5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
84 EXTERN_C const struct regexp_engine my_reg_engine;
89 #include "dquote_static.c"
90 #include "charclass_invlists.h"
91 #include "inline_invlist.c"
92 #include "unicode_constants.h"
94 #define HAS_NONLATIN1_FOLD_CLOSURE(i) \
95 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
96 #define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
97 #define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
100 #define STATIC static
104 struct RExC_state_t {
105 U32 flags; /* RXf_* are we folding, multilining? */
106 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
107 char *precomp; /* uncompiled string. */
108 REGEXP *rx_sv; /* The SV that is the regexp. */
109 regexp *rx; /* perl core regexp structure */
110 regexp_internal *rxi; /* internal data for regexp object
112 char *start; /* Start of input for compile */
113 char *end; /* End of input for compile */
114 char *parse; /* Input-scan pointer. */
115 SSize_t whilem_seen; /* number of WHILEM in this expr */
116 regnode *emit_start; /* Start of emitted-code area */
117 regnode *emit_bound; /* First regnode outside of the
119 regnode *emit; /* Code-emit pointer; if = &emit_dummy,
120 implies compiling, so don't emit */
121 regnode_ssc emit_dummy; /* placeholder for emit to point to;
122 large enough for the largest
123 non-EXACTish node, so can use it as
125 I32 naughty; /* How bad is this pattern? */
126 I32 sawback; /* Did we see \1, ...? */
128 SSize_t size; /* Code size. */
129 I32 npar; /* Capture buffer count, (OPEN) plus
130 one. ("par" 0 is the whole
132 I32 nestroot; /* root parens we are in - used by
136 regnode **open_parens; /* pointers to open parens */
137 regnode **close_parens; /* pointers to close parens */
138 regnode *opend; /* END node in program */
139 I32 utf8; /* whether the pattern is utf8 or not */
140 I32 orig_utf8; /* whether the pattern was originally in utf8 */
141 /* XXX use this for future optimisation of case
142 * where pattern must be upgraded to utf8. */
143 I32 uni_semantics; /* If a d charset modifier should use unicode
144 rules, even if the pattern is not in
146 HV *paren_names; /* Paren names */
148 regnode **recurse; /* Recurse regops */
149 I32 recurse_count; /* Number of recurse regops */
150 U8 *study_chunk_recursed; /* bitmap of which parens we have moved
152 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
156 I32 override_recoding;
157 I32 in_multi_char_class;
158 struct reg_code_block *code_blocks; /* positions of literal (?{})
160 int num_code_blocks; /* size of code_blocks[] */
161 int code_index; /* next code_blocks[] slot */
163 char *starttry; /* -Dr: where regtry was called. */
164 #define RExC_starttry (pRExC_state->starttry)
166 SV *runtime_code_qr; /* qr with the runtime code blocks */
168 const char *lastparse;
170 AV *paren_name_list; /* idx -> name */
171 #define RExC_lastparse (pRExC_state->lastparse)
172 #define RExC_lastnum (pRExC_state->lastnum)
173 #define RExC_paren_name_list (pRExC_state->paren_name_list)
177 #define RExC_flags (pRExC_state->flags)
178 #define RExC_pm_flags (pRExC_state->pm_flags)
179 #define RExC_precomp (pRExC_state->precomp)
180 #define RExC_rx_sv (pRExC_state->rx_sv)
181 #define RExC_rx (pRExC_state->rx)
182 #define RExC_rxi (pRExC_state->rxi)
183 #define RExC_start (pRExC_state->start)
184 #define RExC_end (pRExC_state->end)
185 #define RExC_parse (pRExC_state->parse)
186 #define RExC_whilem_seen (pRExC_state->whilem_seen)
187 #ifdef RE_TRACK_PATTERN_OFFSETS
188 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the
191 #define RExC_emit (pRExC_state->emit)
192 #define RExC_emit_dummy (pRExC_state->emit_dummy)
193 #define RExC_emit_start (pRExC_state->emit_start)
194 #define RExC_emit_bound (pRExC_state->emit_bound)
195 #define RExC_naughty (pRExC_state->naughty)
196 #define RExC_sawback (pRExC_state->sawback)
197 #define RExC_seen (pRExC_state->seen)
198 #define RExC_size (pRExC_state->size)
199 #define RExC_npar (pRExC_state->npar)
200 #define RExC_nestroot (pRExC_state->nestroot)
201 #define RExC_extralen (pRExC_state->extralen)
202 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
203 #define RExC_utf8 (pRExC_state->utf8)
204 #define RExC_uni_semantics (pRExC_state->uni_semantics)
205 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
206 #define RExC_open_parens (pRExC_state->open_parens)
207 #define RExC_close_parens (pRExC_state->close_parens)
208 #define RExC_opend (pRExC_state->opend)
209 #define RExC_paren_names (pRExC_state->paren_names)
210 #define RExC_recurse (pRExC_state->recurse)
211 #define RExC_recurse_count (pRExC_state->recurse_count)
212 #define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
213 #define RExC_study_chunk_recursed_bytes \
214 (pRExC_state->study_chunk_recursed_bytes)
215 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
216 #define RExC_contains_locale (pRExC_state->contains_locale)
217 #define RExC_contains_i (pRExC_state->contains_i)
218 #define RExC_override_recoding (pRExC_state->override_recoding)
219 #define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
222 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
223 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
224 ((*s) == '{' && regcurly(s, FALSE)))
227 * Flags to be passed up and down.
229 #define WORST 0 /* Worst case. */
230 #define HASWIDTH 0x01 /* Known to match non-null strings. */
232 /* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
233 * character. (There needs to be a case: in the switch statement in regexec.c
234 * for any node marked SIMPLE.) Note that this is not the same thing as
237 #define SPSTART 0x04 /* Starts with * or + */
238 #define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
239 #define TRYAGAIN 0x10 /* Weeded out a declaration. */
240 #define RESTART_UTF8 0x20 /* Restart, need to calcuate sizes as UTF-8 */
242 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
244 /* whether trie related optimizations are enabled */
245 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
246 #define TRIE_STUDY_OPT
247 #define FULL_TRIE_STUDY
253 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
254 #define PBITVAL(paren) (1 << ((paren) & 7))
255 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
256 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
257 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
259 #define REQUIRE_UTF8 STMT_START { \
261 *flagp = RESTART_UTF8; \
266 /* This converts the named class defined in regcomp.h to its equivalent class
267 * number defined in handy.h. */
268 #define namedclass_to_classnum(class) ((int) ((class) / 2))
269 #define classnum_to_namedclass(classnum) ((classnum) * 2)
271 #define _invlist_union_complement_2nd(a, b, output) \
272 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
273 #define _invlist_intersection_complement_2nd(a, b, output) \
274 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
276 /* About scan_data_t.
278 During optimisation we recurse through the regexp program performing
279 various inplace (keyhole style) optimisations. In addition study_chunk
280 and scan_commit populate this data structure with information about
281 what strings MUST appear in the pattern. We look for the longest
282 string that must appear at a fixed location, and we look for the
283 longest string that may appear at a floating location. So for instance
288 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
289 strings (because they follow a .* construct). study_chunk will identify
290 both FOO and BAR as being the longest fixed and floating strings respectively.
292 The strings can be composites, for instance
296 will result in a composite fixed substring 'foo'.
298 For each string some basic information is maintained:
300 - offset or min_offset
301 This is the position the string must appear at, or not before.
302 It also implicitly (when combined with minlenp) tells us how many
303 characters must match before the string we are searching for.
304 Likewise when combined with minlenp and the length of the string it
305 tells us how many characters must appear after the string we have
309 Only used for floating strings. This is the rightmost point that
310 the string can appear at. If set to SSize_t_MAX it indicates that the
311 string can occur infinitely far to the right.
