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 "inline_invlist.c"
91 #include "unicode_constants.h"
93 #define HAS_NONLATIN1_FOLD_CLOSURE(i) \
94 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
95 #define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
96 _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
97 #define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
98 #define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
101 #define STATIC static
105 #define MIN(a,b) ((a) < (b) ? (a) : (b))
108 /* this is a chain of data about sub patterns we are processing that
109 need to be handled separately/specially in study_chunk. Its so
110 we can simulate recursion without losing state. */
112 typedef struct scan_frame {
113 regnode *last_regnode; /* last node to process in this frame */
114 regnode *next_regnode; /* next node to process when last is reached */
115 U32 prev_recursed_depth;
116 I32 stopparen; /* what stopparen do we use */
117 U32 is_top_frame; /* what flags do we use? */
119 struct scan_frame *this_prev_frame; /* this previous frame */
120 struct scan_frame *prev_frame; /* previous frame */
121 struct scan_frame *next_frame; /* next frame */
124 /* Certain characters are output as a sequence with the first being a
126 #define isBACKSLASHED_PUNCT(c) \
127 ((c) == '-' || (c) == ']' || (c) == '\\' || (c) == '^')
130 struct RExC_state_t {
131 U32 flags; /* RXf_* are we folding, multilining? */
132 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
133 char *precomp; /* uncompiled string. */
134 REGEXP *rx_sv; /* The SV that is the regexp. */
135 regexp *rx; /* perl core regexp structure */
136 regexp_internal *rxi; /* internal data for regexp object
138 char *start; /* Start of input for compile */
139 char *end; /* End of input for compile */
140 char *parse; /* Input-scan pointer. */
141 SSize_t whilem_seen; /* number of WHILEM in this expr */
142 regnode *emit_start; /* Start of emitted-code area */
143 regnode *emit_bound; /* First regnode outside of the
145 regnode *emit; /* Code-emit pointer; if = &emit_dummy,
146 implies compiling, so don't emit */
147 regnode_ssc emit_dummy; /* placeholder for emit to point to;
148 large enough for the largest
149 non-EXACTish node, so can use it as
151 I32 naughty; /* How bad is this pattern? */
152 I32 sawback; /* Did we see \1, ...? */
154 SSize_t size; /* Code size. */
155 I32 npar; /* Capture buffer count, (OPEN) plus
156 one. ("par" 0 is the whole
158 I32 nestroot; /* root parens we are in - used by
162 regnode **open_parens; /* pointers to open parens */
163 regnode **close_parens; /* pointers to close parens */
164 regnode *opend; /* END node in program */
165 I32 utf8; /* whether the pattern is utf8 or not */
166 I32 orig_utf8; /* whether the pattern was originally in utf8 */
167 /* XXX use this for future optimisation of case
168 * where pattern must be upgraded to utf8. */
169 I32 uni_semantics; /* If a d charset modifier should use unicode
170 rules, even if the pattern is not in
172 HV *paren_names; /* Paren names */
174 regnode **recurse; /* Recurse regops */
175 I32 recurse_count; /* Number of recurse regops */
176 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
178 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
182 I32 override_recoding;
184 I32 recode_x_to_native;
186 I32 in_multi_char_class;
187 struct reg_code_block *code_blocks; /* positions of literal (?{})
189 int num_code_blocks; /* size of code_blocks[] */
190 int code_index; /* next code_blocks[] slot */
191 SSize_t maxlen; /* mininum possible number of chars in string to match */
192 scan_frame *frame_head;
193 scan_frame *frame_last;
196 #ifdef ADD_TO_REGEXEC
197 char *starttry; /* -Dr: where regtry was called. */
198 #define RExC_starttry (pRExC_state->starttry)
200 SV *runtime_code_qr; /* qr with the runtime code blocks */
202 const char *lastparse;
204 AV *paren_name_list; /* idx -> name */
205 U32 study_chunk_recursed_count;
208 #define RExC_lastparse (pRExC_state->lastparse)
209 #define RExC_lastnum (pRExC_state->lastnum)
210 #define RExC_paren_name_list (pRExC_state->paren_name_list)
211 #define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
212 #define RExC_mysv (pRExC_state->mysv1)
213 #define RExC_mysv1 (pRExC_state->mysv1)
214 #define RExC_mysv2 (pRExC_state->mysv2)
219 #define RExC_flags (pRExC_state->flags)
220 #define RExC_pm_flags (pRExC_state->pm_flags)
221 #define RExC_precomp (pRExC_state->precomp)
222 #define RExC_rx_sv (pRExC_state->rx_sv)
223 #define RExC_rx (pRExC_state->rx)
224 #define RExC_rxi (pRExC_state->rxi)
225 #define RExC_start (pRExC_state->start)
226 #define RExC_end (pRExC_state->end)
227 #define RExC_parse (pRExC_state->parse)
228 #define RExC_whilem_seen (pRExC_state->whilem_seen)
229 #ifdef RE_TRACK_PATTERN_OFFSETS
230 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the
233 #define RExC_emit (pRExC_state->emit)
234 #define RExC_emit_dummy (pRExC_state->emit_dummy)
235 #define RExC_emit_start (pRExC_state->emit_start)
236 #define RExC_emit_bound (pRExC_state->emit_bound)
237 #define RExC_sawback (pRExC_state->sawback)
238 #define RExC_seen (pRExC_state->seen)
239 #define RExC_size (pRExC_state->size)
240 #define RExC_maxlen (pRExC_state->maxlen)
241 #define RExC_npar (pRExC_state->npar)
242 #define RExC_nestroot (pRExC_state->nestroot)
243 #define RExC_extralen (pRExC_state->extralen)
244 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
245 #define RExC_utf8 (pRExC_state->utf8)
246 #define RExC_uni_semantics (pRExC_state->uni_semantics)
247 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
248 #define RExC_open_parens (pRExC_state->open_parens)
249 #define RExC_close_parens (pRExC_state->close_parens)
250 #define RExC_opend (pRExC_state->opend)
251 #define RExC_paren_names (pRExC_state->paren_names)
252 #define RExC_recurse (pRExC_state->recurse)
253 #define RExC_recurse_count (pRExC_state->recurse_count)
254 #define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
255 #define RExC_study_chunk_recursed_bytes \
256 (pRExC_state->study_chunk_recursed_bytes)
257 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
258 #define RExC_contains_locale (pRExC_state->contains_locale)
259 #define RExC_contains_i (pRExC_state->contains_i)
260 #define RExC_override_recoding (pRExC_state->override_recoding)
262 # define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
264 #define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
265 #define RExC_frame_head (pRExC_state->frame_head)
266 #define RExC_frame_last (pRExC_state->frame_last)
267 #define RExC_frame_count (pRExC_state->frame_count)
268 #define RExC_strict (pRExC_state->strict)
270 /* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
271 * a flag to disable back-off on the fixed/floating substrings - if it's
272 * a high complexity pattern we assume the benefit of avoiding a full match
273 * is worth the cost of checking for the substrings even if they rarely help.
275 #define RExC_naughty (pRExC_state->naughty)
276 #define TOO_NAUGHTY (10)
277 #define MARK_NAUGHTY(add) \
278 if (RExC_naughty < TOO_NAUGHTY) \
279 RExC_naughty += (add)
280 #define MARK_NAUGHTY_EXP(exp, add) \
281 if (RExC_naughty < TOO_NAUGHTY) \
282 RExC_naughty += RExC_naughty / (exp) + (add)
284 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
285 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
286 ((*s) == '{' && regcurly(s)))
289 * Flags to be passed up and down.
291 #define WORST 0 /* Worst case. */
292 #define HASWIDTH 0x01 /* Known to match non-null strings. */
294 /* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
295 * character. (There needs to be a case: in the switch statement in regexec.c
296 * for any node marked SIMPLE.) Note that this is not the same thing as
299 #define SPSTART 0x04 /* Starts with * or + */
300 #define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
301 #define TRYAGAIN 0x10 /* Weeded out a declaration. */
302 #define RESTART_UTF8 0x20 /* Restart, need to calcuate sizes as UTF-8 */
304 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
306 /* whether trie related optimizations are enabled */
307 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
308 #define TRIE_STUDY_OPT
309 #define FULL_TRIE_STUDY
315 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
316 #define PBITVAL(paren) (1 << ((paren) & 7))
317 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
318 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
319 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
321 #define REQUIRE_UTF8 STMT_START { \
323 *flagp = RESTART_UTF8; \
328 /* This converts the named class defined in regcomp.h to its equivalent class
329 * number defined in handy.h. */
330 #define namedclass_to_classnum(class) ((int) ((class) / 2))
331 #define classnum_to_namedclass(classnum) ((classnum) * 2)
333 #define _invlist_union_complement_2nd(a, b, output) \
334 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
335 #define _invlist_intersection_complement_2nd(a, b, output) \
336 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
338 /* About scan_data_t.
340 During optimisation we recurse through the regexp program performing
341 various inplace (keyhole style) optimisations. In addition study_chunk
342 and scan_commit populate this data structure with information about
343 what strings MUST appear in the pattern. We look for the longest
344 string that must appear at a fixed location, and we look for the
345 longest string that may appear at a floating location. So for instance
350 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
351 strings (because they follow a .* construct). study_chunk will identify
352 both FOO and BAR as being the longest fixed and floating strings respectively.
354 The strings can be composites, for instance
358 will result in a composite fixed substring 'foo'.
360 For each string some basic information is maintained:
362 - offset or min_offset
363 This is the position the string must appear at, or not before.
364 It also implicitly (when combined with minlenp) tells us how many
365 characters must match before the string we are searching for.
366 Likewise when combined with minlenp and the length of the string it
367 tells us how many characters must appear after the string we have
371 Only used for floating strings. This is the rightmost point that
372 the string can appear at. If set to SSize_t_MAX it indicates that the
373 string can occur infinitely far to the right.
376 A pointer to the minimum number of characters of the pattern that the
377 string was found inside. This is important as in the case of positive
378 lookahead or positive lookbehind we can have multiple patterns
383 The minimum length of the pattern overall is 3, the minimum length
384 of the lookahead part is 3, but the minimum length of the part that
385 will actually match is 1. So 'FOO's minimum length is 3, but the
386 minimum length for the F is 1. This is important as the minimum length
387 is used to determine offsets in front of and behind the string being
388 looked for. Since strings can be composites this is the length of the
389 pattern at the time it was committed with a scan_commit. Note that
390 the length is calculated by study_chunk, so that the minimum lengths
391 are not known until the full pattern has been compiled, thus the
392 pointer to the value.
396 In the case of lookbehind the string being searched for can be
397 offset past the start point of the final matching string.
398 If this value was just blithely removed from the min_offset it would
399 invalidate some of the calculations for how many chars must match
400 before or after (as they are derived from min_offset and minlen and
401 the length of the string being searched for).
402 When the final pattern is compiled and the data is moved from the
403 scan_data_t structure into the regexp structure the information
404 about lookbehind is factored in, with the information that would
405 have been lost precalculated in the end_shift field for the
408 The fields pos_min and pos_delta are used to store the minimum offset
409 and the delta to the maximum offset at the current point in the pattern.
413 typedef struct scan_data_t {
414 /*I32 len_min; unused */
415 /*I32 len_delta; unused */
419 SSize_t last_end; /* min value, <0 unless valid. */
420 SSize_t last_start_min;
421 SSize_t last_start_max;
422 SV **longest; /* Either &l_fixed, or &l_float. */
423 SV *longest_fixed; /* longest fixed string found in pattern */
424 SSize_t offset_fixed; /* offset where it starts */
425 SSize_t *minlen_fixed; /* pointer to the minlen relevant to the string */
426 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
427 SV *longest_float; /* longest floating string found in pattern */
428 SSize_t offset_float_min; /* earliest point in string it can appear */
429 SSize_t offset_float_max; /* latest point in string it can appear */
430 SSize_t *minlen_float; /* pointer to the minlen relevant to the string */
431 SSize_t lookbehind_float; /* is the pos of the string modified by LB */
434 SSize_t *last_closep;
435 regnode_ssc *start_class;
439 * Forward declarations for pregcomp()'s friends.
442 static const scan_data_t zero_scan_data =
443 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
445 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
446 #define SF_BEFORE_SEOL 0x0001
447 #define SF_BEFORE_MEOL 0x0002
448 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
449 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
451 #define SF_FIX_SHIFT_EOL (+2)
452 #define SF_FL_SHIFT_EOL (+4)
454 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
455 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
457 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
458 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
459 #define SF_IS_INF 0x0040
460 #define SF_HAS_PAR 0x0080
461 #define SF_IN_PAR 0x0100
462 #define SF_HAS_EVAL 0x0200
463 #define SCF_DO_SUBSTR 0x0400
464 #define SCF_DO_STCLASS_AND 0x0800
465 #define SCF_DO_STCLASS_OR 0x1000
466 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
467 #define SCF_WHILEM_VISITED_POS 0x2000
469 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
470 #define SCF_SEEN_ACCEPT 0x8000
471 #define SCF_TRIE_DOING_RESTUDY 0x10000
472 #define SCF_IN_DEFINE 0x20000
477 #define UTF cBOOL(RExC_utf8)
479 /* The enums for all these are ordered so things work out correctly */
480 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
481 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
482 == REGEX_DEPENDS_CHARSET)
483 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
484 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
485 >= REGEX_UNICODE_CHARSET)
486 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
487 == REGEX_ASCII_RESTRICTED_CHARSET)
488 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
489 >= REGEX_ASCII_RESTRICTED_CHARSET)
490 #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
491 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
493 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
495 /* For programs that want to be strictly Unicode compatible by dying if any
496 * attempt is made to match a non-Unicode code point against a Unicode
498 #define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
500 #define OOB_NAMEDCLASS -1
502 /* There is no code point that is out-of-bounds, so this is problematic. But
503 * its only current use is to initialize a variable that is always set before
505 #define OOB_UNICODE 0xDEADBEEF
507 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
508 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
511 /* length of regex to show in messages that don't mark a position within */
512 #define RegexLengthToShowInErrorMessages 127
515 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
516 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
517 * op/pragma/warn/regcomp.
