5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
84 EXTERN_C const struct regexp_engine my_reg_engine;
89 #include "dquote_static.c"
90 #include "charclass_invlists.h"
91 #include "inline_invlist.c"
92 #include "unicode_constants.h"
94 #define HAS_NONLATIN1_FOLD_CLOSURE(i) \
95 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
96 #define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
97 _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
98 #define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
99 #define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
102 #define STATIC static
106 #define MIN(a,b) ((a) < (b) ? (a) : (b))
109 /* this is a chain of data about sub patterns we are processing that
110 need to be handled separately/specially in study_chunk. Its so
111 we can simulate recursion without losing state. */
113 typedef struct scan_frame {
114 regnode *last_regnode; /* last node to process in this frame */
115 regnode *next_regnode; /* next node to process when last is reached */
116 U32 prev_recursed_depth;
117 I32 stopparen; /* what stopparen do we use */
118 U32 is_top_frame; /* what flags do we use? */
120 struct scan_frame *this_prev_frame; /* this previous frame */
121 struct scan_frame *prev_frame; /* previous frame */
122 struct scan_frame *next_frame; /* next frame */
125 /* Certain characters are output as a sequence with the first being a
127 #define isBACKSLASHED_PUNCT(c) \
128 ((c) == '-' || (c) == ']' || (c) == '\\' || (c) == '^')
131 struct RExC_state_t {
132 U32 flags; /* RXf_* are we folding, multilining? */
133 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
134 char *precomp; /* uncompiled string. */
135 REGEXP *rx_sv; /* The SV that is the regexp. */
136 regexp *rx; /* perl core regexp structure */
137 regexp_internal *rxi; /* internal data for regexp object
139 char *start; /* Start of input for compile */
140 char *end; /* End of input for compile */
141 char *parse; /* Input-scan pointer. */
142 SSize_t whilem_seen; /* number of WHILEM in this expr */
143 regnode *emit_start; /* Start of emitted-code area */
144 regnode *emit_bound; /* First regnode outside of the
146 regnode *emit; /* Code-emit pointer; if = &emit_dummy,
147 implies compiling, so don't emit */
148 regnode_ssc emit_dummy; /* placeholder for emit to point to;
149 large enough for the largest
150 non-EXACTish node, so can use it as
152 I32 naughty; /* How bad is this pattern? */
153 I32 sawback; /* Did we see \1, ...? */
155 SSize_t size; /* Code size. */
156 I32 npar; /* Capture buffer count, (OPEN) plus
157 one. ("par" 0 is the whole
159 I32 nestroot; /* root parens we are in - used by
163 regnode **open_parens; /* pointers to open parens */
164 regnode **close_parens; /* pointers to close parens */
165 regnode *opend; /* END node in program */
166 I32 utf8; /* whether the pattern is utf8 or not */
167 I32 orig_utf8; /* whether the pattern was originally in utf8 */
168 /* XXX use this for future optimisation of case
169 * where pattern must be upgraded to utf8. */
170 I32 uni_semantics; /* If a d charset modifier should use unicode
171 rules, even if the pattern is not in
173 HV *paren_names; /* Paren names */
175 regnode **recurse; /* Recurse regops */
176 I32 recurse_count; /* Number of recurse regops */
177 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
179 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
183 I32 override_recoding;
184 I32 in_multi_char_class;
185 struct reg_code_block *code_blocks; /* positions of literal (?{})
187 int num_code_blocks; /* size of code_blocks[] */
188 int code_index; /* next code_blocks[] slot */
189 SSize_t maxlen; /* mininum possible number of chars in string to match */
190 scan_frame *frame_head;
191 scan_frame *frame_last;
194 #ifdef ADD_TO_REGEXEC
195 char *starttry; /* -Dr: where regtry was called. */
196 #define RExC_starttry (pRExC_state->starttry)
198 SV *runtime_code_qr; /* qr with the runtime code blocks */
200 const char *lastparse;
202 AV *paren_name_list; /* idx -> name */
203 U32 study_chunk_recursed_count;
206 #define RExC_lastparse (pRExC_state->lastparse)
207 #define RExC_lastnum (pRExC_state->lastnum)
208 #define RExC_paren_name_list (pRExC_state->paren_name_list)
209 #define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
210 #define RExC_mysv (pRExC_state->mysv1)
211 #define RExC_mysv1 (pRExC_state->mysv1)
212 #define RExC_mysv2 (pRExC_state->mysv2)
217 #define RExC_flags (pRExC_state->flags)
218 #define RExC_pm_flags (pRExC_state->pm_flags)
219 #define RExC_precomp (pRExC_state->precomp)
220 #define RExC_rx_sv (pRExC_state->rx_sv)
221 #define RExC_rx (pRExC_state->rx)
222 #define RExC_rxi (pRExC_state->rxi)
223 #define RExC_start (pRExC_state->start)
224 #define RExC_end (pRExC_state->end)
225 #define RExC_parse (pRExC_state->parse)
226 #define RExC_whilem_seen (pRExC_state->whilem_seen)
227 #ifdef RE_TRACK_PATTERN_OFFSETS
228 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the
231 #define RExC_emit (pRExC_state->emit)
232 #define RExC_emit_dummy (pRExC_state->emit_dummy)
233 #define RExC_emit_start (pRExC_state->emit_start)
234 #define RExC_emit_bound (pRExC_state->emit_bound)
235 #define RExC_sawback (pRExC_state->sawback)
236 #define RExC_seen (pRExC_state->seen)
237 #define RExC_size (pRExC_state->size)
238 #define RExC_maxlen (pRExC_state->maxlen)
239 #define RExC_npar (pRExC_state->npar)
240 #define RExC_nestroot (pRExC_state->nestroot)
241 #define RExC_extralen (pRExC_state->extralen)
242 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
243 #define RExC_utf8 (pRExC_state->utf8)
244 #define RExC_uni_semantics (pRExC_state->uni_semantics)
245 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
246 #define RExC_open_parens (pRExC_state->open_parens)
247 #define RExC_close_parens (pRExC_state->close_parens)
248 #define RExC_opend (pRExC_state->opend)
249 #define RExC_paren_names (pRExC_state->paren_names)
250 #define RExC_recurse (pRExC_state->recurse)
251 #define RExC_recurse_count (pRExC_state->recurse_count)
252 #define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
253 #define RExC_study_chunk_recursed_bytes \
254 (pRExC_state->study_chunk_recursed_bytes)
255 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
256 #define RExC_contains_locale (pRExC_state->contains_locale)
257 #define RExC_contains_i (pRExC_state->contains_i)
258 #define RExC_override_recoding (pRExC_state->override_recoding)
259 #define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
260 #define RExC_frame_head (pRExC_state->frame_head)
261 #define RExC_frame_last (pRExC_state->frame_last)
262 #define RExC_frame_count (pRExC_state->frame_count)
263 #define RExC_strict (pRExC_state->strict)
265 /* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
266 * a flag to disable back-off on the fixed/floating substrings - if it's
267 * a high complexity pattern we assume the benefit of avoiding a full match
268 * is worth the cost of checking for the substrings even if they rarely help.
270 #define RExC_naughty (pRExC_state->naughty)
271 #define TOO_NAUGHTY (10)
272 #define MARK_NAUGHTY(add) \
273 if (RExC_naughty < TOO_NAUGHTY) \
274 RExC_naughty += (add)
275 #define MARK_NAUGHTY_EXP(exp, add) \
276 if (RExC_naughty < TOO_NAUGHTY) \
277 RExC_naughty += RExC_naughty / (exp) + (add)
279 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
280 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
281 ((*s) == '{' && regcurly(s)))
284 * Flags to be passed up and down.
286 #define WORST 0 /* Worst case. */
287 #define HASWIDTH 0x01 /* Known to match non-null strings. */
289 /* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
290 * character. (There needs to be a case: in the switch statement in regexec.c
291 * for any node marked SIMPLE.) Note that this is not the same thing as
294 #define SPSTART 0x04 /* Starts with * or + */
295 #define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
296 #define TRYAGAIN 0x10 /* Weeded out a declaration. */
297 #define RESTART_UTF8 0x20 /* Restart, need to calcuate sizes as UTF-8 */
299 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
301 /* whether trie related optimizations are enabled */
302 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
303 #define TRIE_STUDY_OPT
304 #define FULL_TRIE_STUDY
310 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
311 #define PBITVAL(paren) (1 << ((paren) & 7))
312 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
313 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
314 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
316 #define REQUIRE_UTF8 STMT_START { \
318 *flagp = RESTART_UTF8; \
323 /* This converts the named class defined in regcomp.h to its equivalent class
324 * number defined in handy.h. */
325 #define namedclass_to_classnum(class) ((int) ((class) / 2))
326 #define classnum_to_namedclass(classnum) ((classnum) * 2)
328 #define _invlist_union_complement_2nd(a, b, output) \
329 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
330 #define _invlist_intersection_complement_2nd(a, b, output) \
331 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
333 /* About scan_data_t.
335 During optimisation we recurse through the regexp program performing
336 various inplace (keyhole style) optimisations. In addition study_chunk
337 and scan_commit populate this data structure with information about
338 what strings MUST appear in the pattern. We look for the longest
339 string that must appear at a fixed location, and we look for the
340 longest string that may appear at a floating location. So for instance
345 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
346 strings (because they follow a .* construct). study_chunk will identify
347 both FOO and BAR as being the longest fixed and floating strings respectively.
349 The strings can be composites, for instance
353 will result in a composite fixed substring 'foo'.
355 For each string some basic information is maintained:
357 - offset or min_offset
358 This is the position the string must appear at, or not before.
359 It also implicitly (when combined with minlenp) tells us how many
360 characters must match before the string we are searching for.
361 Likewise when combined with minlenp and the length of the string it
362 tells us how many characters must appear after the string we have
366 Only used for floating strings. This is the rightmost point that
367 the string can appear at. If set to SSize_t_MAX it indicates that the
368 string can occur infinitely far to the right.
371 A pointer to the minimum number of characters of the pattern that the
372 string was found inside. This is important as in the case of positive
373 lookahead or positive lookbehind we can have multiple patterns
378 The minimum length of the pattern overall is 3, the minimum length
379 of the lookahead part is 3, but the minimum length of the part that
380 will actually match is 1. So 'FOO's minimum length is 3, but the
381 minimum length for the F is 1. This is important as the minimum length
382 is used to determine offsets in front of and behind the string being
383 looked for. Since strings can be composites this is the length of the
384 pattern at the time it was committed with a scan_commit. Note that
385 the length is calculated by study_chunk, so that the minimum lengths
386 are not known until the full pattern has been compiled, thus the
387 pointer to the value.
391 In the case of lookbehind the string being searched for can be
392 offset past the start point of the final matching string.
393 If this value was just blithely removed from the min_offset it would
394 invalidate some of the calculations for how many chars must match
395 before or after (as they are derived from min_offset and minlen and
396 the length of the string being searched for).
397 When the final pattern is compiled and the data is moved from the
398 scan_data_t structure into the regexp structure the information
399 about lookbehind is factored in, with the information that would
400 have been lost precalculated in the end_shift field for the
403 The fields pos_min and pos_delta are used to store the minimum offset
404 and the delta to the maximum offset at the current point in the pattern.
408 typedef struct scan_data_t {
409 /*I32 len_min; unused */
410 /*I32 len_delta; unused */
414 SSize_t last_end; /* min value, <0 unless valid. */
415 SSize_t last_start_min;
416 SSize_t last_start_max;
417 SV **longest; /* Either &l_fixed, or &l_float. */
418 SV *longest_fixed; /* longest fixed string found in pattern */
419 SSize_t offset_fixed; /* offset where it starts */
420 SSize_t *minlen_fixed; /* pointer to the minlen relevant to the string */
421 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
422 SV *longest_float; /* longest floating string found in pattern */
423 SSize_t offset_float_min; /* earliest point in string it can appear */
424 SSize_t offset_float_max; /* latest point in string it can appear */
425 SSize_t *minlen_float; /* pointer to the minlen relevant to the string */
426 SSize_t lookbehind_float; /* is the pos of the string modified by LB */
429 SSize_t *last_closep;
430 regnode_ssc *start_class;
434 * Forward declarations for pregcomp()'s friends.
437 static const scan_data_t zero_scan_data =
438 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
440 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
441 #define SF_BEFORE_SEOL 0x0001
442 #define SF_BEFORE_MEOL 0x0002
443 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
444 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
446 #define SF_FIX_SHIFT_EOL (+2)
447 #define SF_FL_SHIFT_EOL (+4)
449 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
450 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
452 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
453 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
454 #define SF_IS_INF 0x0040
455 #define SF_HAS_PAR 0x0080
456 #define SF_IN_PAR 0x0100
457 #define SF_HAS_EVAL 0x0200
458 #define SCF_DO_SUBSTR 0x0400
459 #define SCF_DO_STCLASS_AND 0x0800
460 #define SCF_DO_STCLASS_OR 0x1000
461 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
462 #define SCF_WHILEM_VISITED_POS 0x2000
464 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
465 #define SCF_SEEN_ACCEPT 0x8000
466 #define SCF_TRIE_DOING_RESTUDY 0x10000
467 #define SCF_IN_DEFINE 0x20000
472 #define UTF cBOOL(RExC_utf8)
474 /* The enums for all these are ordered so things work out correctly */
475 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
476 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
477 == REGEX_DEPENDS_CHARSET)
478 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
479 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
480 >= REGEX_UNICODE_CHARSET)
481 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
482 == REGEX_ASCII_RESTRICTED_CHARSET)
483 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
484 >= REGEX_ASCII_RESTRICTED_CHARSET)
485 #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
486 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
488 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
490 /* For programs that want to be strictly Unicode compatible by dying if any
491 * attempt is made to match a non-Unicode code point against a Unicode
493 #define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
495 #define OOB_NAMEDCLASS -1
497 /* There is no code point that is out-of-bounds, so this is problematic. But
498 * its only current use is to initialize a variable that is always set before
500 #define OOB_UNICODE 0xDEADBEEF
502 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
503 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
506 /* length of regex to show in messages that don't mark a position within */
507 #define RegexLengthToShowInErrorMessages 127
510 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
511 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
512 * op/pragma/warn/regcomp.
