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 struct RExC_state_t {
126 U32 flags; /* RXf_* are we folding, multilining? */
127 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
128 char *precomp; /* uncompiled string. */
129 REGEXP *rx_sv; /* The SV that is the regexp. */
130 regexp *rx; /* perl core regexp structure */
131 regexp_internal *rxi; /* internal data for regexp object
133 char *start; /* Start of input for compile */
134 char *end; /* End of input for compile */
135 char *parse; /* Input-scan pointer. */
136 SSize_t whilem_seen; /* number of WHILEM in this expr */
137 regnode *emit_start; /* Start of emitted-code area */
138 regnode *emit_bound; /* First regnode outside of the
140 regnode *emit; /* Code-emit pointer; if = &emit_dummy,
141 implies compiling, so don't emit */
142 regnode_ssc emit_dummy; /* placeholder for emit to point to;
143 large enough for the largest
144 non-EXACTish node, so can use it as
146 I32 naughty; /* How bad is this pattern? */
147 I32 sawback; /* Did we see \1, ...? */
149 SSize_t size; /* Code size. */
150 I32 npar; /* Capture buffer count, (OPEN) plus
151 one. ("par" 0 is the whole
153 I32 nestroot; /* root parens we are in - used by
157 regnode **open_parens; /* pointers to open parens */
158 regnode **close_parens; /* pointers to close parens */
159 regnode *opend; /* END node in program */
160 I32 utf8; /* whether the pattern is utf8 or not */
161 I32 orig_utf8; /* whether the pattern was originally in utf8 */
162 /* XXX use this for future optimisation of case
163 * where pattern must be upgraded to utf8. */
164 I32 uni_semantics; /* If a d charset modifier should use unicode
165 rules, even if the pattern is not in
167 HV *paren_names; /* Paren names */
169 regnode **recurse; /* Recurse regops */
170 I32 recurse_count; /* Number of recurse regops */
171 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
173 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
177 I32 override_recoding;
178 I32 in_multi_char_class;
179 struct reg_code_block *code_blocks; /* positions of literal (?{})
181 int num_code_blocks; /* size of code_blocks[] */
182 int code_index; /* next code_blocks[] slot */
183 SSize_t maxlen; /* mininum possible number of chars in string to match */
184 scan_frame *frame_head;
185 scan_frame *frame_last;
187 #ifdef ADD_TO_REGEXEC
188 char *starttry; /* -Dr: where regtry was called. */
189 #define RExC_starttry (pRExC_state->starttry)
191 SV *runtime_code_qr; /* qr with the runtime code blocks */
193 const char *lastparse;
195 AV *paren_name_list; /* idx -> name */
196 U32 study_chunk_recursed_count;
199 #define RExC_lastparse (pRExC_state->lastparse)
200 #define RExC_lastnum (pRExC_state->lastnum)
201 #define RExC_paren_name_list (pRExC_state->paren_name_list)
202 #define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
203 #define RExC_mysv (pRExC_state->mysv1)
204 #define RExC_mysv1 (pRExC_state->mysv1)
205 #define RExC_mysv2 (pRExC_state->mysv2)
210 #define RExC_flags (pRExC_state->flags)
211 #define RExC_pm_flags (pRExC_state->pm_flags)
212 #define RExC_precomp (pRExC_state->precomp)
213 #define RExC_rx_sv (pRExC_state->rx_sv)
214 #define RExC_rx (pRExC_state->rx)
215 #define RExC_rxi (pRExC_state->rxi)
216 #define RExC_start (pRExC_state->start)
217 #define RExC_end (pRExC_state->end)
218 #define RExC_parse (pRExC_state->parse)
219 #define RExC_whilem_seen (pRExC_state->whilem_seen)
220 #ifdef RE_TRACK_PATTERN_OFFSETS
221 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the
224 #define RExC_emit (pRExC_state->emit)
225 #define RExC_emit_dummy (pRExC_state->emit_dummy)
226 #define RExC_emit_start (pRExC_state->emit_start)
227 #define RExC_emit_bound (pRExC_state->emit_bound)
228 #define RExC_sawback (pRExC_state->sawback)
229 #define RExC_seen (pRExC_state->seen)
230 #define RExC_size (pRExC_state->size)
231 #define RExC_maxlen (pRExC_state->maxlen)
232 #define RExC_npar (pRExC_state->npar)
233 #define RExC_nestroot (pRExC_state->nestroot)
234 #define RExC_extralen (pRExC_state->extralen)
235 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
236 #define RExC_utf8 (pRExC_state->utf8)
237 #define RExC_uni_semantics (pRExC_state->uni_semantics)
238 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
239 #define RExC_open_parens (pRExC_state->open_parens)
240 #define RExC_close_parens (pRExC_state->close_parens)
241 #define RExC_opend (pRExC_state->opend)
242 #define RExC_paren_names (pRExC_state->paren_names)
243 #define RExC_recurse (pRExC_state->recurse)
244 #define RExC_recurse_count (pRExC_state->recurse_count)
245 #define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
246 #define RExC_study_chunk_recursed_bytes \
247 (pRExC_state->study_chunk_recursed_bytes)
248 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
249 #define RExC_contains_locale (pRExC_state->contains_locale)
250 #define RExC_contains_i (pRExC_state->contains_i)
251 #define RExC_override_recoding (pRExC_state->override_recoding)
252 #define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
253 #define RExC_frame_head (pRExC_state->frame_head)
254 #define RExC_frame_last (pRExC_state->frame_last)
255 #define RExC_frame_count (pRExC_state->frame_count)
257 /* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
258 * a flag to disable back-off on the fixed/floating substrings - if it's
259 * a high complexity pattern we assume the benefit of avoiding a full match
260 * is worth the cost of checking for the substrings even if they rarely help.
262 #define RExC_naughty (pRExC_state->naughty)
263 #define TOO_NAUGHTY (10)
264 #define MARK_NAUGHTY(add) \
265 if (RExC_naughty < TOO_NAUGHTY) \
266 RExC_naughty += (add)
267 #define MARK_NAUGHTY_EXP(exp, add) \
268 if (RExC_naughty < TOO_NAUGHTY) \
269 RExC_naughty += RExC_naughty / (exp) + (add)
271 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
272 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
273 ((*s) == '{' && regcurly(s)))
276 * Flags to be passed up and down.
278 #define WORST 0 /* Worst case. */
279 #define HASWIDTH 0x01 /* Known to match non-null strings. */
281 /* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
282 * character. (There needs to be a case: in the switch statement in regexec.c
283 * for any node marked SIMPLE.) Note that this is not the same thing as
286 #define SPSTART 0x04 /* Starts with * or + */
287 #define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
288 #define TRYAGAIN 0x10 /* Weeded out a declaration. */
289 #define RESTART_UTF8 0x20 /* Restart, need to calcuate sizes as UTF-8 */
291 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
293 /* whether trie related optimizations are enabled */
294 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
295 #define TRIE_STUDY_OPT
296 #define FULL_TRIE_STUDY
302 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
303 #define PBITVAL(paren) (1 << ((paren) & 7))
304 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
305 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
306 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
308 #define REQUIRE_UTF8 STMT_START { \
310 *flagp = RESTART_UTF8; \
315 /* This converts the named class defined in regcomp.h to its equivalent class
316 * number defined in handy.h. */
317 #define namedclass_to_classnum(class) ((int) ((class) / 2))
318 #define classnum_to_namedclass(classnum) ((classnum) * 2)
320 #define _invlist_union_complement_2nd(a, b, output) \
321 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
322 #define _invlist_intersection_complement_2nd(a, b, output) \
323 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
325 /* About scan_data_t.
327 During optimisation we recurse through the regexp program performing
328 various inplace (keyhole style) optimisations. In addition study_chunk
329 and scan_commit populate this data structure with information about
330 what strings MUST appear in the pattern. We look for the longest
331 string that must appear at a fixed location, and we look for the
332 longest string that may appear at a floating location. So for instance
337 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
338 strings (because they follow a .* construct). study_chunk will identify
339 both FOO and BAR as being the longest fixed and floating strings respectively.
341 The strings can be composites, for instance
345 will result in a composite fixed substring 'foo'.
347 For each string some basic information is maintained:
349 - offset or min_offset
350 This is the position the string must appear at, or not before.
351 It also implicitly (when combined with minlenp) tells us how many
352 characters must match before the string we are searching for.
353 Likewise when combined with minlenp and the length of the string it
354 tells us how many characters must appear after the string we have
358 Only used for floating strings. This is the rightmost point that
359 the string can appear at. If set to SSize_t_MAX it indicates that the
360 string can occur infinitely far to the right.
363 A pointer to the minimum number of characters of the pattern that the
364 string was found inside. This is important as in the case of positive
365 lookahead or positive lookbehind we can have multiple patterns
370 The minimum length of the pattern overall is 3, the minimum length
371 of the lookahead part is 3, but the minimum length of the part that
372 will actually match is 1. So 'FOO's minimum length is 3, but the
373 minimum length for the F is 1. This is important as the minimum length
374 is used to determine offsets in front of and behind the string being
375 looked for. Since strings can be composites this is the length of the
376 pattern at the time it was committed with a scan_commit. Note that
377 the length is calculated by study_chunk, so that the minimum lengths
378 are not known until the full pattern has been compiled, thus the
379 pointer to the value.
383 In the case of lookbehind the string being searched for can be
384 offset past the start point of the final matching string.
385 If this value was just blithely removed from the min_offset it would
386 invalidate some of the calculations for how many chars must match
387 before or after (as they are derived from min_offset and minlen and
388 the length of the string being searched for).
389 When the final pattern is compiled and the data is moved from the
390 scan_data_t structure into the regexp structure the information
391 about lookbehind is factored in, with the information that would
392 have been lost precalculated in the end_shift field for the
395 The fields pos_min and pos_delta are used to store the minimum offset
396 and the delta to the maximum offset at the current point in the pattern.
400 typedef struct scan_data_t {
401 /*I32 len_min; unused */
402 /*I32 len_delta; unused */
406 SSize_t last_end; /* min value, <0 unless valid. */
407 SSize_t last_start_min;
408 SSize_t last_start_max;
409 SV **longest; /* Either &l_fixed, or &l_float. */
410 SV *longest_fixed; /* longest fixed string found in pattern */
411 SSize_t offset_fixed; /* offset where it starts */
412 SSize_t *minlen_fixed; /* pointer to the minlen relevant to the string */
413 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
414 SV *longest_float; /* longest floating string found in pattern */
415 SSize_t offset_float_min; /* earliest point in string it can appear */
416 SSize_t offset_float_max; /* latest point in string it can appear */
417 SSize_t *minlen_float; /* pointer to the minlen relevant to the string */
418 SSize_t lookbehind_float; /* is the pos of the string modified by LB */
421 SSize_t *last_closep;
422 regnode_ssc *start_class;
426 * Forward declarations for pregcomp()'s friends.
429 static const scan_data_t zero_scan_data =
430 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
432 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
433 #define SF_BEFORE_SEOL 0x0001
434 #define SF_BEFORE_MEOL 0x0002
435 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
436 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
438 #define SF_FIX_SHIFT_EOL (+2)
439 #define SF_FL_SHIFT_EOL (+4)
441 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
442 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
444 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
445 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
446 #define SF_IS_INF 0x0040
447 #define SF_HAS_PAR 0x0080
448 #define SF_IN_PAR 0x0100
449 #define SF_HAS_EVAL 0x0200
450 #define SCF_DO_SUBSTR 0x0400
451 #define SCF_DO_STCLASS_AND 0x0800
452 #define SCF_DO_STCLASS_OR 0x1000
453 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
454 #define SCF_WHILEM_VISITED_POS 0x2000
456 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
457 #define SCF_SEEN_ACCEPT 0x8000
458 #define SCF_TRIE_DOING_RESTUDY 0x10000
459 #define SCF_IN_DEFINE 0x20000
464 #define UTF cBOOL(RExC_utf8)
466 /* The enums for all these are ordered so things work out correctly */
467 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
468 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
469 == REGEX_DEPENDS_CHARSET)
470 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
471 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
472 >= REGEX_UNICODE_CHARSET)
473 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
474 == REGEX_ASCII_RESTRICTED_CHARSET)
475 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
476 >= REGEX_ASCII_RESTRICTED_CHARSET)
477 #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
478 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
480 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
482 /* For programs that want to be strictly Unicode compatible by dying if any
483 * attempt is made to match a non-Unicode code point against a Unicode
485 #define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
487 #define OOB_NAMEDCLASS -1
489 /* There is no code point that is out-of-bounds, so this is problematic. But
490 * its only current use is to initialize a variable that is always set before
492 #define OOB_UNICODE 0xDEADBEEF
494 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
495 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
498 /* length of regex to show in messages that don't mark a position within */
499 #define RegexLengthToShowInErrorMessages 127
502 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
503 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
504 * op/pragma/warn/regcomp.
506 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
507 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
509 #define REPORT_LOCATION " in regex; marked by " MARKER1 \
510 " in m/%"UTF8f MARKER2 "%"UTF8f"/"
512 #define REPORT_LOCATION_ARGS(offset) \
513 UTF8fARG(UTF, offset, RExC_precomp), \
514 UTF8fARG(UTF, RExC_end - RExC_precomp - offset, RExC_precomp + offset)
517 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
518 * arg. Show regex, up to a maximum length. If it's too long, chop and add
521 #define _FAIL(code) STMT_START { \
522 const char *ellipses = ""; \
523 IV len = RExC_end - RExC_precomp; \
526 SAVEFREESV(RExC_rx_sv); \
527 if (len > RegexLengthToShowInErrorMessages) { \
528 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
529 len = RegexLengthToShowInErrorMessages - 10; \
535 #define FAIL(msg) _FAIL( \
536 Perl_croak(aTHX_ "%s in regex m/%"UTF8f"%s/", \
537 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
539 #define FAIL2(msg,arg) _FAIL( \
540 Perl_croak(aTHX_ msg " in regex m/%"UTF8f"%s/", \
541 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
544 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
546 #define Simple_vFAIL(m) STMT_START { \
548 (RExC_parse > RExC_end ? RExC_end : RExC_parse) - RExC_precomp; \
549 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
550 m, REPORT_LOCATION_ARGS(offset)); \
554 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
556 #define vFAIL(m) STMT_START { \
558 SAVEFREESV(RExC_rx_sv); \
563 * Like Simple_vFAIL(), but accepts two arguments.