519 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
520 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
522 #define REPORT_LOCATION " in regex; marked by " MARKER1 \
523 " in m/%"UTF8f MARKER2 "%"UTF8f"/"
525 #define REPORT_LOCATION_ARGS(offset) \
526 UTF8fARG(UTF, offset, RExC_precomp), \
527 UTF8fARG(UTF, RExC_end - RExC_precomp - offset, RExC_precomp + offset)
529 /* Used to point after bad bytes for an error message, but avoid skipping
530 * past a nul byte. */
531 #define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
534 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
535 * arg. Show regex, up to a maximum length. If it's too long, chop and add
538 #define _FAIL(code) STMT_START { \
539 const char *ellipses = ""; \
540 IV len = RExC_end - RExC_precomp; \
543 SAVEFREESV(RExC_rx_sv); \
544 if (len > RegexLengthToShowInErrorMessages) { \
545 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
546 len = RegexLengthToShowInErrorMessages - 10; \
552 #define FAIL(msg) _FAIL( \
553 Perl_croak(aTHX_ "%s in regex m/%"UTF8f"%s/", \
554 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
556 #define FAIL2(msg,arg) _FAIL( \
557 Perl_croak(aTHX_ msg " in regex m/%"UTF8f"%s/", \
558 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
561 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
563 #define Simple_vFAIL(m) STMT_START { \
565 (RExC_parse > RExC_end ? RExC_end : RExC_parse) - RExC_precomp; \
566 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
567 m, REPORT_LOCATION_ARGS(offset)); \
571 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
573 #define vFAIL(m) STMT_START { \
575 SAVEFREESV(RExC_rx_sv); \
580 * Like Simple_vFAIL(), but accepts two arguments.
582 #define Simple_vFAIL2(m,a1) STMT_START { \
583 const IV offset = RExC_parse - RExC_precomp; \
584 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
585 REPORT_LOCATION_ARGS(offset)); \
589 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
591 #define vFAIL2(m,a1) STMT_START { \
593 SAVEFREESV(RExC_rx_sv); \
594 Simple_vFAIL2(m, a1); \
599 * Like Simple_vFAIL(), but accepts three arguments.
601 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
602 const IV offset = RExC_parse - RExC_precomp; \
603 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
604 REPORT_LOCATION_ARGS(offset)); \
608 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
610 #define vFAIL3(m,a1,a2) STMT_START { \
612 SAVEFREESV(RExC_rx_sv); \
613 Simple_vFAIL3(m, a1, a2); \
617 * Like Simple_vFAIL(), but accepts four arguments.
619 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
620 const IV offset = RExC_parse - RExC_precomp; \
621 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
622 REPORT_LOCATION_ARGS(offset)); \
625 #define vFAIL4(m,a1,a2,a3) STMT_START { \
627 SAVEFREESV(RExC_rx_sv); \
628 Simple_vFAIL4(m, a1, a2, a3); \
631 /* A specialized version of vFAIL2 that works with UTF8f */
632 #define vFAIL2utf8f(m, a1) STMT_START { \
633 const IV offset = RExC_parse - RExC_precomp; \
635 SAVEFREESV(RExC_rx_sv); \
636 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
637 REPORT_LOCATION_ARGS(offset)); \
640 /* These have asserts in them because of [perl #122671] Many warnings in
641 * regcomp.c can occur twice. If they get output in pass1 and later in that
642 * pass, the pattern has to be converted to UTF-8 and the pass restarted, they
643 * would get output again. So they should be output in pass2, and these
644 * asserts make sure new warnings follow that paradigm. */
646 /* m is not necessarily a "literal string", in this macro */
647 #define reg_warn_non_literal_string(loc, m) STMT_START { \
648 const IV offset = loc - RExC_precomp; \
649 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
650 m, REPORT_LOCATION_ARGS(offset)); \
653 #define ckWARNreg(loc,m) STMT_START { \
654 const IV offset = loc - RExC_precomp; \
655 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
656 REPORT_LOCATION_ARGS(offset)); \
659 #define vWARN(loc, m) STMT_START { \
660 const IV offset = loc - RExC_precomp; \
661 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
662 REPORT_LOCATION_ARGS(offset)); \
665 #define vWARN_dep(loc, m) STMT_START { \
666 const IV offset = loc - RExC_precomp; \
667 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), m REPORT_LOCATION, \
668 REPORT_LOCATION_ARGS(offset)); \
671 #define ckWARNdep(loc,m) STMT_START { \
672 const IV offset = loc - RExC_precomp; \
673 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
675 REPORT_LOCATION_ARGS(offset)); \
678 #define ckWARNregdep(loc,m) STMT_START { \
679 const IV offset = loc - RExC_precomp; \
680 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
682 REPORT_LOCATION_ARGS(offset)); \
685 #define ckWARN2reg_d(loc,m, a1) STMT_START { \
686 const IV offset = loc - RExC_precomp; \
687 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
689 a1, REPORT_LOCATION_ARGS(offset)); \
692 #define ckWARN2reg(loc, m, a1) STMT_START { \
693 const IV offset = loc - RExC_precomp; \
694 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
695 a1, REPORT_LOCATION_ARGS(offset)); \
698 #define vWARN3(loc, m, a1, a2) STMT_START { \
699 const IV offset = loc - RExC_precomp; \
700 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
701 a1, a2, REPORT_LOCATION_ARGS(offset)); \
704 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
705 const IV offset = loc - RExC_precomp; \
706 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
707 a1, a2, REPORT_LOCATION_ARGS(offset)); \
710 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
711 const IV offset = loc - RExC_precomp; \
712 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
713 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
716 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
717 const IV offset = loc - RExC_precomp; \
718 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
719 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
722 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
723 const IV offset = loc - RExC_precomp; \
724 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
725 a1, a2, a3, a4, REPORT_LOCATION_ARGS(offset)); \
728 /* Macros for recording node offsets. 20001227 mjd@plover.com
729 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
730 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
731 * Element 0 holds the number n.
732 * Position is 1 indexed.
734 #ifndef RE_TRACK_PATTERN_OFFSETS
735 #define Set_Node_Offset_To_R(node,byte)
736 #define Set_Node_Offset(node,byte)
737 #define Set_Cur_Node_Offset
738 #define Set_Node_Length_To_R(node,len)
739 #define Set_Node_Length(node,len)
740 #define Set_Node_Cur_Length(node,start)
741 #define Node_Offset(n)
742 #define Node_Length(n)
743 #define Set_Node_Offset_Length(node,offset,len)
744 #define ProgLen(ri) ri->u.proglen
745 #define SetProgLen(ri,x) ri->u.proglen = x
747 #define ProgLen(ri) ri->u.offsets[0]
748 #define SetProgLen(ri,x) ri->u.offsets[0] = x
749 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
751 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
752 __LINE__, (int)(node), (int)(byte))); \
754 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
757 RExC_offsets[2*(node)-1] = (byte); \
762 #define Set_Node_Offset(node,byte) \
763 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
764 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
766 #define Set_Node_Length_To_R(node,len) STMT_START { \
768 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
769 __LINE__, (int)(node), (int)(len))); \
771 Perl_croak(aTHX_ "value of node is %d in Length macro", \
774 RExC_offsets[2*(node)] = (len); \
779 #define Set_Node_Length(node,len) \
780 Set_Node_Length_To_R((node)-RExC_emit_start, len)
781 #define Set_Node_Cur_Length(node, start) \
782 Set_Node_Length(node, RExC_parse - start)
784 /* Get offsets and lengths */
785 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
786 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
788 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
789 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
790 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
794 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
795 #define EXPERIMENTAL_INPLACESCAN
796 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
798 #define DEBUG_RExC_seen() \
799 DEBUG_OPTIMISE_MORE_r({ \
800 PerlIO_printf(Perl_debug_log,"RExC_seen: "); \
802 if (RExC_seen & REG_ZERO_LEN_SEEN) \
803 PerlIO_printf(Perl_debug_log,"REG_ZERO_LEN_SEEN "); \
805 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
806 PerlIO_printf(Perl_debug_log,"REG_LOOKBEHIND_SEEN "); \
808 if (RExC_seen & REG_GPOS_SEEN) \
809 PerlIO_printf(Perl_debug_log,"REG_GPOS_SEEN "); \
811 if (RExC_seen & REG_CANY_SEEN) \
812 PerlIO_printf(Perl_debug_log,"REG_CANY_SEEN "); \
814 if (RExC_seen & REG_RECURSE_SEEN) \
815 PerlIO_printf(Perl_debug_log,"REG_RECURSE_SEEN "); \
817 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
818 PerlIO_printf(Perl_debug_log,"REG_TOP_LEVEL_BRANCHES_SEEN "); \
820 if (RExC_seen & REG_VERBARG_SEEN) \
821 PerlIO_printf(Perl_debug_log,"REG_VERBARG_SEEN "); \
823 if (RExC_seen & REG_CUTGROUP_SEEN) \
824 PerlIO_printf(Perl_debug_log,"REG_CUTGROUP_SEEN "); \
826 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
827 PerlIO_printf(Perl_debug_log,"REG_RUN_ON_COMMENT_SEEN "); \
829 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
830 PerlIO_printf(Perl_debug_log,"REG_UNFOLDED_MULTI_SEEN "); \
832 if (RExC_seen & REG_GOSTART_SEEN) \
833 PerlIO_printf(Perl_debug_log,"REG_GOSTART_SEEN "); \
835 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
836 PerlIO_printf(Perl_debug_log,"REG_UNBOUNDED_QUANTIFIER_SEEN "); \
838 PerlIO_printf(Perl_debug_log,"\n"); \
841 #define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
842 if ((flags) & flag) PerlIO_printf(Perl_debug_log, "%s ", #flag)
844 #define DEBUG_SHOW_STUDY_FLAGS(flags,open_str,close_str) \
846 PerlIO_printf(Perl_debug_log, "%s", open_str); \
847 DEBUG_SHOW_STUDY_FLAG(flags,SF_FL_BEFORE_SEOL); \
848 DEBUG_SHOW_STUDY_FLAG(flags,SF_FL_BEFORE_MEOL); \
849 DEBUG_SHOW_STUDY_FLAG(flags,SF_IS_INF); \
850 DEBUG_SHOW_STUDY_FLAG(flags,SF_HAS_PAR); \
851 DEBUG_SHOW_STUDY_FLAG(flags,SF_IN_PAR); \
852 DEBUG_SHOW_STUDY_FLAG(flags,SF_HAS_EVAL); \
853 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_SUBSTR); \
854 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS_AND); \
855 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS_OR); \
856 DEBUG_SHOW_STUDY_FLAG(flags,SCF_DO_STCLASS); \
857 DEBUG_SHOW_STUDY_FLAG(flags,SCF_WHILEM_VISITED_POS); \
858 DEBUG_SHOW_STUDY_FLAG(flags,SCF_TRIE_RESTUDY); \
859 DEBUG_SHOW_STUDY_FLAG(flags,SCF_SEEN_ACCEPT); \
860 DEBUG_SHOW_STUDY_FLAG(flags,SCF_TRIE_DOING_RESTUDY); \
861 DEBUG_SHOW_STUDY_FLAG(flags,SCF_IN_DEFINE); \
862 PerlIO_printf(Perl_debug_log, "%s", close_str); \
866 #define DEBUG_STUDYDATA(str,data,depth) \
867 DEBUG_OPTIMISE_MORE_r(if(data){ \
868 PerlIO_printf(Perl_debug_log, \
869 "%*s" str "Pos:%"IVdf"/%"IVdf \
871 (int)(depth)*2, "", \
872 (IV)((data)->pos_min), \
873 (IV)((data)->pos_delta), \
874 (UV)((data)->flags) \
876 DEBUG_SHOW_STUDY_FLAGS((data)->flags," [ ","]"); \
877 PerlIO_printf(Perl_debug_log, \
878 " Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
879 (IV)((data)->whilem_c), \
880 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
881 is_inf ? "INF " : "" \
883 if ((data)->last_found) \
884 PerlIO_printf(Perl_debug_log, \
885 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
886 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
887 SvPVX_const((data)->last_found), \
888 (IV)((data)->last_end), \
889 (IV)((data)->last_start_min), \
890 (IV)((data)->last_start_max), \
891 ((data)->longest && \
892 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
893 SvPVX_const((data)->longest_fixed), \
894 (IV)((data)->offset_fixed), \
895 ((data)->longest && \
896 (data)->longest==&((data)->longest_float)) ? "*" : "", \
897 SvPVX_const((data)->longest_float), \
898 (IV)((data)->offset_float_min), \
899 (IV)((data)->offset_float_max) \
901 PerlIO_printf(Perl_debug_log,"\n"); \
904 /* is c a control character for which we have a mnemonic? */
905 #define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
908 S_cntrl_to_mnemonic(const U8 c)
910 /* Returns the mnemonic string that represents character 'c', if one
911 * exists; NULL otherwise. The only ones that exist for the purposes of
912 * this routine are a few control characters */
915 case '\a': return "\\a";
916 case '\b': return "\\b";
917 case ESC_NATIVE: return "\\e";
918 case '\f': return "\\f";
919 case '\n': return "\\n";
920 case '\r': return "\\r";
921 case '\t': return "\\t";
927 /* Mark that we cannot extend a found fixed substring at this point.