314 A pointer to the minimum number of characters of the pattern that the
315 string was found inside. This is important as in the case of positive
316 lookahead or positive lookbehind we can have multiple patterns
321 The minimum length of the pattern overall is 3, the minimum length
322 of the lookahead part is 3, but the minimum length of the part that
323 will actually match is 1. So 'FOO's minimum length is 3, but the
324 minimum length for the F is 1. This is important as the minimum length
325 is used to determine offsets in front of and behind the string being
326 looked for. Since strings can be composites this is the length of the
327 pattern at the time it was committed with a scan_commit. Note that
328 the length is calculated by study_chunk, so that the minimum lengths
329 are not known until the full pattern has been compiled, thus the
330 pointer to the value.
334 In the case of lookbehind the string being searched for can be
335 offset past the start point of the final matching string.
336 If this value was just blithely removed from the min_offset it would
337 invalidate some of the calculations for how many chars must match
338 before or after (as they are derived from min_offset and minlen and
339 the length of the string being searched for).
340 When the final pattern is compiled and the data is moved from the
341 scan_data_t structure into the regexp structure the information
342 about lookbehind is factored in, with the information that would
343 have been lost precalculated in the end_shift field for the
346 The fields pos_min and pos_delta are used to store the minimum offset
347 and the delta to the maximum offset at the current point in the pattern.
351 typedef struct scan_data_t {
352 /*I32 len_min; unused */
353 /*I32 len_delta; unused */
357 SSize_t last_end; /* min value, <0 unless valid. */
358 SSize_t last_start_min;
359 SSize_t last_start_max;
360 SV **longest; /* Either &l_fixed, or &l_float. */
361 SV *longest_fixed; /* longest fixed string found in pattern */
362 SSize_t offset_fixed; /* offset where it starts */
363 SSize_t *minlen_fixed; /* pointer to the minlen relevant to the string */
364 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
365 SV *longest_float; /* longest floating string found in pattern */
366 SSize_t offset_float_min; /* earliest point in string it can appear */
367 SSize_t offset_float_max; /* latest point in string it can appear */
368 SSize_t *minlen_float; /* pointer to the minlen relevant to the string */
369 SSize_t lookbehind_float; /* is the pos of the string modified by LB */
372 SSize_t *last_closep;
373 regnode_ssc *start_class;
376 /* The below is perhaps overboard, but this allows us to save a test at the
377 * expense of a mask. This is because on both EBCDIC and ASCII machines, 'A'
378 * and 'a' differ by a single bit; the same with the upper and lower case of
379 * all other ASCII-range alphabetics. On ASCII platforms, they are 32 apart;
380 * on EBCDIC, they are 64. This uses an exclusive 'or' to find that bit and
381 * then inverts it to form a mask, with just a single 0, in the bit position
382 * where the upper- and lowercase differ. XXX There are about 40 other
383 * instances in the Perl core where this micro-optimization could be used.
384 * Should decide if maintenance cost is worse, before changing those
386 * Returns a boolean as to whether or not 'v' is either a lowercase or
387 * uppercase instance of 'c', where 'c' is in [A-Za-z]. If 'c' is a
388 * compile-time constant, the generated code is better than some optimizing
389 * compilers figure out, amounting to a mask and test. The results are
390 * meaningless if 'c' is not one of [A-Za-z] */
391 #define isARG2_lower_or_UPPER_ARG1(c, v) \
392 (((v) & ~('A' ^ 'a')) == ((c) & ~('A' ^ 'a')))
395 * Forward declarations for pregcomp()'s friends.
398 static const scan_data_t zero_scan_data =
399 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
401 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
402 #define SF_BEFORE_SEOL 0x0001
403 #define SF_BEFORE_MEOL 0x0002
404 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
405 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
407 #define SF_FIX_SHIFT_EOL (+2)
408 #define SF_FL_SHIFT_EOL (+4)
410 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
411 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
413 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
414 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
415 #define SF_IS_INF 0x0040
416 #define SF_HAS_PAR 0x0080
417 #define SF_IN_PAR 0x0100
418 #define SF_HAS_EVAL 0x0200
419 #define SCF_DO_SUBSTR 0x0400
420 #define SCF_DO_STCLASS_AND 0x0800
421 #define SCF_DO_STCLASS_OR 0x1000
422 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
423 #define SCF_WHILEM_VISITED_POS 0x2000
425 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
426 #define SCF_SEEN_ACCEPT 0x8000
427 #define SCF_TRIE_DOING_RESTUDY 0x10000
429 #define UTF cBOOL(RExC_utf8)
431 /* The enums for all these are ordered so things work out correctly */
432 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
433 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
434 == REGEX_DEPENDS_CHARSET)
435 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
436 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
437 >= REGEX_UNICODE_CHARSET)
438 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
439 == REGEX_ASCII_RESTRICTED_CHARSET)
440 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
441 >= REGEX_ASCII_RESTRICTED_CHARSET)
442 #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
443 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
445 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
447 /* For programs that want to be strictly Unicode compatible by dying if any
448 * attempt is made to match a non-Unicode code point against a Unicode
450 #define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
452 #define OOB_NAMEDCLASS -1
454 /* There is no code point that is out-of-bounds, so this is problematic. But
455 * its only current use is to initialize a variable that is always set before
457 #define OOB_UNICODE 0xDEADBEEF
459 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
460 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
463 /* length of regex to show in messages that don't mark a position within */
464 #define RegexLengthToShowInErrorMessages 127
467 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
468 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
469 * op/pragma/warn/regcomp.
471 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
472 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
474 #define REPORT_LOCATION " in regex; marked by " MARKER1 \
475 " in m/%"UTF8f MARKER2 "%"UTF8f"/"
477 #define REPORT_LOCATION_ARGS(offset) \
478 UTF8fARG(UTF, offset, RExC_precomp), \
479 UTF8fARG(UTF, RExC_end - RExC_precomp - offset, RExC_precomp + offset)
482 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
483 * arg. Show regex, up to a maximum length. If it's too long, chop and add
486 #define _FAIL(code) STMT_START { \
487 const char *ellipses = ""; \
488 IV len = RExC_end - RExC_precomp; \
491 SAVEFREESV(RExC_rx_sv); \
492 if (len > RegexLengthToShowInErrorMessages) { \
493 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
494 len = RegexLengthToShowInErrorMessages - 10; \
500 #define FAIL(msg) _FAIL( \
501 Perl_croak(aTHX_ "%s in regex m/%"UTF8f"%s/", \
502 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
504 #define FAIL2(msg,arg) _FAIL( \
505 Perl_croak(aTHX_ msg " in regex m/%"UTF8f"%s/", \
506 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
509 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
511 #define Simple_vFAIL(m) STMT_START { \
512 const IV offset = RExC_parse - RExC_precomp; \
513 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
514 m, REPORT_LOCATION_ARGS(offset)); \
518 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
520 #define vFAIL(m) STMT_START { \
522 SAVEFREESV(RExC_rx_sv); \
527 * Like Simple_vFAIL(), but accepts two arguments.
529 #define Simple_vFAIL2(m,a1) STMT_START { \
530 const IV offset = RExC_parse - RExC_precomp; \
531 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
532 REPORT_LOCATION_ARGS(offset)); \
536 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
538 #define vFAIL2(m,a1) STMT_START { \
540 SAVEFREESV(RExC_rx_sv); \
541 Simple_vFAIL2(m, a1); \
546 * Like Simple_vFAIL(), but accepts three arguments.
548 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
549 const IV offset = RExC_parse - RExC_precomp; \
550 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
551 REPORT_LOCATION_ARGS(offset)); \
555 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
557 #define vFAIL3(m,a1,a2) STMT_START { \
559 SAVEFREESV(RExC_rx_sv); \
560 Simple_vFAIL3(m, a1, a2); \
564 * Like Simple_vFAIL(), but accepts four arguments.