565 #define Simple_vFAIL2(m,a1) STMT_START { \
566 const IV offset = RExC_parse - RExC_precomp; \
567 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
568 REPORT_LOCATION_ARGS(offset)); \
572 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
574 #define vFAIL2(m,a1) STMT_START { \
576 SAVEFREESV(RExC_rx_sv); \
577 Simple_vFAIL2(m, a1); \
582 * Like Simple_vFAIL(), but accepts three arguments.
584 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
585 const IV offset = RExC_parse - RExC_precomp; \
586 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
587 REPORT_LOCATION_ARGS(offset)); \
591 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
593 #define vFAIL3(m,a1,a2) STMT_START { \
595 SAVEFREESV(RExC_rx_sv); \
596 Simple_vFAIL3(m, a1, a2); \
600 * Like Simple_vFAIL(), but accepts four arguments.
602 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
603 const IV offset = RExC_parse - RExC_precomp; \
604 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
605 REPORT_LOCATION_ARGS(offset)); \
608 #define vFAIL4(m,a1,a2,a3) STMT_START { \
610 SAVEFREESV(RExC_rx_sv); \
611 Simple_vFAIL4(m, a1, a2, a3); \
614 /* A specialized version of vFAIL2 that works with UTF8f */
615 #define vFAIL2utf8f(m, a1) STMT_START { \
616 const IV offset = RExC_parse - RExC_precomp; \
618 SAVEFREESV(RExC_rx_sv); \
619 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
620 REPORT_LOCATION_ARGS(offset)); \
623 /* These have asserts in them because of [perl #122671] Many warnings in
624 * regcomp.c can occur twice. If they get output in pass1 and later in that
625 * pass, the pattern has to be converted to UTF-8 and the pass restarted, they
626 * would get output again. So they should be output in pass2, and these
627 * asserts make sure new warnings follow that paradigm. */
629 /* m is not necessarily a "literal string", in this macro */
630 #define reg_warn_non_literal_string(loc, m) STMT_START { \
631 const IV offset = loc - RExC_precomp; \
632 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
633 m, REPORT_LOCATION_ARGS(offset)); \
636 #define ckWARNreg(loc,m) STMT_START { \
637 const IV offset = loc - RExC_precomp; \
638 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
639 REPORT_LOCATION_ARGS(offset)); \
642 #define vWARN_dep(loc, m) STMT_START { \
643 const IV offset = loc - RExC_precomp; \
644 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), m REPORT_LOCATION, \
645 REPORT_LOCATION_ARGS(offset)); \
648 #define ckWARNdep(loc,m) STMT_START { \
649 const IV offset = loc - RExC_precomp; \
650 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
652 REPORT_LOCATION_ARGS(offset)); \
655 #define ckWARNregdep(loc,m) STMT_START { \
656 const IV offset = loc - RExC_precomp; \
657 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
659 REPORT_LOCATION_ARGS(offset)); \
662 #define ckWARN2reg_d(loc,m, a1) STMT_START { \
663 const IV offset = loc - RExC_precomp; \
664 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
666 a1, REPORT_LOCATION_ARGS(offset)); \
669 #define ckWARN2reg(loc, m, a1) STMT_START { \
670 const IV offset = loc - RExC_precomp; \
671 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
672 a1, REPORT_LOCATION_ARGS(offset)); \
675 #define vWARN3(loc, m, a1, a2) STMT_START { \
676 const IV offset = loc - RExC_precomp; \
677 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
678 a1, a2, REPORT_LOCATION_ARGS(offset)); \
681 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
682 const IV offset = loc - RExC_precomp; \
683 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
684 a1, a2, REPORT_LOCATION_ARGS(offset)); \
687 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
688 const IV offset = loc - RExC_precomp; \
689 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
690 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
693 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
694 const IV offset = loc - RExC_precomp; \
695 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
696 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
699 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
700 const IV offset = loc - RExC_precomp; \
701 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
702 a1, a2, a3, a4, REPORT_LOCATION_ARGS(offset)); \
706 /* Allow for side effects in s */
707 #define REGC(c,s) STMT_START { \
708 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
711 /* Macros for recording node offsets. 20001227 mjd@plover.com
712 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
713 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
714 * Element 0 holds the number n.
715 * Position is 1 indexed.
717 #ifndef RE_TRACK_PATTERN_OFFSETS
718 #define Set_Node_Offset_To_R(node,byte)
719 #define Set_Node_Offset(node,byte)
720 #define Set_Cur_Node_Offset
721 #define Set_Node_Length_To_R(node,len)
722 #define Set_Node_Length(node,len)
723 #define Set_Node_Cur_Length(node,start)
724 #define Node_Offset(n)
725 #define Node_Length(n)
726 #define Set_Node_Offset_Length(node,offset,len)
727 #define ProgLen(ri) ri->u.proglen
728 #define SetProgLen(ri,x) ri->u.proglen = x
730 #define ProgLen(ri) ri->u.offsets[0]
731 #define SetProgLen(ri,x) ri->u.offsets[0] = x
732 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
734 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
735 __LINE__, (int)(node), (int)(byte))); \
737 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
740 RExC_offsets[2*(node)-1] = (byte); \
745 #define Set_Node_Offset(node,byte) \
746 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
747 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
749 #define Set_Node_Length_To_R(node,len) STMT_START { \
751 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
752 __LINE__, (int)(node), (int)(len))); \
754 Perl_croak(aTHX_ "value of node is %d in Length macro", \
757 RExC_offsets[2*(node)] = (len); \
762 #define Set_Node_Length(node,len) \
763 Set_Node_Length_To_R((node)-RExC_emit_start, len)
764 #define Set_Node_Cur_Length(node, start) \
765 Set_Node_Length(node, RExC_parse - start)
767 /* Get offsets and lengths */
768 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
769 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
771 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
772 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
773 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
777 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
778 #define EXPERIMENTAL_INPLACESCAN
779 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
781 #define DEBUG_RExC_seen() \
782 DEBUG_OPTIMISE_MORE_r({ \
783 PerlIO_printf(Perl_debug_log,"RExC_seen: "); \
785 if (RExC_seen & REG_ZERO_LEN_SEEN) \
786 PerlIO_printf(Perl_debug_log,"REG_ZERO_LEN_SEEN "); \
788 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
789 PerlIO_printf(Perl_debug_log,"REG_LOOKBEHIND_SEEN "); \
791 if (RExC_seen & REG_GPOS_SEEN) \
792 PerlIO_printf(Perl_debug_log,"REG_GPOS_SEEN "); \
794 if (RExC_seen & REG_CANY_SEEN) \
795 PerlIO_printf(Perl_debug_log,"REG_CANY_SEEN "); \
797 if (RExC_seen & REG_RECURSE_SEEN) \
798 PerlIO_printf(Perl_debug_log,"REG_RECURSE_SEEN "); \
800 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
801 PerlIO_printf(Perl_debug_log,"REG_TOP_LEVEL_BRANCHES_SEEN "); \
803 if (RExC_seen & REG_VERBARG_SEEN) \
804 PerlIO_printf(Perl_debug_log,"REG_VERBARG_SEEN "); \
806 if (RExC_seen & REG_CUTGROUP_SEEN) \
807 PerlIO_printf(Perl_debug_log,"REG_CUTGROUP_SEEN "); \
809 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
810 PerlIO_printf(Perl_debug_log,"REG_RUN_ON_COMMENT_SEEN "); \
812 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
813 PerlIO_printf(Perl_debug_log,"REG_UNFOLDED_MULTI_SEEN "); \
815 if (RExC_seen & REG_GOSTART_SEEN) \
816 PerlIO_printf(Perl_debug_log,"REG_GOSTART_SEEN "); \
818 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
819 PerlIO_printf(Perl_debug_log,"REG_UNBOUNDED_QUANTIFIER_SEEN "); \
821 PerlIO_printf(Perl_debug_log,"\n"); \
824 #define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
825 if ((flags) & flag) PerlIO_printf(Perl_debug_log, "%s ", #flag)
827 #define DEBUG_SHOW_STUDY_FLAGS(flags,open_str,close_str) \
829 PerlIO_printf(Perl_debug_log, "%s", open_str); \
830 DEBUG_SHOW_STUDY_FLAG(flags,SF_FL_BEFORE_SEOL); \
831 DEBUG_SHOW_STUDY_FLAG(flags,SF_FL_BEFORE_MEOL); \
832 DEBUG_SHOW_STUDY_FLAG(flags,SF_IS_INF); \
833 DEBUG_SHOW_STUDY_FLAG(flags,SF_HAS_PAR); \
834 DEBUG_SHOW_STUDY_FLAG(flags,SF_IN_PAR); \
835 DEBUG_SHOW_STUDY_FLAG(flags,SF_HAS_EVAL); \
836 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_SUBSTR); \
837 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS_AND); \
838 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS_OR); \
839 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS); \
840 DEBUG_SHOW_STUDY_FLAG(flags,SCF_WHILEM_VISITED_POS); \
841 DEBUG_SHOW_STUDY_FLAG(flags,SCF_TRIE_RESTUDY); \
842 DEBUG_SHOW_STUDY_FLAG(flags,SCF_SEEN_ACCEPT); \
843 DEBUG_SHOW_STUDY_FLAG(flags,SCF_TRIE_DOING_RESTUDY); \
844 DEBUG_SHOW_STUDY_FLAG(flags,SCF_IN_DEFINE); \
845 PerlIO_printf(Perl_debug_log, "%s", close_str); \
849 #define DEBUG_STUDYDATA(str,data,depth) \
850 DEBUG_OPTIMISE_MORE_r(if(data){ \
851 PerlIO_printf(Perl_debug_log, \
852 "%*s" str "Pos:%"IVdf"/%"IVdf \
854 (int)(depth)*2, "", \
855 (IV)((data)->pos_min), \
856 (IV)((data)->pos_delta), \
857 (UV)((data)->flags) \
859 DEBUG_SHOW_STUDY_FLAGS((data)->flags," [ ","]"); \
860 PerlIO_printf(Perl_debug_log, \
861 " Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
862 (IV)((data)->whilem_c), \
863 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
864 is_inf ? "INF " : "" \
866 if ((data)->last_found) \
867 PerlIO_printf(Perl_debug_log, \
868 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
869 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
870 SvPVX_const((data)->last_found), \
871 (IV)((data)->last_end), \
872 (IV)((data)->last_start_min), \
873 (IV)((data)->last_start_max), \
874 ((data)->longest && \
875 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
876 SvPVX_const((data)->longest_fixed), \
877 (IV)((data)->offset_fixed), \
878 ((data)->longest && \
879 (data)->longest==&((data)->longest_float)) ? "*" : "", \
880 SvPVX_const((data)->longest_float), \
881 (IV)((data)->offset_float_min), \
882 (IV)((data)->offset_float_max) \
884 PerlIO_printf(Perl_debug_log,"\n"); \
889 /* is c a control character for which we have a mnemonic? */
890 #define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
893 S_cntrl_to_mnemonic(const U8 c)
895 /* Returns the mnemonic string that represents character 'c', if one
896 * exists; NULL otherwise. The only ones that exist for the purposes of
897 * this routine are a few control characters */
900 case '\a': return "\\a";
901 case '\b': return "\\b";
902 case ESC_NATIVE: return "\\e";
903 case '\f': return "\\f";
904 case '\n': return "\\n";
905 case '\r': return "\\r";
906 case '\t': return "\\t";
914 /* Mark that we cannot extend a found fixed substring at this point.