928 Update the longest found anchored substring and the longest found
929 floating substrings if needed. */
932 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
933 SSize_t *minlenp, int is_inf)
935 const STRLEN l = CHR_SVLEN(data->last_found);
936 const STRLEN old_l = CHR_SVLEN(*data->longest);
937 GET_RE_DEBUG_FLAGS_DECL;
939 PERL_ARGS_ASSERT_SCAN_COMMIT;
941 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
942 SvSetMagicSV(*data->longest, data->last_found);
943 if (*data->longest == data->longest_fixed) {
944 data->offset_fixed = l ? data->last_start_min : data->pos_min;
945 if (data->flags & SF_BEFORE_EOL)
947 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
949 data->flags &= ~SF_FIX_BEFORE_EOL;
950 data->minlen_fixed=minlenp;
951 data->lookbehind_fixed=0;
953 else { /* *data->longest == data->longest_float */
954 data->offset_float_min = l ? data->last_start_min : data->pos_min;
955 data->offset_float_max = (l
956 ? data->last_start_max
957 : (data->pos_delta > SSize_t_MAX - data->pos_min
959 : data->pos_min + data->pos_delta));
961 || (STRLEN)data->offset_float_max > (STRLEN)SSize_t_MAX)
962 data->offset_float_max = SSize_t_MAX;
963 if (data->flags & SF_BEFORE_EOL)
965 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
967 data->flags &= ~SF_FL_BEFORE_EOL;
968 data->minlen_float=minlenp;
969 data->lookbehind_float=0;
972 SvCUR_set(data->last_found, 0);
974 SV * const sv = data->last_found;
975 if (SvUTF8(sv) && SvMAGICAL(sv)) {
976 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
982 data->flags &= ~SF_BEFORE_EOL;
983 DEBUG_STUDYDATA("commit: ",data,0);
986 /* An SSC is just a regnode_charclass_posix with an extra field: the inversion
987 * list that describes which code points it matches */
990 S_ssc_anything(pTHX_ regnode_ssc *ssc)
992 /* Set the SSC 'ssc' to match an empty string or any code point */
994 PERL_ARGS_ASSERT_SSC_ANYTHING;
996 assert(is_ANYOF_SYNTHETIC(ssc));
998 ssc->invlist = sv_2mortal(_new_invlist(2)); /* mortalize so won't leak */
999 _append_range_to_invlist(ssc->invlist, 0, UV_MAX);
1000 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
1004 S_ssc_is_anything(const regnode_ssc *ssc)
1006 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1007 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1008 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1009 * in any way, so there's no point in using it */
1014 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
1016 assert(is_ANYOF_SYNTHETIC(ssc));
1018 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
1022 /* See if the list consists solely of the range 0 - Infinity */
1023 invlist_iterinit(ssc->invlist);
1024 ret = invlist_iternext(ssc->invlist, &start, &end)
1028 invlist_iterfinish(ssc->invlist);
1034 /* If e.g., both \w and \W are set, matches everything */
1035 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1037 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1038 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1048 S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
1050 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1051 * string, any code point, or any posix class under locale */
1053 PERL_ARGS_ASSERT_SSC_INIT;
1055 Zero(ssc, 1, regnode_ssc);
1056 set_ANYOF_SYNTHETIC(ssc);
1057 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
1060 /* If any portion of the regex is to operate under locale rules that aren't
1061 * fully known at compile time, initialization includes it. The reason
1062 * this isn't done for all regexes is that the optimizer was written under
1063 * the assumption that locale was all-or-nothing. Given the complexity and
1064 * lack of documentation in the optimizer, and that there are inadequate
1065 * test cases for locale, many parts of it may not work properly, it is
1066 * safest to avoid locale unless necessary. */
1067 if (RExC_contains_locale) {
1068 ANYOF_POSIXL_SETALL(ssc);
1071 ANYOF_POSIXL_ZERO(ssc);
1076 S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1077 const regnode_ssc *ssc)
1079 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1080 * to the list of code points matched, and locale posix classes; hence does
1081 * not check its flags) */
1086 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1088 assert(is_ANYOF_SYNTHETIC(ssc));
1090 invlist_iterinit(ssc->invlist);
1091 ret = invlist_iternext(ssc->invlist, &start, &end)
1095 invlist_iterfinish(ssc->invlist);
1101 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
1109 S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
1110 const regnode_charclass* const node)
1112 /* Returns a mortal inversion list defining which code points are matched
1113 * by 'node', which is of type ANYOF. Handles complementing the result if
1114 * appropriate. If some code points aren't knowable at this time, the
1115 * returned list must, and will, contain every code point that is a
1118 SV* invlist = sv_2mortal(_new_invlist(0));
1119 SV* only_utf8_locale_invlist = NULL;
1121 const U32 n = ARG(node);
1122 bool new_node_has_latin1 = FALSE;
1124 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1126 /* Look at the data structure created by S_set_ANYOF_arg() */
1127 if (n != ANYOF_ONLY_HAS_BITMAP) {
1128 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1129 AV * const av = MUTABLE_AV(SvRV(rv));
1130 SV **const ary = AvARRAY(av);
1131 assert(RExC_rxi->data->what[n] == 's');
1133 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
1134 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1])));
1136 else if (ary[0] && ary[0] != &PL_sv_undef) {
1138 /* Here, no compile-time swash, and there are things that won't be
1139 * known until runtime -- we have to assume it could be anything */
1140 return _add_range_to_invlist(invlist, 0, UV_MAX);
1142 else if (ary[3] && ary[3] != &PL_sv_undef) {
1144 /* Here no compile-time swash, and no run-time only data. Use the
1145 * node's inversion list */
1146 invlist = sv_2mortal(invlist_clone(ary[3]));
1149 /* Get the code points valid only under UTF-8 locales */
1150 if ((ANYOF_FLAGS(node) & ANYOF_LOC_FOLD)
1151 && ary[2] && ary[2] != &PL_sv_undef)
1153 only_utf8_locale_invlist = ary[2];
1157 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1158 * code points, and an inversion list for the others, but if there are code
1159 * points that should match only conditionally on the target string being
1160 * UTF-8, those are placed in the inversion list, and not the bitmap.
1161 * Since there are circumstances under which they could match, they are
1162 * included in the SSC. But if the ANYOF node is to be inverted, we have
1163 * to exclude them here, so that when we invert below, the end result
1164 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1165 * have to do this here before we add the unconditionally matched code
1167 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1168 _invlist_intersection_complement_2nd(invlist,
1173 /* Add in the points from the bit map */
1174 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
1175 if (ANYOF_BITMAP_TEST(node, i)) {
1176 invlist = add_cp_to_invlist(invlist, i);
1177 new_node_has_latin1 = TRUE;
1181 /* If this can match all upper Latin1 code points, have to add them
1183 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII) {
1184 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1187 /* Similarly for these */
1188 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
1189 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
1192 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1193 _invlist_invert(invlist);
1195 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOF_LOC_FOLD) {
1197 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1198 * locale. We can skip this if there are no 0-255 at all. */
1199 _invlist_union(invlist, PL_Latin1, &invlist);
1202 /* Similarly add the UTF-8 locale possible matches. These have to be
1203 * deferred until after the non-UTF-8 locale ones are taken care of just
1204 * above, or it leads to wrong results under ANYOF_INVERT */
1205 if (only_utf8_locale_invlist) {
1206 _invlist_union_maybe_complement_2nd(invlist,
1207 only_utf8_locale_invlist,
1208 ANYOF_FLAGS(node) & ANYOF_INVERT,
1215 /* These two functions currently do the exact same thing */
1216 #define ssc_init_zero ssc_init
1218 #define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1219 #define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1221 /* 'AND' a given class with another one. Can create false positives. 'ssc'
1222 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1223 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
1226 S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1227 const regnode_charclass *and_with)
1229 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1230 * another SSC or a regular ANYOF class. Can create false positives. */
1235 PERL_ARGS_ASSERT_SSC_AND;
1237 assert(is_ANYOF_SYNTHETIC(ssc));
1239 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1240 * the code point inversion list and just the relevant flags */
1241 if (is_ANYOF_SYNTHETIC(and_with)) {
1242 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
1243 anded_flags = ANYOF_FLAGS(and_with);
1245 /* XXX This is a kludge around what appears to be deficiencies in the
1246 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1247 * there are paths through the optimizer where it doesn't get weeded
1248 * out when it should. And if we don't make some extra provision for
1249 * it like the code just below, it doesn't get added when it should.
1250 * This solution is to add it only when AND'ing, which is here, and
1251 * only when what is being AND'ed is the pristine, original node
1252 * matching anything. Thus it is like adding it to ssc_anything() but
1253 * only when the result is to be AND'ed. Probably the same solution
1254 * could be adopted for the same problem we have with /l matching,
1255 * which is solved differently in S_ssc_init(), and that would lead to
1256 * fewer false positives than that solution has. But if this solution
1257 * creates bugs, the consequences are only that a warning isn't raised
1258 * that should be; while the consequences for having /l bugs is
1259 * incorrect matches */
1260 if (ssc_is_anything((regnode_ssc *)and_with)) {
1261 anded_flags |= ANYOF_WARN_SUPER;
1265 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
1266 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1269 ANYOF_FLAGS(ssc) &= anded_flags;
1271 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1272 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1273 * 'and_with' may be inverted. When not inverted, we have the situation of
1275 * (C1 | P1) & (C2 | P2)
1276 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1277 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1278 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1279 * <= ((C1 & C2) | P1 | P2)
1280 * Alternatively, the last few steps could be:
1281 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1282 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1283 * <= (C1 | C2 | (P1 & P2))
1284 * We favor the second approach if either P1 or P2 is non-empty. This is
1285 * because these components are a barrier to doing optimizations, as what
1286 * they match cannot be known until the moment of matching as they are
1287 * dependent on the current locale, 'AND"ing them likely will reduce or
1289 * But we can do better if we know that C1,P1 are in their initial state (a
1290 * frequent occurrence), each matching everything:
1291 * (<everything>) & (C2 | P2) = C2 | P2
1292 * Similarly, if C2,P2 are in their initial state (again a frequent
1293 * occurrence), the result is a no-op
1294 * (C1 | P1) & (<everything>) = C1 | P1
1297 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1298 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1299 * <= (C1 & ~C2) | (P1 & ~P2)
1302 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
1303 && ! is_ANYOF_SYNTHETIC(and_with))
1307 ssc_intersection(ssc,
1309 FALSE /* Has already been inverted */
1312 /* If either P1 or P2 is empty, the intersection will be also; can skip
1314 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
1315 ANYOF_POSIXL_ZERO(ssc);
1317 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1319 /* Note that the Posix class component P from 'and_with' actually
1321 * P = Pa | Pb | ... | Pn
1322 * where each component is one posix class, such as in [\w\s].
1324 * ~P = ~(Pa | Pb | ... | Pn)
1325 * = ~Pa & ~Pb & ... & ~Pn
1326 * <= ~Pa | ~Pb | ... | ~Pn
1327 * The last is something we can easily calculate, but unfortunately
1328 * is likely to have many false positives. We could do better
1329 * in some (but certainly not all) instances if two classes in
1330 * P have known relationships. For example
1331 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1333 * :lower: & :print: = :lower:
1334 * And similarly for classes that must be disjoint. For example,
1335 * since \s and \w can have no elements in common based on rules in
1336 * the POSIX standard,
1337 * \w & ^\S = nothing
1338 * Unfortunately, some vendor locales do not meet the Posix
1339 * standard, in particular almost everything by Microsoft.
1340 * The loop below just changes e.g., \w into \W and vice versa */
1342 regnode_charclass_posixl temp;
1343 int add = 1; /* To calculate the index of the complement */
1345 ANYOF_POSIXL_ZERO(&temp);
1346 for (i = 0; i < ANYOF_MAX; i++) {
1348 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1349 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
1351 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
1352 ANYOF_POSIXL_SET(&temp, i + add);
1354 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1356 ANYOF_POSIXL_AND(&temp, ssc);
1358 } /* else ssc already has no posixes */
1359 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1360 in its initial state */
1361 else if (! is_ANYOF_SYNTHETIC(and_with)
1362 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
1364 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1365 * copy it over 'ssc' */
1366 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
1367 if (is_ANYOF_SYNTHETIC(and_with)) {
1368 StructCopy(and_with, ssc, regnode_ssc);
1371 ssc->invlist = anded_cp_list;
1372 ANYOF_POSIXL_ZERO(ssc);
1373 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1374 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
1378 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
1379 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
1381 /* One or the other of P1, P2 is non-empty. */
1382 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1383 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1385 ssc_union(ssc, anded_cp_list, FALSE);
1387 else { /* P1 = P2 = empty */
1388 ssc_intersection(ssc, anded_cp_list, FALSE);
1394 S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1395 const regnode_charclass *or_with)
1397 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1398 * another SSC or a regular ANYOF class. Can create false positives if
1399 * 'or_with' is to be inverted. */
1404 PERL_ARGS_ASSERT_SSC_OR;
1406 assert(is_ANYOF_SYNTHETIC(ssc));
1408 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1409 * the code point inversion list and just the relevant flags */
1410 if (is_ANYOF_SYNTHETIC(or_with)) {
1411 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
1412 ored_flags = ANYOF_FLAGS(or_with);
1415 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
1416 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
1419 ANYOF_FLAGS(ssc) |= ored_flags;
1421 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1422 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1423 * 'or_with' may be inverted. When not inverted, we have the simple
1424 * situation of computing:
1425 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1426 * If P1|P2 yields a situation with both a class and its complement are
1427 * set, like having both \w and \W, this matches all code points, and we
1428 * can delete these from the P component of the ssc going forward. XXX We
1429 * might be able to delete all the P components, but I (khw) am not certain
1430 * about this, and it is better to be safe.