514 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
515 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
517 #define REPORT_LOCATION " in regex; marked by " MARKER1 \
518 " in m/%"UTF8f MARKER2 "%"UTF8f"/"
520 #define REPORT_LOCATION_ARGS(offset) \
521 UTF8fARG(UTF, offset, RExC_precomp), \
522 UTF8fARG(UTF, RExC_end - RExC_precomp - offset, RExC_precomp + offset)
525 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
526 * arg. Show regex, up to a maximum length. If it's too long, chop and add
529 #define _FAIL(code) STMT_START { \
530 const char *ellipses = ""; \
531 IV len = RExC_end - RExC_precomp; \
534 SAVEFREESV(RExC_rx_sv); \
535 if (len > RegexLengthToShowInErrorMessages) { \
536 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
537 len = RegexLengthToShowInErrorMessages - 10; \
543 #define FAIL(msg) _FAIL( \
544 Perl_croak(aTHX_ "%s in regex m/%"UTF8f"%s/", \
545 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
547 #define FAIL2(msg,arg) _FAIL( \
548 Perl_croak(aTHX_ msg " in regex m/%"UTF8f"%s/", \
549 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
552 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
554 #define Simple_vFAIL(m) STMT_START { \
556 (RExC_parse > RExC_end ? RExC_end : RExC_parse) - RExC_precomp; \
557 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
558 m, REPORT_LOCATION_ARGS(offset)); \
562 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
564 #define vFAIL(m) STMT_START { \
566 SAVEFREESV(RExC_rx_sv); \
571 * Like Simple_vFAIL(), but accepts two arguments.
573 #define Simple_vFAIL2(m,a1) STMT_START { \
574 const IV offset = RExC_parse - RExC_precomp; \
575 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
576 REPORT_LOCATION_ARGS(offset)); \
580 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
582 #define vFAIL2(m,a1) STMT_START { \
584 SAVEFREESV(RExC_rx_sv); \
585 Simple_vFAIL2(m, a1); \
590 * Like Simple_vFAIL(), but accepts three arguments.
592 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
593 const IV offset = RExC_parse - RExC_precomp; \
594 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
595 REPORT_LOCATION_ARGS(offset)); \
599 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
601 #define vFAIL3(m,a1,a2) STMT_START { \
603 SAVEFREESV(RExC_rx_sv); \
604 Simple_vFAIL3(m, a1, a2); \
608 * Like Simple_vFAIL(), but accepts four arguments.
610 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
611 const IV offset = RExC_parse - RExC_precomp; \
612 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
613 REPORT_LOCATION_ARGS(offset)); \
616 #define vFAIL4(m,a1,a2,a3) STMT_START { \
618 SAVEFREESV(RExC_rx_sv); \
619 Simple_vFAIL4(m, a1, a2, a3); \
622 /* A specialized version of vFAIL2 that works with UTF8f */
623 #define vFAIL2utf8f(m, a1) STMT_START { \
624 const IV offset = RExC_parse - RExC_precomp; \
626 SAVEFREESV(RExC_rx_sv); \
627 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
628 REPORT_LOCATION_ARGS(offset)); \
631 /* These have asserts in them because of [perl #122671] Many warnings in
632 * regcomp.c can occur twice. If they get output in pass1 and later in that
633 * pass, the pattern has to be converted to UTF-8 and the pass restarted, they
634 * would get output again. So they should be output in pass2, and these
635 * asserts make sure new warnings follow that paradigm. */
637 /* m is not necessarily a "literal string", in this macro */
638 #define reg_warn_non_literal_string(loc, m) STMT_START { \
639 const IV offset = loc - RExC_precomp; \
640 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
641 m, REPORT_LOCATION_ARGS(offset)); \
644 #define ckWARNreg(loc,m) STMT_START { \
645 const IV offset = loc - RExC_precomp; \
646 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
647 REPORT_LOCATION_ARGS(offset)); \
650 #define vWARN(loc, m) STMT_START { \
651 const IV offset = loc - RExC_precomp; \
652 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
653 REPORT_LOCATION_ARGS(offset)); \
656 #define vWARN_dep(loc, m) STMT_START { \
657 const IV offset = loc - RExC_precomp; \
658 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), m REPORT_LOCATION, \
659 REPORT_LOCATION_ARGS(offset)); \
662 #define ckWARNdep(loc,m) STMT_START { \
663 const IV offset = loc - RExC_precomp; \
664 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
666 REPORT_LOCATION_ARGS(offset)); \
669 #define ckWARNregdep(loc,m) STMT_START { \
670 const IV offset = loc - RExC_precomp; \
671 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
673 REPORT_LOCATION_ARGS(offset)); \
676 #define ckWARN2reg_d(loc,m, a1) STMT_START { \
677 const IV offset = loc - RExC_precomp; \
678 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
680 a1, REPORT_LOCATION_ARGS(offset)); \
683 #define ckWARN2reg(loc, m, a1) STMT_START { \
684 const IV offset = loc - RExC_precomp; \
685 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
686 a1, REPORT_LOCATION_ARGS(offset)); \
689 #define vWARN3(loc, m, a1, a2) STMT_START { \
690 const IV offset = loc - RExC_precomp; \
691 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
692 a1, a2, REPORT_LOCATION_ARGS(offset)); \
695 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
696 const IV offset = loc - RExC_precomp; \
697 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
698 a1, a2, REPORT_LOCATION_ARGS(offset)); \
701 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
702 const IV offset = loc - RExC_precomp; \
703 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
704 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
707 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
708 const IV offset = loc - RExC_precomp; \
709 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
710 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
713 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
714 const IV offset = loc - RExC_precomp; \
715 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
716 a1, a2, a3, a4, REPORT_LOCATION_ARGS(offset)); \
719 /* Macros for recording node offsets. 20001227 mjd@plover.com
720 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
721 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
722 * Element 0 holds the number n.
723 * Position is 1 indexed.
725 #ifndef RE_TRACK_PATTERN_OFFSETS
726 #define Set_Node_Offset_To_R(node,byte)
727 #define Set_Node_Offset(node,byte)
728 #define Set_Cur_Node_Offset
729 #define Set_Node_Length_To_R(node,len)
730 #define Set_Node_Length(node,len)
731 #define Set_Node_Cur_Length(node,start)
732 #define Node_Offset(n)
733 #define Node_Length(n)
734 #define Set_Node_Offset_Length(node,offset,len)
735 #define ProgLen(ri) ri->u.proglen
736 #define SetProgLen(ri,x) ri->u.proglen = x
738 #define ProgLen(ri) ri->u.offsets[0]
739 #define SetProgLen(ri,x) ri->u.offsets[0] = x
740 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
742 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
743 __LINE__, (int)(node), (int)(byte))); \
745 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
748 RExC_offsets[2*(node)-1] = (byte); \
753 #define Set_Node_Offset(node,byte) \
754 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
755 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
757 #define Set_Node_Length_To_R(node,len) STMT_START { \
759 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
760 __LINE__, (int)(node), (int)(len))); \
762 Perl_croak(aTHX_ "value of node is %d in Length macro", \
765 RExC_offsets[2*(node)] = (len); \
770 #define Set_Node_Length(node,len) \
771 Set_Node_Length_To_R((node)-RExC_emit_start, len)
772 #define Set_Node_Cur_Length(node, start) \
773 Set_Node_Length(node, RExC_parse - start)
775 /* Get offsets and lengths */
776 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
777 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
779 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
780 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
781 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
785 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
786 #define EXPERIMENTAL_INPLACESCAN
787 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
789 #define DEBUG_RExC_seen() \
790 DEBUG_OPTIMISE_MORE_r({ \
791 PerlIO_printf(Perl_debug_log,"RExC_seen: "); \
793 if (RExC_seen & REG_ZERO_LEN_SEEN) \
794 PerlIO_printf(Perl_debug_log,"REG_ZERO_LEN_SEEN "); \
796 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
797 PerlIO_printf(Perl_debug_log,"REG_LOOKBEHIND_SEEN "); \
799 if (RExC_seen & REG_GPOS_SEEN) \
800 PerlIO_printf(Perl_debug_log,"REG_GPOS_SEEN "); \
802 if (RExC_seen & REG_CANY_SEEN) \
803 PerlIO_printf(Perl_debug_log,"REG_CANY_SEEN "); \
805 if (RExC_seen & REG_RECURSE_SEEN) \
806 PerlIO_printf(Perl_debug_log,"REG_RECURSE_SEEN "); \
808 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
809 PerlIO_printf(Perl_debug_log,"REG_TOP_LEVEL_BRANCHES_SEEN "); \
811 if (RExC_seen & REG_VERBARG_SEEN) \
812 PerlIO_printf(Perl_debug_log,"REG_VERBARG_SEEN "); \
814 if (RExC_seen & REG_CUTGROUP_SEEN) \
815 PerlIO_printf(Perl_debug_log,"REG_CUTGROUP_SEEN "); \
817 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
818 PerlIO_printf(Perl_debug_log,"REG_RUN_ON_COMMENT_SEEN "); \
820 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
821 PerlIO_printf(Perl_debug_log,"REG_UNFOLDED_MULTI_SEEN "); \
823 if (RExC_seen & REG_GOSTART_SEEN) \
824 PerlIO_printf(Perl_debug_log,"REG_GOSTART_SEEN "); \
826 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
827 PerlIO_printf(Perl_debug_log,"REG_UNBOUNDED_QUANTIFIER_SEEN "); \
829 PerlIO_printf(Perl_debug_log,"\n"); \
832 #define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
833 if ((flags) & flag) PerlIO_printf(Perl_debug_log, "%s ", #flag)
835 #define DEBUG_SHOW_STUDY_FLAGS(flags,open_str,close_str) \
837 PerlIO_printf(Perl_debug_log, "%s", open_str); \
838 DEBUG_SHOW_STUDY_FLAG(flags,SF_FL_BEFORE_SEOL); \
839 DEBUG_SHOW_STUDY_FLAG(flags,SF_FL_BEFORE_MEOL); \
840 DEBUG_SHOW_STUDY_FLAG(flags,SF_IS_INF); \
841 DEBUG_SHOW_STUDY_FLAG(flags,SF_HAS_PAR); \
842 DEBUG_SHOW_STUDY_FLAG(flags,SF_IN_PAR); \
843 DEBUG_SHOW_STUDY_FLAG(flags,SF_HAS_EVAL); \
844 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_SUBSTR); \
845 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS_AND); \
846 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS_OR); \
847 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS); \
848 DEBUG_SHOW_STUDY_FLAG(flags,SCF_WHILEM_VISITED_POS); \
849 DEBUG_SHOW_STUDY_FLAG(flags,SCF_TRIE_RESTUDY); \
850 DEBUG_SHOW_STUDY_FLAG(flags,SCF_SEEN_ACCEPT); \
851 DEBUG_SHOW_STUDY_FLAG(flags,SCF_TRIE_DOING_RESTUDY); \
852 DEBUG_SHOW_STUDY_FLAG(flags,SCF_IN_DEFINE); \
853 PerlIO_printf(Perl_debug_log, "%s", close_str); \
857 #define DEBUG_STUDYDATA(str,data,depth) \
858 DEBUG_OPTIMISE_MORE_r(if(data){ \
859 PerlIO_printf(Perl_debug_log, \
860 "%*s" str "Pos:%"IVdf"/%"IVdf \
862 (int)(depth)*2, "", \
863 (IV)((data)->pos_min), \
864 (IV)((data)->pos_delta), \
865 (UV)((data)->flags) \
867 DEBUG_SHOW_STUDY_FLAGS((data)->flags," [ ","]"); \
868 PerlIO_printf(Perl_debug_log, \
869 " Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
870 (IV)((data)->whilem_c), \
871 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
872 is_inf ? "INF " : "" \
874 if ((data)->last_found) \
875 PerlIO_printf(Perl_debug_log, \
876 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
877 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
878 SvPVX_const((data)->last_found), \
879 (IV)((data)->last_end), \
880 (IV)((data)->last_start_min), \
881 (IV)((data)->last_start_max), \
882 ((data)->longest && \
883 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
884 SvPVX_const((data)->longest_fixed), \
885 (IV)((data)->offset_fixed), \
886 ((data)->longest && \
887 (data)->longest==&((data)->longest_float)) ? "*" : "", \
888 SvPVX_const((data)->longest_float), \
889 (IV)((data)->offset_float_min), \
890 (IV)((data)->offset_float_max) \
892 PerlIO_printf(Perl_debug_log,"\n"); \
895 /* is c a control character for which we have a mnemonic? */
896 #define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
899 S_cntrl_to_mnemonic(const U8 c)
901 /* Returns the mnemonic string that represents character 'c', if one
902 * exists; NULL otherwise. The only ones that exist for the purposes of
903 * this routine are a few control characters */
906 case '\a': return "\\a";
907 case '\b': return "\\b";
908 case ESC_NATIVE: return "\\e";
909 case '\f': return "\\f";
910 case '\n': return "\\n";
911 case '\r': return "\\r";
912 case '\t': return "\\t";
918 /* Mark that we cannot extend a found fixed substring at this point.