915 Update the longest found anchored substring and the longest found
916 floating substrings if needed. */
919 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
920 SSize_t *minlenp, int is_inf)
922 const STRLEN l = CHR_SVLEN(data->last_found);
923 const STRLEN old_l = CHR_SVLEN(*data->longest);
924 GET_RE_DEBUG_FLAGS_DECL;
926 PERL_ARGS_ASSERT_SCAN_COMMIT;
928 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
929 SvSetMagicSV(*data->longest, data->last_found);
930 if (*data->longest == data->longest_fixed) {
931 data->offset_fixed = l ? data->last_start_min : data->pos_min;
932 if (data->flags & SF_BEFORE_EOL)
934 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
936 data->flags &= ~SF_FIX_BEFORE_EOL;
937 data->minlen_fixed=minlenp;
938 data->lookbehind_fixed=0;
940 else { /* *data->longest == data->longest_float */
941 data->offset_float_min = l ? data->last_start_min : data->pos_min;
942 data->offset_float_max = (l
943 ? data->last_start_max
944 : (data->pos_delta > SSize_t_MAX - data->pos_min
946 : data->pos_min + data->pos_delta));
948 || (STRLEN)data->offset_float_max > (STRLEN)SSize_t_MAX)
949 data->offset_float_max = SSize_t_MAX;
950 if (data->flags & SF_BEFORE_EOL)
952 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
954 data->flags &= ~SF_FL_BEFORE_EOL;
955 data->minlen_float=minlenp;
956 data->lookbehind_float=0;
959 SvCUR_set(data->last_found, 0);
961 SV * const sv = data->last_found;
962 if (SvUTF8(sv) && SvMAGICAL(sv)) {
963 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
969 data->flags &= ~SF_BEFORE_EOL;
970 DEBUG_STUDYDATA("commit: ",data,0);
973 /* An SSC is just a regnode_charclass_posix with an extra field: the inversion
974 * list that describes which code points it matches */
977 S_ssc_anything(pTHX_ regnode_ssc *ssc)
979 /* Set the SSC 'ssc' to match an empty string or any code point */
981 PERL_ARGS_ASSERT_SSC_ANYTHING;
983 assert(is_ANYOF_SYNTHETIC(ssc));
985 ssc->invlist = sv_2mortal(_new_invlist(2)); /* mortalize so won't leak */
986 _append_range_to_invlist(ssc->invlist, 0, UV_MAX);
987 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
991 S_ssc_is_anything(const regnode_ssc *ssc)
993 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
994 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
995 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
996 * in any way, so there's no point in using it */
1001 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
1003 assert(is_ANYOF_SYNTHETIC(ssc));
1005 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
1009 /* See if the list consists solely of the range 0 - Infinity */
1010 invlist_iterinit(ssc->invlist);
1011 ret = invlist_iternext(ssc->invlist, &start, &end)
1015 invlist_iterfinish(ssc->invlist);
1021 /* If e.g., both \w and \W are set, matches everything */
1022 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1024 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1025 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1035 S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
1037 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1038 * string, any code point, or any posix class under locale */
1040 PERL_ARGS_ASSERT_SSC_INIT;
1042 Zero(ssc, 1, regnode_ssc);
1043 set_ANYOF_SYNTHETIC(ssc);
1044 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
1047 /* If any portion of the regex is to operate under locale rules that aren't
1048 * fully known at compile time, initialization includes it. The reason
1049 * this isn't done for all regexes is that the optimizer was written under
1050 * the assumption that locale was all-or-nothing. Given the complexity and
1051 * lack of documentation in the optimizer, and that there are inadequate
1052 * test cases for locale, many parts of it may not work properly, it is
1053 * safest to avoid locale unless necessary. */
1054 if (RExC_contains_locale) {
1055 ANYOF_POSIXL_SETALL(ssc);
1058 ANYOF_POSIXL_ZERO(ssc);
1063 S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1064 const regnode_ssc *ssc)
1066 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1067 * to the list of code points matched, and locale posix classes; hence does
1068 * not check its flags) */
1073 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1075 assert(is_ANYOF_SYNTHETIC(ssc));
1077 invlist_iterinit(ssc->invlist);
1078 ret = invlist_iternext(ssc->invlist, &start, &end)
1082 invlist_iterfinish(ssc->invlist);
1088 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
1096 S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
1097 const regnode_charclass* const node)
1099 /* Returns a mortal inversion list defining which code points are matched
1100 * by 'node', which is of type ANYOF. Handles complementing the result if
1101 * appropriate. If some code points aren't knowable at this time, the
1102 * returned list must, and will, contain every code point that is a
1105 SV* invlist = sv_2mortal(_new_invlist(0));
1106 SV* only_utf8_locale_invlist = NULL;
1108 const U32 n = ARG(node);
1109 bool new_node_has_latin1 = FALSE;
1111 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1113 /* Look at the data structure created by S_set_ANYOF_arg() */
1114 if (n != ANYOF_ONLY_HAS_BITMAP) {
1115 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1116 AV * const av = MUTABLE_AV(SvRV(rv));
1117 SV **const ary = AvARRAY(av);
1118 assert(RExC_rxi->data->what[n] == 's');
1120 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
1121 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1])));
1123 else if (ary[0] && ary[0] != &PL_sv_undef) {
1125 /* Here, no compile-time swash, and there are things that won't be
1126 * known until runtime -- we have to assume it could be anything */
1127 return _add_range_to_invlist(invlist, 0, UV_MAX);
1129 else if (ary[3] && ary[3] != &PL_sv_undef) {
1131 /* Here no compile-time swash, and no run-time only data. Use the
1132 * node's inversion list */
1133 invlist = sv_2mortal(invlist_clone(ary[3]));
1136 /* Get the code points valid only under UTF-8 locales */
1137 if ((ANYOF_FLAGS(node) & ANYOF_LOC_FOLD)
1138 && ary[2] && ary[2] != &PL_sv_undef)
1140 only_utf8_locale_invlist = ary[2];
1144 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1145 * code points, and an inversion list for the others, but if there are code
1146 * points that should match only conditionally on the target string being
1147 * UTF-8, those are placed in the inversion list, and not the bitmap.
1148 * Since there are circumstances under which they could match, they are
1149 * included in the SSC. But if the ANYOF node is to be inverted, we have
1150 * to exclude them here, so that when we invert below, the end result
1151 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1152 * have to do this here before we add the unconditionally matched code
1154 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1155 _invlist_intersection_complement_2nd(invlist,
1160 /* Add in the points from the bit map */
1161 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
1162 if (ANYOF_BITMAP_TEST(node, i)) {
1163 invlist = add_cp_to_invlist(invlist, i);
1164 new_node_has_latin1 = TRUE;
1168 /* If this can match all upper Latin1 code points, have to add them
1170 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII) {
1171 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1174 /* Similarly for these */
1175 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
1176 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
1179 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1180 _invlist_invert(invlist);
1182 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOF_LOC_FOLD) {
1184 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1185 * locale. We can skip this if there are no 0-255 at all. */
1186 _invlist_union(invlist, PL_Latin1, &invlist);
1189 /* Similarly add the UTF-8 locale possible matches. These have to be
1190 * deferred until after the non-UTF-8 locale ones are taken care of just
1191 * above, or it leads to wrong results under ANYOF_INVERT */
1192 if (only_utf8_locale_invlist) {
1193 _invlist_union_maybe_complement_2nd(invlist,
1194 only_utf8_locale_invlist,
1195 ANYOF_FLAGS(node) & ANYOF_INVERT,
1202 /* These two functions currently do the exact same thing */
1203 #define ssc_init_zero ssc_init
1205 #define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1206 #define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1208 /* 'AND' a given class with another one. Can create false positives. 'ssc'
1209 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1210 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
1213 S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1214 const regnode_charclass *and_with)
1216 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1217 * another SSC or a regular ANYOF class. Can create false positives. */
1222 PERL_ARGS_ASSERT_SSC_AND;
1224 assert(is_ANYOF_SYNTHETIC(ssc));
1226 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1227 * the code point inversion list and just the relevant flags */
1228 if (is_ANYOF_SYNTHETIC(and_with)) {
1229 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
1230 anded_flags = ANYOF_FLAGS(and_with);
1232 /* XXX This is a kludge around what appears to be deficiencies in the
1233 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1234 * there are paths through the optimizer where it doesn't get weeded
1235 * out when it should. And if we don't make some extra provision for
1236 * it like the code just below, it doesn't get added when it should.
1237 * This solution is to add it only when AND'ing, which is here, and
1238 * only when what is being AND'ed is the pristine, original node
1239 * matching anything. Thus it is like adding it to ssc_anything() but
1240 * only when the result is to be AND'ed. Probably the same solution
1241 * could be adopted for the same problem we have with /l matching,
1242 * which is solved differently in S_ssc_init(), and that would lead to
1243 * fewer false positives than that solution has. But if this solution
1244 * creates bugs, the consequences are only that a warning isn't raised
1245 * that should be; while the consequences for having /l bugs is
1246 * incorrect matches */
1247 if (ssc_is_anything((regnode_ssc *)and_with)) {
1248 anded_flags |= ANYOF_WARN_SUPER;
1252 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
1253 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1256 ANYOF_FLAGS(ssc) &= anded_flags;
1258 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1259 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1260 * 'and_with' may be inverted. When not inverted, we have the situation of
1262 * (C1 | P1) & (C2 | P2)
1263 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1264 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1265 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1266 * <= ((C1 & C2) | P1 | P2)
1267 * Alternatively, the last few steps could be:
1268 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1269 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1270 * <= (C1 | C2 | (P1 & P2))
1271 * We favor the second approach if either P1 or P2 is non-empty. This is
1272 * because these components are a barrier to doing optimizations, as what
1273 * they match cannot be known until the moment of matching as they are
1274 * dependent on the current locale, 'AND"ing them likely will reduce or
1276 * But we can do better if we know that C1,P1 are in their initial state (a
1277 * frequent occurrence), each matching everything:
1278 * (<everything>) & (C2 | P2) = C2 | P2
1279 * Similarly, if C2,P2 are in their initial state (again a frequent
1280 * occurrence), the result is a no-op
1281 * (C1 | P1) & (<everything>) = C1 | P1
1284 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1285 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1286 * <= (C1 & ~C2) | (P1 & ~P2)
1289 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
1290 && ! is_ANYOF_SYNTHETIC(and_with))
1294 ssc_intersection(ssc,
1296 FALSE /* Has already been inverted */
1299 /* If either P1 or P2 is empty, the intersection will be also; can skip
1301 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
1302 ANYOF_POSIXL_ZERO(ssc);
1304 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1306 /* Note that the Posix class component P from 'and_with' actually
1308 * P = Pa | Pb | ... | Pn
1309 * where each component is one posix class, such as in [\w\s].
1311 * ~P = ~(Pa | Pb | ... | Pn)
1312 * = ~Pa & ~Pb & ... & ~Pn
1313 * <= ~Pa | ~Pb | ... | ~Pn
1314 * The last is something we can easily calculate, but unfortunately
1315 * is likely to have many false positives. We could do better
1316 * in some (but certainly not all) instances if two classes in
1317 * P have known relationships. For example
1318 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1320 * :lower: & :print: = :lower:
1321 * And similarly for classes that must be disjoint. For example,
1322 * since \s and \w can have no elements in common based on rules in
1323 * the POSIX standard,
1324 * \w & ^\S = nothing
1325 * Unfortunately, some vendor locales do not meet the Posix
1326 * standard, in particular almost everything by Microsoft.
1327 * The loop below just changes e.g., \w into \W and vice versa */
1329 regnode_charclass_posixl temp;
1330 int add = 1; /* To calculate the index of the complement */
1332 ANYOF_POSIXL_ZERO(&temp);
1333 for (i = 0; i < ANYOF_MAX; i++) {
1335 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1336 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
1338 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
1339 ANYOF_POSIXL_SET(&temp, i + add);
1341 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1343 ANYOF_POSIXL_AND(&temp, ssc);
1345 } /* else ssc already has no posixes */
1346 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1347 in its initial state */
1348 else if (! is_ANYOF_SYNTHETIC(and_with)
1349 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
1351 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1352 * copy it over 'ssc' */
1353 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
1354 if (is_ANYOF_SYNTHETIC(and_with)) {
1355 StructCopy(and_with, ssc, regnode_ssc);
1358 ssc->invlist = anded_cp_list;
1359 ANYOF_POSIXL_ZERO(ssc);
1360 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1361 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
1365 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
1366 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
1368 /* One or the other of P1, P2 is non-empty. */
1369 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1370 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1372 ssc_union(ssc, anded_cp_list, FALSE);
1374 else { /* P1 = P2 = empty */
1375 ssc_intersection(ssc, anded_cp_list, FALSE);
1381 S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1382 const regnode_charclass *or_with)
1384 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1385 * another SSC or a regular ANYOF class. Can create false positives if
1386 * 'or_with' is to be inverted. */
1391 PERL_ARGS_ASSERT_SSC_OR;
1393 assert(is_ANYOF_SYNTHETIC(ssc));
1395 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1396 * the code point inversion list and just the relevant flags */
1397 if (is_ANYOF_SYNTHETIC(or_with)) {
1398 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
1399 ored_flags = ANYOF_FLAGS(or_with);
1402 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
1403 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
1406 ANYOF_FLAGS(ssc) |= ored_flags;
1408 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1409 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1410 * 'or_with' may be inverted. When not inverted, we have the simple
1411 * situation of computing:
1412 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1413 * If P1|P2 yields a situation with both a class and its complement are
1414 * set, like having both \w and \W, this matches all code points, and we
1415 * can delete these from the P component of the ssc going forward. XXX We
1416 * might be able to delete all the P components, but I (khw) am not certain
1417 * about this, and it is better to be safe.
1420 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1421 * <= (C1 | P1) | ~C2
1422 * <= (C1 | ~C2) | P1
1423 * (which results in actually simpler code than the non-inverted case)
1426 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
1427 && ! is_ANYOF_SYNTHETIC(or_with))
1429 /* We ignore P2, leaving P1 going forward */
1430 } /* else Not inverted */
1431 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
1432 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
1433 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1435 for (i = 0; i < ANYOF_MAX; i += 2) {
1436 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1438 ssc_match_all_cp(ssc);
1439 ANYOF_POSIXL_CLEAR(ssc, i);
1440 ANYOF_POSIXL_CLEAR(ssc, i+1);
1448 FALSE /* Already has been inverted */
1452 PERL_STATIC_INLINE void
1453 S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1455 PERL_ARGS_ASSERT_SSC_UNION;
1457 assert(is_ANYOF_SYNTHETIC(ssc));
1459 _invlist_union_maybe_complement_2nd(ssc->invlist,
1465 PERL_STATIC_INLINE void
1466 S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1468 const bool invert2nd)
1470 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1472 assert(is_ANYOF_SYNTHETIC(ssc));
1474 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1480 PERL_STATIC_INLINE void
1481 S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1483 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1485 assert(is_ANYOF_SYNTHETIC(ssc));
1487 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1490 PERL_STATIC_INLINE void
1491 S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1493 /* AND just the single code point 'cp' into the SSC 'ssc' */
1495 SV* cp_list = _new_invlist(2);
1497 PERL_ARGS_ASSERT_SSC_CP_AND;
1499 assert(is_ANYOF_SYNTHETIC(ssc));
1501 cp_list = add_cp_to_invlist(cp_list, cp);
1502 ssc_intersection(ssc, cp_list,
1503 FALSE /* Not inverted */
1505 SvREFCNT_dec_NN(cp_list);
1508 PERL_STATIC_INLINE void
1509 S_ssc_clear_locale(regnode_ssc *ssc)
1511 /* Set the SSC 'ssc' to not match any locale things */
1512 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1514 assert(is_ANYOF_SYNTHETIC(ssc));
1516 ANYOF_POSIXL_ZERO(ssc);
1517 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1520 #define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
1523 S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
1525 /* The synthetic start class is used to hopefully quickly winnow down
1526 * places where a pattern could start a match in the target string. If it
1527 * doesn't really narrow things down that much, there isn't much point to
1528 * having the overhead of using it. This function uses some very crude
1529 * heuristics to decide if to use the ssc or not.