1433 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1434 * <= (C1 | P1) | ~C2
1435 * <= (C1 | ~C2) | P1
1436 * (which results in actually simpler code than the non-inverted case)
1439 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
1440 && ! is_ANYOF_SYNTHETIC(or_with))
1442 /* We ignore P2, leaving P1 going forward */
1443 } /* else Not inverted */
1444 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
1445 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
1446 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1448 for (i = 0; i < ANYOF_MAX; i += 2) {
1449 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1451 ssc_match_all_cp(ssc);
1452 ANYOF_POSIXL_CLEAR(ssc, i);
1453 ANYOF_POSIXL_CLEAR(ssc, i+1);
1461 FALSE /* Already has been inverted */
1465 PERL_STATIC_INLINE void
1466 S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1468 PERL_ARGS_ASSERT_SSC_UNION;
1470 assert(is_ANYOF_SYNTHETIC(ssc));
1472 _invlist_union_maybe_complement_2nd(ssc->invlist,
1478 PERL_STATIC_INLINE void
1479 S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1481 const bool invert2nd)
1483 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1485 assert(is_ANYOF_SYNTHETIC(ssc));
1487 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1493 PERL_STATIC_INLINE void
1494 S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1496 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1498 assert(is_ANYOF_SYNTHETIC(ssc));
1500 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1503 PERL_STATIC_INLINE void
1504 S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1506 /* AND just the single code point 'cp' into the SSC 'ssc' */
1508 SV* cp_list = _new_invlist(2);
1510 PERL_ARGS_ASSERT_SSC_CP_AND;
1512 assert(is_ANYOF_SYNTHETIC(ssc));
1514 cp_list = add_cp_to_invlist(cp_list, cp);
1515 ssc_intersection(ssc, cp_list,
1516 FALSE /* Not inverted */
1518 SvREFCNT_dec_NN(cp_list);
1521 PERL_STATIC_INLINE void
1522 S_ssc_clear_locale(regnode_ssc *ssc)
1524 /* Set the SSC 'ssc' to not match any locale things */
1525 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1527 assert(is_ANYOF_SYNTHETIC(ssc));
1529 ANYOF_POSIXL_ZERO(ssc);
1530 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1533 #define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
1536 S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
1538 /* The synthetic start class is used to hopefully quickly winnow down
1539 * places where a pattern could start a match in the target string. If it
1540 * doesn't really narrow things down that much, there isn't much point to
1541 * having the overhead of using it. This function uses some very crude
1542 * heuristics to decide if to use the ssc or not.
1544 * It returns TRUE if 'ssc' rules out more than half what it considers to
1545 * be the "likely" possible matches, but of course it doesn't know what the
1546 * actual things being matched are going to be; these are only guesses
1548 * For /l matches, it assumes that the only likely matches are going to be
1549 * in the 0-255 range, uniformly distributed, so half of that is 127
1550 * For /a and /d matches, it assumes that the likely matches will be just
1551 * the ASCII range, so half of that is 63
1552 * For /u and there isn't anything matching above the Latin1 range, it
1553 * assumes that that is the only range likely to be matched, and uses
1554 * half that as the cut-off: 127. If anything matches above Latin1,
1555 * it assumes that all of Unicode could match (uniformly), except for
1556 * non-Unicode code points and things in the General Category "Other"
1557 * (unassigned, private use, surrogates, controls and formats). This
1558 * is a much large number. */
1560 const U32 max_match = (LOC)
1564 : (invlist_highest(ssc->invlist) < 256)
1566 : ((NON_OTHER_COUNT + 1) / 2) - 1;
1567 U32 count = 0; /* Running total of number of code points matched by
1569 UV start, end; /* Start and end points of current range in inversion
1572 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
1574 invlist_iterinit(ssc->invlist);
1575 while (invlist_iternext(ssc->invlist, &start, &end)) {
1577 /* /u is the only thing that we expect to match above 255; so if not /u
1578 * and even if there are matches above 255, ignore them. This catches
1579 * things like \d under /d which does match the digits above 255, but
1580 * since the pattern is /d, it is not likely to be expecting them */
1581 if (! UNI_SEMANTICS) {
1585 end = MIN(end, 255);
1587 count += end - start + 1;
1588 if (count > max_match) {
1589 invlist_iterfinish(ssc->invlist);
1599 S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
1601 /* The inversion list in the SSC is marked mortal; now we need a more
1602 * permanent copy, which is stored the same way that is done in a regular
1603 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
1606 SV* invlist = invlist_clone(ssc->invlist);
1608 PERL_ARGS_ASSERT_SSC_FINALIZE;
1610 assert(is_ANYOF_SYNTHETIC(ssc));
1612 /* The code in this file assumes that all but these flags aren't relevant
1613 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
1614 * by the time we reach here */
1615 assert(! (ANYOF_FLAGS(ssc) & ~ANYOF_COMMON_FLAGS));
1617 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
1619 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
1620 NULL, NULL, NULL, FALSE);
1622 /* Make sure is clone-safe */
1623 ssc->invlist = NULL;
1625 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1626 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
1629 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
1632 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
1633 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
1634 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
1635 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
1636 ? (TRIE_LIST_CUR( idx ) - 1) \
1642 dump_trie(trie,widecharmap,revcharmap)
1643 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
1644 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
1646 These routines dump out a trie in a somewhat readable format.
1647 The _interim_ variants are used for debugging the interim
1648 tables that are used to generate the final compressed
1649 representation which is what dump_trie expects.
1651 Part of the reason for their existence is to provide a form
1652 of documentation as to how the different representations function.
1657 Dumps the final compressed table form of the trie to Perl_debug_log.
1658 Used for debugging make_trie().
1662 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
1663 AV *revcharmap, U32 depth)
1666 SV *sv=sv_newmortal();
1667 int colwidth= widecharmap ? 6 : 4;
1669 GET_RE_DEBUG_FLAGS_DECL;
1671 PERL_ARGS_ASSERT_DUMP_TRIE;
1673 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
1674 (int)depth * 2 + 2,"",
1675 "Match","Base","Ofs" );
1677 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
1678 SV ** const tmp = av_fetch( revcharmap, state, 0);
1680 PerlIO_printf( Perl_debug_log, "%*s",
1682 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1683 PL_colors[0], PL_colors[1],
1684 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1685 PERL_PV_ESCAPE_FIRSTCHAR
1690 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
1691 (int)depth * 2 + 2,"");
1693 for( state = 0 ; state < trie->uniquecharcount ; state++ )
1694 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
1695 PerlIO_printf( Perl_debug_log, "\n");
1697 for( state = 1 ; state < trie->statecount ; state++ ) {
1698 const U32 base = trie->states[ state ].trans.base;
1700 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|",
1701 (int)depth * 2 + 2,"", (UV)state);
1703 if ( trie->states[ state ].wordnum ) {
1704 PerlIO_printf( Perl_debug_log, " W%4X",
1705 trie->states[ state ].wordnum );
1707 PerlIO_printf( Perl_debug_log, "%6s", "" );
1710 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
1715 while( ( base + ofs < trie->uniquecharcount ) ||
1716 ( base + ofs - trie->uniquecharcount < trie->lasttrans
1717 && trie->trans[ base + ofs - trie->uniquecharcount ].check
1721 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
1723 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1724 if ( ( base + ofs >= trie->uniquecharcount )
1725 && ( base + ofs - trie->uniquecharcount
1727 && trie->trans[ base + ofs
1728 - trie->uniquecharcount ].check == state )
1730 PerlIO_printf( Perl_debug_log, "%*"UVXf,
1732 (UV)trie->trans[ base + ofs
1733 - trie->uniquecharcount ].next );
1735 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
1739 PerlIO_printf( Perl_debug_log, "]");
1742 PerlIO_printf( Perl_debug_log, "\n" );
1744 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=",
1746 for (word=1; word <= trie->wordcount; word++) {
1747 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
1748 (int)word, (int)(trie->wordinfo[word].prev),
1749 (int)(trie->wordinfo[word].len));
1751 PerlIO_printf(Perl_debug_log, "\n" );
1754 Dumps a fully constructed but uncompressed trie in list form.
1755 List tries normally only are used for construction when the number of
1756 possible chars (trie->uniquecharcount) is very high.
1757 Used for debugging make_trie().
1760 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
1761 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1765 SV *sv=sv_newmortal();
1766 int colwidth= widecharmap ? 6 : 4;
1767 GET_RE_DEBUG_FLAGS_DECL;
1769 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1771 /* print out the table precompression. */
1772 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1773 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1774 "------:-----+-----------------\n" );
1776 for( state=1 ; state < next_alloc ; state ++ ) {
1779 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1780 (int)depth * 2 + 2,"", (UV)state );
1781 if ( ! trie->states[ state ].wordnum ) {
1782 PerlIO_printf( Perl_debug_log, "%5s| ","");
1784 PerlIO_printf( Perl_debug_log, "W%4x| ",
1785 trie->states[ state ].wordnum
1788 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1789 SV ** const tmp = av_fetch( revcharmap,
1790 TRIE_LIST_ITEM(state,charid).forid, 0);
1792 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1794 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
1796 PL_colors[0], PL_colors[1],
1797 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
1798 | PERL_PV_ESCAPE_FIRSTCHAR
1800 TRIE_LIST_ITEM(state,charid).forid,
1801 (UV)TRIE_LIST_ITEM(state,charid).newstate
1804 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1805 (int)((depth * 2) + 14), "");
1808 PerlIO_printf( Perl_debug_log, "\n");
1813 Dumps a fully constructed but uncompressed trie in table form.
1814 This is the normal DFA style state transition table, with a few
1815 twists to facilitate compression later.
1816 Used for debugging make_trie().
1819 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1820 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1825 SV *sv=sv_newmortal();
1826 int colwidth= widecharmap ? 6 : 4;
1827 GET_RE_DEBUG_FLAGS_DECL;
1829 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1832 print out the table precompression so that we can do a visual check
1833 that they are identical.
1836 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1838 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1839 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1841 PerlIO_printf( Perl_debug_log, "%*s",
1843 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1844 PL_colors[0], PL_colors[1],
1845 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1846 PERL_PV_ESCAPE_FIRSTCHAR
1852 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1854 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1855 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1858 PerlIO_printf( Perl_debug_log, "\n" );
1860 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1862 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1863 (int)depth * 2 + 2,"",
1864 (UV)TRIE_NODENUM( state ) );
1866 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1867 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1869 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1871 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1873 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1874 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n",
1875 (UV)trie->trans[ state ].check );
1877 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n",
1878 (UV)trie->trans[ state ].check,
1879 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1887 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1888 startbranch: the first branch in the whole branch sequence
1889 first : start branch of sequence of branch-exact nodes.
1890 May be the same as startbranch
1891 last : Thing following the last branch.
1892 May be the same as tail.
1893 tail : item following the branch sequence
1894 count : words in the sequence
1895 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
1896 depth : indent depth
1898 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1900 A trie is an N'ary tree where the branches are determined by digital
1901 decomposition of the key. IE, at the root node you look up the 1st character and
1902 follow that branch repeat until you find the end of the branches. Nodes can be
1903 marked as "accepting" meaning they represent a complete word. Eg:
1907 would convert into the following structure. Numbers represent states, letters
1908 following numbers represent valid transitions on the letter from that state, if
1909 the number is in square brackets it represents an accepting state, otherwise it
1910 will be in parenthesis.
1912 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1916 (1) +-i->(6)-+-s->[7]
1918 +-s->(3)-+-h->(4)-+-e->[5]
1920 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1922 This shows that when matching against the string 'hers' we will begin at state 1
1923 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1924 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1925 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1926 single traverse. We store a mapping from accepting to state to which word was
1927 matched, and then when we have multiple possibilities we try to complete the
1928 rest of the regex in the order in which they occurred in the alternation.
1930 The only prior NFA like behaviour that would be changed by the TRIE support is
1931 the silent ignoring of duplicate alternations which are of the form:
1933 / (DUPE|DUPE) X? (?{ ... }) Y /x
1935 Thus EVAL blocks following a trie may be called a different number of times with
1936 and without the optimisation. With the optimisations dupes will be silently
1937 ignored. This inconsistent behaviour of EVAL type nodes is well established as
1938 the following demonstrates:
1940 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1942 which prints out 'word' three times, but
1944 'words'=~/(word|word|word)(?{ print $1 })S/
1946 which doesnt print it out at all. This is due to other optimisations kicking in.
