) == LNBREAK) {
3428 if (flags & SCF_DO_STCLASS) {
3430 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3431 if (flags & SCF_DO_STCLASS_AND) {
3432 for (value = 0; value < 256; value++)
3433 if (!is_VERTWS_cp(value))
3434 ANYOF_BITMAP_CLEAR(data->start_class, value);
3437 for (value = 0; value < 256; value++)
3438 if (is_VERTWS_cp(value))
3439 ANYOF_BITMAP_SET(data->start_class, value);
3441 if (flags & SCF_DO_STCLASS_OR)
3442 cl_and(data->start_class, and_withp);
3443 flags &= ~SCF_DO_STCLASS;
3447 if (flags & SCF_DO_SUBSTR) {
3448 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3450 data->pos_delta += 1;
3451 data->longest = &(data->longest_float);
3455 else if (OP(scan) == FOLDCHAR) {
3456 int d = ARG(scan)==0xDF ? 1 : 2;
3457 flags &= ~SCF_DO_STCLASS;
3460 if (flags & SCF_DO_SUBSTR) {
3461 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3463 data->pos_delta += d;
3464 data->longest = &(data->longest_float);
3467 else if (strchr((const char*)PL_simple,OP(scan))) {
3470 if (flags & SCF_DO_SUBSTR) {
3471 SCAN_COMMIT(pRExC_state,data,minlenp);
3475 if (flags & SCF_DO_STCLASS) {
3476 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3478 /* Some of the logic below assumes that switching
3479 locale on will only add false positives. */
3480 switch (PL_regkind[OP(scan)]) {
3484 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3485 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3486 cl_anything(pRExC_state, data->start_class);
3489 if (OP(scan) == SANY)
3491 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3492 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3493 || (data->start_class->flags & ANYOF_CLASS));
3494 cl_anything(pRExC_state, data->start_class);
3496 if (flags & SCF_DO_STCLASS_AND || !value)
3497 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3500 if (flags & SCF_DO_STCLASS_AND)
3501 cl_and(data->start_class,
3502 (struct regnode_charclass_class*)scan);
3504 cl_or(pRExC_state, data->start_class,
3505 (struct regnode_charclass_class*)scan);
3508 if (flags & SCF_DO_STCLASS_AND) {
3509 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3510 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3511 for (value = 0; value < 256; value++)
3512 if (!isALNUM(value))
3513 ANYOF_BITMAP_CLEAR(data->start_class, value);
3517 if (data->start_class->flags & ANYOF_LOCALE)
3518 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3520 for (value = 0; value < 256; value++)
3522 ANYOF_BITMAP_SET(data->start_class, value);
3527 if (flags & SCF_DO_STCLASS_AND) {
3528 if (data->start_class->flags & ANYOF_LOCALE)
3529 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3532 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3533 data->start_class->flags |= ANYOF_LOCALE;
3537 if (flags & SCF_DO_STCLASS_AND) {
3538 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3539 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3540 for (value = 0; value < 256; value++)
3542 ANYOF_BITMAP_CLEAR(data->start_class, value);
3546 if (data->start_class->flags & ANYOF_LOCALE)
3547 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3549 for (value = 0; value < 256; value++)
3550 if (!isALNUM(value))
3551 ANYOF_BITMAP_SET(data->start_class, value);
3556 if (flags & SCF_DO_STCLASS_AND) {
3557 if (data->start_class->flags & ANYOF_LOCALE)
3558 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3561 data->start_class->flags |= ANYOF_LOCALE;
3562 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3566 if (flags & SCF_DO_STCLASS_AND) {
3567 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3568 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3569 for (value = 0; value < 256; value++)
3570 if (!isSPACE(value))
3571 ANYOF_BITMAP_CLEAR(data->start_class, value);
3575 if (data->start_class->flags & ANYOF_LOCALE)
3576 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3578 for (value = 0; value < 256; value++)
3580 ANYOF_BITMAP_SET(data->start_class, value);
3585 if (flags & SCF_DO_STCLASS_AND) {
3586 if (data->start_class->flags & ANYOF_LOCALE)
3587 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3590 data->start_class->flags |= ANYOF_LOCALE;
3591 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3595 if (flags & SCF_DO_STCLASS_AND) {
3596 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3597 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3598 for (value = 0; value < 256; value++)
3600 ANYOF_BITMAP_CLEAR(data->start_class, value);
3604 if (data->start_class->flags & ANYOF_LOCALE)
3605 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3607 for (value = 0; value < 256; value++)
3608 if (!isSPACE(value))
3609 ANYOF_BITMAP_SET(data->start_class, value);
3614 if (flags & SCF_DO_STCLASS_AND) {
3615 if (data->start_class->flags & ANYOF_LOCALE) {
3616 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3617 for (value = 0; value < 256; value++)
3618 if (!isSPACE(value))
3619 ANYOF_BITMAP_CLEAR(data->start_class, value);
3623 data->start_class->flags |= ANYOF_LOCALE;
3624 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3628 if (flags & SCF_DO_STCLASS_AND) {
3629 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3630 for (value = 0; value < 256; value++)
3631 if (!isDIGIT(value))
3632 ANYOF_BITMAP_CLEAR(data->start_class, value);
3635 if (data->start_class->flags & ANYOF_LOCALE)
3636 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3638 for (value = 0; value < 256; value++)
3640 ANYOF_BITMAP_SET(data->start_class, value);
3645 if (flags & SCF_DO_STCLASS_AND) {
3646 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3647 for (value = 0; value < 256; value++)
3649 ANYOF_BITMAP_CLEAR(data->start_class, value);
3652 if (data->start_class->flags & ANYOF_LOCALE)
3653 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3655 for (value = 0; value < 256; value++)
3656 if (!isDIGIT(value))
3657 ANYOF_BITMAP_SET(data->start_class, value);
3661 CASE_SYNST_FNC(VERTWS);
3662 CASE_SYNST_FNC(HORIZWS);
3665 if (flags & SCF_DO_STCLASS_OR)
3666 cl_and(data->start_class, and_withp);
3667 flags &= ~SCF_DO_STCLASS;
3670 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3671 data->flags |= (OP(scan) == MEOL
3675 else if ( PL_regkind[OP(scan)] == BRANCHJ
3676 /* Lookbehind, or need to calculate parens/evals/stclass: */
3677 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3678 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3679 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3680 || OP(scan) == UNLESSM )