566 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
567 const IV offset = RExC_parse - RExC_precomp; \
568 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
569 REPORT_LOCATION_ARGS(offset)); \
572 #define vFAIL4(m,a1,a2,a3) STMT_START { \
574 SAVEFREESV(RExC_rx_sv); \
575 Simple_vFAIL4(m, a1, a2, a3); \
578 /* A specialized version of vFAIL2 that works with UTF8f */
579 #define vFAIL2utf8f(m, a1) STMT_START { \
580 const IV offset = RExC_parse - RExC_precomp; \
582 SAVEFREESV(RExC_rx_sv); \
583 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
584 REPORT_LOCATION_ARGS(offset)); \
588 /* m is not necessarily a "literal string", in this macro */
589 #define reg_warn_non_literal_string(loc, m) STMT_START { \
590 const IV offset = loc - RExC_precomp; \
591 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
592 m, REPORT_LOCATION_ARGS(offset)); \
595 #define ckWARNreg(loc,m) STMT_START { \
596 const IV offset = loc - RExC_precomp; \
597 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
598 REPORT_LOCATION_ARGS(offset)); \
601 #define vWARN_dep(loc, m) STMT_START { \
602 const IV offset = loc - RExC_precomp; \
603 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), m REPORT_LOCATION, \
604 REPORT_LOCATION_ARGS(offset)); \
607 #define ckWARNdep(loc,m) STMT_START { \
608 const IV offset = loc - RExC_precomp; \
609 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
611 REPORT_LOCATION_ARGS(offset)); \
614 #define ckWARNregdep(loc,m) STMT_START { \
615 const IV offset = loc - RExC_precomp; \
616 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
618 REPORT_LOCATION_ARGS(offset)); \
621 #define ckWARN2reg_d(loc,m, a1) STMT_START { \
622 const IV offset = loc - RExC_precomp; \
623 Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
625 a1, REPORT_LOCATION_ARGS(offset)); \
628 #define ckWARN2reg(loc, m, a1) STMT_START { \
629 const IV offset = loc - RExC_precomp; \
630 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
631 a1, REPORT_LOCATION_ARGS(offset)); \
634 #define vWARN3(loc, m, a1, a2) STMT_START { \
635 const IV offset = loc - RExC_precomp; \
636 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
637 a1, a2, REPORT_LOCATION_ARGS(offset)); \
640 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
641 const IV offset = loc - RExC_precomp; \
642 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
643 a1, a2, REPORT_LOCATION_ARGS(offset)); \
646 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
647 const IV offset = loc - RExC_precomp; \
648 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
649 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
652 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
653 const IV offset = loc - RExC_precomp; \
654 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
655 a1, a2, a3, REPORT_LOCATION_ARGS(offset)); \
658 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
659 const IV offset = loc - RExC_precomp; \
660 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
661 a1, a2, a3, a4, REPORT_LOCATION_ARGS(offset)); \
665 /* Allow for side effects in s */
666 #define REGC(c,s) STMT_START { \
667 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
670 /* Macros for recording node offsets. 20001227 mjd@plover.com
671 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
672 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
673 * Element 0 holds the number n.
674 * Position is 1 indexed.
676 #ifndef RE_TRACK_PATTERN_OFFSETS
677 #define Set_Node_Offset_To_R(node,byte)
678 #define Set_Node_Offset(node,byte)
679 #define Set_Cur_Node_Offset
680 #define Set_Node_Length_To_R(node,len)
681 #define Set_Node_Length(node,len)
682 #define Set_Node_Cur_Length(node,start)
683 #define Node_Offset(n)
684 #define Node_Length(n)
685 #define Set_Node_Offset_Length(node,offset,len)
686 #define ProgLen(ri) ri->u.proglen
687 #define SetProgLen(ri,x) ri->u.proglen = x
689 #define ProgLen(ri) ri->u.offsets[0]
690 #define SetProgLen(ri,x) ri->u.offsets[0] = x
691 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
693 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
694 __LINE__, (int)(node), (int)(byte))); \
696 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
699 RExC_offsets[2*(node)-1] = (byte); \
704 #define Set_Node_Offset(node,byte) \
705 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
706 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
708 #define Set_Node_Length_To_R(node,len) STMT_START { \
710 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
711 __LINE__, (int)(node), (int)(len))); \
713 Perl_croak(aTHX_ "value of node is %d in Length macro", \
716 RExC_offsets[2*(node)] = (len); \
721 #define Set_Node_Length(node,len) \
722 Set_Node_Length_To_R((node)-RExC_emit_start, len)
723 #define Set_Node_Cur_Length(node, start) \
724 Set_Node_Length(node, RExC_parse - start)
726 /* Get offsets and lengths */
727 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
728 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
730 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
731 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
732 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
736 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
737 #define EXPERIMENTAL_INPLACESCAN
738 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
740 #define DEBUG_RExC_seen() \
741 DEBUG_OPTIMISE_MORE_r({ \
742 PerlIO_printf(Perl_debug_log,"RExC_seen: "); \
744 if (RExC_seen & REG_SEEN_ZERO_LEN) \
745 PerlIO_printf(Perl_debug_log,"REG_SEEN_ZERO_LEN "); \
747 if (RExC_seen & REG_SEEN_LOOKBEHIND) \
748 PerlIO_printf(Perl_debug_log,"REG_SEEN_LOOKBEHIND "); \
750 if (RExC_seen & REG_SEEN_GPOS) \
751 PerlIO_printf(Perl_debug_log,"REG_SEEN_GPOS "); \
753 if (RExC_seen & REG_SEEN_CANY) \
754 PerlIO_printf(Perl_debug_log,"REG_SEEN_CANY "); \
756 if (RExC_seen & REG_SEEN_RECURSE) \
757 PerlIO_printf(Perl_debug_log,"REG_SEEN_RECURSE "); \
759 if (RExC_seen & REG_TOP_LEVEL_BRANCHES) \
760 PerlIO_printf(Perl_debug_log,"REG_TOP_LEVEL_BRANCHES "); \
762 if (RExC_seen & REG_SEEN_VERBARG) \
763 PerlIO_printf(Perl_debug_log,"REG_SEEN_VERBARG "); \
765 if (RExC_seen & REG_SEEN_CUTGROUP) \
766 PerlIO_printf(Perl_debug_log,"REG_SEEN_CUTGROUP "); \
768 if (RExC_seen & REG_SEEN_RUN_ON_COMMENT) \
769 PerlIO_printf(Perl_debug_log,"REG_SEEN_RUN_ON_COMMENT "); \
771 if (RExC_seen & REG_SEEN_EXACTF_SHARP_S) \
772 PerlIO_printf(Perl_debug_log,"REG_SEEN_EXACTF_SHARP_S "); \
774 if (RExC_seen & REG_SEEN_GOSTART) \
775 PerlIO_printf(Perl_debug_log,"REG_SEEN_GOSTART "); \
777 PerlIO_printf(Perl_debug_log,"\n"); \
780 #define DEBUG_STUDYDATA(str,data,depth) \
781 DEBUG_OPTIMISE_MORE_r(if(data){ \
782 PerlIO_printf(Perl_debug_log, \
783 "%*s" str "Pos:%"IVdf"/%"IVdf \
784 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
785 (int)(depth)*2, "", \
786 (IV)((data)->pos_min), \
787 (IV)((data)->pos_delta), \
788 (UV)((data)->flags), \
789 (IV)((data)->whilem_c), \
790 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
791 is_inf ? "INF " : "" \
793 if ((data)->last_found) \
794 PerlIO_printf(Perl_debug_log, \
795 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
796 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
797 SvPVX_const((data)->last_found), \
798 (IV)((data)->last_end), \
799 (IV)((data)->last_start_min), \
800 (IV)((data)->last_start_max), \
801 ((data)->longest && \
802 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
803 SvPVX_const((data)->longest_fixed), \
804 (IV)((data)->offset_fixed), \
805 ((data)->longest && \
806 (data)->longest==&((data)->longest_float)) ? "*" : "", \
807 SvPVX_const((data)->longest_float), \
808 (IV)((data)->offset_float_min), \
809 (IV)((data)->offset_float_max) \
811 PerlIO_printf(Perl_debug_log,"\n"); \
814 /* Mark that we cannot extend a found fixed substring at this point.