1948 Example of what happens on a structural level:
1950 The regexp /(ac|ad|ab)+/ will produce the following debug output:
1952 1: CURLYM[1] {1,32767}(18)
1963 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1964 and should turn into:
1966 1: CURLYM[1] {1,32767}(18)
1968 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1976 Cases where tail != last would be like /(?foo|bar)baz/:
1986 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1987 and would end up looking like:
1990 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1997 d = uvchr_to_utf8_flags(d, uv, 0);
1999 is the recommended Unicode-aware way of saying
2004 #define TRIE_STORE_REVCHAR(val) \
2007 SV *zlopp = newSV(7); /* XXX: optimize me */ \
2008 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
2009 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
2010 SvCUR_set(zlopp, kapow - flrbbbbb); \
2013 av_push(revcharmap, zlopp); \
2015 char ooooff = (char)val; \
2016 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2020 /* This gets the next character from the input, folding it if not already
2022 #define TRIE_READ_CHAR STMT_START { \
2025 /* if it is UTF then it is either already folded, or does not need \
2027 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
2029 else if (folder == PL_fold_latin1) { \
2030 /* This folder implies Unicode rules, which in the range expressible \
2031 * by not UTF is the lower case, with the two exceptions, one of \
2032 * which should have been taken care of before calling this */ \
2033 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2034 uvc = toLOWER_L1(*uc); \
2035 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2038 /* raw data, will be folded later if needed */ \
2046 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2047 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
2048 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
2049 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
2051 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2052 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2053 TRIE_LIST_CUR( state )++; \
2056 #define TRIE_LIST_NEW(state) STMT_START { \
2057 Newxz( trie->states[ state ].trans.list, \
2058 4, reg_trie_trans_le ); \
2059 TRIE_LIST_CUR( state ) = 1; \
2060 TRIE_LIST_LEN( state ) = 4; \
2063 #define TRIE_HANDLE_WORD(state) STMT_START { \
2064 U16 dupe= trie->states[ state ].wordnum; \
2065 regnode * const noper_next = regnext( noper ); \
2068 /* store the word for dumping */ \
2070 if (OP(noper) != NOTHING) \
2071 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
2073 tmp = newSVpvn_utf8( "", 0, UTF ); \
2074 av_push( trie_words, tmp ); \
2078 trie->wordinfo[curword].prev = 0; \
2079 trie->wordinfo[curword].len = wordlen; \
2080 trie->wordinfo[curword].accept = state; \
2082 if ( noper_next < tail ) { \
2084 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2086 trie->jump[curword] = (U16)(noper_next - convert); \
2088 jumper = noper_next; \
2090 nextbranch= regnext(cur); \
2094 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2095 /* chain, so that when the bits of chain are later */\
2096 /* linked together, the dups appear in the chain */\
2097 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2098 trie->wordinfo[dupe].prev = curword; \
2100 /* we haven't inserted this word yet. */ \
2101 trie->states[ state ].wordnum = curword; \
2106 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2107 ( ( base + charid >= ucharcount \
2108 && base + charid < ubound \
2109 && state == trie->trans[ base - ucharcount + charid ].check \
2110 && trie->trans[ base - ucharcount + charid ].next ) \
2111 ? trie->trans[ base - ucharcount + charid ].next \
2112 : ( state==1 ? special : 0 ) \
2116 #define MADE_JUMP_TRIE 2
2117 #define MADE_EXACT_TRIE 4
2120 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2121 regnode *first, regnode *last, regnode *tail,
2122 U32 word_count, U32 flags, U32 depth)
2124 /* first pass, loop through and scan words */
2125 reg_trie_data *trie;
2126 HV *widecharmap = NULL;
2127 AV *revcharmap = newAV();
2133 regnode *jumper = NULL;
2134 regnode *nextbranch = NULL;
2135 regnode *convert = NULL;
2136 U32 *prev_states; /* temp array mapping each state to previous one */
2137 /* we just use folder as a flag in utf8 */
2138 const U8 * folder = NULL;
2141 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuuu"));
2142 AV *trie_words = NULL;
2143 /* along with revcharmap, this only used during construction but both are
2144 * useful during debugging so we store them in the struct when debugging.
2147 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
2148 STRLEN trie_charcount=0;
2150 SV *re_trie_maxbuff;
2151 GET_RE_DEBUG_FLAGS_DECL;
2153 PERL_ARGS_ASSERT_MAKE_TRIE;
2155 PERL_UNUSED_ARG(depth);
2159 case EXACT: case EXACTL: break;
2163 case EXACTFLU8: folder = PL_fold_latin1; break;
2164 case EXACTF: folder = PL_fold; break;
2165 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
2168 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
2170 trie->startstate = 1;
2171 trie->wordcount = word_count;
2172 RExC_rxi->data->data[ data_slot ] = (void*)trie;
2173 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
2174 if (flags == EXACT || flags == EXACTL)
2175 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2176 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2177 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2180 trie_words = newAV();
2183 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2184 assert(re_trie_maxbuff);
2185 if (!SvIOK(re_trie_maxbuff)) {
2186 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2188 DEBUG_TRIE_COMPILE_r({
2189 PerlIO_printf( Perl_debug_log,
2190 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2191 (int)depth * 2 + 2, "",
2192 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
2193 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
2196 /* Find the node we are going to overwrite */
2197 if ( first == startbranch && OP( last ) != BRANCH ) {
2198 /* whole branch chain */
2201 /* branch sub-chain */
2202 convert = NEXTOPER( first );
2205 /* -- First loop and Setup --
2207 We first traverse the branches and scan each word to determine if it
2208 contains widechars, and how many unique chars there are, this is
2209 important as we have to build a table with at least as many columns as we
2212 We use an array of integers to represent the character codes 0..255
2213 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2214 the native representation of the character value as the key and IV's for
2217 *TODO* If we keep track of how many times each character is used we can
2218 remap the columns so that the table compression later on is more
2219 efficient in terms of memory by ensuring the most common value is in the
2220 middle and the least common are on the outside. IMO this would be better
2221 than a most to least common mapping as theres a decent chance the most
2222 common letter will share a node with the least common, meaning the node
2223 will not be compressible. With a middle is most common approach the worst
2224 case is when we have the least common nodes twice.
2228 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2229 regnode *noper = NEXTOPER( cur );
2230 const U8 *uc = (U8*)STRING( noper );
2231 const U8 *e = uc + STR_LEN( noper );
2233 U32 wordlen = 0; /* required init */
2234 STRLEN minchars = 0;
2235 STRLEN maxchars = 0;
2236 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2239 if (OP(noper) == NOTHING) {
2240 regnode *noper_next= regnext(noper);
2241 if (noper_next != tail && OP(noper_next) == flags) {
2243 uc= (U8*)STRING(noper);
2244 e= uc + STR_LEN(noper);
2245 trie->minlen= STR_LEN(noper);
2252 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
2253 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2254 regardless of encoding */
2255 if (OP( noper ) == EXACTFU_SS) {
2256 /* false positives are ok, so just set this */
2257 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
2260 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2262 TRIE_CHARCOUNT(trie)++;
2265 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2266 * is in effect. Under /i, this character can match itself, or
2267 * anything that folds to it. If not under /i, it can match just
2268 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2269 * all fold to k, and all are single characters. But some folds
2270 * expand to more than one character, so for example LATIN SMALL
2271 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2272 * the string beginning at 'uc' is 'ffi', it could be matched by
2273 * three characters, or just by the one ligature character. (It
2274 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2275 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2276 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2277 * match.) The trie needs to know the minimum and maximum number
2278 * of characters that could match so that it can use size alone to
2279 * quickly reject many match attempts. The max is simple: it is
2280 * the number of folded characters in this branch (since a fold is
2281 * never shorter than what folds to it. */
2285 /* And the min is equal to the max if not under /i (indicated by
2286 * 'folder' being NULL), or there are no multi-character folds. If
2287 * there is a multi-character fold, the min is incremented just
2288 * once, for the character that folds to the sequence. Each
2289 * character in the sequence needs to be added to the list below of
2290 * characters in the trie, but we count only the first towards the
2291 * min number of characters needed. This is done through the
2292 * variable 'foldlen', which is returned by the macros that look
2293 * for these sequences as the number of bytes the sequence
2294 * occupies. Each time through the loop, we decrement 'foldlen' by
2295 * how many bytes the current char occupies. Only when it reaches
2296 * 0 do we increment 'minchars' or look for another multi-character
2298 if (folder == NULL) {
2301 else if (foldlen > 0) {
2302 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
2307 /* See if *uc is the beginning of a multi-character fold. If
2308 * so, we decrement the length remaining to look at, to account
2309 * for the current character this iteration. (We can use 'uc'
2310 * instead of the fold returned by TRIE_READ_CHAR because for
2311 * non-UTF, the latin1_safe macro is smart enough to account
2312 * for all the unfolded characters, and because for UTF, the
2313 * string will already have been folded earlier in the
2314 * compilation process */
2316 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2317 foldlen -= UTF8SKIP(uc);
2320 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2325 /* The current character (and any potential folds) should be added
2326 * to the possible matching characters for this position in this
2330 U8 folded= folder[ (U8) uvc ];
2331 if ( !trie->charmap[ folded ] ) {
2332 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2333 TRIE_STORE_REVCHAR( folded );
2336 if ( !trie->charmap[ uvc ] ) {
2337 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
2338 TRIE_STORE_REVCHAR( uvc );
2341 /* store the codepoint in the bitmap, and its folded
2343 TRIE_BITMAP_SET(trie, uvc);
2345 /* store the folded codepoint */
2346 if ( folder ) TRIE_BITMAP_SET(trie, folder[(U8) uvc ]);
2349 /* store first byte of utf8 representation of
2350 variant codepoints */
2351 if (! UVCHR_IS_INVARIANT(uvc)) {
2352 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
2355 set_bit = 0; /* We've done our bit :-) */
2359 /* XXX We could come up with the list of code points that fold
2360 * to this using PL_utf8_foldclosures, except not for
2361 * multi-char folds, as there may be multiple combinations
2362 * there that could work, which needs to wait until runtime to
2363 * resolve (The comment about LIGATURE FFI above is such an
2368 widecharmap = newHV();
2370 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
2373 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
2375 if ( !SvTRUE( *svpp ) ) {
2376 sv_setiv( *svpp, ++trie->uniquecharcount );
2377 TRIE_STORE_REVCHAR(uvc);
2380 } /* end loop through characters in this branch of the trie */
2382 /* We take the min and max for this branch and combine to find the min
2383 * and max for all branches processed so far */
2384 if( cur == first ) {
2385 trie->minlen = minchars;
2386 trie->maxlen = maxchars;
2387 } else if (minchars < trie->minlen) {
2388 trie->minlen = minchars;
2389 } else if (maxchars > trie->maxlen) {
2390 trie->maxlen = maxchars;
2392 } /* end first pass */
2393 DEBUG_TRIE_COMPILE_r(
2394 PerlIO_printf( Perl_debug_log,
2395 "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2396 (int)depth * 2 + 2,"",
2397 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
2398 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2399 (int)trie->minlen, (int)trie->maxlen )
2403 We now know what we are dealing with in terms of unique chars and
2404 string sizes so we can calculate how much memory a naive
2405 representation using a flat table will take. If it's over a reasonable
2406 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
2407 conservative but potentially much slower representation using an array
2410 At the end we convert both representations into the same compressed
2411 form that will be used in regexec.c for matching with. The latter
2412 is a form that cannot be used to construct with but has memory
2413 properties similar to the list form and access properties similar
2414 to the table form making it both suitable for fast searches and
2415 small enough that its feasable to store for the duration of a program.
2417 See the comment in the code where the compressed table is produced
2418 inplace from the flat tabe representation for an explanation of how
2419 the compression works.
2424 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2427 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2428 > SvIV(re_trie_maxbuff) )
2431 Second Pass -- Array Of Lists Representation
2433 Each state will be represented by a list of charid:state records
2434 (reg_trie_trans_le) the first such element holds the CUR and LEN
2435 points of the allocated array. (See defines above).
2437 We build the initial structure using the lists, and then convert
2438 it into the compressed table form which allows faster lookups
2439 (but cant be modified once converted).
2442 STRLEN transcount = 1;
2444 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2445 "%*sCompiling trie using list compiler\n",
2446 (int)depth * 2 + 2, ""));
2448 trie->states = (reg_trie_state *)
2449 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2450 sizeof(reg_trie_state) );
2454 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2456 regnode *noper = NEXTOPER( cur );
2457 U8 *uc = (U8*)STRING( noper );
2458 const U8 *e = uc + STR_LEN( noper );
2459 U32 state = 1; /* required init */
2460 U16 charid = 0; /* sanity init */
2461 U32 wordlen = 0; /* required init */
2463 if (OP(noper) == NOTHING) {
2464 regnode *noper_next= regnext(noper);
2465 if (noper_next != tail && OP(noper_next) == flags) {
2467 uc= (U8*)STRING(noper);
2468 e= uc + STR_LEN(noper);
2472 if (OP(noper) != NOTHING) {
2473 for ( ; uc < e ; uc += len ) {
2478 charid = trie->charmap[ uvc ];
2480 SV** const svpp = hv_fetch( widecharmap,
2487 charid=(U16)SvIV( *svpp );
2490 /* charid is now 0 if we dont know the char read, or
2491 * nonzero if we do */
2498 if ( !trie->states[ state ].trans.list ) {
2499 TRIE_LIST_NEW( state );
2502 check <= TRIE_LIST_USED( state );
2505 if ( TRIE_LIST_ITEM( state, check ).forid
2508 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2513 newstate = next_alloc++;
2514 prev_states[newstate] = state;
2515 TRIE_LIST_PUSH( state, charid, newstate );
2520 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2524 TRIE_HANDLE_WORD(state);
2526 } /* end second pass */
2528 /* next alloc is the NEXT state to be allocated */
2529 trie->statecount = next_alloc;
2530 trie->states = (reg_trie_state *)
2531 PerlMemShared_realloc( trie->states,
2533 * sizeof(reg_trie_state) );
2535 /* and now dump it out before we compress it */
2536 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2537 revcharmap, next_alloc,
2541 trie->trans = (reg_trie_trans *)
2542 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
2549 for( state=1 ; state < next_alloc ; state ++ ) {
2553 DEBUG_TRIE_COMPILE_MORE_r(
2554 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
2558 if (trie->states[state].trans.list) {
2559 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
2563 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2564 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
2565 if ( forid < minid ) {
2567 } else if ( forid > maxid ) {
2571 if ( transcount < tp + maxid - minid + 1) {
2573 trie->trans = (reg_trie_trans *)
2574 PerlMemShared_realloc( trie->trans,
2576 * sizeof(reg_trie_trans) );
2577 Zero( trie->trans + (transcount / 2),
2581 base = trie->uniquecharcount + tp - minid;
2582 if ( maxid == minid ) {
2584 for ( ; zp < tp ; zp++ ) {
2585 if ( ! trie->trans[ zp ].next ) {
2586 base = trie->uniquecharcount + zp - minid;
2587 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
2589 trie->trans[ zp ].check = state;
2595 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
2597 trie->trans[ tp ].check = state;
2602 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2603 const U32 tid = base
2604 - trie->uniquecharcount
2605 + TRIE_LIST_ITEM( state, idx ).forid;
2606 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
2608 trie->trans[ tid ].check = state;
2610 tp += ( maxid - minid + 1 );
2612 Safefree(trie->states[ state ].trans.list);
2615 DEBUG_TRIE_COMPILE_MORE_r(
2616 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
2619 trie->states[ state ].trans.base=base;
2621 trie->lasttrans = tp + 1;
2625 Second Pass -- Flat Table Representation.
2627 we dont use the 0 slot of either trans[] or states[] so we add 1 to
2628 each. We know that we will need Charcount+1 trans at most to store
2629 the data (one row per char at worst case) So we preallocate both
2630 structures assuming worst case.