1531 * It returns TRUE if 'ssc' rules out more than half what it considers to
1532 * be the "likely" possible matches, but of course it doesn't know what the
1533 * actual things being matched are going to be; these are only guesses
1535 * For /l matches, it assumes that the only likely matches are going to be
1536 * in the 0-255 range, uniformly distributed, so half of that is 127
1537 * For /a and /d matches, it assumes that the likely matches will be just
1538 * the ASCII range, so half of that is 63
1539 * For /u and there isn't anything matching above the Latin1 range, it
1540 * assumes that that is the only range likely to be matched, and uses
1541 * half that as the cut-off: 127. If anything matches above Latin1,
1542 * it assumes that all of Unicode could match (uniformly), except for
1543 * non-Unicode code points and things in the General Category "Other"
1544 * (unassigned, private use, surrogates, controls and formats). This
1545 * is a much large number. */
1547 const U32 max_match = (LOC)
1551 : (invlist_highest(ssc->invlist) < 256)
1553 : ((NON_OTHER_COUNT + 1) / 2) - 1;
1554 U32 count = 0; /* Running total of number of code points matched by
1556 UV start, end; /* Start and end points of current range in inversion
1559 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
1561 invlist_iterinit(ssc->invlist);
1562 while (invlist_iternext(ssc->invlist, &start, &end)) {
1564 /* /u is the only thing that we expect to match above 255; so if not /u
1565 * and even if there are matches above 255, ignore them. This catches
1566 * things like \d under /d which does match the digits above 255, but
1567 * since the pattern is /d, it is not likely to be expecting them */
1568 if (! UNI_SEMANTICS) {
1572 end = MIN(end, 255);
1574 count += end - start + 1;
1575 if (count > max_match) {
1576 invlist_iterfinish(ssc->invlist);
1586 S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
1588 /* The inversion list in the SSC is marked mortal; now we need a more
1589 * permanent copy, which is stored the same way that is done in a regular
1590 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
1593 SV* invlist = invlist_clone(ssc->invlist);
1595 PERL_ARGS_ASSERT_SSC_FINALIZE;
1597 assert(is_ANYOF_SYNTHETIC(ssc));
1599 /* The code in this file assumes that all but these flags aren't relevant
1600 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
1601 * by the time we reach here */
1602 assert(! (ANYOF_FLAGS(ssc) & ~ANYOF_COMMON_FLAGS));
1604 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
1606 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
1607 NULL, NULL, NULL, FALSE);
1609 /* Make sure is clone-safe */
1610 ssc->invlist = NULL;
1612 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1613 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
1616 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
1619 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
1620 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
1621 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
1622 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
1623 ? (TRIE_LIST_CUR( idx ) - 1) \
1629 dump_trie(trie,widecharmap,revcharmap)
1630 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
1631 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
1633 These routines dump out a trie in a somewhat readable format.
1634 The _interim_ variants are used for debugging the interim
1635 tables that are used to generate the final compressed
1636 representation which is what dump_trie expects.
1638 Part of the reason for their existence is to provide a form
1639 of documentation as to how the different representations function.
1644 Dumps the final compressed table form of the trie to Perl_debug_log.
1645 Used for debugging make_trie().
1649 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
1650 AV *revcharmap, U32 depth)
1653 SV *sv=sv_newmortal();
1654 int colwidth= widecharmap ? 6 : 4;
1656 GET_RE_DEBUG_FLAGS_DECL;
1658 PERL_ARGS_ASSERT_DUMP_TRIE;
1660 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
1661 (int)depth * 2 + 2,"",
1662 "Match","Base","Ofs" );
1664 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
1665 SV ** const tmp = av_fetch( revcharmap, state, 0);
1667 PerlIO_printf( Perl_debug_log, "%*s",
1669 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1670 PL_colors[0], PL_colors[1],
1671 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1672 PERL_PV_ESCAPE_FIRSTCHAR
1677 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
1678 (int)depth * 2 + 2,"");
1680 for( state = 0 ; state < trie->uniquecharcount ; state++ )
1681 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
1682 PerlIO_printf( Perl_debug_log, "\n");
1684 for( state = 1 ; state < trie->statecount ; state++ ) {
1685 const U32 base = trie->states[ state ].trans.base;
1687 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|",
1688 (int)depth * 2 + 2,"", (UV)state);
1690 if ( trie->states[ state ].wordnum ) {
1691 PerlIO_printf( Perl_debug_log, " W%4X",
1692 trie->states[ state ].wordnum );
1694 PerlIO_printf( Perl_debug_log, "%6s", "" );
1697 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
1702 while( ( base + ofs < trie->uniquecharcount ) ||
1703 ( base + ofs - trie->uniquecharcount < trie->lasttrans
1704 && trie->trans[ base + ofs - trie->uniquecharcount ].check
1708 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
1710 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1711 if ( ( base + ofs >= trie->uniquecharcount )
1712 && ( base + ofs - trie->uniquecharcount
1714 && trie->trans[ base + ofs
1715 - trie->uniquecharcount ].check == state )
1717 PerlIO_printf( Perl_debug_log, "%*"UVXf,
1719 (UV)trie->trans[ base + ofs
1720 - trie->uniquecharcount ].next );
1722 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
1726 PerlIO_printf( Perl_debug_log, "]");
1729 PerlIO_printf( Perl_debug_log, "\n" );
1731 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=",
1733 for (word=1; word <= trie->wordcount; word++) {
1734 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
1735 (int)word, (int)(trie->wordinfo[word].prev),
1736 (int)(trie->wordinfo[word].len));
1738 PerlIO_printf(Perl_debug_log, "\n" );
1741 Dumps a fully constructed but uncompressed trie in list form.
1742 List tries normally only are used for construction when the number of
1743 possible chars (trie->uniquecharcount) is very high.
1744 Used for debugging make_trie().
1747 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
1748 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1752 SV *sv=sv_newmortal();
1753 int colwidth= widecharmap ? 6 : 4;
1754 GET_RE_DEBUG_FLAGS_DECL;
1756 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1758 /* print out the table precompression. */
1759 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1760 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1761 "------:-----+-----------------\n" );
1763 for( state=1 ; state < next_alloc ; state ++ ) {
1766 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1767 (int)depth * 2 + 2,"", (UV)state );
1768 if ( ! trie->states[ state ].wordnum ) {
1769 PerlIO_printf( Perl_debug_log, "%5s| ","");
1771 PerlIO_printf( Perl_debug_log, "W%4x| ",
1772 trie->states[ state ].wordnum
1775 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1776 SV ** const tmp = av_fetch( revcharmap,
1777 TRIE_LIST_ITEM(state,charid).forid, 0);
1779 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1781 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
1783 PL_colors[0], PL_colors[1],
1784 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
1785 | PERL_PV_ESCAPE_FIRSTCHAR
1787 TRIE_LIST_ITEM(state,charid).forid,
1788 (UV)TRIE_LIST_ITEM(state,charid).newstate
1791 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1792 (int)((depth * 2) + 14), "");
1795 PerlIO_printf( Perl_debug_log, "\n");
1800 Dumps a fully constructed but uncompressed trie in table form.
1801 This is the normal DFA style state transition table, with a few
1802 twists to facilitate compression later.
1803 Used for debugging make_trie().
1806 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1807 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1812 SV *sv=sv_newmortal();
1813 int colwidth= widecharmap ? 6 : 4;
1814 GET_RE_DEBUG_FLAGS_DECL;
1816 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1819 print out the table precompression so that we can do a visual check
1820 that they are identical.
1823 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1825 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1826 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1828 PerlIO_printf( Perl_debug_log, "%*s",
1830 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1831 PL_colors[0], PL_colors[1],
1832 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1833 PERL_PV_ESCAPE_FIRSTCHAR
1839 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1841 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1842 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1845 PerlIO_printf( Perl_debug_log, "\n" );
1847 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1849 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1850 (int)depth * 2 + 2,"",
1851 (UV)TRIE_NODENUM( state ) );
1853 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1854 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1856 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1858 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1860 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1861 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n",
1862 (UV)trie->trans[ state ].check );
1864 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n",
1865 (UV)trie->trans[ state ].check,
1866 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1874 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1875 startbranch: the first branch in the whole branch sequence
1876 first : start branch of sequence of branch-exact nodes.
1877 May be the same as startbranch
1878 last : Thing following the last branch.
1879 May be the same as tail.
1880 tail : item following the branch sequence
1881 count : words in the sequence
1882 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
1883 depth : indent depth
1885 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1887 A trie is an N'ary tree where the branches are determined by digital
1888 decomposition of the key. IE, at the root node you look up the 1st character and
1889 follow that branch repeat until you find the end of the branches. Nodes can be
1890 marked as "accepting" meaning they represent a complete word. Eg:
1894 would convert into the following structure. Numbers represent states, letters
1895 following numbers represent valid transitions on the letter from that state, if
1896 the number is in square brackets it represents an accepting state, otherwise it
1897 will be in parenthesis.
1899 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1903 (1) +-i->(6)-+-s->[7]
1905 +-s->(3)-+-h->(4)-+-e->[5]
1907 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1909 This shows that when matching against the string 'hers' we will begin at state 1
1910 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1911 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1912 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1913 single traverse. We store a mapping from accepting to state to which word was
1914 matched, and then when we have multiple possibilities we try to complete the
1915 rest of the regex in the order in which they occured in the alternation.
1917 The only prior NFA like behaviour that would be changed by the TRIE support is
1918 the silent ignoring of duplicate alternations which are of the form:
1920 / (DUPE|DUPE) X? (?{ ... }) Y /x
1922 Thus EVAL blocks following a trie may be called a different number of times with
1923 and without the optimisation. With the optimisations dupes will be silently
1924 ignored. This inconsistent behaviour of EVAL type nodes is well established as
1925 the following demonstrates:
1927 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1929 which prints out 'word' three times, but
1931 'words'=~/(word|word|word)(?{ print $1 })S/
1933 which doesnt print it out at all. This is due to other optimisations kicking in.
1935 Example of what happens on a structural level:
1937 The regexp /(ac|ad|ab)+/ will produce the following debug output:
1939 1: CURLYM[1] {1,32767}(18)
1950 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1951 and should turn into:
1953 1: CURLYM[1] {1,32767}(18)
1955 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1963 Cases where tail != last would be like /(?foo|bar)baz/:
1973 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1974 and would end up looking like:
1977 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1984 d = uvchr_to_utf8_flags(d, uv, 0);
1986 is the recommended Unicode-aware way of saying
1991 #define TRIE_STORE_REVCHAR(val) \
1994 SV *zlopp = newSV(7); /* XXX: optimize me */ \
1995 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1996 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
1997 SvCUR_set(zlopp, kapow - flrbbbbb); \
2000 av_push(revcharmap, zlopp); \
2002 char ooooff = (char)val; \
2003 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2007 /* This gets the next character from the input, folding it if not already
2009 #define TRIE_READ_CHAR STMT_START { \
2012 /* if it is UTF then it is either already folded, or does not need \
2014 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
2016 else if (folder == PL_fold_latin1) { \
2017 /* This folder implies Unicode rules, which in the range expressible \
2018 * by not UTF is the lower case, with the two exceptions, one of \
2019 * which should have been taken care of before calling this */ \
2020 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2021 uvc = toLOWER_L1(*uc); \
2022 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2025 /* raw data, will be folded later if needed */ \
2033 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2034 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
2035 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
2036 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
2038 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2039 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2040 TRIE_LIST_CUR( state )++; \
2043 #define TRIE_LIST_NEW(state) STMT_START { \
2044 Newxz( trie->states[ state ].trans.list, \
2045 4, reg_trie_trans_le ); \
2046 TRIE_LIST_CUR( state ) = 1; \
2047 TRIE_LIST_LEN( state ) = 4; \
2050 #define TRIE_HANDLE_WORD(state) STMT_START { \
2051 U16 dupe= trie->states[ state ].wordnum; \
2052 regnode * const noper_next = regnext( noper ); \
2055 /* store the word for dumping */ \
2057 if (OP(noper) != NOTHING) \
2058 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
2060 tmp = newSVpvn_utf8( "", 0, UTF ); \
2061 av_push( trie_words, tmp ); \
2065 trie->wordinfo[curword].prev = 0; \
2066 trie->wordinfo[curword].len = wordlen; \
2067 trie->wordinfo[curword].accept = state; \
2069 if ( noper_next < tail ) { \
2071 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2073 trie->jump[curword] = (U16)(noper_next - convert); \
2075 jumper = noper_next; \
2077 nextbranch= regnext(cur); \
2081 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2082 /* chain, so that when the bits of chain are later */\
2083 /* linked together, the dups appear in the chain */\
2084 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2085 trie->wordinfo[dupe].prev = curword; \
2087 /* we haven't inserted this word yet. */ \
2088 trie->states[ state ].wordnum = curword; \
2093 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2094 ( ( base + charid >= ucharcount \
2095 && base + charid < ubound \
2096 && state == trie->trans[ base - ucharcount + charid ].check \
2097 && trie->trans[ base - ucharcount + charid ].next ) \
2098 ? trie->trans[ base - ucharcount + charid ].next \
2099 : ( state==1 ? special : 0 ) \
2103 #define MADE_JUMP_TRIE 2
2104 #define MADE_EXACT_TRIE 4
2107 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2108 regnode *first, regnode *last, regnode *tail,
2109 U32 word_count, U32 flags, U32 depth)
2111 /* first pass, loop through and scan words */
2112 reg_trie_data *trie;
2113 HV *widecharmap = NULL;
2114 AV *revcharmap = newAV();
2120 regnode *jumper = NULL;
2121 regnode *nextbranch = NULL;
2122 regnode *convert = NULL;
2123 U32 *prev_states; /* temp array mapping each state to previous one */
2124 /* we just use folder as a flag in utf8 */
2125 const U8 * folder = NULL;
2128 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuuu"));
2129 AV *trie_words = NULL;
2130 /* along with revcharmap, this only used during construction but both are
2131 * useful during debugging so we store them in the struct when debugging.