3682 /* Negative Lookahead/lookbehind
3683 In this case we can't do fixed string optimisation.
3686 I32 deltanext, minnext, fake = 0;
3688 struct regnode_charclass_class intrnl;
3691 data_fake.flags = 0;
3693 data_fake.whilem_c = data->whilem_c;
3694 data_fake.last_closep = data->last_closep;
3697 data_fake.last_closep = &fake;
3698 data_fake.pos_delta = delta;
3699 if ( flags & SCF_DO_STCLASS && !scan->flags
3700 && OP(scan) == IFMATCH ) { /* Lookahead */
3701 cl_init(pRExC_state, &intrnl);
3702 data_fake.start_class = &intrnl;
3703 f |= SCF_DO_STCLASS_AND;
3705 if (flags & SCF_WHILEM_VISITED_POS)
3706 f |= SCF_WHILEM_VISITED_POS;
3707 next = regnext(scan);
3708 nscan = NEXTOPER(NEXTOPER(scan));
3709 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3710 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3713 FAIL("Variable length lookbehind not implemented");
3715 else if (minnext > (I32)U8_MAX) {
3716 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3718 scan->flags = (U8)minnext;
3721 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3723 if (data_fake.flags & SF_HAS_EVAL)
3724 data->flags |= SF_HAS_EVAL;
3725 data->whilem_c = data_fake.whilem_c;
3727 if (f & SCF_DO_STCLASS_AND) {
3728 const int was = (data->start_class->flags & ANYOF_EOS);
3730 cl_and(data->start_class, &intrnl);
3732 data->start_class->flags |= ANYOF_EOS;
3735 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3737 /* Positive Lookahead/lookbehind
3738 In this case we can do fixed string optimisation,
3739 but we must be careful about it. Note in the case of
3740 lookbehind the positions will be offset by the minimum
3741 length of the pattern, something we won't know about
3742 until after the recurse.
3744 I32 deltanext, fake = 0;
3746 struct regnode_charclass_class intrnl;
3748 /* We use SAVEFREEPV so that when the full compile
3749 is finished perl will clean up the allocated
3750 minlens when its all done. This was we don't
3751 have to worry about freeing them when we know
3752 they wont be used, which would be a pain.
3755 Newx( minnextp, 1, I32 );
3756 SAVEFREEPV(minnextp);
3759 StructCopy(data, &data_fake, scan_data_t);
3760 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3763 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3764 data_fake.last_found=newSVsv(data->last_found);
3768 data_fake.last_closep = &fake;
3769 data_fake.flags = 0;
3770 data_fake.pos_delta = delta;
3772 data_fake.flags |= SF_IS_INF;
3773 if ( flags & SCF_DO_STCLASS && !scan->flags
3774 && OP(scan) == IFMATCH ) { /* Lookahead */
3775 cl_init(pRExC_state, &intrnl);
3776 data_fake.start_class = &intrnl;
3777 f |= SCF_DO_STCLASS_AND;
3779 if (flags & SCF_WHILEM_VISITED_POS)
3780 f |= SCF_WHILEM_VISITED_POS;
3781 next = regnext(scan);
3782 nscan = NEXTOPER(NEXTOPER(scan));
3784 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3785 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3788 FAIL("Variable length lookbehind not implemented");
3790 else if (*minnextp > (I32)U8_MAX) {
3791 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3793 scan->flags = (U8)*minnextp;
3798 if (f & SCF_DO_STCLASS_AND) {
3799 const int was = (data->start_class->flags & ANYOF_EOS);
3801 cl_and(data->start_class, &intrnl);
3803 data->start_class->flags |= ANYOF_EOS;
3806 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3808 if (data_fake.flags & SF_HAS_EVAL)
3809 data->flags |= SF_HAS_EVAL;
3810 data->whilem_c = data_fake.whilem_c;
3811 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3812 if (RExC_rx->minlen<*minnextp)
https://perl5.git.perl.org / perl5.git / blob