2632 We then construct the trie using only the .next slots of the entry
2635 We use the .check field of the first entry of the node temporarily
2636 to make compression both faster and easier by keeping track of how
2637 many non zero fields are in the node.
2639 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
2642 There are two terms at use here: state as a TRIE_NODEIDX() which is
2643 a number representing the first entry of the node, and state as a
2644 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
2645 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
2646 if there are 2 entrys per node. eg:
2654 The table is internally in the right hand, idx form. However as we
2655 also have to deal with the states array which is indexed by nodenum
2656 we have to use TRIE_NODENUM() to convert.
2659 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2660 "%*sCompiling trie using table compiler\n",
2661 (int)depth * 2 + 2, ""));
2663 trie->trans = (reg_trie_trans *)
2664 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
2665 * trie->uniquecharcount + 1,
2666 sizeof(reg_trie_trans) );
2667 trie->states = (reg_trie_state *)
2668 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2669 sizeof(reg_trie_state) );
2670 next_alloc = trie->uniquecharcount + 1;
2673 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2675 regnode *noper = NEXTOPER( cur );
2676 const U8 *uc = (U8*)STRING( noper );
2677 const U8 *e = uc + STR_LEN( noper );
2679 U32 state = 1; /* required init */
2681 U16 charid = 0; /* sanity init */
2682 U32 accept_state = 0; /* sanity init */
2684 U32 wordlen = 0; /* required init */
2686 if (OP(noper) == NOTHING) {
2687 regnode *noper_next= regnext(noper);
2688 if (noper_next != tail && OP(noper_next) == flags) {
2690 uc= (U8*)STRING(noper);
2691 e= uc + STR_LEN(noper);
2695 if ( OP(noper) != NOTHING ) {
2696 for ( ; uc < e ; uc += len ) {
2701 charid = trie->charmap[ uvc ];
2703 SV* const * const svpp = hv_fetch( widecharmap,
2707 charid = svpp ? (U16)SvIV(*svpp) : 0;
2711 if ( !trie->trans[ state + charid ].next ) {
2712 trie->trans[ state + charid ].next = next_alloc;
2713 trie->trans[ state ].check++;
2714 prev_states[TRIE_NODENUM(next_alloc)]
2715 = TRIE_NODENUM(state);
2716 next_alloc += trie->uniquecharcount;
2718 state = trie->trans[ state + charid ].next;
2720 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2722 /* charid is now 0 if we dont know the char read, or
2723 * nonzero if we do */
2726 accept_state = TRIE_NODENUM( state );
2727 TRIE_HANDLE_WORD(accept_state);
2729 } /* end second pass */
2731 /* and now dump it out before we compress it */
2732 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
2734 next_alloc, depth+1));
2738 * Inplace compress the table.*
2740 For sparse data sets the table constructed by the trie algorithm will
2741 be mostly 0/FAIL transitions or to put it another way mostly empty.
2742 (Note that leaf nodes will not contain any transitions.)
2744 This algorithm compresses the tables by eliminating most such
2745 transitions, at the cost of a modest bit of extra work during lookup:
2747 - Each states[] entry contains a .base field which indicates the
2748 index in the state[] array wheres its transition data is stored.
2750 - If .base is 0 there are no valid transitions from that node.
2752 - If .base is nonzero then charid is added to it to find an entry in
2755 -If trans[states[state].base+charid].check!=state then the
2756 transition is taken to be a 0/Fail transition. Thus if there are fail
2757 transitions at the front of the node then the .base offset will point
2758 somewhere inside the previous nodes data (or maybe even into a node
2759 even earlier), but the .check field determines if the transition is
2763 The following process inplace converts the table to the compressed
2764 table: We first do not compress the root node 1,and mark all its
2765 .check pointers as 1 and set its .base pointer as 1 as well. This
2766 allows us to do a DFA construction from the compressed table later,
2767 and ensures that any .base pointers we calculate later are greater
2770 - We set 'pos' to indicate the first entry of the second node.
2772 - We then iterate over the columns of the node, finding the first and
2773 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
2774 and set the .check pointers accordingly, and advance pos
2775 appropriately and repreat for the next node. Note that when we copy
2776 the next pointers we have to convert them from the original
2777 NODEIDX form to NODENUM form as the former is not valid post
2780 - If a node has no transitions used we mark its base as 0 and do not
2781 advance the pos pointer.
2783 - If a node only has one transition we use a second pointer into the
2784 structure to fill in allocated fail transitions from other states.
2785 This pointer is independent of the main pointer and scans forward
2786 looking for null transitions that are allocated to a state. When it
2787 finds one it writes the single transition into the "hole". If the
2788 pointer doesnt find one the single transition is appended as normal.
2790 - Once compressed we can Renew/realloc the structures to release the
2793 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
2794 specifically Fig 3.47 and the associated pseudocode.
2798 const U32 laststate = TRIE_NODENUM( next_alloc );
2801 trie->statecount = laststate;
2803 for ( state = 1 ; state < laststate ; state++ ) {
2805 const U32 stateidx = TRIE_NODEIDX( state );
2806 const U32 o_used = trie->trans[ stateidx ].check;
2807 U32 used = trie->trans[ stateidx ].check;
2808 trie->trans[ stateidx ].check = 0;
2811 used && charid < trie->uniquecharcount;
2814 if ( flag || trie->trans[ stateidx + charid ].next ) {
2815 if ( trie->trans[ stateidx + charid ].next ) {
2817 for ( ; zp < pos ; zp++ ) {
2818 if ( ! trie->trans[ zp ].next ) {
2822 trie->states[ state ].trans.base
2824 + trie->uniquecharcount
2826 trie->trans[ zp ].next
2827 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
2829 trie->trans[ zp ].check = state;
2830 if ( ++zp > pos ) pos = zp;
2837 trie->states[ state ].trans.base
2838 = pos + trie->uniquecharcount - charid ;
2840 trie->trans[ pos ].next
2841 = SAFE_TRIE_NODENUM(
2842 trie->trans[ stateidx + charid ].next );
2843 trie->trans[ pos ].check = state;
2848 trie->lasttrans = pos + 1;
2849 trie->states = (reg_trie_state *)
2850 PerlMemShared_realloc( trie->states, laststate
2851 * sizeof(reg_trie_state) );
2852 DEBUG_TRIE_COMPILE_MORE_r(
2853 PerlIO_printf( Perl_debug_log,
2854 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
2855 (int)depth * 2 + 2,"",
2856 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
2860 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
2863 } /* end table compress */
2865 DEBUG_TRIE_COMPILE_MORE_r(
2866 PerlIO_printf(Perl_debug_log,
2867 "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
2868 (int)depth * 2 + 2, "",
2869 (UV)trie->statecount,
2870 (UV)trie->lasttrans)
2872 /* resize the trans array to remove unused space */
2873 trie->trans = (reg_trie_trans *)
2874 PerlMemShared_realloc( trie->trans, trie->lasttrans
2875 * sizeof(reg_trie_trans) );
2877 { /* Modify the program and insert the new TRIE node */
2878 U8 nodetype =(U8)(flags & 0xFF);
2882 regnode *optimize = NULL;
2883 #ifdef RE_TRACK_PATTERN_OFFSETS
2886 U32 mjd_nodelen = 0;
2887 #endif /* RE_TRACK_PATTERN_OFFSETS */
2888 #endif /* DEBUGGING */
2890 This means we convert either the first branch or the first Exact,
2891 depending on whether the thing following (in 'last') is a branch
2892 or not and whther first is the startbranch (ie is it a sub part of
2893 the alternation or is it the whole thing.)
2894 Assuming its a sub part we convert the EXACT otherwise we convert
2895 the whole branch sequence, including the first.
2897 /* Find the node we are going to overwrite */
2898 if ( first != startbranch || OP( last ) == BRANCH ) {
2899 /* branch sub-chain */
2900 NEXT_OFF( first ) = (U16)(last - first);
2901 #ifdef RE_TRACK_PATTERN_OFFSETS
2903 mjd_offset= Node_Offset((convert));
2904 mjd_nodelen= Node_Length((convert));
2907 /* whole branch chain */
2909 #ifdef RE_TRACK_PATTERN_OFFSETS
2912 const regnode *nop = NEXTOPER( convert );
2913 mjd_offset= Node_Offset((nop));
2914 mjd_nodelen= Node_Length((nop));
2918 PerlIO_printf(Perl_debug_log,
2919 "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2920 (int)depth * 2 + 2, "",
2921 (UV)mjd_offset, (UV)mjd_nodelen)
2924 /* But first we check to see if there is a common prefix we can
2925 split out as an EXACT and put in front of the TRIE node. */
2926 trie->startstate= 1;
2927 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2929 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2933 const U32 base = trie->states[ state ].trans.base;
2935 if ( trie->states[state].wordnum )
2938 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2939 if ( ( base + ofs >= trie->uniquecharcount ) &&
2940 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2941 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2943 if ( ++count > 1 ) {
2944 SV **tmp = av_fetch( revcharmap, ofs, 0);
2945 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2946 if ( state == 1 ) break;
2948 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2950 PerlIO_printf(Perl_debug_log,
2951 "%*sNew Start State=%"UVuf" Class: [",
2952 (int)depth * 2 + 2, "",
2955 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2956 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2958 TRIE_BITMAP_SET(trie,*ch);
2960 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2962 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2966 TRIE_BITMAP_SET(trie,*ch);
2968 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2969 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2975 SV **tmp = av_fetch( revcharmap, idx, 0);
2977 char *ch = SvPV( *tmp, len );
2979 SV *sv=sv_newmortal();
2980 PerlIO_printf( Perl_debug_log,
2981 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2982 (int)depth * 2 + 2, "",
2984 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2985 PL_colors[0], PL_colors[1],
2986 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2987 PERL_PV_ESCAPE_FIRSTCHAR
2992 OP( convert ) = nodetype;
2993 str=STRING(convert);
2996 STR_LEN(convert) += len;
3002 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
3007 trie->prefixlen = (state-1);
3009 regnode *n = convert+NODE_SZ_STR(convert);
3010 NEXT_OFF(convert) = NODE_SZ_STR(convert);
3011 trie->startstate = state;
3012 trie->minlen -= (state - 1);
3013 trie->maxlen -= (state - 1);
3015 /* At least the UNICOS C compiler choked on this
3016 * being argument to DEBUG_r(), so let's just have
3019 #ifdef PERL_EXT_RE_BUILD
3025 regnode *fix = convert;
3026 U32 word = trie->wordcount;
3028 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
3029 while( ++fix < n ) {
3030 Set_Node_Offset_Length(fix, 0, 0);
3033 SV ** const tmp = av_fetch( trie_words, word, 0 );
3035 if ( STR_LEN(convert) <= SvCUR(*tmp) )
3036 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
3038 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
3046 NEXT_OFF(convert) = (U16)(tail - convert);
3047 DEBUG_r(optimize= n);
3053 if ( trie->maxlen ) {
3054 NEXT_OFF( convert ) = (U16)(tail - convert);
3055 ARG_SET( convert, data_slot );
3056 /* Store the offset to the first unabsorbed branch in
3057 jump[0], which is otherwise unused by the jump logic.
3058 We use this when dumping a trie and during optimisation. */
3060 trie->jump[0] = (U16)(nextbranch - convert);
3062 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
3063 * and there is a bitmap
3064 * and the first "jump target" node we found leaves enough room
3065 * then convert the TRIE node into a TRIEC node, with the bitmap
3066 * embedded inline in the opcode - this is hypothetically faster.
3068 if ( !trie->states[trie->startstate].wordnum
3070 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
3072 OP( convert ) = TRIEC;
3073 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
3074 PerlMemShared_free(trie->bitmap);
3077 OP( convert ) = TRIE;
3079 /* store the type in the flags */
3080 convert->flags = nodetype;
3084 + regarglen[ OP( convert ) ];
3086 /* XXX We really should free up the resource in trie now,
3087 as we won't use them - (which resources?) dmq */
3089 /* needed for dumping*/
3090 DEBUG_r(if (optimize) {
3091 regnode *opt = convert;
3093 while ( ++opt < optimize) {
3094 Set_Node_Offset_Length(opt,0,0);
3097 Try to clean up some of the debris left after the
3100 while( optimize < jumper ) {
3101 mjd_nodelen += Node_Length((optimize));
3102 OP( optimize ) = OPTIMIZED;
3103 Set_Node_Offset_Length(optimize,0,0);
3106 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
3108 } /* end node insert */
3110 /* Finish populating the prev field of the wordinfo array. Walk back
3111 * from each accept state until we find another accept state, and if
3112 * so, point the first word's .prev field at the second word. If the
3113 * second already has a .prev field set, stop now. This will be the
3114 * case either if we've already processed that word's accept state,
3115 * or that state had multiple words, and the overspill words were
3116 * already linked up earlier.
3123 for (word=1; word <= trie->wordcount; word++) {
3125 if (trie->wordinfo[word].prev)
3127 state = trie->wordinfo[word].accept;
3129 state = prev_states[state];
3132 prev = trie->states[state].wordnum;
3136 trie->wordinfo[word].prev = prev;
3138 Safefree(prev_states);
3142 /* and now dump out the compressed format */
3143 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
3145 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
3147 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
3148 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
3150 SvREFCNT_dec_NN(revcharmap);
3154 : trie->startstate>1
3160 S_construct_ahocorasick_from_trie(pTHX_ RExC_state_t *pRExC_state, regnode *source, U32 depth)
3162 /* The Trie is constructed and compressed now so we can build a fail array if
3165 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and
3167 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi,
3171 We find the fail state for each state in the trie, this state is the longest
3172 proper suffix of the current state's 'word' that is also a proper prefix of
3173 another word in our trie. State 1 represents the word '' and is thus the
3174 default fail state. This allows the DFA not to have to restart after its
3175 tried and failed a word at a given point, it simply continues as though it
3176 had been matching the other word in the first place.