2134 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
2135 STRLEN trie_charcount=0;
2137 SV *re_trie_maxbuff;
2138 GET_RE_DEBUG_FLAGS_DECL;
2140 PERL_ARGS_ASSERT_MAKE_TRIE;
2142 PERL_UNUSED_ARG(depth);
2146 case EXACT: case EXACTL: break;
2150 case EXACTFLU8: folder = PL_fold_latin1; break;
2151 case EXACTF: folder = PL_fold; break;
2152 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
2155 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
2157 trie->startstate = 1;
2158 trie->wordcount = word_count;
2159 RExC_rxi->data->data[ data_slot ] = (void*)trie;
2160 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
2161 if (flags == EXACT || flags == EXACTL)
2162 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2163 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2164 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2167 trie_words = newAV();
2170 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2171 assert(re_trie_maxbuff);
2172 if (!SvIOK(re_trie_maxbuff)) {
2173 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2175 DEBUG_TRIE_COMPILE_r({
2176 PerlIO_printf( Perl_debug_log,
2177 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2178 (int)depth * 2 + 2, "",
2179 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
2180 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
2183 /* Find the node we are going to overwrite */
2184 if ( first == startbranch && OP( last ) != BRANCH ) {
2185 /* whole branch chain */
2188 /* branch sub-chain */
2189 convert = NEXTOPER( first );
2192 /* -- First loop and Setup --
2194 We first traverse the branches and scan each word to determine if it
2195 contains widechars, and how many unique chars there are, this is
2196 important as we have to build a table with at least as many columns as we
2199 We use an array of integers to represent the character codes 0..255
2200 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2201 the native representation of the character value as the key and IV's for
2204 *TODO* If we keep track of how many times each character is used we can
2205 remap the columns so that the table compression later on is more
2206 efficient in terms of memory by ensuring the most common value is in the
2207 middle and the least common are on the outside. IMO this would be better
2208 than a most to least common mapping as theres a decent chance the most
2209 common letter will share a node with the least common, meaning the node
2210 will not be compressible. With a middle is most common approach the worst
2211 case is when we have the least common nodes twice.
2215 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2216 regnode *noper = NEXTOPER( cur );
2217 const U8 *uc = (U8*)STRING( noper );
2218 const U8 *e = uc + STR_LEN( noper );
2220 U32 wordlen = 0; /* required init */
2221 STRLEN minchars = 0;
2222 STRLEN maxchars = 0;
2223 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2226 if (OP(noper) == NOTHING) {
2227 regnode *noper_next= regnext(noper);
2228 if (noper_next != tail && OP(noper_next) == flags) {
2230 uc= (U8*)STRING(noper);
2231 e= uc + STR_LEN(noper);
2232 trie->minlen= STR_LEN(noper);
2239 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
2240 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2241 regardless of encoding */
2242 if (OP( noper ) == EXACTFU_SS) {
2243 /* false positives are ok, so just set this */
2244 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
2247 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2249 TRIE_CHARCOUNT(trie)++;
2252 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2253 * is in effect. Under /i, this character can match itself, or
2254 * anything that folds to it. If not under /i, it can match just
2255 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2256 * all fold to k, and all are single characters. But some folds
2257 * expand to more than one character, so for example LATIN SMALL
2258 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2259 * the string beginning at 'uc' is 'ffi', it could be matched by
2260 * three characters, or just by the one ligature character. (It
2261 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2262 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2263 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2264 * match.) The trie needs to know the minimum and maximum number
2265 * of characters that could match so that it can use size alone to
2266 * quickly reject many match attempts. The max is simple: it is
2267 * the number of folded characters in this branch (since a fold is
2268 * never shorter than what folds to it. */
2272 /* And the min is equal to the max if not under /i (indicated by
2273 * 'folder' being NULL), or there are no multi-character folds. If
2274 * there is a multi-character fold, the min is incremented just
2275 * once, for the character that folds to the sequence. Each
2276 * character in the sequence needs to be added to the list below of
2277 * characters in the trie, but we count only the first towards the
2278 * min number of characters needed. This is done through the
2279 * variable 'foldlen', which is returned by the macros that look
2280 * for these sequences as the number of bytes the sequence
2281 * occupies. Each time through the loop, we decrement 'foldlen' by
2282 * how many bytes the current char occupies. Only when it reaches
2283 * 0 do we increment 'minchars' or look for another multi-character
2285 if (folder == NULL) {
2288 else if (foldlen > 0) {
2289 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
2294 /* See if *uc is the beginning of a multi-character fold. If
2295 * so, we decrement the length remaining to look at, to account
2296 * for the current character this iteration. (We can use 'uc'
2297 * instead of the fold returned by TRIE_READ_CHAR because for
2298 * non-UTF, the latin1_safe macro is smart enough to account
2299 * for all the unfolded characters, and because for UTF, the
2300 * string will already have been folded earlier in the
2301 * compilation process */
2303 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2304 foldlen -= UTF8SKIP(uc);
2307 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2312 /* The current character (and any potential folds) should be added
2313 * to the possible matching characters for this position in this
2317 U8 folded= folder[ (U8) uvc ];
2318 if ( !trie->charmap[ folded ] ) {
2319 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2320 TRIE_STORE_REVCHAR( folded );
2323 if ( !trie->charmap[ uvc ] ) {
2324 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
2325 TRIE_STORE_REVCHAR( uvc );
2328 /* store the codepoint in the bitmap, and its folded
2330 TRIE_BITMAP_SET(trie, uvc);
2332 /* store the folded codepoint */
2333 if ( folder ) TRIE_BITMAP_SET(trie, folder[(U8) uvc ]);
2336 /* store first byte of utf8 representation of
2337 variant codepoints */
2338 if (! UVCHR_IS_INVARIANT(uvc)) {
2339 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
2342 set_bit = 0; /* We've done our bit :-) */
2346 /* XXX We could come up with the list of code points that fold
2347 * to this using PL_utf8_foldclosures, except not for
2348 * multi-char folds, as there may be multiple combinations
2349 * there that could work, which needs to wait until runtime to
2350 * resolve (The comment about LIGATURE FFI above is such an
2355 widecharmap = newHV();
2357 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
2360 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
2362 if ( !SvTRUE( *svpp ) ) {
2363 sv_setiv( *svpp, ++trie->uniquecharcount );
2364 TRIE_STORE_REVCHAR(uvc);
2367 } /* end loop through characters in this branch of the trie */
2369 /* We take the min and max for this branch and combine to find the min
2370 * and max for all branches processed so far */
2371 if( cur == first ) {
2372 trie->minlen = minchars;
2373 trie->maxlen = maxchars;
2374 } else if (minchars < trie->minlen) {
2375 trie->minlen = minchars;
2376 } else if (maxchars > trie->maxlen) {
2377 trie->maxlen = maxchars;
2379 } /* end first pass */
2380 DEBUG_TRIE_COMPILE_r(
2381 PerlIO_printf( Perl_debug_log,
2382 "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2383 (int)depth * 2 + 2,"",
2384 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
2385 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2386 (int)trie->minlen, (int)trie->maxlen )
2390 We now know what we are dealing with in terms of unique chars and
2391 string sizes so we can calculate how much memory a naive
2392 representation using a flat table will take. If it's over a reasonable
2393 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
2394 conservative but potentially much slower representation using an array
2397 At the end we convert both representations into the same compressed
2398 form that will be used in regexec.c for matching with. The latter
2399 is a form that cannot be used to construct with but has memory
2400 properties similar to the list form and access properties similar
2401 to the table form making it both suitable for fast searches and
2402 small enough that its feasable to store for the duration of a program.
2404 See the comment in the code where the compressed table is produced
2405 inplace from the flat tabe representation for an explanation of how
2406 the compression works.
2411 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2414 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2415 > SvIV(re_trie_maxbuff) )
2418 Second Pass -- Array Of Lists Representation
2420 Each state will be represented by a list of charid:state records
2421 (reg_trie_trans_le) the first such element holds the CUR and LEN
2422 points of the allocated array. (See defines above).
2424 We build the initial structure using the lists, and then convert
2425 it into the compressed table form which allows faster lookups
2426 (but cant be modified once converted).
2429 STRLEN transcount = 1;
2431 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2432 "%*sCompiling trie using list compiler\n",
2433 (int)depth * 2 + 2, ""));
2435 trie->states = (reg_trie_state *)
2436 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2437 sizeof(reg_trie_state) );
2441 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2443 regnode *noper = NEXTOPER( cur );
2444 U8 *uc = (U8*)STRING( noper );
2445 const U8 *e = uc + STR_LEN( noper );
2446 U32 state = 1; /* required init */
2447 U16 charid = 0; /* sanity init */
2448 U32 wordlen = 0; /* required init */
2450 if (OP(noper) == NOTHING) {
2451 regnode *noper_next= regnext(noper);
2452 if (noper_next != tail && OP(noper_next) == flags) {
2454 uc= (U8*)STRING(noper);
2455 e= uc + STR_LEN(noper);
2459 if (OP(noper) != NOTHING) {
2460 for ( ; uc < e ; uc += len ) {
2465 charid = trie->charmap[ uvc ];
2467 SV** const svpp = hv_fetch( widecharmap,
2474 charid=(U16)SvIV( *svpp );
2477 /* charid is now 0 if we dont know the char read, or
2478 * nonzero if we do */
2485 if ( !trie->states[ state ].trans.list ) {
2486 TRIE_LIST_NEW( state );
2489 check <= TRIE_LIST_USED( state );
2492 if ( TRIE_LIST_ITEM( state, check ).forid
2495 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2500 newstate = next_alloc++;
2501 prev_states[newstate] = state;
2502 TRIE_LIST_PUSH( state, charid, newstate );
2507 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2511 TRIE_HANDLE_WORD(state);
2513 } /* end second pass */
2515 /* next alloc is the NEXT state to be allocated */
2516 trie->statecount = next_alloc;
2517 trie->states = (reg_trie_state *)
2518 PerlMemShared_realloc( trie->states,
2520 * sizeof(reg_trie_state) );
2522 /* and now dump it out before we compress it */
2523 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2524 revcharmap, next_alloc,
2528 trie->trans = (reg_trie_trans *)
2529 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
2536 for( state=1 ; state < next_alloc ; state ++ ) {
2540 DEBUG_TRIE_COMPILE_MORE_r(
2541 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
2545 if (trie->states[state].trans.list) {
2546 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
2550 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2551 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
2552 if ( forid < minid ) {
2554 } else if ( forid > maxid ) {
2558 if ( transcount < tp + maxid - minid + 1) {
2560 trie->trans = (reg_trie_trans *)
2561 PerlMemShared_realloc( trie->trans,
2563 * sizeof(reg_trie_trans) );
2564 Zero( trie->trans + (transcount / 2),
2568 base = trie->uniquecharcount + tp - minid;
2569 if ( maxid == minid ) {
2571 for ( ; zp < tp ; zp++ ) {
2572 if ( ! trie->trans[ zp ].next ) {
2573 base = trie->uniquecharcount + zp - minid;
2574 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
2576 trie->trans[ zp ].check = state;
2582 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
2584 trie->trans[ tp ].check = state;
2589 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2590 const U32 tid = base
2591 - trie->uniquecharcount
2592 + TRIE_LIST_ITEM( state, idx ).forid;
2593 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
2595 trie->trans[ tid ].check = state;
2597 tp += ( maxid - minid + 1 );
2599 Safefree(trie->states[ state ].trans.list);
2602 DEBUG_TRIE_COMPILE_MORE_r(
2603 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
2606 trie->states[ state ].trans.base=base;
2608 trie->lasttrans = tp + 1;
2612 Second Pass -- Flat Table Representation.
2614 we dont use the 0 slot of either trans[] or states[] so we add 1 to
2615 each. We know that we will need Charcount+1 trans at most to store
2616 the data (one row per char at worst case) So we preallocate both
2617 structures assuming worst case.
2619 We then construct the trie using only the .next slots of the entry
2622 We use the .check field of the first entry of the node temporarily
2623 to make compression both faster and easier by keeping track of how
2624 many non zero fields are in the node.
2626 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
2629 There are two terms at use here: state as a TRIE_NODEIDX() which is
2630 a number representing the first entry of the node, and state as a
2631 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
2632 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
2633 if there are 2 entrys per node. eg:
2641 The table is internally in the right hand, idx form. However as we
2642 also have to deal with the states array which is indexed by nodenum
2643 we have to use TRIE_NODENUM() to convert.