919 Update the longest found anchored substring and the longest found
920 floating substrings if needed. */
923 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
924 SSize_t *minlenp, int is_inf)
926 const STRLEN l = CHR_SVLEN(data->last_found);
927 const STRLEN old_l = CHR_SVLEN(*data->longest);
928 GET_RE_DEBUG_FLAGS_DECL;
930 PERL_ARGS_ASSERT_SCAN_COMMIT;
932 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
933 SvSetMagicSV(*data->longest, data->last_found);
934 if (*data->longest == data->longest_fixed) {
935 data->offset_fixed = l ? data->last_start_min : data->pos_min;
936 if (data->flags & SF_BEFORE_EOL)
938 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
940 data->flags &= ~SF_FIX_BEFORE_EOL;
941 data->minlen_fixed=minlenp;
942 data->lookbehind_fixed=0;
944 else { /* *data->longest == data->longest_float */
945 data->offset_float_min = l ? data->last_start_min : data->pos_min;
946 data->offset_float_max = (l
947 ? data->last_start_max
948 : (data->pos_delta > SSize_t_MAX - data->pos_min
950 : data->pos_min + data->pos_delta));
952 || (STRLEN)data->offset_float_max > (STRLEN)SSize_t_MAX)
953 data->offset_float_max = SSize_t_MAX;
954 if (data->flags & SF_BEFORE_EOL)
956 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
958 data->flags &= ~SF_FL_BEFORE_EOL;
959 data->minlen_float=minlenp;
960 data->lookbehind_float=0;
963 SvCUR_set(data->last_found, 0);
965 SV * const sv = data->last_found;
966 if (SvUTF8(sv) && SvMAGICAL(sv)) {
967 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
973 data->flags &= ~SF_BEFORE_EOL;
974 DEBUG_STUDYDATA("commit: ",data,0);
977 /* An SSC is just a regnode_charclass_posix with an extra field: the inversion
978 * list that describes which code points it matches */
981 S_ssc_anything(pTHX_ regnode_ssc *ssc)
983 /* Set the SSC 'ssc' to match an empty string or any code point */
985 PERL_ARGS_ASSERT_SSC_ANYTHING;
987 assert(is_ANYOF_SYNTHETIC(ssc));
989 ssc->invlist = sv_2mortal(_new_invlist(2)); /* mortalize so won't leak */
990 _append_range_to_invlist(ssc->invlist, 0, UV_MAX);
991 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
995 S_ssc_is_anything(const regnode_ssc *ssc)
997 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
998 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
999 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1000 * in any way, so there's no point in using it */
1005 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
1007 assert(is_ANYOF_SYNTHETIC(ssc));
1009 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
1013 /* See if the list consists solely of the range 0 - Infinity */
1014 invlist_iterinit(ssc->invlist);
1015 ret = invlist_iternext(ssc->invlist, &start, &end)
1019 invlist_iterfinish(ssc->invlist);
1025 /* If e.g., both \w and \W are set, matches everything */
1026 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1028 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1029 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1039 S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
1041 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1042 * string, any code point, or any posix class under locale */
1044 PERL_ARGS_ASSERT_SSC_INIT;
1046 Zero(ssc, 1, regnode_ssc);
1047 set_ANYOF_SYNTHETIC(ssc);
1048 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
1051 /* If any portion of the regex is to operate under locale rules that aren't
1052 * fully known at compile time, initialization includes it. The reason
1053 * this isn't done for all regexes is that the optimizer was written under
1054 * the assumption that locale was all-or-nothing. Given the complexity and
1055 * lack of documentation in the optimizer, and that there are inadequate
1056 * test cases for locale, many parts of it may not work properly, it is
1057 * safest to avoid locale unless necessary. */
1058 if (RExC_contains_locale) {
1059 ANYOF_POSIXL_SETALL(ssc);
1062 ANYOF_POSIXL_ZERO(ssc);
1067 S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1068 const regnode_ssc *ssc)
1070 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1071 * to the list of code points matched, and locale posix classes; hence does
1072 * not check its flags) */
1077 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1079 assert(is_ANYOF_SYNTHETIC(ssc));
1081 invlist_iterinit(ssc->invlist);
1082 ret = invlist_iternext(ssc->invlist, &start, &end)
1086 invlist_iterfinish(ssc->invlist);
1092 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
1100 S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
1101 const regnode_charclass* const node)
1103 /* Returns a mortal inversion list defining which code points are matched
1104 * by 'node', which is of type ANYOF. Handles complementing the result if
1105 * appropriate. If some code points aren't knowable at this time, the
1106 * returned list must, and will, contain every code point that is a
1109 SV* invlist = sv_2mortal(_new_invlist(0));
1110 SV* only_utf8_locale_invlist = NULL;
1112 const U32 n = ARG(node);
1113 bool new_node_has_latin1 = FALSE;
1115 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1117 /* Look at the data structure created by S_set_ANYOF_arg() */
1118 if (n != ANYOF_ONLY_HAS_BITMAP) {
1119 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1120 AV * const av = MUTABLE_AV(SvRV(rv));
1121 SV **const ary = AvARRAY(av);
1122 assert(RExC_rxi->data->what[n] == 's');
1124 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
1125 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1])));
1127 else if (ary[0] && ary[0] != &PL_sv_undef) {
1129 /* Here, no compile-time swash, and there are things that won't be
1130 * known until runtime -- we have to assume it could be anything */
1131 return _add_range_to_invlist(invlist, 0, UV_MAX);
1133 else if (ary[3] && ary[3] != &PL_sv_undef) {
1135 /* Here no compile-time swash, and no run-time only data. Use the
1136 * node's inversion list */
1137 invlist = sv_2mortal(invlist_clone(ary[3]));
1140 /* Get the code points valid only under UTF-8 locales */
1141 if ((ANYOF_FLAGS(node) & ANYOF_LOC_FOLD)
1142 && ary[2] && ary[2] != &PL_sv_undef)
1144 only_utf8_locale_invlist = ary[2];
1148 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1149 * code points, and an inversion list for the others, but if there are code
1150 * points that should match only conditionally on the target string being
1151 * UTF-8, those are placed in the inversion list, and not the bitmap.
1152 * Since there are circumstances under which they could match, they are
1153 * included in the SSC. But if the ANYOF node is to be inverted, we have
1154 * to exclude them here, so that when we invert below, the end result
1155 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1156 * have to do this here before we add the unconditionally matched code
1158 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1159 _invlist_intersection_complement_2nd(invlist,
1164 /* Add in the points from the bit map */
1165 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
1166 if (ANYOF_BITMAP_TEST(node, i)) {
1167 invlist = add_cp_to_invlist(invlist, i);
1168 new_node_has_latin1 = TRUE;
1172 /* If this can match all upper Latin1 code points, have to add them
1174 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII) {
1175 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1178 /* Similarly for these */
1179 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
1180 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
1183 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1184 _invlist_invert(invlist);
1186 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOF_LOC_FOLD) {
1188 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1189 * locale. We can skip this if there are no 0-255 at all. */
1190 _invlist_union(invlist, PL_Latin1, &invlist);
1193 /* Similarly add the UTF-8 locale possible matches. These have to be
1194 * deferred until after the non-UTF-8 locale ones are taken care of just
1195 * above, or it leads to wrong results under ANYOF_INVERT */
1196 if (only_utf8_locale_invlist) {
1197 _invlist_union_maybe_complement_2nd(invlist,
1198 only_utf8_locale_invlist,
1199 ANYOF_FLAGS(node) & ANYOF_INVERT,
1206 /* These two functions currently do the exact same thing */
1207 #define ssc_init_zero ssc_init
1209 #define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1210 #define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1212 /* 'AND' a given class with another one. Can create false positives. 'ssc'
1213 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1214 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
1217 S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1218 const regnode_charclass *and_with)
1220 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1221 * another SSC or a regular ANYOF class. Can create false positives. */
1226 PERL_ARGS_ASSERT_SSC_AND;
1228 assert(is_ANYOF_SYNTHETIC(ssc));
1230 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1231 * the code point inversion list and just the relevant flags */
1232 if (is_ANYOF_SYNTHETIC(and_with)) {
1233 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
1234 anded_flags = ANYOF_FLAGS(and_with);
1236 /* XXX This is a kludge around what appears to be deficiencies in the
1237 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1238 * there are paths through the optimizer where it doesn't get weeded
1239 * out when it should. And if we don't make some extra provision for
1240 * it like the code just below, it doesn't get added when it should.
1241 * This solution is to add it only when AND'ing, which is here, and
1242 * only when what is being AND'ed is the pristine, original node
1243 * matching anything. Thus it is like adding it to ssc_anything() but
1244 * only when the result is to be AND'ed. Probably the same solution
1245 * could be adopted for the same problem we have with /l matching,
1246 * which is solved differently in S_ssc_init(), and that would lead to
1247 * fewer false positives than that solution has. But if this solution
1248 * creates bugs, the consequences are only that a warning isn't raised
1249 * that should be; while the consequences for having /l bugs is
1250 * incorrect matches */
1251 if (ssc_is_anything((regnode_ssc *)and_with)) {
1252 anded_flags |= ANYOF_WARN_SUPER;
1256 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
1257 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1260 ANYOF_FLAGS(ssc) &= anded_flags;
1262 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1263 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1264 * 'and_with' may be inverted. When not inverted, we have the situation of
1266 * (C1 | P1) & (C2 | P2)
1267 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1268 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1269 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1270 * <= ((C1 & C2) | P1 | P2)
1271 * Alternatively, the last few steps could be:
1272 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1273 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1274 * <= (C1 | C2 | (P1 & P2))
1275 * We favor the second approach if either P1 or P2 is non-empty. This is
1276 * because these components are a barrier to doing optimizations, as what
1277 * they match cannot be known until the moment of matching as they are
1278 * dependent on the current locale, 'AND"ing them likely will reduce or
1280 * But we can do better if we know that C1,P1 are in their initial state (a
1281 * frequent occurrence), each matching everything:
1282 * (<everything>) & (C2 | P2) = C2 | P2
1283 * Similarly, if C2,P2 are in their initial state (again a frequent
1284 * occurrence), the result is a no-op
1285 * (C1 | P1) & (<everything>) = C1 | P1
1288 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1289 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1290 * <= (C1 & ~C2) | (P1 & ~P2)
1293 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
1294 && ! is_ANYOF_SYNTHETIC(and_with))
1298 ssc_intersection(ssc,
1300 FALSE /* Has already been inverted */
1303 /* If either P1 or P2 is empty, the intersection will be also; can skip
1305 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
1306 ANYOF_POSIXL_ZERO(ssc);
1308 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1310 /* Note that the Posix class component P from 'and_with' actually
1312 * P = Pa | Pb | ... | Pn
1313 * where each component is one posix class, such as in [\w\s].
1315 * ~P = ~(Pa | Pb | ... | Pn)
1316 * = ~Pa & ~Pb & ... & ~Pn
1317 * <= ~Pa | ~Pb | ... | ~Pn
1318 * The last is something we can easily calculate, but unfortunately
1319 * is likely to have many false positives. We could do better
1320 * in some (but certainly not all) instances if two classes in
1321 * P have known relationships. For example
1322 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1324 * :lower: & :print: = :lower:
1325 * And similarly for classes that must be disjoint. For example,
1326 * since \s and \w can have no elements in common based on rules in
1327 * the POSIX standard,
1328 * \w & ^\S = nothing
1329 * Unfortunately, some vendor locales do not meet the Posix
1330 * standard, in particular almost everything by Microsoft.
1331 * The loop below just changes e.g., \w into \W and vice versa */
1333 regnode_charclass_posixl temp;
1334 int add = 1; /* To calculate the index of the complement */
1336 ANYOF_POSIXL_ZERO(&temp);
1337 for (i = 0; i < ANYOF_MAX; i++) {
1339 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1340 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
1342 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
1343 ANYOF_POSIXL_SET(&temp, i + add);
1345 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1347 ANYOF_POSIXL_AND(&temp, ssc);
1349 } /* else ssc already has no posixes */
1350 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1351 in its initial state */
1352 else if (! is_ANYOF_SYNTHETIC(and_with)
1353 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
1355 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1356 * copy it over 'ssc' */
1357 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
1358 if (is_ANYOF_SYNTHETIC(and_with)) {
1359 StructCopy(and_with, ssc, regnode_ssc);
1362 ssc->invlist = anded_cp_list;
1363 ANYOF_POSIXL_ZERO(ssc);
1364 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1365 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
1369 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
1370 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
1372 /* One or the other of P1, P2 is non-empty. */
1373 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1374 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1376 ssc_union(ssc, anded_cp_list, FALSE);
1378 else { /* P1 = P2 = empty */
1379 ssc_intersection(ssc, anded_cp_list, FALSE);
1385 S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1386 const regnode_charclass *or_with)
1388 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1389 * another SSC or a regular ANYOF class. Can create false positives if
1390 * 'or_with' is to be inverted. */
1395 PERL_ARGS_ASSERT_SSC_OR;
1397 assert(is_ANYOF_SYNTHETIC(ssc));
1399 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1400 * the code point inversion list and just the relevant flags */
1401 if (is_ANYOF_SYNTHETIC(or_with)) {
1402 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
1403 ored_flags = ANYOF_FLAGS(or_with);
1406 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
1407 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
1410 ANYOF_FLAGS(ssc) |= ored_flags;
1412 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1413 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1414 * 'or_with' may be inverted. When not inverted, we have the simple
1415 * situation of computing:
1416 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1417 * If P1|P2 yields a situation with both a class and its complement are
1418 * set, like having both \w and \W, this matches all code points, and we
1419 * can delete these from the P component of the ssc going forward. XXX We
1420 * might be able to delete all the P components, but I (khw) am not certain
1421 * about this, and it is better to be safe.