3178 'abcdgu'=~/abcdefg|cdgu/
3179 When we get to 'd' we are still matching the first word, we would encounter
3180 'g' which would fail, which would bring us to the state representing 'd' in
3181 the second word where we would try 'g' and succeed, proceeding to match
3184 /* add a fail transition */
3185 const U32 trie_offset = ARG(source);
3186 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
3188 const U32 ucharcount = trie->uniquecharcount;
3189 const U32 numstates = trie->statecount;
3190 const U32 ubound = trie->lasttrans + ucharcount;
3194 U32 base = trie->states[ 1 ].trans.base;
3197 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T"));
3199 GET_RE_DEBUG_FLAGS_DECL;
3201 PERL_ARGS_ASSERT_CONSTRUCT_AHOCORASICK_FROM_TRIE;
3202 PERL_UNUSED_CONTEXT;
3204 PERL_UNUSED_ARG(depth);
3207 if ( OP(source) == TRIE ) {
3208 struct regnode_1 *op = (struct regnode_1 *)
3209 PerlMemShared_calloc(1, sizeof(struct regnode_1));
3210 StructCopy(source,op,struct regnode_1);
3211 stclass = (regnode *)op;
3213 struct regnode_charclass *op = (struct regnode_charclass *)
3214 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
3215 StructCopy(source,op,struct regnode_charclass);
3216 stclass = (regnode *)op;
3218 OP(stclass)+=2; /* convert the TRIE type to its AHO-CORASICK equivalent */
3220 ARG_SET( stclass, data_slot );
3221 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
3222 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3223 aho->trie=trie_offset;
3224 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
3225 Copy( trie->states, aho->states, numstates, reg_trie_state );
3226 Newxz( q, numstates, U32);
3227 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
3230 /* initialize fail[0..1] to be 1 so that we always have
3231 a valid final fail state */
3232 fail[ 0 ] = fail[ 1 ] = 1;
3234 for ( charid = 0; charid < ucharcount ; charid++ ) {
3235 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
3237 q[ q_write ] = newstate;
3238 /* set to point at the root */
3239 fail[ q[ q_write++ ] ]=1;
3242 while ( q_read < q_write) {
3243 const U32 cur = q[ q_read++ % numstates ];
3244 base = trie->states[ cur ].trans.base;
3246 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
3247 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
3249 U32 fail_state = cur;
3252 fail_state = fail[ fail_state ];
3253 fail_base = aho->states[ fail_state ].trans.base;
3254 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
3256 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
3257 fail[ ch_state ] = fail_state;
3258 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
3260 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
3262 q[ q_write++ % numstates] = ch_state;
3266 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
3267 when we fail in state 1, this allows us to use the
3268 charclass scan to find a valid start char. This is based on the principle
3269 that theres a good chance the string being searched contains lots of stuff
3270 that cant be a start char.
3272 fail[ 0 ] = fail[ 1 ] = 0;
3273 DEBUG_TRIE_COMPILE_r({
3274 PerlIO_printf(Perl_debug_log,
3275 "%*sStclass Failtable (%"UVuf" states): 0",
3276 (int)(depth * 2), "", (UV)numstates
3278 for( q_read=1; q_read<numstates; q_read++ ) {
3279 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
3281 PerlIO_printf(Perl_debug_log, "\n");
3284 /*RExC_seen |= REG_TRIEDFA_SEEN;*/
3289 #define DEBUG_PEEP(str,scan,depth) \
3290 DEBUG_OPTIMISE_r({if (scan){ \
3291 regnode *Next = regnext(scan); \
3292 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state); \
3293 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)", \
3294 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),\
3295 Next ? (REG_NODE_NUM(Next)) : 0 ); \
3296 DEBUG_SHOW_STUDY_FLAGS(flags," [ ","]");\
3297 PerlIO_printf(Perl_debug_log, "\n"); \
3300 /* The below joins as many adjacent EXACTish nodes as possible into a single
3301 * one. The regop may be changed if the node(s) contain certain sequences that
3302 * require special handling. The joining is only done if:
3303 * 1) there is room in the current conglomerated node to entirely contain the
3305 * 2) they are the exact same node type
3307 * The adjacent nodes actually may be separated by NOTHING-kind nodes, and
3308 * these get optimized out
3310 * If a node is to match under /i (folded), the number of characters it matches
3311 * can be different than its character length if it contains a multi-character
3312 * fold. *min_subtract is set to the total delta number of characters of the
3315 * And *unfolded_multi_char is set to indicate whether or not the node contains
3316 * an unfolded multi-char fold. This happens when whether the fold is valid or
3317 * not won't be known until runtime; namely for EXACTF nodes that contain LATIN
3318 * SMALL LETTER SHARP S, as only if the target string being matched against
3319 * turns out to be UTF-8 is that fold valid; and also for EXACTFL nodes whose
3320 * folding rules depend on the locale in force at runtime. (Multi-char folds
3321 * whose components are all above the Latin1 range are not run-time locale
3322 * dependent, and have already been folded by the time this function is
3325 * This is as good a place as any to discuss the design of handling these
3326 * multi-character fold sequences. It's been wrong in Perl for a very long
3327 * time. There are three code points in Unicode whose multi-character folds
3328 * were long ago discovered to mess things up. The previous designs for
3329 * dealing with these involved assigning a special node for them. This
3330 * approach doesn't always work, as evidenced by this example:
3331 * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches
3332 * Both sides fold to "sss", but if the pattern is parsed to create a node that
3333 * would match just the \xDF, it won't be able to handle the case where a
3334 * successful match would have to cross the node's boundary. The new approach
3335 * that hopefully generally solves the problem generates an EXACTFU_SS node
3336 * that is "sss" in this case.
3338 * It turns out that there are problems with all multi-character folds, and not
3339 * just these three. Now the code is general, for all such cases. The
3340 * approach taken is:
3341 * 1) This routine examines each EXACTFish node that could contain multi-
3342 * character folded sequences. Since a single character can fold into
3343 * such a sequence, the minimum match length for this node is less than
3344 * the number of characters in the node. This routine returns in
3345 * *min_subtract how many characters to subtract from the the actual
3346 * length of the string to get a real minimum match length; it is 0 if
3347 * there are no multi-char foldeds. This delta is used by the caller to
3348 * adjust the min length of the match, and the delta between min and max,
3349 * so that the optimizer doesn't reject these possibilities based on size
3351 * 2) For the sequence involving the Sharp s (\xDF), the node type EXACTFU_SS
3352 * is used for an EXACTFU node that contains at least one "ss" sequence in
3353 * it. For non-UTF-8 patterns and strings, this is the only case where
3354 * there is a possible fold length change. That means that a regular
3355 * EXACTFU node without UTF-8 involvement doesn't have to concern itself
3356 * with length changes, and so can be processed faster. regexec.c takes
3357 * advantage of this. Generally, an EXACTFish node that is in UTF-8 is
3358 * pre-folded by regcomp.c (except EXACTFL, some of whose folds aren't
3359 * known until runtime). This saves effort in regex matching. However,
3360 * the pre-folding isn't done for non-UTF8 patterns because the fold of
3361 * the MICRO SIGN requires UTF-8, and we don't want to slow things down by
3362 * forcing the pattern into UTF8 unless necessary. Also what EXACTF (and,
3363 * again, EXACTFL) nodes fold to isn't known until runtime. The fold
3364 * possibilities for the non-UTF8 patterns are quite simple, except for
3365 * the sharp s. All the ones that don't involve a UTF-8 target string are
3366 * members of a fold-pair, and arrays are set up for all of them so that
3367 * the other member of the pair can be found quickly. Code elsewhere in
3368 * this file makes sure that in EXACTFU nodes, the sharp s gets folded to
3369 * 'ss', even if the pattern isn't UTF-8. This avoids the issues
3370 * described in the next item.
3371 * 3) A problem remains for unfolded multi-char folds. (These occur when the
3372 * validity of the fold won't be known until runtime, and so must remain
3373 * unfolded for now. This happens for the sharp s in EXACTF and EXACTFA
3374 * nodes when the pattern isn't in UTF-8. (Note, BTW, that there cannot
3375 * be an EXACTF node with a UTF-8 pattern.) They also occur for various
3376 * folds in EXACTFL nodes, regardless of the UTF-ness of the pattern.)
3377 * The reason this is a problem is that the optimizer part of regexec.c
3378 * (probably unwittingly, in Perl_regexec_flags()) makes an assumption
3379 * that a character in the pattern corresponds to at most a single
3380 * character in the target string. (And I do mean character, and not byte
3381 * here, unlike other parts of the documentation that have never been
3382 * updated to account for multibyte Unicode.) sharp s in EXACTF and
3383 * EXACTFL nodes can match the two character string 'ss'; in EXACTFA nodes
3384 * it can match "\x{17F}\x{17F}". These, along with other ones in EXACTFL
3385 * nodes, violate the assumption, and they are the only instances where it
3386 * is violated. I'm reluctant to try to change the assumption, as the
3387 * code involved is impenetrable to me (khw), so instead the code here
3388 * punts. This routine examines EXACTFL nodes, and (when the pattern
3389 * isn't UTF-8) EXACTF and EXACTFA for such unfolded folds, and returns a
3390 * boolean indicating whether or not the node contains such a fold. When
3391 * it is true, the caller sets a flag that later causes the optimizer in
3392 * this file to not set values for the floating and fixed string lengths,
3393 * and thus avoids the optimizer code in regexec.c that makes the invalid
3394 * assumption. Thus, there is no optimization based on string lengths for
3395 * EXACTFL nodes that contain these few folds, nor for non-UTF8-pattern
3396 * EXACTF and EXACTFA nodes that contain the sharp s. (The reason the
3397 * assumption is wrong only in these cases is that all other non-UTF-8
3398 * folds are 1-1; and, for UTF-8 patterns, we pre-fold all other folds to
3399 * their expanded versions. (Again, we can't prefold sharp s to 'ss' in
3400 * EXACTF nodes because we don't know at compile time if it actually
3401 * matches 'ss' or not. For EXACTF nodes it will match iff the target
3402 * string is in UTF-8. This is in contrast to EXACTFU nodes, where it
3403 * always matches; and EXACTFA where it never does. In an EXACTFA node in
3404 * a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the
3405 * problem; but in a non-UTF8 pattern, folding it to that above-Latin1
3406 * string would require the pattern to be forced into UTF-8, the overhead
3407 * of which we want to avoid. Similarly the unfolded multi-char folds in
3408 * EXACTFL nodes will match iff the locale at the time of match is a UTF-8
3411 * Similarly, the code that generates tries doesn't currently handle
3412 * not-already-folded multi-char folds, and it looks like a pain to change
3413 * that. Therefore, trie generation of EXACTFA nodes with the sharp s
3414 * doesn't work. Instead, such an EXACTFA is turned into a new regnode,
3415 * EXACTFA_NO_TRIE, which the trie code knows not to handle. Most people
3416 * using /iaa matching will be doing so almost entirely with ASCII
3417 * strings, so this should rarely be encountered in practice */
3419 #define JOIN_EXACT(scan,min_subtract,unfolded_multi_char, flags) \
3420 if (PL_regkind[OP(scan)] == EXACT) \
3421 join_exact(pRExC_state,(scan),(min_subtract),unfolded_multi_char, (flags),NULL,depth+1)
3424 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
3425 UV *min_subtract, bool *unfolded_multi_char,
3426 U32 flags,regnode *val, U32 depth)
3428 /* Merge several consecutive EXACTish nodes into one. */
3429 regnode *n = regnext(scan);
3431 regnode *next = scan + NODE_SZ_STR(scan);
3435 regnode *stop = scan;
3436 GET_RE_DEBUG_FLAGS_DECL;
3438 PERL_UNUSED_ARG(depth);
3441 PERL_ARGS_ASSERT_JOIN_EXACT;
3442 #ifndef EXPERIMENTAL_INPLACESCAN
3443 PERL_UNUSED_ARG(flags);
3444 PERL_UNUSED_ARG(val);
3446 DEBUG_PEEP("join",scan,depth);
3448 /* Look through the subsequent nodes in the chain. Skip NOTHING, merge
3449 * EXACT ones that are mergeable to the current one. */
3451 && (PL_regkind[OP(n)] == NOTHING
3452 || (stringok && OP(n) == OP(scan)))
3454 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX)
3457 if (OP(n) == TAIL || n > next)
3459 if (PL_regkind[OP(n)] == NOTHING) {
3460 DEBUG_PEEP("skip:",n,depth);
3461 NEXT_OFF(scan) += NEXT_OFF(n);
3462 next = n + NODE_STEP_REGNODE;
3469 else if (stringok) {
3470 const unsigned int oldl = STR_LEN(scan);
3471 regnode * const nnext = regnext(n);
3473 /* XXX I (khw) kind of doubt that this works on platforms (should
3474 * Perl ever run on one) where U8_MAX is above 255 because of lots
3475 * of other assumptions */
3476 /* Don't join if the sum can't fit into a single node */
3477 if (oldl + STR_LEN(n) > U8_MAX)
3480 DEBUG_PEEP("merg",n,depth);
3483 NEXT_OFF(scan) += NEXT_OFF(n);
3484 STR_LEN(scan) += STR_LEN(n);
3485 next = n + NODE_SZ_STR(n);
3486 /* Now we can overwrite *n : */
3487 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
3495 #ifdef EXPERIMENTAL_INPLACESCAN
3496 if (flags && !NEXT_OFF(n)) {
3497 DEBUG_PEEP("atch", val, depth);
3498 if (reg_off_by_arg[OP(n)]) {
3499 ARG_SET(n, val - n);
3502 NEXT_OFF(n) = val - n;
3510 *unfolded_multi_char = FALSE;
3512 /* Here, all the adjacent mergeable EXACTish nodes have been merged. We
3513 * can now analyze for sequences of problematic code points. (Prior to
3514 * this final joining, sequences could have been split over boundaries, and
3515 * hence missed). The sequences only happen in folding, hence for any
3516 * non-EXACT EXACTish node */
3517 if (OP(scan) != EXACT && OP(scan) != EXACTL) {
3518 U8* s0 = (U8*) STRING(scan);
3520 U8* s_end = s0 + STR_LEN(scan);
3522 int total_count_delta = 0; /* Total delta number of characters that
3523 multi-char folds expand to */
3525 /* One pass is made over the node's string looking for all the
3526 * possibilities. To avoid some tests in the loop, there are two main
3527 * cases, for UTF-8 patterns (which can't have EXACTF nodes) and
3532 if (OP(scan) == EXACTFL) {
3535 /* An EXACTFL node would already have been changed to another
3536 * node type unless there is at least one character in it that
3537 * is problematic; likely a character whose fold definition
3538 * won't be known until runtime, and so has yet to be folded.