2646 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2647 "%*sCompiling trie using table compiler\n",
2648 (int)depth * 2 + 2, ""));
2650 trie->trans = (reg_trie_trans *)
2651 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
2652 * trie->uniquecharcount + 1,
2653 sizeof(reg_trie_trans) );
2654 trie->states = (reg_trie_state *)
2655 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2656 sizeof(reg_trie_state) );
2657 next_alloc = trie->uniquecharcount + 1;
2660 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2662 regnode *noper = NEXTOPER( cur );
2663 const U8 *uc = (U8*)STRING( noper );
2664 const U8 *e = uc + STR_LEN( noper );
2666 U32 state = 1; /* required init */
2668 U16 charid = 0; /* sanity init */
2669 U32 accept_state = 0; /* sanity init */
2671 U32 wordlen = 0; /* required init */
2673 if (OP(noper) == NOTHING) {
2674 regnode *noper_next= regnext(noper);
2675 if (noper_next != tail && OP(noper_next) == flags) {
2677 uc= (U8*)STRING(noper);
2678 e= uc + STR_LEN(noper);
2682 if ( OP(noper) != NOTHING ) {
2683 for ( ; uc < e ; uc += len ) {
2688 charid = trie->charmap[ uvc ];
2690 SV* const * const svpp = hv_fetch( widecharmap,
2694 charid = svpp ? (U16)SvIV(*svpp) : 0;
2698 if ( !trie->trans[ state + charid ].next ) {
2699 trie->trans[ state + charid ].next = next_alloc;
2700 trie->trans[ state ].check++;
2701 prev_states[TRIE_NODENUM(next_alloc)]
2702 = TRIE_NODENUM(state);
2703 next_alloc += trie->uniquecharcount;
2705 state = trie->trans[ state + charid ].next;
2707 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2709 /* charid is now 0 if we dont know the char read, or
2710 * nonzero if we do */
2713 accept_state = TRIE_NODENUM( state );
2714 TRIE_HANDLE_WORD(accept_state);
2716 } /* end second pass */
2718 /* and now dump it out before we compress it */
2719 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
2721 next_alloc, depth+1));
2725 * Inplace compress the table.*
2727 For sparse data sets the table constructed by the trie algorithm will
2728 be mostly 0/FAIL transitions or to put it another way mostly empty.
2729 (Note that leaf nodes will not contain any transitions.)
2731 This algorithm compresses the tables by eliminating most such
2732 transitions, at the cost of a modest bit of extra work during lookup:
2734 - Each states[] entry contains a .base field which indicates the
2735 index in the state[] array wheres its transition data is stored.
2737 - If .base is 0 there are no valid transitions from that node.
2739 - If .base is nonzero then charid is added to it to find an entry in
2742 -If trans[states[state].base+charid].check!=state then the
2743 transition is taken to be a 0/Fail transition. Thus if there are fail
2744 transitions at the front of the node then the .base offset will point
2745 somewhere inside the previous nodes data (or maybe even into a node
2746 even earlier), but the .check field determines if the transition is
2750 The following process inplace converts the table to the compressed
2751 table: We first do not compress the root node 1,and mark all its
2752 .check pointers as 1 and set its .base pointer as 1 as well. This
2753 allows us to do a DFA construction from the compressed table later,
2754 and ensures that any .base pointers we calculate later are greater
2757 - We set 'pos' to indicate the first entry of the second node.
2759 - We then iterate over the columns of the node, finding the first and
2760 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
2761 and set the .check pointers accordingly, and advance pos
2762 appropriately and repreat for the next node. Note that when we copy
2763 the next pointers we have to convert them from the original
2764 NODEIDX form to NODENUM form as the former is not valid post
2767 - If a node has no transitions used we mark its base as 0 and do not
2768 advance the pos pointer.
2770 - If a node only has one transition we use a second pointer into the
2771 structure to fill in allocated fail transitions from other states.
2772 This pointer is independent of the main pointer and scans forward
2773 looking for null transitions that are allocated to a state. When it
2774 finds one it writes the single transition into the "hole". If the
2775 pointer doesnt find one the single transition is appended as normal.
2777 - Once compressed we can Renew/realloc the structures to release the
2780 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
2781 specifically Fig 3.47 and the associated pseudocode.
2785 const U32 laststate = TRIE_NODENUM( next_alloc );
2788 trie->statecount = laststate;
2790 for ( state = 1 ; state < laststate ; state++ ) {
2792 const U32 stateidx = TRIE_NODEIDX( state );
2793 const U32 o_used = trie->trans[ stateidx ].check;
2794 U32 used = trie->trans[ stateidx ].check;
2795 trie->trans[ stateidx ].check = 0;
2798 used && charid < trie->uniquecharcount;
2801 if ( flag || trie->trans[ stateidx + charid ].next ) {
2802 if ( trie->trans[ stateidx + charid ].next ) {
2804 for ( ; zp < pos ; zp++ ) {
2805 if ( ! trie->trans[ zp ].next ) {
2809 trie->states[ state ].trans.base
2811 + trie->uniquecharcount
2813 trie->trans[ zp ].next
2814 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
2816 trie->trans[ zp ].check = state;
2817 if ( ++zp > pos ) pos = zp;
2824 trie->states[ state ].trans.base
2825 = pos + trie->uniquecharcount - charid ;
2827 trie->trans[ pos ].next
2828 = SAFE_TRIE_NODENUM(
2829 trie->trans[ stateidx + charid ].next );
2830 trie->trans[ pos ].check = state;
2835 trie->lasttrans = pos + 1;
2836 trie->states = (reg_trie_state *)
2837 PerlMemShared_realloc( trie->states, laststate
2838 * sizeof(reg_trie_state) );
2839 DEBUG_TRIE_COMPILE_MORE_r(
2840 PerlIO_printf( Perl_debug_log,
2841 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
2842 (int)depth * 2 + 2,"",
2843 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
2847 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
2850 } /* end table compress */
2852 DEBUG_TRIE_COMPILE_MORE_r(
2853 PerlIO_printf(Perl_debug_log,
2854 "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
2855 (int)depth * 2 + 2, "",
2856 (UV)trie->statecount,
2857 (UV)trie->lasttrans)
2859 /* resize the trans array to remove unused space */
2860 trie->trans = (reg_trie_trans *)
2861 PerlMemShared_realloc( trie->trans, trie->lasttrans
2862 * sizeof(reg_trie_trans) );
2864 { /* Modify the program and insert the new TRIE node */
2865 U8 nodetype =(U8)(flags & 0xFF);
2869 regnode *optimize = NULL;
2870 #ifdef RE_TRACK_PATTERN_OFFSETS
2873 U32 mjd_nodelen = 0;
2874 #endif /* RE_TRACK_PATTERN_OFFSETS */
2875 #endif /* DEBUGGING */
2877 This means we convert either the first branch or the first Exact,
2878 depending on whether the thing following (in 'last') is a branch
2879 or not and whther first is the startbranch (ie is it a sub part of
2880 the alternation or is it the whole thing.)
2881 Assuming its a sub part we convert the EXACT otherwise we convert
2882 the whole branch sequence, including the first.
2884 /* Find the node we are going to overwrite */
2885 if ( first != startbranch || OP( last ) == BRANCH ) {
2886 /* branch sub-chain */
2887 NEXT_OFF( first ) = (U16)(last - first);
2888 #ifdef RE_TRACK_PATTERN_OFFSETS
2890 mjd_offset= Node_Offset((convert));
2891 mjd_nodelen= Node_Length((convert));
2894 /* whole branch chain */
2896 #ifdef RE_TRACK_PATTERN_OFFSETS
2899 const regnode *nop = NEXTOPER( convert );
2900 mjd_offset= Node_Offset((nop));
2901 mjd_nodelen= Node_Length((nop));
2905 PerlIO_printf(Perl_debug_log,
2906 "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2907 (int)depth * 2 + 2, "",
2908 (UV)mjd_offset, (UV)mjd_nodelen)
2911 /* But first we check to see if there is a common prefix we can
2912 split out as an EXACT and put in front of the TRIE node. */
2913 trie->startstate= 1;
2914 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2916 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2920 const U32 base = trie->states[ state ].trans.base;
2922 if ( trie->states[state].wordnum )
2925 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2926 if ( ( base + ofs >= trie->uniquecharcount ) &&
2927 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2928 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2930 if ( ++count > 1 ) {
2931 SV **tmp = av_fetch( revcharmap, ofs, 0);
2932 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2933 if ( state == 1 ) break;
2935 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2937 PerlIO_printf(Perl_debug_log,
2938 "%*sNew Start State=%"UVuf" Class: [",
2939 (int)depth * 2 + 2, "",
2942 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2943 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2945 TRIE_BITMAP_SET(trie,*ch);
2947 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2949 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2953 TRIE_BITMAP_SET(trie,*ch);
2955 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2956 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2962 SV **tmp = av_fetch( revcharmap, idx, 0);
2964 char *ch = SvPV( *tmp, len );
2966 SV *sv=sv_newmortal();
2967 PerlIO_printf( Perl_debug_log,
2968 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2969 (int)depth * 2 + 2, "",
2971 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2972 PL_colors[0], PL_colors[1],
2973 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2974 PERL_PV_ESCAPE_FIRSTCHAR
2979 OP( convert ) = nodetype;
2980 str=STRING(convert);
2983 STR_LEN(convert) += len;
2989 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2994 trie->prefixlen = (state-1);
2996 regnode *n = convert+NODE_SZ_STR(convert);
2997 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2998 trie->startstate = state;
2999 trie->minlen -= (state - 1);
3000 trie->maxlen -= (state - 1);
3002 /* At least the UNICOS C compiler choked on this
3003 * being argument to DEBUG_r(), so let's just have
3006 #ifdef PERL_EXT_RE_BUILD
3012 regnode *fix = convert;
3013 U32 word = trie->wordcount;
3015 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
3016 while( ++fix < n ) {
3017 Set_Node_Offset_Length(fix, 0, 0);
3020 SV ** const tmp = av_fetch( trie_words, word, 0 );
3022 if ( STR_LEN(convert) <= SvCUR(*tmp) )
3023 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
3025 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
3033 NEXT_OFF(convert) = (U16)(tail - convert);
3034 DEBUG_r(optimize= n);
3040 if ( trie->maxlen ) {
3041 NEXT_OFF( convert ) = (U16)(tail - convert);
3042 ARG_SET( convert, data_slot );
3043 /* Store the offset to the first unabsorbed branch in
3044 jump[0], which is otherwise unused by the jump logic.
3045 We use this when dumping a trie and during optimisation. */
3047 trie->jump[0] = (U16)(nextbranch - convert);
3049 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
3050 * and there is a bitmap
3051 * and the first "jump target" node we found leaves enough room
3052 * then convert the TRIE node into a TRIEC node, with the bitmap
3053 * embedded inline in the opcode - this is hypothetically faster.
3055 if ( !trie->states[trie->startstate].wordnum
3057 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
3059 OP( convert ) = TRIEC;
3060 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
3061 PerlMemShared_free(trie->bitmap);
3064 OP( convert ) = TRIE;
3066 /* store the type in the flags */
3067 convert->flags = nodetype;
3071 + regarglen[ OP( convert ) ];
3073 /* XXX We really should free up the resource in trie now,
3074 as we won't use them - (which resources?) dmq */
3076 /* needed for dumping*/
3077 DEBUG_r(if (optimize) {
3078 regnode *opt = convert;
3080 while ( ++opt < optimize) {
3081 Set_Node_Offset_Length(opt,0,0);
3084 Try to clean up some of the debris left after the
3087 while( optimize < jumper ) {
3088 mjd_nodelen += Node_Length((optimize));
3089 OP( optimize ) = OPTIMIZED;
3090 Set_Node_Offset_Length(optimize,0,0);
3093 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
3095 } /* end node insert */
3097 /* Finish populating the prev field of the wordinfo array. Walk back
3098 * from each accept state until we find another accept state, and if
3099 * so, point the first word's .prev field at the second word. If the
3100 * second already has a .prev field set, stop now. This will be the
3101 * case either if we've already processed that word's accept state,
3102 * or that state had multiple words, and the overspill words were
3103 * already linked up earlier.
3110 for (word=1; word <= trie->wordcount; word++) {
3112 if (trie->wordinfo[word].prev)
3114 state = trie->wordinfo[word].accept;
3116 state = prev_states[state];
3119 prev = trie->states[state].wordnum;
3123 trie->wordinfo[word].prev = prev;
3125 Safefree(prev_states);
3129 /* and now dump out the compressed format */
3130 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
3132 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
3134 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
3135 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
3137 SvREFCNT_dec_NN(revcharmap);
3141 : trie->startstate>1
3147 S_construct_ahocorasick_from_trie(pTHX_ RExC_state_t *pRExC_state, regnode *source, U32 depth)
3149 /* The Trie is constructed and compressed now so we can build a fail array if
3152 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and
3154 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi,
3158 We find the fail state for each state in the trie, this state is the longest
3159 proper suffix of the current state's 'word' that is also a proper prefix of
3160 another word in our trie. State 1 represents the word '' and is thus the
3161 default fail state. This allows the DFA not to have to restart after its
3162 tried and failed a word at a given point, it simply continues as though it
3163 had been matching the other word in the first place.