815 Update the longest found anchored substring and the longest found
816 floating substrings if needed. */
819 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
820 SSize_t *minlenp, int is_inf)
822 const STRLEN l = CHR_SVLEN(data->last_found);
823 const STRLEN old_l = CHR_SVLEN(*data->longest);
824 GET_RE_DEBUG_FLAGS_DECL;
826 PERL_ARGS_ASSERT_SCAN_COMMIT;
828 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
829 SvSetMagicSV(*data->longest, data->last_found);
830 if (*data->longest == data->longest_fixed) {
831 data->offset_fixed = l ? data->last_start_min : data->pos_min;
832 if (data->flags & SF_BEFORE_EOL)
834 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
836 data->flags &= ~SF_FIX_BEFORE_EOL;
837 data->minlen_fixed=minlenp;
838 data->lookbehind_fixed=0;
840 else { /* *data->longest == data->longest_float */
841 data->offset_float_min = l ? data->last_start_min : data->pos_min;
842 data->offset_float_max = (l
843 ? data->last_start_max
844 : (data->pos_delta == SSize_t_MAX
846 : data->pos_min + data->pos_delta));
848 || (STRLEN)data->offset_float_max > (STRLEN)SSize_t_MAX)
849 data->offset_float_max = SSize_t_MAX;
850 if (data->flags & SF_BEFORE_EOL)
852 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
854 data->flags &= ~SF_FL_BEFORE_EOL;
855 data->minlen_float=minlenp;
856 data->lookbehind_float=0;
859 SvCUR_set(data->last_found, 0);
861 SV * const sv = data->last_found;
862 if (SvUTF8(sv) && SvMAGICAL(sv)) {
863 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
869 data->flags &= ~SF_BEFORE_EOL;
870 DEBUG_STUDYDATA("commit: ",data,0);
873 /* An SSC is just a regnode_charclass_posix with an extra field: the inversion
874 * list that describes which code points it matches */
877 S_ssc_anything(pTHX_ regnode_ssc *ssc)
879 /* Set the SSC 'ssc' to match an empty string or any code point */
881 PERL_ARGS_ASSERT_SSC_ANYTHING;
883 assert(is_ANYOF_SYNTHETIC(ssc));
885 ssc->invlist = sv_2mortal(_new_invlist(2)); /* mortalize so won't leak */
886 _append_range_to_invlist(ssc->invlist, 0, UV_MAX);
887 ANYOF_FLAGS(ssc) |= ANYOF_EMPTY_STRING; /* Plus match empty string */
891 S_ssc_is_anything(pTHX_ const regnode_ssc *ssc)
893 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
894 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
895 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
896 * in any way, so there's no point in using it */
901 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
903 assert(is_ANYOF_SYNTHETIC(ssc));
905 if (! (ANYOF_FLAGS(ssc) & ANYOF_EMPTY_STRING)) {
909 /* See if the list consists solely of the range 0 - Infinity */
910 invlist_iterinit(ssc->invlist);
911 ret = invlist_iternext(ssc->invlist, &start, &end)
915 invlist_iterfinish(ssc->invlist);
921 /* If e.g., both \w and \W are set, matches everything */
922 if (ANYOF_FLAGS(ssc) & ANYOF_POSIXL) {
924 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
925 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
935 S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
937 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
938 * string, any code point, or any posix class under locale */
940 PERL_ARGS_ASSERT_SSC_INIT;
942 Zero(ssc, 1, regnode_ssc);
943 set_ANYOF_SYNTHETIC(ssc);
944 ARG_SET(ssc, ANYOF_NONBITMAP_EMPTY);
947 /* If any portion of the regex is to operate under locale rules,
948 * initialization includes it. The reason this isn't done for all regexes
949 * is that the optimizer was written under the assumption that locale was
950 * all-or-nothing. Given the complexity and lack of documentation in the
951 * optimizer, and that there are inadequate test cases for locale, many
952 * parts of it may not work properly, it is safest to avoid locale unless
954 if (RExC_contains_locale) {
955 ANYOF_POSIXL_SETALL(ssc);
956 ANYOF_FLAGS(ssc) |= ANYOF_LOCALE|ANYOF_POSIXL;
957 if (RExC_contains_i) {
958 ANYOF_FLAGS(ssc) |= ANYOF_LOC_FOLD;
962 ANYOF_POSIXL_ZERO(ssc);
967 S_ssc_is_cp_posixl_init(pTHX_ const RExC_state_t *pRExC_state,
968 const regnode_ssc *ssc)
970 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
971 * to the list of code points matched, and locale posix classes; hence does
972 * not check its flags) */
977 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
979 assert(is_ANYOF_SYNTHETIC(ssc));
981 invlist_iterinit(ssc->invlist);
982 ret = invlist_iternext(ssc->invlist, &start, &end)
986 invlist_iterfinish(ssc->invlist);
992 if (RExC_contains_locale) {
993 if (! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE)
994 || ! (ANYOF_FLAGS(ssc) & ANYOF_POSIXL)
995 || ! ANYOF_POSIXL_TEST_ALL_SET(ssc))
999 if (RExC_contains_i && ! (ANYOF_FLAGS(ssc) & ANYOF_LOC_FOLD)) {
1008 S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
1009 const regnode_charclass_posixl* const node)
1011 /* Returns a mortal inversion list defining which code points are matched
1012 * by 'node', which is of type ANYOF. Handles complementing the result if
1013 * appropriate. If some code points aren't knowable at this time, the
1014 * returned list must, and will, contain every possible code point. */
1016 SV* invlist = sv_2mortal(_new_invlist(0));
1018 const U32 n = ARG(node);
1020 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1022 /* Look at the data structure created by S_set_ANYOF_arg() */
1023 if (n != ANYOF_NONBITMAP_EMPTY) {
1024 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1025 AV * const av = MUTABLE_AV(SvRV(rv));
1026 SV **const ary = AvARRAY(av);
1027 assert(RExC_rxi->data->what[n] == 's');
1029 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
1030 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1])));
1032 else if (ary[0] && ary[0] != &PL_sv_undef) {
1034 /* Here, no compile-time swash, and there are things that won't be
1035 * known until runtime -- we have to assume it could be anything */
1036 return _add_range_to_invlist(invlist, 0, UV_MAX);
1040 /* Here no compile-time swash, and no run-time only data. Use the
1041 * node's inversion list */
1042 invlist = sv_2mortal(invlist_clone(ary[2]));
1046 /* An ANYOF node contains a bitmap for the first 256 code points, and an
1047 * inversion list for the others, but if there are code points that should
1048 * match only conditionally on the target string being UTF-8, those are
1049 * placed in the inversion list, and not the bitmap. Since there are
1050 * circumstances under which they could match, they are included in the
1051 * SSC. But if the ANYOF node is to be inverted, we have to exclude them
1052 * here, so that when we invert below, the end result actually does include
1053 * them. (Think about "\xe0" =~ /[^\xc0]/di;). We have to do this here
1054 * before we add the unconditionally matched code points */
1055 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1056 _invlist_intersection_complement_2nd(invlist,
1061 /* Add in the points from the bit map */
1062 for (i = 0; i < 256; i++) {
1063 if (ANYOF_BITMAP_TEST(node, i)) {
1064 invlist = add_cp_to_invlist(invlist, i);
1068 /* If this can match all upper Latin1 code points, have to add them
1070 if (ANYOF_FLAGS(node) & ANYOF_NON_UTF8_NON_ASCII_ALL) {
1071 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1074 /* Similarly for these */
1075 if (ANYOF_FLAGS(node) & ANYOF_ABOVE_LATIN1_ALL) {
1076 invlist = _add_range_to_invlist(invlist, 256, UV_MAX);
1079 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1080 _invlist_invert(invlist);
1086 /* These two functions currently do the exact same thing */
1087 #define ssc_init_zero ssc_init
1089 #define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1090 #define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1093 S_ssc_flags_and(regnode_ssc *ssc, const U8 and_with)
1095 /* Take the flags 'and_with' and accumulate them anded into the flags for
1096 * the SSC 'ssc'. The non-SSC related flags in 'and_with' are ignored.