1424 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1425 * <= (C1 | P1) | ~C2
1426 * <= (C1 | ~C2) | P1
1427 * (which results in actually simpler code than the non-inverted case)
1430 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
1431 && ! is_ANYOF_SYNTHETIC(or_with))
1433 /* We ignore P2, leaving P1 going forward */
1434 } /* else Not inverted */
1435 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
1436 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
1437 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1439 for (i = 0; i < ANYOF_MAX; i += 2) {
1440 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1442 ssc_match_all_cp(ssc);
1443 ANYOF_POSIXL_CLEAR(ssc, i);
1444 ANYOF_POSIXL_CLEAR(ssc, i+1);
1452 FALSE /* Already has been inverted */
1456 PERL_STATIC_INLINE void
1457 S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1459 PERL_ARGS_ASSERT_SSC_UNION;
1461 assert(is_ANYOF_SYNTHETIC(ssc));
1463 _invlist_union_maybe_complement_2nd(ssc->invlist,
1469 PERL_STATIC_INLINE void
1470 S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1472 const bool invert2nd)
1474 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1476 assert(is_ANYOF_SYNTHETIC(ssc));
1478 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1484 PERL_STATIC_INLINE void
1485 S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1487 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1489 assert(is_ANYOF_SYNTHETIC(ssc));
1491 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1494 PERL_STATIC_INLINE void
1495 S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1497 /* AND just the single code point 'cp' into the SSC 'ssc' */
1499 SV* cp_list = _new_invlist(2);
1501 PERL_ARGS_ASSERT_SSC_CP_AND;
1503 assert(is_ANYOF_SYNTHETIC(ssc));
1505 cp_list = add_cp_to_invlist(cp_list, cp);
1506 ssc_intersection(ssc, cp_list,
1507 FALSE /* Not inverted */
1509 SvREFCNT_dec_NN(cp_list);
1512 PERL_STATIC_INLINE void
1513 S_ssc_clear_locale(regnode_ssc *ssc)
1515 /* Set the SSC 'ssc' to not match any locale things */
1516 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1518 assert(is_ANYOF_SYNTHETIC(ssc));
1520 ANYOF_POSIXL_ZERO(ssc);
1521 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1524 #define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
1527 S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
1529 /* The synthetic start class is used to hopefully quickly winnow down
1530 * places where a pattern could start a match in the target string. If it
1531 * doesn't really narrow things down that much, there isn't much point to
1532 * having the overhead of using it. This function uses some very crude
1533 * heuristics to decide if to use the ssc or not.
1535 * It returns TRUE if 'ssc' rules out more than half what it considers to
1536 * be the "likely" possible matches, but of course it doesn't know what the
1537 * actual things being matched are going to be; these are only guesses
1539 * For /l matches, it assumes that the only likely matches are going to be
1540 * in the 0-255 range, uniformly distributed, so half of that is 127
1541 * For /a and /d matches, it assumes that the likely matches will be just
1542 * the ASCII range, so half of that is 63
1543 * For /u and there isn't anything matching above the Latin1 range, it
1544 * assumes that that is the only range likely to be matched, and uses
1545 * half that as the cut-off: 127. If anything matches above Latin1,
1546 * it assumes that all of Unicode could match (uniformly), except for
1547 * non-Unicode code points and things in the General Category "Other"
1548 * (unassigned, private use, surrogates, controls and formats). This
1549 * is a much large number. */
1551 const U32 max_match = (LOC)
1555 : (invlist_highest(ssc->invlist) < 256)
1557 : ((NON_OTHER_COUNT + 1) / 2) - 1;
1558 U32 count = 0; /* Running total of number of code points matched by
1560 UV start, end; /* Start and end points of current range in inversion
1563 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
1565 invlist_iterinit(ssc->invlist);
1566 while (invlist_iternext(ssc->invlist, &start, &end)) {
1568 /* /u is the only thing that we expect to match above 255; so if not /u
1569 * and even if there are matches above 255, ignore them. This catches
1570 * things like \d under /d which does match the digits above 255, but
1571 * since the pattern is /d, it is not likely to be expecting them */
1572 if (! UNI_SEMANTICS) {
1576 end = MIN(end, 255);
1578 count += end - start + 1;
1579 if (count > max_match) {
1580 invlist_iterfinish(ssc->invlist);
1590 S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
1592 /* The inversion list in the SSC is marked mortal; now we need a more
1593 * permanent copy, which is stored the same way that is done in a regular
1594 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
1597 SV* invlist = invlist_clone(ssc->invlist);
1599 PERL_ARGS_ASSERT_SSC_FINALIZE;
1601 assert(is_ANYOF_SYNTHETIC(ssc));
1603 /* The code in this file assumes that all but these flags aren't relevant
1604 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
1605 * by the time we reach here */
1606 assert(! (ANYOF_FLAGS(ssc) & ~ANYOF_COMMON_FLAGS));
1608 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
1610 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
1611 NULL, NULL, NULL, FALSE);
1613 /* Make sure is clone-safe */
1614 ssc->invlist = NULL;
1616 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1617 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
1620 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
1623 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
1624 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
1625 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
1626 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
1627 ? (TRIE_LIST_CUR( idx ) - 1) \
1633 dump_trie(trie,widecharmap,revcharmap)
1634 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
1635 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
1637 These routines dump out a trie in a somewhat readable format.
1638 The _interim_ variants are used for debugging the interim
1639 tables that are used to generate the final compressed
1640 representation which is what dump_trie expects.
1642 Part of the reason for their existence is to provide a form
1643 of documentation as to how the different representations function.
1648 Dumps the final compressed table form of the trie to Perl_debug_log.
1649 Used for debugging make_trie().
1653 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
1654 AV *revcharmap, U32 depth)
1657 SV *sv=sv_newmortal();
1658 int colwidth= widecharmap ? 6 : 4;
1660 GET_RE_DEBUG_FLAGS_DECL;
1662 PERL_ARGS_ASSERT_DUMP_TRIE;
1664 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
1665 (int)depth * 2 + 2,"",
1666 "Match","Base","Ofs" );
1668 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
1669 SV ** const tmp = av_fetch( revcharmap, state, 0);
1671 PerlIO_printf( Perl_debug_log, "%*s",
1673 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1674 PL_colors[0], PL_colors[1],
1675 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1676 PERL_PV_ESCAPE_FIRSTCHAR
1681 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
1682 (int)depth * 2 + 2,"");
1684 for( state = 0 ; state < trie->uniquecharcount ; state++ )
1685 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
1686 PerlIO_printf( Perl_debug_log, "\n");
1688 for( state = 1 ; state < trie->statecount ; state++ ) {
1689 const U32 base = trie->states[ state ].trans.base;
1691 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|",
1692 (int)depth * 2 + 2,"", (UV)state);
1694 if ( trie->states[ state ].wordnum ) {
1695 PerlIO_printf( Perl_debug_log, " W%4X",
1696 trie->states[ state ].wordnum );
1698 PerlIO_printf( Perl_debug_log, "%6s", "" );
1701 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
1706 while( ( base + ofs < trie->uniquecharcount ) ||
1707 ( base + ofs - trie->uniquecharcount < trie->lasttrans
1708 && trie->trans[ base + ofs - trie->uniquecharcount ].check
1712 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
1714 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1715 if ( ( base + ofs >= trie->uniquecharcount )
1716 && ( base + ofs - trie->uniquecharcount
1718 && trie->trans[ base + ofs
1719 - trie->uniquecharcount ].check == state )
1721 PerlIO_printf( Perl_debug_log, "%*"UVXf,
1723 (UV)trie->trans[ base + ofs
1724 - trie->uniquecharcount ].next );
1726 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
1730 PerlIO_printf( Perl_debug_log, "]");
1733 PerlIO_printf( Perl_debug_log, "\n" );
1735 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=",
1737 for (word=1; word <= trie->wordcount; word++) {
1738 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
1739 (int)word, (int)(trie->wordinfo[word].prev),
1740 (int)(trie->wordinfo[word].len));
1742 PerlIO_printf(Perl_debug_log, "\n" );
1745 Dumps a fully constructed but uncompressed trie in list form.
1746 List tries normally only are used for construction when the number of
1747 possible chars (trie->uniquecharcount) is very high.
1748 Used for debugging make_trie().
1751 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
1752 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1756 SV *sv=sv_newmortal();
1757 int colwidth= widecharmap ? 6 : 4;
1758 GET_RE_DEBUG_FLAGS_DECL;
1760 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1762 /* print out the table precompression. */
1763 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1764 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1765 "------:-----+-----------------\n" );
1767 for( state=1 ; state < next_alloc ; state ++ ) {
1770 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1771 (int)depth * 2 + 2,"", (UV)state );
1772 if ( ! trie->states[ state ].wordnum ) {
1773 PerlIO_printf( Perl_debug_log, "%5s| ","");
1775 PerlIO_printf( Perl_debug_log, "W%4x| ",
1776 trie->states[ state ].wordnum
1779 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1780 SV ** const tmp = av_fetch( revcharmap,
1781 TRIE_LIST_ITEM(state,charid).forid, 0);
1783 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1785 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
1787 PL_colors[0], PL_colors[1],
1788 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
1789 | PERL_PV_ESCAPE_FIRSTCHAR
1791 TRIE_LIST_ITEM(state,charid).forid,
1792 (UV)TRIE_LIST_ITEM(state,charid).newstate
1795 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1796 (int)((depth * 2) + 14), "");
1799 PerlIO_printf( Perl_debug_log, "\n");
1804 Dumps a fully constructed but uncompressed trie in table form.
1805 This is the normal DFA style state transition table, with a few
1806 twists to facilitate compression later.
1807 Used for debugging make_trie().
1810 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1811 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1816 SV *sv=sv_newmortal();
1817 int colwidth= widecharmap ? 6 : 4;
1818 GET_RE_DEBUG_FLAGS_DECL;
1820 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1823 print out the table precompression so that we can do a visual check
1824 that they are identical.
1827 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1829 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1830 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1832 PerlIO_printf( Perl_debug_log, "%*s",
1834 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1835 PL_colors[0], PL_colors[1],
1836 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1837 PERL_PV_ESCAPE_FIRSTCHAR
1843 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1845 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1846 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1849 PerlIO_printf( Perl_debug_log, "\n" );
1851 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1853 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1854 (int)depth * 2 + 2,"",
1855 (UV)TRIE_NODENUM( state ) );
1857 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1858 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1860 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1862 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1864 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1865 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n",
1866 (UV)trie->trans[ state ].check );
1868 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n",
1869 (UV)trie->trans[ state ].check,
1870 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1878 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1879 startbranch: the first branch in the whole branch sequence
1880 first : start branch of sequence of branch-exact nodes.
1881 May be the same as startbranch
1882 last : Thing following the last branch.
1883 May be the same as tail.
1884 tail : item following the branch sequence
1885 count : words in the sequence
1886 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
1887 depth : indent depth
1889 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1891 A trie is an N'ary tree where the branches are determined by digital
1892 decomposition of the key. IE, at the root node you look up the 1st character and
1893 follow that branch repeat until you find the end of the branches. Nodes can be
1894 marked as "accepting" meaning they represent a complete word. Eg:
1898 would convert into the following structure. Numbers represent states, letters
1899 following numbers represent valid transitions on the letter from that state, if
1900 the number is in square brackets it represents an accepting state, otherwise it
1901 will be in parenthesis.
1903 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1907 (1) +-i->(6)-+-s->[7]
1909 +-s->(3)-+-h->(4)-+-e->[5]
1911 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1913 This shows that when matching against the string 'hers' we will begin at state 1
1914 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1915 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1916 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1917 single traverse. We store a mapping from accepting to state to which word was
1918 matched, and then when we have multiple possibilities we try to complete the
1919 rest of the regex in the order in which they occured in the alternation.
1921 The only prior NFA like behaviour that would be changed by the TRIE support is
1922 the silent ignoring of duplicate alternations which are of the form:
1924 / (DUPE|DUPE) X? (?{ ... }) Y /x
1926 Thus EVAL blocks following a trie may be called a different number of times with
1927 and without the optimisation. With the optimisations dupes will be silently
1928 ignored. This inconsistent behaviour of EVAL type nodes is well established as
1929 the following demonstrates:
1931 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1933 which prints out 'word' three times, but
1935 'words'=~/(word|word|word)(?{ print $1 })S/
1937 which doesnt print it out at all. This is due to other optimisations kicking in.
1939 Example of what happens on a structural level:
1941 The regexp /(ac|ad|ab)+/ will produce the following debug output:
1943 1: CURLYM[1] {1,32767}(18)
1954 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1955 and should turn into:
1957 1: CURLYM[1] {1,32767}(18)
1959 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1967 Cases where tail != last would be like /(?foo|bar)baz/:
1977 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1978 and would end up looking like:
1981 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1988 d = uvchr_to_utf8_flags(d, uv, 0);
1990 is the recommended Unicode-aware way of saying
1995 #define TRIE_STORE_REVCHAR(val) \
1998 SV *zlopp = newSV(7); /* XXX: optimize me */ \
1999 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
2000 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
2001 SvCUR_set(zlopp, kapow - flrbbbbb); \
2004 av_push(revcharmap, zlopp); \
2006 char ooooff = (char)val; \
2007 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2011 /* This gets the next character from the input, folding it if not already
2013 #define TRIE_READ_CHAR STMT_START { \
2016 /* if it is UTF then it is either already folded, or does not need \
2018 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
2020 else if (folder == PL_fold_latin1) { \
2021 /* This folder implies Unicode rules, which in the range expressible \
2022 * by not UTF is the lower case, with the two exceptions, one of \
2023 * which should have been taken care of before calling this */ \
2024 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2025 uvc = toLOWER_L1(*uc); \
2026 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2029 /* raw data, will be folded later if needed */ \
2037 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2038 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
2039 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
2040 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
2042 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2043 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2044 TRIE_LIST_CUR( state )++; \
2047 #define TRIE_LIST_NEW(state) STMT_START { \
2048 Newxz( trie->states[ state ].trans.list, \
2049 4, reg_trie_trans_le ); \
2050 TRIE_LIST_CUR( state ) = 1; \
2051 TRIE_LIST_LEN( state ) = 4; \
2054 #define TRIE_HANDLE_WORD(state) STMT_START { \
2055 U16 dupe= trie->states[ state ].wordnum; \
2056 regnode * const noper_next = regnext( noper ); \
2059 /* store the word for dumping */ \
2061 if (OP(noper) != NOTHING) \
2062 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
2064 tmp = newSVpvn_utf8( "", 0, UTF ); \
2065 av_push( trie_words, tmp ); \
2069 trie->wordinfo[curword].prev = 0; \
2070 trie->wordinfo[curword].len = wordlen; \
2071 trie->wordinfo[curword].accept = state; \
2073 if ( noper_next < tail ) { \
2075 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2077 trie->jump[curword] = (U16)(noper_next - convert); \
2079 jumper = noper_next; \
2081 nextbranch= regnext(cur); \
2085 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2086 /* chain, so that when the bits of chain are later */\
2087 /* linked together, the dups appear in the chain */\
2088 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2089 trie->wordinfo[dupe].prev = curword; \
2091 /* we haven't inserted this word yet. */ \
2092 trie->states[ state ].wordnum = curword; \
2097 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2098 ( ( base + charid >= ucharcount \
2099 && base + charid < ubound \
2100 && state == trie->trans[ base - ucharcount + charid ].check \
2101 && trie->trans[ base - ucharcount + charid ].next ) \
2102 ? trie->trans[ base - ucharcount + charid ].next \
2103 : ( state==1 ? special : 0 ) \
2107 #define MADE_JUMP_TRIE 2
2108 #define MADE_EXACT_TRIE 4
2111 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2112 regnode *first, regnode *last, regnode *tail,
2113 U32 word_count, U32 flags, U32 depth)
2115 /* first pass, loop through and scan words */
2116 reg_trie_data *trie;
2117 HV *widecharmap = NULL;
2118 AV *revcharmap = newAV();
2124 regnode *jumper = NULL;
2125 regnode *nextbranch = NULL;
2126 regnode *convert = NULL;
2127 U32 *prev_states; /* temp array mapping each state to previous one */
2128 /* we just use folder as a flag in utf8 */
2129 const U8 * folder = NULL;
2132 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuuu"));
2133 AV *trie_words = NULL;
2134 /* along with revcharmap, this only used during construction but both are
2135 * useful during debugging so we store them in the struct when debugging.