3539 * For all but the UTF-8 locale, folds are 1-1 in length, but
3540 * to handle the UTF-8 case, we need to create a temporary
3541 * folded copy using UTF-8 locale rules in order to analyze it.
3542 * This is because our macros that look to see if a sequence is
3543 * a multi-char fold assume everything is folded (otherwise the
3544 * tests in those macros would be too complicated and slow).
3545 * Note that here, the non-problematic folds will have already
3546 * been done, so we can just copy such characters. We actually
3547 * don't completely fold the EXACTFL string. We skip the
3548 * unfolded multi-char folds, as that would just create work
3549 * below to figure out the size they already are */
3551 Newx(folded, UTF8_MAX_FOLD_CHAR_EXPAND * STR_LEN(scan) + 1, U8);
3554 STRLEN s_len = UTF8SKIP(s);
3555 if (! is_PROBLEMATIC_LOCALE_FOLD_utf8(s)) {
3556 Copy(s, d, s_len, U8);
3559 else if (is_FOLDS_TO_MULTI_utf8(s)) {
3560 *unfolded_multi_char = TRUE;
3561 Copy(s, d, s_len, U8);
3564 else if (isASCII(*s)) {
3565 *(d++) = toFOLD(*s);
3569 _to_utf8_fold_flags(s, d, &len, FOLD_FLAGS_FULL);
3575 /* Point the remainder of the routine to look at our temporary
3579 } /* End of creating folded copy of EXACTFL string */
3581 /* Examine the string for a multi-character fold sequence. UTF-8
3582 * patterns have all characters pre-folded by the time this code is
3584 while (s < s_end - 1) /* Can stop 1 before the end, as minimum
3585 length sequence we are looking for is 2 */
3587 int count = 0; /* How many characters in a multi-char fold */
3588 int len = is_MULTI_CHAR_FOLD_utf8_safe(s, s_end);
3589 if (! len) { /* Not a multi-char fold: get next char */
3594 /* Nodes with 'ss' require special handling, except for
3595 * EXACTFA-ish for which there is no multi-char fold to this */
3596 if (len == 2 && *s == 's' && *(s+1) == 's'
3597 && OP(scan) != EXACTFA
3598 && OP(scan) != EXACTFA_NO_TRIE)
3601 if (OP(scan) != EXACTFL) {
3602 OP(scan) = EXACTFU_SS;
3606 else { /* Here is a generic multi-char fold. */
3607 U8* multi_end = s + len;
3609 /* Count how many characters are in it. In the case of
3610 * /aa, no folds which contain ASCII code points are
3611 * allowed, so check for those, and skip if found. */
3612 if (OP(scan) != EXACTFA && OP(scan) != EXACTFA_NO_TRIE) {
3613 count = utf8_length(s, multi_end);
3617 while (s < multi_end) {
3620 goto next_iteration;
3630 /* The delta is how long the sequence is minus 1 (1 is how long
3631 * the character that folds to the sequence is) */
3632 total_count_delta += count - 1;
3636 /* We created a temporary folded copy of the string in EXACTFL
3637 * nodes. Therefore we need to be sure it doesn't go below zero,
3638 * as the real string could be shorter */
3639 if (OP(scan) == EXACTFL) {
3640 int total_chars = utf8_length((U8*) STRING(scan),
3641 (U8*) STRING(scan) + STR_LEN(scan));
3642 if (total_count_delta > total_chars) {
3643 total_count_delta = total_chars;
3647 *min_subtract += total_count_delta;
3650 else if (OP(scan) == EXACTFA) {
3652 /* Non-UTF-8 pattern, EXACTFA node. There can't be a multi-char
3653 * fold to the ASCII range (and there are no existing ones in the
3654 * upper latin1 range). But, as outlined in the comments preceding
3655 * this function, we need to flag any occurrences of the sharp s.
3656 * This character forbids trie formation (because of added
3659 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
3660 OP(scan) = EXACTFA_NO_TRIE;
3661 *unfolded_multi_char = TRUE;
3670 /* Non-UTF-8 pattern, not EXACTFA node. Look for the multi-char
3671 * folds that are all Latin1. As explained in the comments
3672 * preceding this function, we look also for the sharp s in EXACTF
3673 * and EXACTFL nodes; it can be in the final position. Otherwise
3674 * we can stop looking 1 byte earlier because have to find at least
3675 * two characters for a multi-fold */
3676 const U8* upper = (OP(scan) == EXACTF || OP(scan) == EXACTFL)
3681 int len = is_MULTI_CHAR_FOLD_latin1_safe(s, s_end);
3682 if (! len) { /* Not a multi-char fold. */
3683 if (*s == LATIN_SMALL_LETTER_SHARP_S
3684 && (OP(scan) == EXACTF || OP(scan) == EXACTFL))
3686 *unfolded_multi_char = TRUE;
3693 && isALPHA_FOLD_EQ(*s, 's')
3694 && isALPHA_FOLD_EQ(*(s+1), 's'))
3697 /* EXACTF nodes need to know that the minimum length
3698 * changed so that a sharp s in the string can match this
3699 * ss in the pattern, but they remain EXACTF nodes, as they
3700 * won't match this unless the target string is is UTF-8,
3701 * which we don't know until runtime. EXACTFL nodes can't
3702 * transform into EXACTFU nodes */
3703 if (OP(scan) != EXACTF && OP(scan) != EXACTFL) {
3704 OP(scan) = EXACTFU_SS;
3708 *min_subtract += len - 1;
3715 /* Allow dumping but overwriting the collection of skipped
3716 * ops and/or strings with fake optimized ops */
3717 n = scan + NODE_SZ_STR(scan);
3725 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
3729 /* REx optimizer. Converts nodes into quicker variants "in place".
3730 Finds fixed substrings. */
3732 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
3733 to the position after last scanned or to NULL. */
3735 #define INIT_AND_WITHP \
3736 assert(!and_withp); \
3737 Newx(and_withp,1, regnode_ssc); \
3738 SAVEFREEPV(and_withp)
3742 S_unwind_scan_frames(pTHX_ const void *p)
3744 scan_frame *f= (scan_frame *)p;
3746 scan_frame *n= f->next_frame;
3754 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
3755 SSize_t *minlenp, SSize_t *deltap,
3760 regnode_ssc *and_withp,
3761 U32 flags, U32 depth)
3762 /* scanp: Start here (read-write). */
3763 /* deltap: Write maxlen-minlen here. */
3764 /* last: Stop before this one. */
3765 /* data: string data about the pattern */
3766 /* stopparen: treat close N as END */
3767 /* recursed: which subroutines have we recursed into */
3768 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
3770 /* There must be at least this number of characters to match */
3773 regnode *scan = *scanp, *next;
3775 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
3776 int is_inf_internal = 0; /* The studied chunk is infinite */
3777 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
3778 scan_data_t data_fake;
3779 SV *re_trie_maxbuff = NULL;
3780 regnode *first_non_open = scan;
3781 SSize_t stopmin = SSize_t_MAX;
3782 scan_frame *frame = NULL;
3783 GET_RE_DEBUG_FLAGS_DECL;
3785 PERL_ARGS_ASSERT_STUDY_CHUNK;
3789 while (first_non_open && OP(first_non_open) == OPEN)
3790 first_non_open=regnext(first_non_open);
3796 RExC_study_chunk_recursed_count++;
3798 DEBUG_OPTIMISE_MORE_r(
3800 PerlIO_printf(Perl_debug_log,
3801 "%*sstudy_chunk stopparen=%ld recursed_count=%lu depth=%lu recursed_depth=%lu scan=%p last=%p",
3802 (int)(depth*2), "", (long)stopparen,
3803 (unsigned long)RExC_study_chunk_recursed_count,
3804 (unsigned long)depth, (unsigned long)recursed_depth,
3807 if (recursed_depth) {
3810 for ( j = 0 ; j < recursed_depth ; j++ ) {
3811 for ( i = 0 ; i < (U32)RExC_npar ; i++ ) {
3813 PAREN_TEST(RExC_study_chunk_recursed +
3814 ( j * RExC_study_chunk_recursed_bytes), i )
3817 !PAREN_TEST(RExC_study_chunk_recursed +
3818 (( j - 1 ) * RExC_study_chunk_recursed_bytes), i)
3821 PerlIO_printf(Perl_debug_log," %d",(int)i);
3825 if ( j + 1 < recursed_depth ) {
3826 PerlIO_printf(Perl_debug_log, ",");
3830 PerlIO_printf(Perl_debug_log,"\n");
3833 while ( scan && OP(scan) != END && scan < last ){
3834 UV min_subtract = 0; /* How mmany chars to subtract from the minimum
3835 node length to get a real minimum (because
3836 the folded version may be shorter) */
3837 bool unfolded_multi_char = FALSE;
3838 /* Peephole optimizer: */
3839 DEBUG_STUDYDATA("Peep:", data, depth);
3840 DEBUG_PEEP("Peep", scan, depth);
3843 /* The reason we do this here we need to deal with things like /(?:f)(?:o)(?:o)/
3844 * which cant be dealt with by the normal EXACT parsing code, as each (?:..) is handled
3845 * by a different invocation of reg() -- Yves
3847 JOIN_EXACT(scan,&min_subtract, &unfolded_multi_char, 0);
3849 /* Follow the next-chain of the current node and optimize
3850 away all the NOTHINGs from it. */
3851 if (OP(scan) != CURLYX) {
3852 const int max = (reg_off_by_arg[OP(scan)]
3854 /* I32 may be smaller than U16 on CRAYs! */
3855 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
3856 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
3860 /* Skip NOTHING and LONGJMP. */
3861 while ((n = regnext(n))
3862 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
3863 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
3864 && off + noff < max)
3866 if (reg_off_by_arg[OP(scan)])
3869 NEXT_OFF(scan) = off;
3872 /* The principal pseudo-switch. Cannot be a switch, since we
3873 look into several different things. */
3874 if ( OP(scan) == DEFINEP ) {
3876 SSize_t deltanext = 0;
3877 SSize_t fake_last_close = 0;
3878 I32 f = SCF_IN_DEFINE;
3880 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
3881 scan = regnext(scan);
3882 assert( OP(scan) == IFTHEN );
3883 DEBUG_PEEP("expect IFTHEN", scan, depth);
3885 data_fake.last_closep= &fake_last_close;
3887 next = regnext(scan);
3888 scan = NEXTOPER(NEXTOPER(scan));
3889 DEBUG_PEEP("scan", scan, depth);
3890 DEBUG_PEEP("next", next, depth);
3892 /* we suppose the run is continuous, last=next...
3893 * NOTE we dont use the return here! */
3894 (void)study_chunk(pRExC_state, &scan, &minlen,
3895 &deltanext, next, &data_fake, stopparen,
3896 recursed_depth, NULL, f, depth+1);
3901 OP(scan) == BRANCH ||
3902 OP(scan) == BRANCHJ ||
3905 next = regnext(scan);
3908 /* The op(next)==code check below is to see if we
3909 * have "BRANCH-BRANCH", "BRANCHJ-BRANCHJ", "IFTHEN-IFTHEN"
3910 * IFTHEN is special as it might not appear in pairs.
3911 * Not sure whether BRANCH-BRANCHJ is possible, regardless
3912 * we dont handle it cleanly. */
3913 if (OP(next) == code || code == IFTHEN) {
3914 /* NOTE - There is similar code to this block below for
3915 * handling TRIE nodes on a re-study. If you change stuff here
3916 * check there too. */
3917 SSize_t max1 = 0, min1 = SSize_t_MAX, num = 0;
3919 regnode * const startbranch=scan;
3921 if (flags & SCF_DO_SUBSTR) {
3922 /* Cannot merge strings after this. */
3923 scan_commit(pRExC_state, data, minlenp, is_inf);
3926 if (flags & SCF_DO_STCLASS)
3927 ssc_init_zero(pRExC_state, &accum);
3929 while (OP(scan) == code) {
3930 SSize_t deltanext, minnext, fake;
3932 regnode_ssc this_class;
3934 DEBUG_PEEP("Branch", scan, depth);
3937 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
3939 data_fake.whilem_c = data->whilem_c;
3940 data_fake.last_closep = data->last_closep;
3943 data_fake.last_closep = &fake;
3945 data_fake.pos_delta = delta;
3946 next = regnext(scan);
3948 scan = NEXTOPER(scan); /* everything */
3949 if (code != BRANCH) /* everything but BRANCH */
3950 scan = NEXTOPER(scan);
3952 if (flags & SCF_DO_STCLASS) {
3953 ssc_init(pRExC_state, &this_class);
3954 data_fake.start_class = &this_class;
3955 f = SCF_DO_STCLASS_AND;
3957 if (flags & SCF_WHILEM_VISITED_POS)
3958 f |= SCF_WHILEM_VISITED_POS;
3960 /* we suppose the run is continuous, last=next...*/
3961 minnext = study_chunk(pRExC_state,&nbs
https://perl5.git.perl.org / perl5.git / blob