1097 * The flags 'and_with' should not come from another SSC (otherwise the
1098 * EMPTY_STRING flag won't work) */
1100 const U8 ssc_only_flags = ANYOF_FLAGS(ssc) & ~ANYOF_COMMON_FLAGS;
1102 PERL_ARGS_ASSERT_SSC_FLAGS_AND;
1104 /* Use just the SSC-related flags from 'and_with' */
1105 ANYOF_FLAGS(ssc) &= (and_with & ANYOF_COMMON_FLAGS);
1106 ANYOF_FLAGS(ssc) |= ssc_only_flags;
1109 /* 'AND' a given class with another one. Can create false positives. 'ssc'
1110 * should not be inverted. 'and_with->flags & ANYOF_POSIXL' should be 0 if
1111 * 'and_with' is a regnode_charclass instead of a regnode_ssc. */
1114 S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1115 const regnode_ssc *and_with)
1117 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1118 * another SSC or a regular ANYOF class. Can create false positives. */
1123 PERL_ARGS_ASSERT_SSC_AND;
1125 assert(is_ANYOF_SYNTHETIC(ssc));
1127 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1128 * the code point inversion list and just the relevant flags */
1129 if (is_ANYOF_SYNTHETIC(and_with)) {
1130 anded_cp_list = and_with->invlist;
1131 anded_flags = ANYOF_FLAGS(and_with);
1133 /* XXX This is a kludge around what appears to be deficiencies in the
1134 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1135 * there are paths through the optimizer where it doesn't get weeded
1136 * out when it should. And if we don't make some extra provision for
1137 * it like the code just below, it doesn't get added when it should.
1138 * This solution is to add it only when AND'ing, which is here, and
1139 * only when what is being AND'ed is the pristine, original node
1140 * matching anything. Thus it is like adding it to ssc_anything() but
1141 * only when the result is to be AND'ed. Probably the same solution
1142 * could be adopted for the same problem we have with /l matching,
1143 * which is solved differently in S_ssc_init(), and that would lead to
1144 * fewer false positives than that solution has. But if this solution
1145 * creates bugs, the consequences are only that a warning isn't raised
1146 * that should be; while the consequences for having /l bugs is
1147 * incorrect matches */
1148 if (ssc_is_anything(and_with)) {
1149 anded_flags |= ANYOF_WARN_SUPER;
1153 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state,
1154 (regnode_charclass_posixl*) and_with);
1155 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1158 ANYOF_FLAGS(ssc) &= anded_flags;
1160 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1161 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1162 * 'and_with' may be inverted. When not inverted, we have the situation of
1164 * (C1 | P1) & (C2 | P2)
1165 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1166 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1167 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1168 * <= ((C1 & C2) | P1 | P2)
1169 * Alternatively, the last few steps could be:
1170 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1171 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1172 * <= (C1 | C2 | (P1 & P2))
1173 * We favor the second approach if either P1 or P2 is non-empty. This is
1174 * because these components are a barrier to doing optimizations, as what
1175 * they match cannot be known until the moment of matching as they are
1176 * dependent on the current locale, 'AND"ing them likely will reduce or
1178 * But we can do better if we know that C1,P1 are in their initial state (a
1179 * frequent occurrence), each matching everything:
1180 * (<everything>) & (C2 | P2) = C2 | P2
1181 * Similarly, if C2,P2 are in their initial state (again a frequent
1182 * occurrence), the result is a no-op
1183 * (C1 | P1) & (<everything>) = C1 | P1
1186 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1187 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1188 * <= (C1 & ~C2) | (P1 & ~P2)
1191 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
1192 && ! is_ANYOF_SYNTHETIC(and_with))
1196 ssc_intersection(ssc,
1198 FALSE /* Has already been inverted */
1201 /* If either P1 or P2 is empty, the intersection will be also; can skip
1203 if (! (ANYOF_FLAGS(and_with) & ANYOF_POSIXL)) {
1204 ANYOF_POSIXL_ZERO(ssc);
1206 else if (ANYOF_POSIXL_TEST_ANY_SET(ssc)) {
1208 /* Note that the Posix class component P from 'and_with' actually
1210 * P = Pa | Pb | ... | Pn
1211 * where each component is one posix class, such as in [\w\s].
1213 * ~P = ~(Pa | Pb | ... | Pn)
1214 * = ~Pa & ~Pb & ... & ~Pn
1215 * <= ~Pa | ~Pb | ... | ~Pn
1216 * The last is something we can easily calculate, but unfortunately
1217 * is likely to have many false positives. We could do better
1218 * in some (but certainly not all) instances if two classes in
1219 * P have known relationships. For example
1220 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1222 * :lower: & :print: = :lower:
1223 * And similarly for classes that must be disjoint. For example,
1224 * since \s and \w can have no elements in common based on rules in
1225 * the POSIX standard,
1226 * \w & ^\S = nothing
1227 * Unfortunately, some vendor locales do not meet the Posix
1228 * standard, in particular almost everything by Microsoft.
1229 * The loop below just changes e.g., \w into \W and vice versa */
1231 regnode_charclass_posixl temp;
1232 int add = 1; /* To calculate the index of the complement */
1234 ANYOF_POSIXL_ZERO(&temp);
1235 for (i = 0; i < ANYOF_MAX; i++) {
1237 || ! ANYOF_POSIXL_TEST(and_with, i)
1238 || ! ANYOF_POSIXL_TEST(and_with, i + 1));
1240 if (ANYOF_POSIXL_TEST(and_with, i)) {
1241 ANYOF_POSIXL_SET(&temp, i + add);
1243 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1245 ANYOF_POSIXL_AND(&temp, ssc);
1247 } /* else ssc already has no posixes */
1248 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1249 in its initial state */
1250 else if (! is_ANYOF_SYNTHETIC(and_with)
1251 || ! ssc_is_cp_posixl_init(pRExC_state, and_with))
1253 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1254 * copy it over 'ssc' */
1255 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
1256 if (is_ANYOF_SYNTHETIC(and_with)) {
1257 StructCopy(and_with, ssc, regnode_ssc);
1260 ssc->invlist = anded_cp_list;
1261 ANYOF_POSIXL_ZERO(ssc);
1262 if (ANYOF_FLAGS(and_with) & ANYOF_POSIXL) {
1263 ANYOF_POSIXL_OR(and_with, ssc);
1267 else if ((ANYOF_FLAGS(ssc) & ANYOF_POSIXL)
1268 || (ANYOF_FLAGS(and_with) & ANYOF_POSIXL))
1270 /* One or the other of P1, P2 is non-empty. */
1271 ANYOF_POSIXL_AND(and_with, ssc);
1272 ssc_union(ssc, anded_cp_list, FALSE);
1274 else { /* P1 = P2 = empty */
1275 ssc_intersection(ssc, anded_cp_list, FALSE);
1281 S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1282 const regnode_ssc *or_with)
1284 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1285 * another SSC or a regular ANYOF class. Can create false positives if
1286 * 'or_with' is to be inverted. */
1291 PERL_ARGS_ASSERT_SSC_OR;
1293 assert(is_ANYOF_SYNTHETIC(ssc));
1295 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1296 * the code point inversion list and just the relevant flags */
1297 if (is_ANYOF_SYNTHETIC(or_with)) {
1298 ored_cp_list = or_with->invlist;
1299 ored_flags = ANYOF_FLAGS(or_with);
1302 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state,
1303 (regnode_charclass_posixl*) or_with);
1304 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
1307 ANYOF_FLAGS(ssc) |= ored_flags;
1309 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1310 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1311 * 'or_with' may be inverted. When not inverted, we have the simple
1312 * situation of computing:
1313 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1314 * If P1|P2 yields a situation with both a class and its complement are
1315 * set, like having both \w and \W, this matches all code points, and we
1316 * can delete these from the P component of the ssc going forward. XXX We
1317 * might be able to delete all the P components, but I (khw) am not certain
1318 * about this, and it is better to be safe.