2138 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
2139 STRLEN trie_charcount=0;
2141 SV *re_trie_maxbuff;
2142 GET_RE_DEBUG_FLAGS_DECL;
2144 PERL_ARGS_ASSERT_MAKE_TRIE;
2146 PERL_UNUSED_ARG(depth);
2150 case EXACT: case EXACTL: break;
2154 case EXACTFLU8: folder = PL_fold_latin1; break;
2155 case EXACTF: folder = PL_fold; break;
2156 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
2159 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
2161 trie->startstate = 1;
2162 trie->wordcount = word_count;
2163 RExC_rxi->data->data[ data_slot ] = (void*)trie;
2164 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
2165 if (flags == EXACT || flags == EXACTL)
2166 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2167 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2168 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2171 trie_words = newAV();
2174 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2175 assert(re_trie_maxbuff);
2176 if (!SvIOK(re_trie_maxbuff)) {
2177 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2179 DEBUG_TRIE_COMPILE_r({
2180 PerlIO_printf( Perl_debug_log,
2181 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2182 (int)depth * 2 + 2, "",
2183 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
2184 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
2187 /* Find the node we are going to overwrite */
2188 if ( first == startbranch && OP( last ) != BRANCH ) {
2189 /* whole branch chain */
2192 /* branch sub-chain */
2193 convert = NEXTOPER( first );
2196 /* -- First loop and Setup --
2198 We first traverse the branches and scan each word to determine if it
2199 contains widechars, and how many unique chars there are, this is
2200 important as we have to build a table with at least as many columns as we
2203 We use an array of integers to represent the character codes 0..255
2204 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2205 the native representation of the character value as the key and IV's for
2208 *TODO* If we keep track of how many times each character is used we can
2209 remap the columns so that the table compression later on is more
2210 efficient in terms of memory by ensuring the most common value is in the
2211 middle and the least common are on the outside. IMO this would be better
2212 than a most to least common mapping as theres a decent chance the most
2213 common letter will share a node with the least common, meaning the node
2214 will not be compressible. With a middle is most common approach the worst
2215 case is when we have the least common nodes twice.
2219 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2220 regnode *noper = NEXTOPER( cur );
2221 const U8 *uc = (U8*)STRING( noper );
2222 const U8 *e = uc + STR_LEN( noper );
2224 U32 wordlen = 0; /* required init */
2225 STRLEN minchars = 0;
2226 STRLEN maxchars = 0;
2227 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2230 if (OP(noper) == NOTHING) {
2231 regnode *noper_next= regnext(noper);
2232 if (noper_next != tail && OP(noper_next) == flags) {
2234 uc= (U8*)STRING(noper);
2235 e= uc + STR_LEN(noper);
2236 trie->minlen= STR_LEN(noper);
2243 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
2244 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2245 regardless of encoding */
2246 if (OP( noper ) == EXACTFU_SS) {
2247 /* false positives are ok, so just set this */
2248 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
2251 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2253 TRIE_CHARCOUNT(trie)++;
2256 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2257 * is in effect. Under /i, this character can match itself, or
2258 * anything that folds to it. If not under /i, it can match just
2259 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2260 * all fold to k, and all are single characters. But some folds
2261 * expand to more than one character, so for example LATIN SMALL
2262 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2263 * the string beginning at 'uc' is 'ffi', it could be matched by
2264 * three characters, or just by the one ligature character. (It
2265 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2266 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2267 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2268 * match.) The trie needs to know the minimum and maximum number
2269 * of characters that could match so that it can use size alone to
2270 * quickly reject many match attempts. The max is simple: it is
2271 * the number of folded characters in this branch (since a fold is
2272 * never shorter than what folds to it. */
2276 /* And the min is equal to the max if not under /i (indicated by
2277 * 'folder' being NULL), or there are no multi-character folds. If
2278 * there is a multi-character fold, the min is incremented just
2279 * once, for the character that folds to the sequence. Each
2280 * character in the sequence needs to be added to the list below of
2281 * characters in the trie, but we count only the first towards the
2282 * min number of characters needed. This is done through the
2283 * variable 'foldlen', which is returned by the macros that look
2284 * for these sequences as the number of bytes the sequence
2285 * occupies. Each time through the loop, we decrement 'foldlen' by
2286 * how many bytes the current char occupies. Only when it reaches
2287 * 0 do we increment 'minchars' or look for another multi-character
2289 if (folder == NULL) {
2292 else if (foldlen > 0) {
2293 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
2298 /* See if *uc is the beginning of a multi-character fold. If
2299 * so, we decrement the length remaining to look at, to account
2300 * for the current character this iteration. (We can use 'uc'
2301 * instead of the fold returned by TRIE_READ_CHAR because for
2302 * non-UTF, the latin1_safe macro is smart enough to account
2303 * for all the unfolded characters, and because for UTF, the
2304 * string will already have been folded earlier in the
2305 * compilation process */
2307 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2308 foldlen -= UTF8SKIP(uc);
2311 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2316 /* The current character (and any potential folds) should be added
2317 * to the possible matching characters for this position in this
2321 U8 folded= folder[ (U8) uvc ];
2322 if ( !trie->charmap[ folded ] ) {
2323 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2324 TRIE_STORE_REVCHAR( folded );
2327 if ( !trie->charmap[ uvc ] ) {
2328 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
2329 TRIE_STORE_REVCHAR( uvc );
2332 /* store the codepoint in the bitmap, and its folded
2334 TRIE_BITMAP_SET(trie, uvc);
2336 /* store the folded codepoint */
2337 if ( folder ) TRIE_BITMAP_SET(trie, folder[(U8) uvc ]);
2340 /* store first byte of utf8 representation of
2341 variant codepoints */
2342 if (! UVCHR_IS_INVARIANT(uvc)) {
2343 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
2346 set_bit = 0; /* We've done our bit :-) */
2350 /* XXX We could come up with the list of code points that fold
2351 * to this using PL_utf8_foldclosures, except not for
2352 * multi-char folds, as there may be multiple combinations
2353 * there that could work, which needs to wait until runtime to
2354 * resolve (The comment about LIGATURE FFI above is such an
2359 widecharmap = newHV();
2361 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
2364 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
2366 if ( !SvTRUE( *svpp ) ) {
2367 sv_setiv( *svpp, ++trie->uniquecharcount );
2368 TRIE_STORE_REVCHAR(uvc);
2371 } /* end loop through characters in this branch of the trie */
2373 /* We take the min and max for this branch and combine to find the min
2374 * and max for all branches processed so far */
2375 if( cur == first ) {
2376 trie->minlen = minchars;
2377 trie->maxlen = maxchars;
2378 } else if (minchars < trie->minlen) {
2379 trie->minlen = minchars;
2380 } else if (maxchars > trie->maxlen) {
2381 trie->maxlen = maxchars;
2383 } /* end first pass */
2384 DEBUG_TRIE_COMPILE_r(
2385 PerlIO_printf( Perl_debug_log,
2386 "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2387 (int)depth * 2 + 2,"",
2388 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
2389 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2390 (int)trie->minlen, (int)trie->maxlen )
2394 We now know what we are dealing with in terms of unique chars and
2395 string sizes so we can calculate how much memory a naive
2396 representation using a flat table will take. If it's over a reasonable
2397 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
2398 conservative but potentially much slower representation using an array
2401 At the end we convert both representations into the same compressed
2402 form that will be used in regexec.c for matching with. The latter
2403 is a form that cannot be used to construct with but has memory
2404 properties similar to the list form and access properties similar
2405 to the table form making it both suitable for fast searches and
2406 small enough that its feasable to store for the duration of a program.
2408 See the comment in the code where the compressed table is produced
2409 inplace from the flat tabe representation for an explanation of how
2410 the compression works.
2415 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2418 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2419 > SvIV(re_trie_maxbuff) )
2422 Second Pass -- Array Of Lists Representation
2424 Each state will be represented by a list of charid:state records
2425 (reg_trie_trans_le) the first such element holds the CUR and LEN
2426 points of the allocated array. (See defines above).
2428 We build the initial structure using the lists, and then convert
2429 it into the compressed table form which allows faster lookups
2430 (but cant be modified once converted).
2433 STRLEN transcount = 1;
2435 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2436 "%*sCompiling trie using list compiler\n",
2437 (int)depth * 2 + 2, ""));
2439 trie->states = (reg_trie_state *)
2440 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2441 sizeof(reg_trie_state) );
2445 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2447 regnode *noper = NEXTOPER( cur );
2448 U8 *uc = (U8*)STRING( noper );
2449 const U8 *e = uc + STR_LEN( noper );
2450 U32 state = 1; /* required init */
2451 U16 charid = 0; /* sanity init */
2452 U32 wordlen = 0; /* required init */
2454 if (OP(noper) == NOTHING) {
2455 regnode *noper_next= regnext(noper);
2456 if (noper_next != tail && OP(noper_next) == flags) {
2458 uc= (U8*)STRING(noper);
2459 e= uc + STR_LEN(noper);
2463 if (OP(noper) != NOTHING) {
2464 for ( ; uc < e ; uc += len ) {
2469 charid = trie->charmap[ uvc ];
2471 SV** const svpp = hv_fetch( widecharmap,
2478 charid=(U16)SvIV( *svpp );
2481 /* charid is now 0 if we dont know the char read, or
2482 * nonzero if we do */
2489 if ( !trie->states[ state ].trans.list ) {
2490 TRIE_LIST_NEW( state );
2493 check <= TRIE_LIST_USED( state );
2496 if ( TRIE_LIST_ITEM( state, check ).forid
2499 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2504 newstate = next_alloc++;
2505 prev_states[newstate] = state;
2506 TRIE_LIST_PUSH( state, charid, newstate );
2511 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2515 TRIE_HANDLE_WORD(state);
2517 } /* end second pass */
2519 /* next alloc is the NEXT state to be allocated */
2520 trie->statecount = next_alloc;
2521 trie->states = (reg_trie_state *)
2522 PerlMemShared_realloc( trie->states,
2524 * sizeof(reg_trie_state) );
2526 /* and now dump it out before we compress it */
2527 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2528 revcharmap, next_alloc,
2532 trie->trans = (reg_trie_trans *)
2533 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
2540 for( state=1 ; state < next_alloc ; state ++ ) {
2544 DEBUG_TRIE_COMPILE_MORE_r(
2545 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
2549 if (trie->states[state].trans.list) {
2550 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
2554 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2555 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
2556 if ( forid < minid ) {
2558 } else if ( forid > maxid ) {
2562 if ( transcount < tp + maxid - minid + 1) {
2564 trie->trans = (reg_trie_trans *)
2565 PerlMemShared_realloc( trie->trans,
2567 * sizeof(reg_trie_trans) );
2568 Zero( trie->trans + (transcount / 2),
2572 base = trie->uniquecharcount + tp - minid;
2573 if ( maxid == minid ) {
2575 for ( ; zp < tp ; zp++ ) {
2576 if ( ! trie->trans[ zp ].next ) {
2577 base = trie->uniquecharcount + zp - minid;
2578 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
2580 trie->trans[ zp ].check = state;
2586 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
2588 trie->trans[ tp ].check = state;
2593 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2594 const U32 tid = base
2595 - trie->uniquecharcount
2596 + TRIE_LIST_ITEM( state, idx ).forid;
2597 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
2599 trie->trans[ tid ].check = state;
2601 tp += ( maxid - minid + 1 );
2603 Safefree(trie->states[ state ].trans.list);
2606 DEBUG_TRIE_COMPILE_MORE_r(
2607 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
2610 trie->states[ state ].trans.base=base;
2612 trie->lasttrans = tp + 1;
2616 Second Pass -- Flat Table Representation.
2618 we dont use the 0 slot of either trans[] or states[] so we add 1 to
2619 each. We know that we will need Charcount+1 trans at most to store
2620 the data (one row per char at worst case) So we preallocate both
2621 structures assuming worst case.
2623 We then construct the trie using only the .next slots of the entry
2626 We use the .check field of the first entry of the node temporarily
2627 to make compression both faster and easier by keeping track of how
2628 many non zero fields are in the node.