3165 'abcdgu'=~/abcdefg|cdgu/
3166 When we get to 'd' we are still matching the first word, we would encounter
3167 'g' which would fail, which would bring us to the state representing 'd' in
3168 the second word where we would try 'g' and succeed, proceeding to match
3171 /* add a fail transition */
3172 const U32 trie_offset = ARG(source);
3173 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
3175 const U32 ucharcount = trie->uniquecharcount;
3176 const U32 numstates = trie->statecount;
3177 const U32 ubound = trie->lasttrans + ucharcount;
3181 U32 base = trie->states[ 1 ].trans.base;
3184 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T"));
3186 GET_RE_DEBUG_FLAGS_DECL;
3188 PERL_ARGS_ASSERT_CONSTRUCT_AHOCORASICK_FROM_TRIE;
3189 PERL_UNUSED_CONTEXT;
3191 PERL_UNUSED_ARG(depth);
3194 if ( OP(source) == TRIE ) {
3195 struct regnode_1 *op = (struct regnode_1 *)
3196 PerlMemShared_calloc(1, sizeof(struct regnode_1));
3197 StructCopy(source,op,struct regnode_1);
3198 stclass = (regnode *)op;
3200 struct regnode_charclass *op = (struct regnode_charclass *)
3201 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
3202 StructCopy(source,op,struct regnode_charclass);
3203 stclass = (regnode *)op;
3205 OP(stclass)+=2; /* convert the TRIE type to its AHO-CORASICK equivalent */
3207 ARG_SET( stclass, data_slot );
3208 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
3209 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3210 aho->trie=trie_offset;
3211 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
3212 Copy( trie->states, aho->states, numstates, reg_trie_state );
3213 Newxz( q, numstates, U32);
3214 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
3217 /* initialize fail[0..1] to be 1 so that we always have
3218 a valid final fail state */
3219 fail[ 0 ] = fail[ 1 ] = 1;
3221 for ( charid = 0; charid < ucharcount ; charid++ ) {
3222 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
3224 q[ q_write ] = newstate;
3225 /* set to point at the root */
3226 fail[ q[ q_write++ ] ]=1;
3229 while ( q_read < q_write) {
3230 const U32 cur = q[ q_read++ % numstates ];
3231 base = trie->states[ cur ].trans.base;
3233 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
3234 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
3236 U32 fail_state = cur;
3239 fail_state = fail[ fail_state ];
3240 fail_base = aho->states[ fail_state ].trans.base;
3241 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
3243 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
3244 fail[ ch_state ] = fail_state;
3245 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
3247 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
3249 q[ q_write++ % numstates] = ch_state;
3253 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
3254 when we fail in state 1, this allows us to use the
3255 charclass scan to find a valid start char. This is based on the principle
3256 that theres a good chance the string being searched contains lots of stuff
3257 that cant be a start char.
3259 fail[ 0 ] = fail[ 1 ] = 0;
3260 DEBUG_TRIE_COMPILE_r({
3261 PerlIO_printf(Perl_debug_log,
3262 "%*sStclass Failtable (%"UVuf" states): 0",
3263 (int)(depth * 2), "", (UV)numstates
3265 for( q_read=1; q_read<numstates; q_read++ ) {
3266 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
3268 PerlIO_printf(Perl_debug_log, "\n");
3271 /*RExC_seen |= REG_TRIEDFA_SEEN;*/
3276 #define DEBUG_PEEP(str,scan,depth) \
3277 DEBUG_OPTIMISE_r({if (scan){ \
3278 regnode *Next = regnext(scan); \
3279 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state); \
3280 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)", \
3281 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),\
3282 Next ? (REG_NODE_NUM(Next)) : 0 ); \
3283 DEBUG_SHOW_STUDY_FLAGS(flags," [ ","]");\
3284 PerlIO_printf(Perl_debug_log, "\n"); \
3287 /* The below joins as many adjacent EXACTish nodes as possible into a single
3288 * one. The regop may be changed if the node(s) contain certain sequences that
3289 * require special handling. The joining is only done if:
3290 * 1) there is room in the current conglomerated node to entirely contain the
3292 * 2) they are the exact same node type
3294 * The adjacent nodes actually may be separated by NOTHING-kind nodes, and
3295 * these get optimized out
3297 * If a node is to match under /i (folded), the number of characters it matches
3298 * can be different than its character length if it contains a multi-character
3299 * fold. *min_subtract is set to the total delta number of characters of the
3302 * And *unfolded_multi_char is set to indicate whether or not the node contains
3303 * an unfolded multi-char fold. This happens when whether the fold is valid or
3304 * not won't be known until runtime; namely for EXACTF nodes that contain LATIN
3305 * SMALL LETTER SHARP S, as only if the target string being matched against
3306 * turns out to be UTF-8 is that fold valid; and also for EXACTFL nodes whose
3307 * folding rules depend on the locale in force at runtime. (Multi-char folds
3308 * whose components are all above the Latin1 range are not run-time locale
3309 * dependent, and have already been folded by the time this function is
3312 * This is as good a place as any to discuss the design of handling these
3313 * multi-character fold sequences. It's been wrong in Perl for a very long
3314 * time. There are three code points in Unicode whose multi-character folds
3315 * were long ago discovered to mess things up. The previous designs for
3316 * dealing with these involved assigning a special node for them. This
3317 * approach doesn't always work, as evidenced by this example:
3318 * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches
3319 * Both sides fold to "sss", but if the pattern is parsed to create a node that
3320 * would match just the \xDF, it won't be able to handle the case where a
3321 * successful match would have to cross the node's boundary. The new approach
3322 * that hopefully generally solves the problem generates an EXACTFU_SS node
3323 * that is "sss" in this case.
3325 * It turns out that there are problems with all multi-character folds, and not
3326 * just these three. Now the code is general, for all such cases. The
3327 * approach taken is:
3328 * 1) This routine examines each EXACTFish node that could contain multi-
3329 * character folded sequences. Since a single character can fold into
3330 * such a sequence, the minimum match length for this node is less than
3331 * the number of characters in the node. This routine returns in
3332 * *min_subtract how many characters to subtract from the the actual
3333 * length of the string to get a real minimum match length; it is 0 if
3334 * there are no multi-char foldeds. This delta is used by the caller to
3335 * adjust the min length of the match, and the delta between min and max,
3336 * so that the optimizer doesn't reject these possibilities based on size
3338 * 2) For the sequence involving the Sharp s (\xDF), the node type EXACTFU_SS
3339 * is used for an EXACTFU node that contains at least one "ss" sequence in
3340 * it. For non-UTF-8 patterns and strings, this is the only case where
3341 * there is a possible fold length change. That means that a regular
3342 * EXACTFU node without UTF-8 involvement doesn't have to concern itself
3343 * with length changes, and so can be processed faster. regexec.c takes
3344 * advantage of this. Generally, an EXACTFish node that is in UTF-8 is
3345 * pre-folded by regcomp.c (except EXACTFL, some of whose folds aren't
3346 * known until runtime). This saves effort in regex matching. However,
3347 * the pre-folding isn't done for non-UTF8 patterns because the fold of
3348 * the MICRO SIGN requires UTF-8, and we don't want to slow things down by
3349 * forcing the pattern into UTF8 unless necessary. Also what EXACTF (and,
3350 * again, EXACTFL) nodes fold to isn't known until runtime. The fold
3351 * possibilities for the non-UTF8 patterns are quite simple, except for
3352 * the sharp s. All the ones that don't involve a UTF-8 target string are
3353 * members of a fold-pair, and arrays are set up for all of them so that
3354 * the other member of the pair can be found quickly. Code elsewhere in
3355 * this file makes sure that in EXACTFU nodes, the sharp s gets folded to
3356 * 'ss', even if the pattern isn't UTF-8. This avoids the issues
3357 * described in the next item.
3358 * 3) A problem remains for unfolded multi-char folds. (These occur when the
3359 * validity of the fold won't be known until runtime, and so must remain
3360 * unfolded for now. This happens for the sharp s in EXACTF and EXACTFA
3361 * nodes when the pattern isn't in UTF-8. (Note, BTW, that there cannot
3362 * be an EXACTF node with a UTF-8 pattern.) They also occur for various
3363 * folds in EXACTFL nodes, regardless of the UTF-ness of the pattern.)
3364 * The reason this is a problem is that the optimizer part of regexec.c
3365 * (probably unwittingly, in Perl_regexec_flags()) makes an assumption
3366 * that a character in the pattern corresponds to at most a single
3367 * character in the target string. (And I do mean character, and not byte
3368 * here, unlike other parts of the documentation that have never been
3369 * updated to account for multibyte Unicode.) sharp s in EXACTF and
3370 * EXACTFL nodes can match the two character string 'ss'; in EXACTFA nodes
3371 * it can match "\x{17F}\x{17F}". These, along with other ones in EXACTFL
3372 * nodes, violate the assumption, and they are the only instances where it
3373 * is violated. I'm reluctant to try to change the assumption, as the
3374 * code involved is impenetrable to me (khw), so instead the code here
3375 * punts. This routine examines EXACTFL nodes, and (when the pattern
3376 * isn't UTF-8) EXACTF and EXACTFA for such unfolded folds, and returns a
3377 * boolean indicating whether or not the node contains such a fold. When
3378 * it is true, the caller sets a flag that later causes the optimizer in
3379 * this file to not set values for the floating and fixed string lengths,
3380 * and thus avoids the optimizer code in regexec.c that makes the invalid
3381 * assumption. Thus, there is no optimization based on string lengths for
3382 * EXACTFL nodes that contain these few folds, nor for non-UTF8-pattern
3383 * EXACTF and EXACTFA nodes that contain the sharp s. (The reason the
3384 * assumption is wrong only in these cases is that all other non-UTF-8
3385 * folds are 1-1; and, for UTF-8 patterns, we pre-fold all other folds to
3386 * their expanded versions. (Again, we can't prefold sharp s to 'ss' in
3387 * EXACTF nodes because we don't know at compile time if it actually
3388 * matches 'ss' or not. For EXACTF nodes it will match iff the target
3389 * string is in UTF-8. This is in contrast to EXACTFU nodes, where it
3390 * always matches; and EXACTFA where it never does. In an EXACTFA node in
3391 * a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the
3392 * problem; but in a non-UTF8 pattern, folding it to that above-Latin1
3393 * string would require the pattern to be forced into UTF-8, the overhead
3394 * of which we want to avoid. Similarly the unfolded multi-char folds in
3395 * EXACTFL nodes will match iff the locale at the time of match is a UTF-8
3398 * Similarly, the code that generates tries doesn't currently handle
3399 * not-already-folded multi-char folds, and it looks like a pain to change
3400 * that. Therefore, trie generation of EXACTFA nodes with the sharp s
3401 * doesn't work. Instead, such an EXACTFA is turned into a new regnode,
3402 * EXACTFA_NO_TRIE, which the trie code knows not to handle. Most people
3403 * using /iaa matching will be doing so almost entirely with ASCII
3404 * strings, so this should rarely be encountered in practice */
3406 #define JOIN_EXACT(scan,min_subtract,unfolded_multi_char, flags) \
3407 if (PL_regkind[OP(scan)] == EXACT) \
3408 join_exact(pRExC_state,(scan),(min_subtract),unfolded_multi_char, (flags),NULL,depth+1)
3411 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
3412 UV *min_subtract, bool *unfolded_multi_char,
3413 U32 flags,regnode *val, U32 depth)
3415 /* Merge several consecutive EXACTish nodes into one. */
3416 regnode *n = regnext(scan);
3418 regnode *next = scan + NODE_SZ_STR(scan);
3422 regnode *stop = scan;
3423 GET_RE_DEBUG_FLAGS_DECL;
3425 PERL_UNUSED_ARG(depth);
3428 PERL_ARGS_ASSERT_JOIN_EXACT;
3429 #ifndef EXPERIMENTAL_INPLACESCAN
3430 PERL_UNUSED_ARG(flags);
3431 PERL_UNUSED_ARG(val);
3433 DEBUG_PEEP("join",scan,depth);
3435 /* Look through the subsequent nodes in the chain. Skip NOTHING, merge
3436 * EXACT ones that are mergeable to the current one. */
3438 && (PL_regkind[OP(n)] == NOTHING
3439 || (stringok && OP(n) == OP(scan)))
3441 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX)
3444 if (OP(n) == TAIL || n > next)
3446 if (PL_regkind[OP(n)] == NOTHING) {
3447 DEBUG_PEEP("skip:",n,depth);
3448 NEXT_OFF(scan) += NEXT_OFF(n);
3449 next = n + NODE_STEP_REGNODE;
3456 else if (stringok) {
3457 const unsigned int oldl = STR_LEN(scan);
3458 regnode * const nnext = regnext(n);
3460 /* XXX I (khw) kind of doubt that this works on platforms (should
3461 * Perl ever run on one) where U8_MAX is above 255 because of lots
3462 * of other assumptions */
3463 /* Don't join if the sum can't fit into a single node */
3464 if (oldl + STR_LEN(n) > U8_MAX)
3467 DEBUG_PEEP("merg",n,depth);
3470 NEXT_OFF(scan) += NEXT_OFF(n);
3471 STR_LEN(scan) += STR_LEN(n);
3472 next = n + NODE_SZ_STR(n);
3473 /* Now we can overwrite *n : */
3474 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
3482 #ifdef EXPERIMENTAL_INPLACESCAN
3483 if (flags && !NEXT_OFF(n)) {
3484 DEBUG_PEEP("atch", val, depth);
3485 if (reg_off_by_arg[OP(n)]) {
3486 ARG_SET(n, val - n);
3489 NEXT_OFF(n) = val - n;
3497 *unfolded_multi_char = FALSE;
3499 /* Here, all the adjacent mergeable EXACTish nodes have been merged. We
3500 * can now analyze for sequences of problematic code points. (Prior to
3501 * this final joining, sequences could have been split over boundaries, and
3502 * hence missed). The sequences only happen in folding, hence for any
3503 * non-EXACT EXACTish node */
3504 if (OP(scan) != EXACT && OP(scan) != EXACTL) {
3505 U8* s0 = (U8*) STRING(scan);
3507 U8* s_end = s0 + STR_LEN(scan);
3509 int total_count_delta = 0; /* Total delta number of characters that
3510 multi-char folds expand to */
3512 /* One pass is made over the node's string looking for all the
3513 * possibilities. To avoid some tests in the loop, there are two main
3514 * cases, for UTF-8 patterns (which can't have EXACTF nodes) and
3519 if (OP(scan) == EXACTFL) {
3522 /* An EXACTFL node would already have been changed to another
3523 * node type unless there is at least one character in it that
3524 * is problematic; likely a character whose fold definition
3525 * won't be known until runtime, and so has yet to be folded.