1321 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1322 * <= (C1 | P1) | ~C2
1323 * <= (C1 | ~C2) | P1
1324 * (which results in actually simpler code than the non-inverted case)
1327 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
1328 && ! is_ANYOF_SYNTHETIC(or_with))
1330 /* We ignore P2, leaving P1 going forward */
1332 else { /* Not inverted */
1333 ANYOF_POSIXL_OR(or_with, ssc);
1334 if (ANYOF_POSIXL_TEST_ANY_SET(ssc)) {
1336 for (i = 0; i < ANYOF_MAX; i += 2) {
1337 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1339 ssc_match_all_cp(ssc);
1340 ANYOF_POSIXL_CLEAR(ssc, i);
1341 ANYOF_POSIXL_CLEAR(ssc, i+1);
1342 if (! ANYOF_POSIXL_TEST_ANY_SET(ssc)) {
1343 ANYOF_FLAGS(ssc) &= ~ANYOF_POSIXL;
1352 FALSE /* Already has been inverted */
1356 PERL_STATIC_INLINE void
1357 S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1359 PERL_ARGS_ASSERT_SSC_UNION;
1361 assert(is_ANYOF_SYNTHETIC(ssc));
1363 _invlist_union_maybe_complement_2nd(ssc->invlist,
1369 PERL_STATIC_INLINE void
1370 S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1372 const bool invert2nd)
1374 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1376 assert(is_ANYOF_SYNTHETIC(ssc));
1378 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1384 PERL_STATIC_INLINE void
1385 S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1387 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1389 assert(is_ANYOF_SYNTHETIC(ssc));
1391 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1394 PERL_STATIC_INLINE void
1395 S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1397 /* AND just the single code point 'cp' into the SSC 'ssc' */
1399 SV* cp_list = _new_invlist(2);
1401 PERL_ARGS_ASSERT_SSC_CP_AND;
1403 assert(is_ANYOF_SYNTHETIC(ssc));
1405 cp_list = add_cp_to_invlist(cp_list, cp);
1406 ssc_intersection(ssc, cp_list,
1407 FALSE /* Not inverted */
1409 SvREFCNT_dec_NN(cp_list);
1412 PERL_STATIC_INLINE void
1413 S_ssc_clear_locale(pTHX_ regnode_ssc *ssc)
1415 /* Set the SSC 'ssc' to not match any locale things */
1417 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1419 assert(is_ANYOF_SYNTHETIC(ssc));
1421 ANYOF_POSIXL_ZERO(ssc);
1422 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1426 S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
1428 /* The inversion list in the SSC is marked mortal; now we need a more
1429 * permanent copy, which is stored the same way that is done in a regular
1430 * ANYOF node, with the first 256 code points in a bit map */
1432 SV* invlist = invlist_clone(ssc->invlist);
1434 PERL_ARGS_ASSERT_SSC_FINALIZE;
1436 assert(is_ANYOF_SYNTHETIC(ssc));
1438 /* The code in this file assumes that all but these flags aren't relevant
1439 * to the SSC, except ANYOF_EMPTY_STRING, which should be cleared by the
1440 * time we reach here */
1441 assert(! (ANYOF_FLAGS(ssc) & ~ANYOF_COMMON_FLAGS));
1443 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
1445 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist, NULL, NULL, FALSE);
1447 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE) || RExC_contains_locale);
1450 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
1451 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
1452 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
1453 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
1454 ? (TRIE_LIST_CUR( idx ) - 1) \
1460 dump_trie(trie,widecharmap,revcharmap)
1461 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
1462 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
1464 These routines dump out a trie in a somewhat readable format.
1465 The _interim_ variants are used for debugging the interim
1466 tables that are used to generate the final compressed
1467 representation which is what dump_trie expects.
1469 Part of the reason for their existence is to provide a form
1470 of documentation as to how the different representations function.
1475 Dumps the final compressed table form of the trie to Perl_debug_log.
1476 Used for debugging make_trie().
1480 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
1481 AV *revcharmap, U32 depth)
1484 SV *sv=sv_newmortal();
1485 int colwidth= widecharmap ? 6 : 4;
1487 GET_RE_DEBUG_FLAGS_DECL;
1489 PERL_ARGS_ASSERT_DUMP_TRIE;
1491 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
1492 (int)depth * 2 + 2,"",
1493 "Match","Base","Ofs" );
1495 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
1496 SV ** const tmp = av_fetch( revcharmap, state, 0);
1498 PerlIO_printf( Perl_debug_log, "%*s",
1500 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1501 PL_colors[0], PL_colors[1],
1502 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1503 PERL_PV_ESCAPE_FIRSTCHAR
1508 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
1509 (int)depth * 2 + 2,"");
1511 for( state = 0 ; state < trie->uniquecharcount ; state++ )
1512 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
1513 PerlIO_printf( Perl_debug_log, "\n");
1515 for( state = 1 ; state < trie->statecount ; state++ ) {
1516 const U32 base = trie->states[ state ].trans.base;
1518 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|",
1519 (int)depth * 2 + 2,"", (UV)state);
1521 if ( trie->states[ state ].wordnum ) {
1522 PerlIO_printf( Perl_debug_log, " W%4X",
1523 trie->states[ state ].wordnum );
1525 PerlIO_printf( Perl_debug_log, "%6s", "" );
1528 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
1533 while( ( base + ofs < trie->uniquecharcount ) ||
1534 ( base + ofs - trie->uniquecharcount < trie->lasttrans
1535 && trie->trans[ base + ofs - trie->uniquecharcount ].check
1539 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
1541 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1542 if ( ( base + ofs >= trie->uniquecharcount )
1543 && ( base + ofs - trie->uniquecharcount
1545 && trie->trans[ base + ofs
1546 - trie->uniquecharcount ].check == state )
1548 PerlIO_printf( Perl_debug_log, "%*"UVXf,
1550 (UV)trie->trans[ base + ofs
1551 - trie->uniquecharcount ].next );
1553 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
1557 PerlIO_printf( Perl_debug_log, "]");
1560 PerlIO_printf( Perl_debug_log, "\n" );
1562 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=",
1564 for (word=1; word <= trie->wordcount; word++) {
1565 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
1566 (int)word, (int)(trie->wordinfo[word].prev),
1567 (int)(trie->wordinfo[word].len));
1569 PerlIO_printf(Perl_debug_log, "\n" );
1572 Dumps a fully constructed but uncompressed trie in list form.
1573 List tries normally only are used for construction when the number of
1574 possible chars (trie->uniquecharcount) is very high.