2630 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
2633 There are two terms at use here: state as a TRIE_NODEIDX() which is
2634 a number representing the first entry of the node, and state as a
2635 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
2636 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
2637 if there are 2 entrys per node. eg:
2645 The table is internally in the right hand, idx form. However as we
2646 also have to deal with the states array which is indexed by nodenum
2647 we have to use TRIE_NODENUM() to convert.
2650 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2651 "%*sCompiling trie using table compiler\n",
2652 (int)depth * 2 + 2, ""));
2654 trie->trans = (reg_trie_trans *)
2655 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
2656 * trie->uniquecharcount + 1,
2657 sizeof(reg_trie_trans) );
2658 trie->states = (reg_trie_state *)
2659 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2660 sizeof(reg_trie_state) );
2661 next_alloc = trie->uniquecharcount + 1;
2664 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2666 regnode *noper = NEXTOPER( cur );
2667 const U8 *uc = (U8*)STRING( noper );
2668 const U8 *e = uc + STR_LEN( noper );
2670 U32 state = 1; /* required init */
2672 U16 charid = 0; /* sanity init */
2673 U32 accept_state = 0; /* sanity init */
2675 U32 wordlen = 0; /* required init */
2677 if (OP(noper) == NOTHING) {
2678 regnode *noper_next= regnext(noper);
2679 if (noper_next != tail && OP(noper_next) == flags) {
2681 uc= (U8*)STRING(noper);
2682 e= uc + STR_LEN(noper);
2686 if ( OP(noper) != NOTHING ) {
2687 for ( ; uc < e ; uc += len ) {
2692 charid = trie->charmap[ uvc ];
2694 SV* const * const svpp = hv_fetch( widecharmap,
2698 charid = svpp ? (U16)SvIV(*svpp) : 0;
2702 if ( !trie->trans[ state + charid ].next ) {
2703 trie->trans[ state + charid ].next = next_alloc;
2704 trie->trans[ state ].check++;
2705 prev_states[TRIE_NODENUM(next_alloc)]
2706 = TRIE_NODENUM(state);
2707 next_alloc += trie->uniquecharcount;
2709 state = trie->trans[ state + charid ].next;
2711 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2713 /* charid is now 0 if we dont know the char read, or
2714 * nonzero if we do */
2717 accept_state = TRIE_NODENUM( state );
2718 TRIE_HANDLE_WORD(accept_state);
2720 } /* end second pass */
2722 /* and now dump it out before we compress it */
2723 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
2725 next_alloc, depth+1));
2729 * Inplace compress the table.*
2731 For sparse data sets the table constructed by the trie algorithm will
2732 be mostly 0/FAIL transitions or to put it another way mostly empty.
2733 (Note that leaf nodes will not contain any transitions.)
2735 This algorithm compresses the tables by eliminating most such
2736 transitions, at the cost of a modest bit of extra work during lookup:
2738 - Each states[] entry contains a .base field which indicates the
2739 index in the state[] array wheres its transition data is stored.
2741 - If .base is 0 there are no valid transitions from that node.
2743 - If .base is nonzero then charid is added to it to find an entry in
2746 -If trans[states[state].base+charid].check!=state then the
2747 transition is taken to be a 0/Fail transition. Thus if there are fail
2748 transitions at the front of the node then the .base offset will point
2749 somewhere inside the previous nodes data (or maybe even into a node
2750 even earlier), but the .check field determines if the transition is
2754 The following process inplace converts the table to the compressed
2755 table: We first do not compress the root node 1,and mark all its
2756 .check pointers as 1 and set its .base pointer as 1 as well. This
2757 allows us to do a DFA construction from the compressed table later,
2758 and ensures that any .base pointers we calculate later are greater
2761 - We set 'pos' to indicate the first entry of the second node.
2763 - We then iterate over the columns of the node, finding the first and
2764 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
2765 and set the .check pointers accordingly, and advance pos
2766 appropriately and repreat for the next node. Note that when we copy
2767 the next pointers we have to convert them from the original
2768 NODEIDX form to NODENUM form as the former is not valid post
2771 - If a node has no transitions used we mark its base as 0 and do not
2772 advance the pos pointer.
2774 - If a node only has one transition we use a second pointer into the
2775 structure to fill in allocated fail transitions from other states.
2776 This pointer is independent of the main pointer and scans forward
2777 looking for null transitions that are allocated to a state. When it
2778 finds one it writes the single transition into the "hole". If the
2779 pointer doesnt find one the single transition is appended as normal.
2781 - Once compressed we can Renew/realloc the structures to release the
2784 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
2785 specifically Fig 3.47 and the associated pseudocode.
2789 const U32 laststate = TRIE_NODENUM( next_alloc );
2792 trie->statecount = laststate;
2794 for ( state = 1 ; state < laststate ; state++ ) {
2796 const U32 stateidx = TRIE_NODEIDX( state );
2797 const U32 o_used = trie->trans[ stateidx ].check;
2798 U32 used = trie->trans[ stateidx ].check;
2799 trie->trans[ stateidx ].check = 0;
2802 used && charid < trie->uniquecharcount;
2805 if ( flag || trie->trans[ stateidx + charid ].next ) {
2806 if ( trie->trans[ stateidx + charid ].next ) {
2808 for ( ; zp < pos ; zp++ ) {
2809 if ( ! trie->trans[ zp ].next ) {
2813 trie->states[ state ].trans.base
2815 + trie->uniquecharcount
2817 trie->trans[ zp ].next
2818 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
2820 trie->trans[ zp ].check = state;
2821 if ( ++zp > pos ) pos = zp;
2828 trie->states[ state ].trans.base
2829 = pos + trie->uniquecharcount - charid ;
2831 trie->trans[ pos ].next
2832 = SAFE_TRIE_NODENUM(
2833 trie->trans[ stateidx + charid ].next );
2834 trie->trans[ pos ].check = state;
2839 trie->lasttrans = pos + 1;
2840 trie->states = (reg_trie_state *)
2841 PerlMemShared_realloc( trie->states, laststate
2842 * sizeof(reg_trie_state) );
2843 DEBUG_TRIE_COMPILE_MORE_r(
2844 PerlIO_printf( Perl_debug_log,
2845 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
2846 (int)depth * 2 + 2,"",
2847 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
2851 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
2854 } /* end table compress */
2856 DEBUG_TRIE_COMPILE_MORE_r(
2857 PerlIO_printf(Perl_debug_log,
2858 "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
2859 (int)depth * 2 + 2, "",
2860 (UV)trie->statecount,
2861 (UV)trie->lasttrans)
2863 /* resize the trans array to remove unused space */
2864 trie->trans = (reg_trie_trans *)
2865 PerlMemShared_realloc( trie->trans, trie->lasttrans
2866 * sizeof(reg_trie_trans) );
2868 { /* Modify the program and insert the new TRIE node */
2869 U8 nodetype =(U8)(flags & 0xFF);
2873 regnode *optimize = NULL;
2874 #ifdef RE_TRACK_PATTERN_OFFSETS
2877 U32 mjd_nodelen = 0;
2878 #endif /* RE_TRACK_PATTERN_OFFSETS */
2879 #endif /* DEBUGGING */
2881 This means we convert either the first branch or the first Exact,
2882 depending on whether the thing following (in 'last') is a branch
2883 or not and whther first is the startbranch (ie is it a sub part of
2884 the alternation or is it the whole thing.)
2885 Assuming its a sub part we convert the EXACT otherwise we convert
2886 the whole branch sequence, including the first.
2888 /* Find the node we are going to overwrite */
2889 if ( first != startbranch || OP( last ) == BRANCH ) {
2890 /* branch sub-chain */
2891 NEXT_OFF( first ) = (U16)(last - first);
2892 #ifdef RE_TRACK_PATTERN_OFFSETS
2894 mjd_offset= Node_Offset((convert));
2895 mjd_nodelen= Node_Length((convert));
2898 /* whole branch chain */
2900 #ifdef RE_TRACK_PATTERN_OFFSETS
2903 const regnode *nop = NEXTOPER( convert );
2904 mjd_offset= Node_Offset((nop));
2905 mjd_nodelen= Node_Length((nop));
2909 PerlIO_printf(Perl_debug_log,
2910 "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2911 (int)depth * 2 + 2, "",
2912 (UV)mjd_offset, (UV)mjd_nodelen)
2915 /* But first we check to see if there is a common prefix we can
2916 split out as an EXACT and put in front of the TRIE node. */
2917 trie->startstate= 1;
2918 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2920 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2924 const U32 base = trie->states[ state ].trans.base;
2926 if ( trie->states[state].wordnum )
2929 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2930 if ( ( base + ofs >= trie->uniquecharcount ) &&
2931 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2932 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2934 if ( ++count > 1 ) {
2935 SV **tmp = av_fetch( revcharmap, ofs, 0);
2936 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2937 if ( state == 1 ) break;
2939 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2941 PerlIO_printf(Perl_debug_log,
2942 "%*sNew Start State=%"UVuf" Class: [",
2943 (int)depth * 2 + 2, "",
2946 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2947 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2949 TRIE_BITMAP_SET(trie,*ch);
2951 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2953 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2957 TRIE_BITMAP_SET(trie,*ch);
2959 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2960 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2966 SV **tmp = av_fetch( revcharmap, idx, 0);
2968 char *ch = SvPV( *tmp, len );
2970 SV *sv=sv_newmortal();
2971 PerlIO_printf( Perl_debug_log,
2972 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2973 (int)depth * 2 + 2, "",
2975 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2976 PL_colors[0], PL_colors[1],
2977 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2978 PERL_PV_ESCAPE_FIRSTCHAR
2983 OP( convert ) = nodetype;
2984 str=STRING(convert);
2987 STR_LEN(convert) += len;
2993 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2998 trie->prefixlen = (state-1);
3000 regnode *n = convert+NODE_SZ_STR(convert);
3001 NEXT_OFF(convert) = NODE_SZ_STR(convert);
3002 trie->startstate = state;
3003 trie->minlen -= (state - 1);
3004 trie->maxlen -= (state - 1);
3006 /* At least the UNICOS C compiler choked on this
3007 * being argument to DEBUG_r(), so let's just have
3010 #ifdef PERL_EXT_RE_BUILD
3016 regnode *fix = convert;
3017 U32 word = trie->wordcount;
3019 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
3020 while( ++fix < n ) {
3021 Set_Node_Offset_Length(fix, 0, 0);
3024 SV ** const tmp = av_fetch( trie_words, word, 0 );
3026 if ( STR_LEN(convert) <= SvCUR(*tmp) )
3027 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
3029 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
3037 NEXT_OFF(convert) = (U16)(tail - convert);
3038 DEBUG_r(optimize= n);
3044 if ( trie->maxlen ) {
3045 NEXT_OFF( convert ) = (U16)(tail - convert);
3046 ARG_SET( convert, data_slot );
3047 /* Store the offset to the first unabsorbed branch in
3048 jump[0], which is otherwise unused by the jump logic.
3049 We use this when dumping a trie and during optimisation. */
3051 trie->jump[0] = (U16)(nextbranch - convert);
3053 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
3054 * and there is a bitmap
3055 * and the first "jump target" node we found leaves enough room
3056 * then convert the TRIE node into a TRIEC node, with the bitmap
3057 * embedded inline in the opcode - this is hypothetically faster.
3059 if ( !trie->states[trie->startstate].wordnum
3061 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
3063 OP( convert ) = TRIEC;
3064 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
3065 PerlMemShared_free(trie->bitmap);
3068 OP( convert ) = TRIE;
3070 /* store the type in the flags */
3071 convert->flags = nodetype;
3075 + regarglen[ OP( convert ) ];
3077 /* XXX We really should free up the resource in trie now,
3078 as we won't use them - (which resources?) dmq */
3080 /* needed for dumping*/
3081 DEBUG_r(if (optimize) {
3082 regnode *opt = convert;
3084 while ( ++opt < optimize) {
3085 Set_Node_Offset_Length(opt,0,0);
3088 Try to clean up some of the debris left after the
3091 while( optimize < jumper ) {
3092 mjd_nodelen += Node_Length((optimize));
3093 OP( optimize ) = OPTIMIZED;
3094 Set_Node_Offset_Length(optimize,0,0);
3097 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
3099 } /* end node insert */
3101 /* Finish populating the prev field of the wordinfo array. Walk back
3102 * from each accept state until we find another accept state, and if
3103 * so, point the first word's .prev field at the second word. If the
3104 * second already has a .prev field set, stop now. This will be the
3105 * case either if we've already processed that word's accept state,
3106 * or that state had multiple words, and the overspill words were
3107 * already linked up earlier.
3114 for (word=1; word <= trie->wordcount; word++) {
3116 if (trie->wordinfo[word].prev)
3118 state = trie->wordinfo[word].accept;
3120 state = prev_states[state];
3123 prev = trie->states[state].wordnum;
3127 trie->wordinfo[word].prev = prev;
3129 Safefree(prev_states);
3133 /* and now dump out the compressed format */
3134 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
3136 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
3138 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
3139 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
3141 SvREFCNT_dec_NN(revcharmap);
3145 : trie->startstate>1
3151 S_construct_ahocorasick_from_trie(pTHX_ RExC_state_t *pRExC_state, regnode *source, U32 depth)
3153 /* The Trie is constructed and compressed now so we can build a fail array if
3156 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and
3158 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi,
3162 We find the fail state for each state in the trie, this state is the longest
3163 proper suffix of the current state's 'word' that is also a proper prefix of
3164 another word in our trie. State 1 represents the word '' and is thus the
3165 default fail state. This allows the DFA not to have to restart after its
3166 tried and failed a word at a given point, it simply continues as though it
3167 had been matching the other word in the first place.