3526 * For all but the UTF-8 locale, folds are 1-1 in length, but
3527 * to handle the UTF-8 case, we need to create a temporary
3528 * folded copy using UTF-8 locale rules in order to analyze it.
3529 * This is because our macros that look to see if a sequence is
3530 * a multi-char fold assume everything is folded (otherwise the
3531 * tests in those macros would be too complicated and slow).
3532 * Note that here, the non-problematic folds will have already
3533 * been done, so we can just copy such characters. We actually
3534 * don't completely fold the EXACTFL string. We skip the
3535 * unfolded multi-char folds, as that would just create work
3536 * below to figure out the size they already are */
3538 Newx(folded, UTF8_MAX_FOLD_CHAR_EXPAND * STR_LEN(scan) + 1, U8);
3541 STRLEN s_len = UTF8SKIP(s);
3542 if (! is_PROBLEMATIC_LOCALE_FOLD_utf8(s)) {
3543 Copy(s, d, s_len, U8);
3546 else if (is_FOLDS_TO_MULTI_utf8(s)) {
3547 *unfolded_multi_char = TRUE;
3548 Copy(s, d, s_len, U8);
3551 else if (isASCII(*s)) {
3552 *(d++) = toFOLD(*s);
3556 _to_utf8_fold_flags(s, d, &len, FOLD_FLAGS_FULL);
3562 /* Point the remainder of the routine to look at our temporary
3566 } /* End of creating folded copy of EXACTFL string */
3568 /* Examine the string for a multi-character fold sequence. UTF-8
3569 * patterns have all characters pre-folded by the time this code is
3571 while (s < s_end - 1) /* Can stop 1 before the end, as minimum
3572 length sequence we are looking for is 2 */
3574 int count = 0; /* How many characters in a multi-char fold */
3575 int len = is_MULTI_CHAR_FOLD_utf8_safe(s, s_end);
3576 if (! len) { /* Not a multi-char fold: get next char */
3581 /* Nodes with 'ss' require special handling, except for
3582 * EXACTFA-ish for which there is no multi-char fold to this */
3583 if (len == 2 && *s == 's' && *(s+1) == 's'
3584 && OP(scan) != EXACTFA
3585 && OP(scan) != EXACTFA_NO_TRIE)
3588 if (OP(scan) != EXACTFL) {
3589 OP(scan) = EXACTFU_SS;
3593 else { /* Here is a generic multi-char fold. */
3594 U8* multi_end = s + len;
3596 /* Count how many characters are in it. In the case of
3597 * /aa, no folds which contain ASCII code points are
3598 * allowed, so check for those, and skip if found. */
3599 if (OP(scan) != EXACTFA && OP(scan) != EXACTFA_NO_TRIE) {
3600 count = utf8_length(s, multi_end);
3604 while (s < multi_end) {
3607 goto next_iteration;
3617 /* The delta is how long the sequence is minus 1 (1 is how long
3618 * the character that folds to the sequence is) */
3619 total_count_delta += count - 1;
3623 /* We created a temporary folded copy of the string in EXACTFL
3624 * nodes. Therefore we need to be sure it doesn't go below zero,
3625 * as the real string could be shorter */
3626 if (OP(scan) == EXACTFL) {
3627 int total_chars = utf8_length((U8*) STRING(scan),
3628 (U8*) STRING(scan) + STR_LEN(scan));
3629 if (total_count_delta > total_chars) {
3630 total_count_delta = total_chars;
3634 *min_subtract += total_count_delta;
3637 else if (OP(scan) == EXACTFA) {
3639 /* Non-UTF-8 pattern, EXACTFA node. There can't be a multi-char
3640 * fold to the ASCII range (and there are no existing ones in the
3641 * upper latin1 range). But, as outlined in the comments preceding
3642 * this function, we need to flag any occurrences of the sharp s.
3643 * This character forbids trie formation (because of added
3646 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
3647 OP(scan) = EXACTFA_NO_TRIE;
3648 *unfolded_multi_char = TRUE;
3657 /* Non-UTF-8 pattern, not EXACTFA node. Look for the multi-char
3658 * folds that are all Latin1. As explained in the comments
3659 * preceding this function, we look also for the sharp s in EXACTF
3660 * and EXACTFL nodes; it can be in the final position. Otherwise
3661 * we can stop looking 1 byte earlier because have to find at least
3662 * two characters for a multi-fold */
3663 const U8* upper = (OP(scan) == EXACTF || OP(scan) == EXACTFL)
3668 int len = is_MULTI_CHAR_FOLD_latin1_safe(s, s_end);
3669 if (! len) { /* Not a multi-char fold. */
3670 if (*s == LATIN_SMALL_LETTER_SHARP_S
3671 && (OP(scan) == EXACTF || OP(scan) == EXACTFL))
3673 *unfolded_multi_char = TRUE;
3680 && isALPHA_FOLD_EQ(*s, 's')
3681 && isALPHA_FOLD_EQ(*(s+1), 's'))
3684 /* EXACTF nodes need to know that the minimum length
3685 * changed so that a sharp s in the string can match this
3686 * ss in the pattern, but they remain EXACTF nodes, as they
3687 * won't match this unless the target string is is UTF-8,
3688 * which we don't know until runtime. EXACTFL nodes can't
3689 * transform into EXACTFU nodes */
3690 if (OP(scan) != EXACTF && OP(scan) != EXACTFL) {
3691 OP(scan) = EXACTFU_SS;
3695 *min_subtract += len - 1;
3702 /* Allow dumping but overwriting the collection of skipped
3703 * ops and/or strings with fake optimized ops */
3704 n = scan + NODE_SZ_STR(scan);
3712 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
3716 /* REx optimizer. Converts nodes into quicker variants "in place".
3717 Finds fixed substrings. */
3719 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
3720 to the position after last scanned or to NULL. */
3722 #define INIT_AND_WITHP \
3723 assert(!and_withp); \
3724 Newx(and_withp,1, regnode_ssc); \
3725 SAVEFREEPV(and_withp)
3729 S_unwind_scan_frames(pTHX_ const void *p)
3731 scan_frame *f= (scan_frame *)p;
3733 scan_frame *n= f->next_frame;
3741 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
3742 SSize_t *minlenp, SSize_t *deltap,
3747 regnode_ssc *and_withp,
3748 U32 flags, U32 depth)
3749 /* scanp: Start here (read-write). */
3750 /* deltap: Write maxlen-minlen here. */
3751 /* last: Stop before this one. */
3752 /* data: string data about the pattern */
3753 /* stopparen: treat close N as END */
3754 /* recursed: which subroutines have we recursed into */
3755 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
3757 /* There must be at least this number of characters to match */
3760 regnode *scan = *scanp, *next;
3762 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
3763 int is_inf_internal = 0; /* The studied chunk is infinite */
3764 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
3765 scan_data_t data_fake;
3766 SV *re_trie_maxbuff = NULL;
3767 regnode *first_non_open = scan;
3768 SSize_t stopmin = SSize_t_MAX;
3769 scan_frame *frame = NULL;
3770 GET_RE_DEBUG_FLAGS_DECL;
3772 PERL_ARGS_ASSERT_STUDY_CHUNK;
3776 while (first_non_open && OP(first_non_open) == OPEN)
3777 first_non_open=regnext(first_non_open);
3783 RExC_study_chunk_recursed_count++;
3785 DEBUG_OPTIMISE_MORE_r(
3787 PerlIO_printf(Perl_debug_log,
3788 "%*sstudy_chunk stopparen=%ld recursed_count=%lu depth=%lu recursed_depth=%lu scan=%p last=%p",
3789 (int)(depth*2), "", (long)stopparen,
3790 (unsigned long)RExC_study_chunk_recursed_count,
3791 (unsigned long)depth, (unsigned long)recursed_depth,
3794 if (recursed_depth) {
3797 for ( j = 0 ; j < recursed_depth ; j++ ) {
3798 for ( i = 0 ; i < (U32)RExC_npar ; i++ ) {
3800 PAREN_TEST(RExC_study_chunk_recursed +
3801 ( j * RExC_study_chunk_recursed_bytes), i )
3804 !PAREN_TEST(RExC_study_chunk_recursed +
3805 (( j - 1 ) * RExC_study_chunk_recursed_bytes), i)
3808 PerlIO_printf(Perl_debug_log," %d",(int)i);
3812 if ( j + 1 < recursed_depth ) {
3813 PerlIO_printf(Perl_debug_log, ",");
3817 PerlIO_printf(Perl_debug_log,"\n");
3820 while ( scan && OP(scan) != END && scan < last ){
3821 UV min_subtract = 0; /* How mmany chars to subtract from the minimum
3822 node length to get a real minimum (because
3823 the folded version may be shorter) */
3824 bool unfolded_multi_char = FALSE;
3825 /* Peephole optimizer: */
3826 DEBUG_STUDYDATA("Peep:", data, depth);
3827 DEBUG_PEEP("Peep", scan, depth);
3830 /* The reason we do this here we need to deal with things like /(?:f)(?:o)(?:o)/
3831 * which cant be dealt with by the normal EXACT parsing code, as each (?:..) is handled
3832 * by a different invocation of reg() -- Yves
3834 JOIN_EXACT(scan,&min_subtract, &unfolded_multi_char, 0);
3836 /* Follow the next-chain of the current node and optimize
3837 away all the NOTHINGs from it. */
3838 if (OP(scan) != CURLYX) {
3839 const int max = (reg_off_by_arg[OP(scan)]
3841 /* I32 may be smaller than U16 on CRAYs! */
3842 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
3843 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
3847 /* Skip NOTHING and LONGJMP. */
3848 while ((n = regnext(n))
3849 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
3850 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
3851 && off + noff < max)
3853 if (reg_off_by_arg[OP(scan)])
3856 NEXT_OFF(scan) = off;
3859 /* The principal pseudo-switch. Cannot be a switch, since we
3860 look into several different things. */
3861 if ( OP(scan) == DEFINEP ) {
3863 SSize_t deltanext = 0;
3864 SSize_t fake_last_close = 0;
3865 I32 f = SCF_IN_DEFINE;
3867 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
3868 scan = regnext(scan);
3869 assert( OP(scan) == IFTHEN );
3870 DEBUG_PEEP("expect IFTHEN", scan, depth);
3872 data_fake.last_closep= &fake_last_close;
3874 next = regnext(scan);
3875 scan = NEXTOPER(NEXTOPER(scan));
3876 DEBUG_PEEP("scan", scan, depth);
3877 DEBUG_PEEP("next", next, depth);
3879 /* we suppose the run is continuous, last=next...
3880 * NOTE we dont use the return here! */
3881 (void)study_chunk(pRExC_state, &scan, &minlen,
3882 &deltanext, next, &data_fake, stopparen,
3883 recursed_depth, NULL, f, depth+1);
3888 OP(scan) == BRANCH ||
3889 OP(scan) == BRANCHJ ||
3892 next = regnext(scan);
3895 /* The op(next)==code check below is to see if we
3896 * have "BRANCH-BRANCH", "BRANCHJ-BRANCHJ", "IFTHEN-IFTHEN"
3897 * IFTHEN is special as it might not appear in pairs.
3898 * Not sure whether BRANCH-BRANCHJ is possible, regardless
3899 * we dont handle it cleanly. */
3900 if (OP(next) == code || code == IFTHEN) {
3901 /* NOTE - There is similar code to this block below for
3902 * handling TRIE nodes on a re-study. If you change stuff here
3903 * check there too. */
3904 SSize_t max1 = 0, min1 = SSize_t_MAX, num = 0;
3906 regnode * const startbranch=scan;
3908 if (flags & SCF_DO_SUBSTR) {
3909 /* Cannot merge strings after this. */
3910 scan_commit(pRExC_state, data, minlenp, is_inf);
3913 if (flags & SCF_DO_STCLASS)
3914 ssc_init_zero(pRExC_state, &accum);
3916 while (OP(scan) == code) {
3917 SSize_t deltanext, minnext, fake;
3919 regnode_ssc this_class;
3921 DEBUG_PEEP("Branch", scan, depth);
3924 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
3926 data_fake.whilem_c = data->whilem_c;
3927 data_fake.last_closep = data->last_closep;
3930 data_fake.last_closep = &fake;
3932 data_fake.pos_delta = delta;
3933 next = regnext(scan);
3935 scan = NEXTOPER(scan); /* everything */
3936 if (code != BRANCH) /* everything but BRANCH */
3937 scan = NEXTOPER(scan);
3939 if (flags & SCF_DO_STCLASS) {
3940 ssc_init(pRExC_state, &this_class);
3941 data_fake.start_class = &this_class;
3942 f = SCF_DO_STCLASS_AND;
3944 if (flags & SCF_WHILEM_VISITED_POS)
3945 f |= SCF_WHILEM_VISITED_POS;
3947 /* we suppose the run is continuous, last=next...*/
3948 minnext = study_chunk(pRExC_state, &scan, minlenp,
3949 &deltanext, next, &data_fake, stopparen,
3950 recursed_depth, NULL, f,depth+1);
3954 if (deltanext == SSize_t_MAX) {
3955 is_inf = is_inf_internal = 1;
3957 } else if (max1 < minnext + deltanext)
3958 max1 = minnext + deltanext;
3960 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3962 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3963 if ( stopmin > minnext)
3964 stopmin = min + min1;
3965 flags &= ~SCF_DO_SUBSTR;
3967 data->flags |= SCF_SEEN_ACCEPT;
https://perl5.git.perl.org / perl5.git / blob