1575 Used for debugging make_trie().
1578 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
1579 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1583 SV *sv=sv_newmortal();
1584 int colwidth= widecharmap ? 6 : 4;
1585 GET_RE_DEBUG_FLAGS_DECL;
1587 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1589 /* print out the table precompression. */
1590 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1591 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1592 "------:-----+-----------------\n" );
1594 for( state=1 ; state < next_alloc ; state ++ ) {
1597 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1598 (int)depth * 2 + 2,"", (UV)state );
1599 if ( ! trie->states[ state ].wordnum ) {
1600 PerlIO_printf( Perl_debug_log, "%5s| ","");
1602 PerlIO_printf( Perl_debug_log, "W%4x| ",
1603 trie->states[ state ].wordnum
1606 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1607 SV ** const tmp = av_fetch( revcharmap,
1608 TRIE_LIST_ITEM(state,charid).forid, 0);
1610 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1612 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
1614 PL_colors[0], PL_colors[1],
1615 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
1616 | PERL_PV_ESCAPE_FIRSTCHAR
1618 TRIE_LIST_ITEM(state,charid).forid,
1619 (UV)TRIE_LIST_ITEM(state,charid).newstate
1622 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1623 (int)((depth * 2) + 14), "");
1626 PerlIO_printf( Perl_debug_log, "\n");
1631 Dumps a fully constructed but uncompressed trie in table form.
1632 This is the normal DFA style state transition table, with a few
1633 twists to facilitate compression later.
1634 Used for debugging make_trie().
1637 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1638 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1643 SV *sv=sv_newmortal();
1644 int colwidth= widecharmap ? 6 : 4;
1645 GET_RE_DEBUG_FLAGS_DECL;
1647 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1650 print out the table precompression so that we can do a visual check
1651 that they are identical.
1654 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1656 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1657 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1659 PerlIO_printf( Perl_debug_log, "%*s",
1661 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1662 PL_colors[0], PL_colors[1],
1663 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1664 PERL_PV_ESCAPE_FIRSTCHAR
1670 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1672 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1673 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1676 PerlIO_printf( Perl_debug_log, "\n" );
1678 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1680 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1681 (int)depth * 2 + 2,"",
1682 (UV)TRIE_NODENUM( state ) );
1684 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1685 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1687 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1689 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1691 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1692 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n",
1693 (UV)trie->trans[ state ].check );
1695 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n",
1696 (UV)trie->trans[ state ].check,
1697 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1705 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1706 startbranch: the first branch in the whole branch sequence
1707 first : start branch of sequence of branch-exact nodes.
1708 May be the same as startbranch
1709 last : Thing following the last branch.
1710 May be the same as tail.
1711 tail : item following the branch sequence
1712 count : words in the sequence
1713 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1714 depth : indent depth
1716 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1718 A trie is an N'ary tree where the branches are determined by digital
1719 decomposition of the key. IE, at the root node you look up the 1st character and
1720 follow that branch repeat until you find the end of the branches. Nodes can be
1721 marked as "accepting" meaning they represent a complete word. Eg:
1725 would convert into the following structure. Numbers represent states, letters
1726 following numbers represent valid transitions on the letter from that state, if
1727 the number is in square brackets it represents an accepting state, otherwise it
1728 will be in parenthesis.
1730 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1734 (1) +-i->(6)-+-s->[7]
1736 +-s->(3)-+-h->(4)-+-e->[5]
1738 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1740 This shows that when matching against the string 'hers' we will begin at state 1
1741 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1742 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1743 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1744 single traverse. We store a mapping from accepting to state to which word was
1745 matched, and then when we have multiple possibilities we try to complete the
1746 rest of the regex in the order in which they occured in the alternation.
1748 The only prior NFA like behaviour that would be changed by the TRIE support is
1749 the silent ignoring of duplicate alternations which are of the form:
1751 / (DUPE|DUPE) X? (?{ ... }) Y /x
1753 Thus EVAL blocks following a trie may be called a different number of times with
1754 and without the optimisation. With the optimisations dupes will be silently
1755 ignored. This inconsistent behaviour of EVAL type nodes is well established as
1756 the following demonstrates:
1758 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1760 which prints out 'word' three times, but
1762 'words'=~/(word|word|word)(?{ print $1 })S/
1764 which doesnt print it out at all. This is due to other optimisations kicking in.
1766 Example of what happens on a structural level:
1768 The regexp /(ac|ad|ab)+/ will produce the following debug output:
1770 1: CURLYM[1] {1,32767}(18)
1781 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1782 and should turn into:
1784 1: CURLYM[1] {1,32767}(18)
1786 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1794 Cases where tail != last would be like /(?foo|bar)baz/:
1804 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1805 and would end up looking like:
1808 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1815 d = uvchr_to_utf8_flags(d, uv, 0);
1817 is the recommended Unicode-aware way of saying
1822 #define TRIE_STORE_REVCHAR(val) \
1825 SV *zlopp = newSV(7); /* XXX: optimize me */ \
1826 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1827 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
1828 SvCUR_set(zlopp, kapow - flrbbbbb); \
1831 av_push(revcharmap, zlopp); \
1833 char ooooff = (char)val; \
1834 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1838 /* This gets the next character from the input, folding it if not already
1840 #define TRIE_READ_CHAR STMT_START { \
1843 /* if it is UTF then it is either already folded, or does not need \
1845 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
1847 else if (folder == PL_fold_latin1) { \
1848 /* This folder implies Unicode rules, which in the range expressible \
1849 * by not UTF is the lower case, with the two exceptions, one of \
1850 * which should have been taken care of before calling this */ \
1851 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
1852 uvc = toLOWER_L1(*uc); \
1853 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
1856 /* raw data, will be folded later if needed */ \
1864 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1865 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1866 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1867 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1869 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1870 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1871 TRIE_LIST_CUR( state )++; \
1874 #define TRIE_LIST_NEW(state) STMT_START { \
1875 Newxz( trie->states[ state ].trans.list, \
1876 4, reg_trie_trans_le ); \
1877 TRIE_LIST_CUR( state ) = 1; \
1878 TRIE_LIST_LEN( state ) = 4; \
1881 #define TRIE_HANDLE_WORD(state) STMT_START { \
1882 U16 dupe= trie->states[ state ].wordnum; \
1883 regnode * const noper_next = regnext( noper ); \
1886 /* store the word for dumping */ \
1888 if (OP(noper) != NOTHING) \
1889 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1891 tmp = newSVpvn_utf8( "", 0, UTF ); \
1892 av_push( trie_words, tmp ); \
1896 trie->wordinfo[curword].prev = 0; \
1897 trie->wordinfo[curword].len = wordlen; \
1898 trie->wordinfo[curword].accept = state; \
1900 if ( noper_next < tail ) { \
1902 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
1904 trie->jump[curword] = (U16)(noper_next - convert); \
1906 jumper = noper_next; \
1908 nextbranch= regnext(cur); \
1912 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1913 /* chain, so that when the bits of chain are later */\
1914 /* linked together, the dups appear in the chain */\
1915 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1916 trie->wordinfo[dupe].prev = curword; \
1918 /* we haven't inserted this word yet. */ \
1919 trie->states[ state ].wordnum = curword; \
1924 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1925 ( ( base + charid >= ucharcount \
1926 && base + charid < ubound \
1927 && state == trie->trans[ base - ucharcount + charid ].check \
1928 && trie->trans[ base - ucharcount + charid ].next ) \
1929 ? trie->trans[ base - ucharcount + charid ].next \
1930 : ( state==1 ? special : 0 ) \
1934 #define MADE_JUMP_TRIE 2
1935 #define MADE_EXACT_TRIE 4
1938 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
1939 regnode *first, regnode *last, regnode *tail,
1940 U32 word_count, U32 flags, U32 depth)
1943 /* first pass, loop through and scan words */
1944 reg_trie_data *trie;
1945 HV *widecharmap = NULL;
1946 AV *revcharmap = newAV();
1952 regnode *jumper = NULL;
1953 regnode *nextbranch = NULL;
1954 regnode *convert = NULL;
1955 U32 *prev_states; /* temp array mapping each state to previous one */
1956 /* we just use folder as a flag in utf8 */
1957 const U8 * folder = NULL;
1960 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuuu"));
1961 AV *trie_words = NULL;
1962 /* along with revcharmap, this only used during construction but both are
1963 * useful during debugging so we store them in the struct when debugging.