3169 'abcdgu'=~/abcdefg|cdgu/
3170 When we get to 'd' we are still matching the first word, we would encounter
3171 'g' which would fail, which would bring us to the state representing 'd' in
3172 the second word where we would try 'g' and succeed, proceeding to match
3175 /* add a fail transition */
3176 const U32 trie_offset = ARG(source);
3177 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
3179 const U32 ucharcount = trie->uniquecharcount;
3180 const U32 numstates = trie->statecount;
3181 const U32 ubound = trie->lasttrans + ucharcount;
3185 U32 base = trie->states[ 1 ].trans.base;
3188 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T"));
3190 GET_RE_DEBUG_FLAGS_DECL;
3192 PERL_ARGS_ASSERT_CONSTRUCT_AHOCORASICK_FROM_TRIE;
3193 PERL_UNUSED_CONTEXT;
3195 PERL_UNUSED_ARG(depth);
3198 if ( OP(source) == TRIE ) {
3199 struct regnode_1 *op = (struct regnode_1 *)
3200 PerlMemShared_calloc(1, sizeof(struct regnode_1));
3201 StructCopy(source,op,struct regnode_1);
3202 stclass = (regnode *)op;
3204 struct regnode_charclass *op = (struct regnode_charclass *)
3205 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
3206 StructCopy(source,op,struct regnode_charclass);
3207 stclass = (regnode *)op;
3209 OP(stclass)+=2; /* convert the TRIE type to its AHO-CORASICK equivalent */
3211 ARG_SET( stclass, data_slot );
3212 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
3213 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3214 aho->trie=trie_offset;
3215 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
3216 Copy( trie->states, aho->states, numstates, reg_trie_state );
3217 Newxz( q, numstates, U32);
3218 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
3221 /* initialize fail[0..1] to be 1 so that we always have
3222 a valid final fail state */
3223 fail[ 0 ] = fail[ 1 ] = 1;
3225 for ( charid = 0; charid < ucharcount ; charid++ ) {
3226 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
3228 q[ q_write ] = newstate;
3229 /* set to point at the root */
3230 fail[ q[ q_write++ ] ]=1;
3233 while ( q_read < q_write) {
3234 const U32 cur = q[ q_read++ % numstates ];
3235 base = trie->states[ cur ].trans.base;
3237 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
3238 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
3240 U32 fail_state = cur;
3243 fail_state = fail[ fail_state ];
3244 fail_base = aho->states[ fail_state ].trans.base;
3245 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
3247 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
3248 fail[ ch_state ] = fail_state;
3249 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
3251 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
3253 q[ q_write++ % numstates] = ch_state;
3257 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
3258 when we fail in state 1, this allows us to use the
3259 charclass scan to find a valid start char. This is based on the principle
3260 that theres a good chance the string being searched contains lots of stuff
3261 that cant be a start char.
3263 fail[ 0 ] = fail[ 1 ] = 0;
3264 DEBUG_TRIE_COMPILE_r({
3265 PerlIO_printf(Perl_debug_log,
3266 "%*sStclass Failtable (%"UVuf" states): 0",
3267 (int)(depth * 2), "", (UV)numstates
3269 for( q_read=1; q_read<numstates; q_read++ ) {
3270 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
3272 PerlIO_printf(Perl_debug_log, "\n");
3275 /*RExC_seen |= REG_TRIEDFA_SEEN;*/
3280 #define DEBUG_PEEP(str,scan,depth) \
3281 DEBUG_OPTIMISE_r({if (scan){ \
3282 regnode *Next = regnext(scan); \
3283 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state); \
3284 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)", \
3285 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),\
3286 Next ? (REG_NODE_NUM(Next)) : 0 ); \
3287 DEBUG_SHOW_STUDY_FLAGS(flags," [ ","]");\
3288 PerlIO_printf(Perl_debug_log, "\n"); \
3291 /* The below joins as many adjacent EXACTish nodes as possible into a single
3292 * one. The regop may be changed if the node(s) contain certain sequences that
3293 * require special handling. The joining is only done if:
3294 * 1) there is room in the current conglomerated node to entirely contain the
3296 * 2) they are the exact same node type
3298 * The adjacent nodes actually may be separated by NOTHING-kind nodes, and
3299 * these get optimized out
3301 * If a node is to match under /i (folded), the number of characters it matches
3302 * can be different than its character length if it contains a multi-character
3303 * fold. *min_subtract is set to the total delta number of characters of the
3306 * And *unfolded_multi_char is set to indicate whether or not the node contains
3307 * an unfolded multi-char fold. This happens when whether the fold is valid or
3308 * not won't be known until runtime; namely for EXACTF nodes that contain LATIN
3309 * SMALL LETTER SHARP S, as only if the target string being matched against
3310 * turns out to be UTF-8 is that fold valid; and also for EXACTFL nodes whose
3311 * folding rules depend on the locale in force at runtime. (Multi-char folds
3312 * whose components are all above the Latin1 range are not run-time locale
3313 * dependent, and have already been folded by the time this function is
3316 * This is as good a place as any to discuss the design of handling these
3317 * multi-character fold sequences. It's been wrong in Perl for a very long
3318 * time. There are three code points in Unicode whose multi-character folds
3319 * were long ago discovered to mess things up. The previous designs for
3320 * dealing with these involved assigning a special node for them. This
3321 * approach doesn't always work, as evidenced by this example:
3322 * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches
3323 * Both sides fold to "sss", but if the pattern is parsed to create a node that
3324 * would match just the \xDF, it won't be able to handle the case where a
3325 * successful match would have to cross the node's boundary. The new approach
3326 * that hopefully generally solves the problem generates an EXACTFU_SS node
3327 * that is "sss" in this case.
3329 * It turns out that there are problems with all multi-character folds, and not
3330 * just these three. Now the code is general, for all such cases. The
3331 * approach taken is:
3332 * 1) This routine examines each EXACTFish node that could contain multi-
3333 * character folded sequences. Since a single character can fold into
3334 * such a sequence, the minimum match length for this node is less than
3335 * the number of characters in the node. This routine returns in
3336 * *min_subtract how many characters to subtract from the the actual
3337 * length of the string to get a real minimum match length; it is 0 if
3338 * there are no multi-char foldeds. This delta is used by the caller to
3339 * adjust the min length of the match, and the delta between min and max,
3340 * so that the optimizer doesn't reject these possibilities based on size
3342 * 2) For the sequence involving the Sharp s (\xDF), the node type EXACTFU_SS
3343 * is used for an EXACTFU node that contains at least one "ss" sequence in
3344 * it. For non-UTF-8 patterns and strings, this is the only case where
3345 * there is a possible fold length change. That means that a regular
3346 * EXACTFU node without UTF-8 involvement doesn't have to concern itself
3347 * with length changes, and so can be processed faster. regexec.c takes
3348 * advantage of this. Generally, an EXACTFish node that is in UTF-8 is
3349 * pre-folded by regcomp.c (except EXACTFL, some of whose folds aren't
3350 * known until runtime). This saves effort in regex matching. However,
3351 * the pre-folding isn't done for non-UTF8 patterns because the fold of
3352 * the MICRO SIGN requires UTF-8, and we don't want to slow things down by
3353 * forcing the pattern into UTF8 unless necessary. Also what EXACTF (and,
3354 * again, EXACTFL) nodes fold to isn't known until runtime. The fold
3355 * possibilities for the non-UTF8 patterns are quite simple, except for
3356 * the sharp s. All the ones that don't involve a UTF-8 target string are
3357 * members of a fold-pair, and arrays are set up for all of them so that
3358 * the other member of the pair can be found quickly. Code elsewhere in
3359 * this file makes sure that in EXACTFU nodes, the sharp s gets folded to
3360 * 'ss', even if the pattern isn't UTF-8. This avoids the issues
3361 * described in the next item.
3362 * 3) A problem remains for unfolded multi-char folds. (These occur when the
3363 * validity of the fold won't be known until runtime, and so must remain
3364 * unfolded for now. This happens for the sharp s in EXACTF and EXACTFA
3365 * nodes when the pattern isn't in UTF-8. (Note, BTW, that there cannot
3366 * be an EXACTF node with a UTF-8 pattern.) They also occur for various
3367 * folds in EXACTFL nodes, regardless of the UTF-ness of the pattern.)
3368 * The reason this is a problem is that the optimizer part of regexec.c
3369 * (probably unwittingly, in Perl_regexec_flags()) makes an assumption
3370 * that a character in the pattern corresponds to at most a single
3371 * character in the target string. (And I do mean character, and not byte
3372 * here, unlike other parts of the documentation that have never been
3373 * updated to account for multibyte Unicode.) sharp s in EXACTF and
3374 * EXACTFL nodes can match the two character string 'ss'; in EXACTFA nodes
3375 * it can match "\x{17F}\x{17F}". These, along with other ones in EXACTFL
3376 * nodes, violate the assumption, and they are the only instances where it
3377 * is violated. I'm reluctant to try to change the assumption, as the
3378 * code involved is impenetrable to me (khw), so instead the code here
3379 * punts. This routine examines EXACTFL nodes, and (when the pattern
3380 * isn't UTF-8) EXACTF and EXACTFA for such unfolded folds, and returns a
3381 * boolean indicating whether or not the node contains such a fold. When
3382 * it is true, the caller sets a flag that later causes the optimizer in
3383 * this file to not set values for the floating and fixed string lengths,
3384 * and thus avoids the optimizer code in regexec.c that makes the invalid
3385 * assumption. Thus, there is no optimization based on string lengths for
3386 * EXACTFL nodes that contain these few folds, nor for non-UTF8-pattern
3387 * EXACTF and EXACTFA nodes that contain the sharp s. (The reason the
3388 * assumption is wrong only in these cases is that all other non-UTF-8
3389 * folds are 1-1; and, for UTF-8 patterns, we pre-fold all other folds to
3390 * their expanded versions. (Again, we can't prefold sharp s to 'ss' in
3391 * EXACTF nodes because we don't know at compile time if it actually
3392 * matches 'ss' or not. For EXACTF nodes it will match iff the target
3393 * string is in UTF-8. This is in contrast to EXACTFU nodes, where it
3394 * always matches; and EXACTFA where it never does. In an EXACTFA node in
3395 * a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the
3396 * problem; but in a non-UTF8 pattern, folding it to that above-Latin1
3397 * string would require the pattern to be forced into UTF-8, the overhead
3398 * of which we want to avoid. Similarly the unfolded multi-char folds in
3399 * EXACTFL nodes will match iff the locale at the time of match is a UTF-8
3402 * Similarly, the code that generates tries doesn't currently handle
3403 * not-already-folded multi-char folds, and it looks like a pain to change
3404 * that. Therefore, trie generation of EXACTFA nodes with the sharp s
3405 * doesn't work. Instead, such an EXACTFA is turned into a new regnode,
3406 * EXACTFA_NO_TRIE, which the trie code knows not to handle. Most people
3407 * using /iaa matching will be doing so almost entirely with ASCII
3408 * strings, so this should rarely be encountered in practice */
3410 #define JOIN_EXACT(scan,min_subtract,unfolded_multi_char, flags) \
3411 if (PL_regkind[OP(scan)] == EXACT) \
3412 join_exact(pRExC_state,(scan),(min_subtract),unfolded_multi_char, (flags),NULL,depth+1)
3415 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
3416 UV *min_subtract, bool *unfolded_multi_char,
3417 U32 flags,regnode *val, U32 depth)
3419 /* Merge several consecutive EXACTish nodes into one. */
3420 regnode *n = regnext(scan);
3422 regnode *next = scan + NODE_SZ_STR(scan);
3426 regnode *stop = scan;
3427 GET_RE_DEBUG_FLAGS_DECL;
3429 PERL_UNUSED_ARG(depth);
3432 PERL_ARGS_ASSERT_JOIN_EXACT;
3433 #ifndef EXPERIMENTAL_INPLACESCAN
3434 PERL_UNUSED_ARG(flags);
3435 PERL_UNUSED_ARG(val);
3437 DEBUG_PEEP("join",scan,depth);
3439 /* Look through the subsequent nodes in the chain. Skip NOTHING, merge
3440 * EXACT ones that are mergeable to the current one. */
3442 && (PL_regkind[OP(n)] == NOTHING
3443 || (stringok && OP(n) == OP(scan)))
3445 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX)
3448 if (OP(n) == TAIL || n > next)
3450 if (PL_regkind[OP(n)] == NOTHING) {
3451 DEBUG_PEEP("skip:",n,depth);
3452 NEXT_OFF(scan) += NEXT_OFF(n);
3453 next = n + NODE_STEP_REGNODE;
3460 else if (stringok) {
3461 const unsigned int oldl = STR_LEN(scan);
3462 regnode * const nnext = regnext(n);
3464 /* XXX I (khw) kind of doubt that this works on platforms (should
3465 * Perl ever run on one) where U8_MAX is above 255 because of lots
3466 * of other assumptions */
3467 /* Don't join if the sum can't fit into a single node */
3468 if (oldl + STR_LEN(n) > U8_MAX)
3471 DEBUG_PEEP("merg",n,depth);
3474 NEXT_OFF(scan) += NEXT_OFF(n);
3475 STR_LEN(scan) += STR_LEN(n);
3476 next = n + NODE_SZ_STR(n);
3477 /* Now we can overwrite *n : */
3478 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
3486 #ifdef EXPERIMENTAL_INPLACESCAN
3487 if (flags && !NEXT_OFF(n)) {
3488 DEBUG_PEEP("atch", val, depth);
3489 if (reg_off_by_arg[OP(n)]) {
3490 ARG_SET(n, val - n);
3493 NEXT_OFF(n) = val - n;
3501 *unfolded_multi_char = FALSE;
3503 /* Here, all the adjacent mergeable EXACTish nodes have been merged. We
3504 * can now analyze for sequences of problematic code points. (Prior to
3505 * this final joining, sequences could have been split over boundaries, and
3506 * hence missed). The sequences only happen in folding, hence for any
3507 * non-EXACT EXACTish node */
3508 if (OP(scan) != EXACT && OP(scan) != EXACTL) {
3509 U8* s0 = (U8*) STRING(scan);
3511 U8* s_end = s0 + STR_LEN(scan);
3513 int total_count_delta = 0; /* Total delta number of characters that
3514 multi-char folds expand to */
3516 /* One pass is made over the node's string looking for all the
3517 * possibilities. To avoid some tests in the loop, there are two main
3518 * cases, for UTF-8 patterns (which can't have EXACTF nodes) and
3523 if (OP(scan) == EXACTFL) {
3526 /* An EXACTFL node would already have been changed to another
3527 * node type unless there is at least one character in it that
3528 * is problematic; likely a character whose fold definition
3529 * won't be known until runtime, and so has yet to be folded.
3530 * For all but the UTF-8 locale, folds are 1-1 in length, but
3531 * to handle the UTF-8 case, we need to create a temporary
3532 * folded copy using UTF-8 locale rules in order to analyze it.
3533 * This is because our macros that look to see if a sequence is
3534 * a multi-char fold assume everything is folded (otherwise the
3535 * tests in those macros would be too complicated and slow).
3536 * Note that here, the non-problematic folds will have already
3537 * been done, so we can just copy such characters. We actually
3538 * don't completely fold the EXACTFL string. We skip the
3539 * unfolded multi-char folds, as that would just create work
3540 * below to figure out the size they already are */
3542 Newx(folded, UTF8_MAX_FOLD_CHAR_EXPAND * STR_LEN(scan) + 1, U8);
3545 STRLEN s_len = UTF8SKIP(s);
3546 if (! is_PROBLEMATIC_LOCALE_FOLD_utf8(s)) {
3547 Copy(s, d, s_len, U8);
3550 else if (is_FOLDS_TO_MULTI_utf8(s)) {
3551 *unfolded_multi_char = TRUE;
3552 Copy(s, d, s_len, U8);
3555 else if (isASCII(*s)) {
3556 *(d++) = toFOLD(*s);
3560 _to_utf8_fold_flags(s, d, &len, FOLD_FLAGS_FULL);
3566 /* Point the remainder of the routine to look at our temporary
3570 } /* End of creating folded copy of EXACTFL string */
3572 /* Examine the string for a multi-character fold sequence. UTF-8
3573 * patterns have all characters pre-folded by the time this code is
3575 while (s < s_end - 1) /* Can stop 1 before the end, as minimum
3576 length sequence we are looking for is 2 */
3578 int count = 0; /* How many characters in a multi-char fold */
3579 int len = is_MULTI_CHAR_FOLD_utf8_safe(s, s_end);
3580 if (! len) { /* Not a multi-char fold: get next char */
3585 /* Nodes with 'ss' require special handling, except for
3586 * EXACTFA-ish for which there is no multi-char fold to this */
3587 if (len == 2 && *s == 's' && *(s+1) == 's'
3588 && OP(scan) != EXACTFA
3589 && OP(scan) != EXACTFA_NO_TRIE)
3592 if (OP(scan) != EXACTFL) {
3593 OP(scan) = EXACTFU_SS;
3597 else { /* Here is a generic multi-char fold. */
3598 U8* multi_end = s + len;
3600 /* Count how many characters are in it. In the case of
3601 * /aa, no folds which contain ASCII code points are
3602 * allowed, so check for those, and skip if found. */
3603 if (OP(scan) != EXACTFA && OP(scan) != EXACTFA_NO_TRIE) {
3604 count = utf8_length(s, multi_end);
3608 while (s < multi_end) {
3611 goto next_iteration;
3621 /* The delta is how long the sequence is minus 1 (1 is how long
3622 * the character that folds to the sequence is) */
3623 total_count_delta += count - 1;
3627 /* We created a temporary folded copy of the string in EXACTFL
3628 * nodes. Therefore we need to be sure it doesn't go below zero,
3629 * as the real string could be shorter */
3630 if (OP(scan) == EXACTFL) {
3631 int total_chars = utf8_length((U8*) STRING(scan),
3632 (U8*) STRING(scan) + STR_LEN(scan));
3633 if (total_count_delta > total_chars) {
3634 total_count_delta = total_chars;
3638 *min_subtract += total_count_delta;
3641 else if (OP(scan) == EXACTFA) {
3643 /* Non-UTF-8 pattern, EXACTFA node. There can't be a multi-char
3644 * fold to the ASCII range (and there are no existing ones in the
3645 * upper latin1 range). But, as outlined in the comments preceding
3646 * this function, we need to flag any occurrences of the sharp s.
3647 * This character forbids trie formation (because of added
3650 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
3651 OP(scan) = EXACTFA_NO_TRIE;
3652 *unfolded_multi_char = TRUE;
3661 /* Non-UTF-8 pattern, not EXACTFA node. Look for the multi-char
3662 * folds that are all Latin1. As explained in the comments
3663 * preceding this function, we look also for the sharp s in EXACTF
3664 * and EXACTFL nodes; it can be in the final position. Otherwise
3665 * we can stop looking 1 byte earlier because have to find at least
3666 * two characters for a multi-fold */
3667 const U8* upper = (OP(scan) == EXACTF || OP(scan) == EXACTFL)
3672 int len = is_MULTI_CHAR_FOLD_latin1_safe(s, s_end);
3673 if (! len) { /* Not a multi-char fold. */
3674 if (*s == LATIN_SMALL_LETTER_SHARP_S
3675 && (OP(scan) == EXACTF || OP(scan) == EXACTFL))
3677 *unfolded_multi_char = TRUE;
3684 && isALPHA_FOLD_EQ(*s, 's')
3685 && isALPHA_FOLD_EQ(*(s+1), 's'))
3688 /* EXACTF nodes need to know that the minimum length
3689 * changed so that a sharp s in the string can match this
3690 * ss in the pattern, but they remain EXACTF nodes, as they
3691 * won't match this unless the target string is is UTF-8,
3692 * which we don't know until runtime. EXACTFL nodes can't
3693 * transform into EXACTFU nodes */
3694 if (OP(scan) != EXACTF && OP(scan) != EXACTFL) {
3695 OP(scan) = EXACTFU_SS;
3699 *min_subtract += len - 1;
3706 /* Allow dumping but overwriting the collection of skipped
3707 * ops and/or strings with fake optimized ops */
3708 n = scan + NODE_SZ_STR(scan);
3716 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
3720 /* REx optimizer. Converts nodes into quicker variants "in place".
3721 Finds fixed substrings. */
3723 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
3724 to the position after last scanned or to NULL. */
3726 #define INIT_AND_WITHP \
3727 assert(!and_withp); \
3728 Newx(and_withp,1, regnode_ssc); \
3729 SAVEFREEPV(and_withp)
3733 S_unwind_scan_frames(pTHX_ const void *p)
3735 scan_frame *f= (scan_frame *)p;
3737 scan_frame *n= f->next_frame;
3745 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
3746 SSize_t *minlenp, SSize_t *deltap,
3751 regnode_ssc *and_withp,
3752 U32 flags, U32 depth)
3753 /* scanp: Start here (read-write). */
3754 /* deltap: Write maxlen-minlen here. */
3755 /* last: Stop before this one. */
3756 /* data: string data about the pattern */
3757 /* stopparen: treat close N as END */
3758 /* recursed: which subroutines have we recursed into */
3759 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
3761 /* There must be at least this number of characters to match */
3764 regnode *scan = *scanp, *next;
3766 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
3767 int is_inf_internal = 0; /* The studied chunk is infinite */
3768 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
3769 scan_data_t data_fake;
3770 SV *re_trie_maxbuff = NULL;
3771 regnode *first_non_open = scan;
3772 SSize_t stopmin = SSize_t_MAX;
3773 scan_frame *frame = NULL;
3774 GET_RE_DEBUG_FLAGS_DECL;
3776 PERL_ARGS_ASSERT_STUDY_CHUNK;
3780 while (first_non_open && OP(first_non_open) == OPEN)
3781 first_non_open=regnext(first_non_open);
3787 RExC_study_chunk_recursed_count++;
3789 DEBUG_OPTIMISE_MORE_r(
3791 PerlIO_printf(Perl_debug_log,
3792 "%*sstudy_chunk stopparen=%ld recursed_count=%lu depth=%lu recursed_depth=%lu scan=%p last=%p",
3793 (int)(depth*2), "", (long)stopparen,
3794 (unsigned long)RExC_study_chunk_recursed_count,
3795 (unsigned long)depth, (unsigned long)recursed_depth,
3798 if (recursed_depth) {
3801 for ( j = 0 ; j < recursed_depth ; j++ ) {
3802 for ( i = 0 ; i < (U32)RExC_npar ; i++ ) {
3804 PAREN_TEST(RExC_study_chunk_recursed +
3805 ( j * RExC_study_chunk_recursed_bytes), i )
3808 !PAREN_TEST(RExC_study_chunk_recursed +
3809 (( j - 1 ) * RExC_study_chunk_recursed_bytes), i)
3812 PerlIO_printf(Perl_debug_log," %d",(int)i);
3816 if ( j + 1 < recursed_depth ) {
3817 PerlIO_printf(Perl_debug_log, ",");
3821 PerlIO_printf(Perl_debug_log,"\n");
3824 while ( scan && OP(scan) != END && scan < last ){
3825 UV min_subtract = 0; /* How mmany chars to subtract from the minimum
3826 node length to get a real minimum (because
3827 the folded version may be shorter) */
3828 bool unfolded_multi_char = FALSE;
3829 /* Peephole optimizer: */
3830 DEBUG_STUDYDATA("Peep:", data, depth);
3831 DEBUG_PEEP("Peep", scan, depth);
3834 /* The reason we do this here we need to deal with things like /(?:f)(?:o)(?:o)/
3835 * which cant be dealt with by the normal EXACT parsing code, as each (?:..) is handled
3836 * by a different invocation of reg() -- Yves
3838 JOIN_EXACT(scan,&min_subtract, &unfolded_multi_char, 0);
3840 /* Follow the next-chain of the current node and optimize
3841 away all the NOTHINGs from it. */
3842 if (OP(scan) != CURLYX) {
3843 const int max = (reg_off_by_arg[OP(scan)]
3845 /* I32 may be smaller than U16 on CRAYs! */
3846 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
3847 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
3851 /* Skip NOTHING and LONGJMP. */
3852 while ((n = regnext(n))
3853 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
3854 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
3855 && off + noff < max)
3857 if (reg_off_by_arg[OP(scan)])
3860 NEXT_OFF(scan) = off;
3863 /* The principal pseudo-switch. Cannot be a switch, since we
3864 look into several different things. */
3865 if ( OP(scan) == DEFINEP ) {
3867 SSize_t deltanext = 0;
3868 SSize_t fake_last_close = 0;
3869 I32 f = SCF_IN_DEFINE;
3871 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
3872 scan = regnext(scan);
3873 assert( OP(scan) == IFTHEN );
3874 DEBUG_PEEP("expect IFTHEN", scan, depth);
3876 data_fake.last_closep= &fake_last_close;
3878 next = regnext(scan);
3879 scan = NEXTOPER(NEXTOPER(scan));
3880 DEBUG_PEEP("scan", scan, depth);
3881 DEBUG_PEEP("next", next, depth);
3883 /* we suppose the run is continuous, last=next...
3884 * NOTE we dont use the return here! */
3885 (void)study_chunk(pRExC_state, &scan, &minlen,
3886 &deltanext, next, &data_fake, stopparen,
3887 recursed_depth, NULL, f, depth+1);
3892 OP(scan) == BRANCH ||
3893 OP(scan) == BRANCHJ ||
3896 next = regnext(scan);
3899 /* The op(next)==code check below is to see if we
3900 * have "BRANCH-BRANCH", "BRANCHJ-BRANCHJ", "IFTHEN-IFTHEN"
3901 * IFTHEN is special as it might not appear in pairs.
3902 * Not sure whether BRANCH-BRANCHJ is possible, regardless
3903 * we dont handle it cleanly. */
3904 if (OP(next) == code || code == IFTHEN) {
3905 /* NOTE - There is similar code to this block below for
3906 * handling TRIE nodes on a re-study. If you change stuff here
3907 * check there too. */
3908 SSize_t max1 = 0, min1 = SSize_t_MAX, num = 0;
3910 regnode * const startbranch=scan;
3912 if (flags & SCF_DO_SUBSTR) {
3913 /* Cannot merge strings after this. */
3914 scan_commit(pRExC_state, data, minlenp, is_inf);
3917 if (flags & SCF_DO_STCLASS)
3918 ssc_init_zero(pRExC_state, &accum);
3920 while (OP(scan) == code) {
3921 SSize_t deltanext, minnext, fake;
3923 regnode_ssc this_class;
3925 DEBUG_PEEP("Branch", scan, depth);
3928 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
3930 data_fake.whilem_c = data->whilem_c;
3931 data_fake.last_closep = data->last_closep;
3934 data_fake.last_closep = &fake;
3936 data_fake.pos_delta = delta;
3937 next = regnext(scan);
3939 scan = NEXTOPER(scan); /* everything */
3940 if (code != BRANCH) /* everything but BRANCH */
3941 scan = NEXTOPER(scan);
3943 if (flags & SCF_DO_STCLASS) {
3944 ssc_init(pRExC_state, &this_class);
3945 data_fake.start_class = &this_class;
3946 f = SCF_DO_STCLASS_AND;
3948 if (flags & SCF_WHILEM_VISITED_POS)
3949 f |= SCF_WHILEM_VISITED_POS;
3951 /* we suppose the run is continuous, last=next...*/
3952 minnext = study_chunk(pRExC_state, &scan, minlenp,
3953 &deltanext, next, &data_fake, stopparen,
3954 recursed_depth, NULL, f,depth+1);
3958 if (deltanext == SSize_t_MAX) {
3959 is_inf = is_inf_internal = 1;
3961 } else if (max1 < minnext + deltanext)
3962 max1 = minnext + deltanext;
3964 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3966 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3967 if ( stopmin > minnext)
https://perl5.git.perl.org / perl5.git / blob