1966 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
1967 STRLEN trie_charcount=0;
1969 SV *re_trie_maxbuff;
1970 GET_RE_DEBUG_FLAGS_DECL;
1972 PERL_ARGS_ASSERT_MAKE_TRIE;
1974 PERL_UNUSED_ARG(depth);
1981 case EXACTFU: folder = PL_fold_latin1; break;
1982 case EXACTF: folder = PL_fold; break;
1983 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
1986 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1988 trie->startstate = 1;
1989 trie->wordcount = word_count;
1990 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1991 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1993 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1994 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1995 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1998 trie_words = newAV();
2001 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2002 if (!SvIOK(re_trie_maxbuff)) {
2003 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2005 DEBUG_TRIE_COMPILE_r({
2006 PerlIO_printf( Perl_debug_log,
2007 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2008 (int)depth * 2 + 2, "",
2009 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
2010 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
2013 /* Find the node we are going to overwrite */
2014 if ( first == startbranch && OP( last ) != BRANCH ) {
2015 /* whole branch chain */
2018 /* branch sub-chain */
2019 convert = NEXTOPER( first );
2022 /* -- First loop and Setup --
2024 We first traverse the branches and scan each word to determine if it
2025 contains widechars, and how many unique chars there are, this is
2026 important as we have to build a table with at least as many columns as we
2029 We use an array of integers to represent the character codes 0..255
2030 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2031 the native representation of the character value as the key and IV's for
2034 *TODO* If we keep track of how many times each character is used we can
2035 remap the columns so that the table compression later on is more
2036 efficient in terms of memory by ensuring the most common value is in the
2037 middle and the least common are on the outside. IMO this would be better
2038 than a most to least common mapping as theres a decent chance the most
2039 common letter will share a node with the least common, meaning the node
2040 will not be compressible. With a middle is most common approach the worst
2041 case is when we have the least common nodes twice.
2045 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2046 regnode *noper = NEXTOPER( cur );
2047 const U8 *uc = (U8*)STRING( noper );
2048 const U8 *e = uc + STR_LEN( noper );
2050 U32 wordlen = 0; /* required init */
2051 STRLEN minbytes = 0;
2052 STRLEN maxbytes = 0;
2053 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2056 if (OP(noper) == NOTHING) {
2057 regnode *noper_next= regnext(noper);
2058 if (noper_next != tail && OP(noper_next) == flags) {
2060 uc= (U8*)STRING(noper);
2061 e= uc + STR_LEN(noper);
2062 trie->minlen= STR_LEN(noper);
2069 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
2070 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2071 regardless of encoding */
2072 if (OP( noper ) == EXACTFU_SS) {
2073 /* false positives are ok, so just set this */
2074 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
2077 for ( ; uc < e ; uc += len ) {
2078 TRIE_CHARCOUNT(trie)++;
2081 /* Acummulate to the current values, the range in the number of
2082 * bytes that this character could match. The max is presumed to
2083 * be the same as the folded input (which TRIE_READ_CHAR returns),
2084 * except that when this is not in UTF-8, it could be matched
2085 * against a string which is UTF-8, and the variant characters
2086 * could be 2 bytes instead of the 1 here. Likewise, for the
2087 * minimum number of bytes when not folded. When folding, the min
2088 * is assumed to be 1 byte could fold to match the single character
2089 * here, or in the case of a multi-char fold, 1 byte can fold to
2090 * the whole sequence. 'foldlen' is used to denote whether we are
2091 * in such a sequence, skipping the min setting if so. XXX TODO
2092 * Use the exact list of what folds to each character, from
2093 * PL_utf8_foldclosures */
2095 maxbytes += UTF8SKIP(uc);
2097 /* A non-UTF-8 string could be 1 byte to match our 2 */
2098 minbytes += (UTF8_IS_DOWNGRADEABLE_START(*uc))
2104 foldlen -= UTF8SKIP(uc);
2107 foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e);
2113 maxbytes += (UNI_IS_INVARIANT(*uc))
2124 foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e);
2131 U8 folded= folder[ (U8) uvc ];
2132 if ( !trie->charmap[ folded ] ) {
2133 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2134 TRIE_STORE_REVCHAR( folded );
2137 if ( !trie->charmap[ uvc ] ) {
2138 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
2139 TRIE_STORE_REVCHAR( uvc );
2142 /* store the codepoint in the bitmap, and its folded
2144 TRIE_BITMAP_SET(trie, uvc);
2146 /* store the folded codepoint */
2147 if ( folder ) TRIE_BITMAP_SET(trie, folder[(U8) uvc ]);
2150 /* store first byte of utf8 representation of
2151 variant codepoints */
2152 if (! UVCHR_IS_INVARIANT(uvc)) {
2153 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
2156 set_bit = 0; /* We've done our bit :-) */
2161 widecharmap = newHV();
2163 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
2166 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
2168 if ( !SvTRUE( *svpp ) ) {
2169 sv_setiv( *svpp, ++trie->uniquecharcount );
2170 TRIE_STORE_REVCHAR(uvc);
2174 if( cur == first ) {
2175 trie->minlen = minbytes;
2176 trie->maxlen = maxbytes;
2177 } else if (minbytes < trie->minlen) {
2178 trie->minlen = minbytes;
2179 } else if (maxbytes > trie->maxlen) {
2180 trie->maxlen = maxbytes;
2182 } /* end first pass */
2183 DEBUG_TRIE_COMPILE_r(
2184 PerlIO_printf( Perl_debug_log,
2185 "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2186 (int)depth * 2 + 2,"",
2187 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
2188 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2189 (int)trie->minlen, (int)trie->maxlen )
2193 We now know what we are dealing with in terms of unique chars and
2194 string sizes so we can calculate how much memory a naive
2195 representation using a flat table will take. If it's over a reasonable
2196 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
2197 conservative but potentially much slower representation using an array
2200 At the end we convert both representations into the same compressed
2201 form that will be used in regexec.c for matching with. The latter
2202 is a form that cannot be used to construct with but has memory
2203 properties similar to the list form and access properties similar
2204 to the table form making it both suitable for fast searches and
2205 small enough that its feasable to store for the duration of a program.
2207 See the comment in the code where the compressed table is produced
2208 inplace from the flat tabe representation for an explanation of how
2209 the compression works.
2214 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2217 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2218 > SvIV(re_trie_maxbuff) )
2221 Second Pass -- Array Of Lists Representation
2223 Each state will be represented by a list of charid:state records
2224 (reg_trie_trans_le) the first such element holds the CUR and LEN
2225 points of the allocated array. (See defines above).
2227 We build the initial structure using the lists, and then convert
2228 it into the compressed table form which allows faster lookups
2229 (but cant be modified once converted).
2232 STRLEN transcount = 1;
2234 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2235 "%*sCompiling trie using list compiler\n",
2236 (int)depth * 2 + 2, ""));
2238 trie->states = (reg_trie_state *)
2239 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2240 sizeof(reg_trie_state) );
2244 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2246 regnode *noper = NEXTOPER( cur );
2247 U8 *uc = (U8*)STRING( noper );
2248 const U8 *e = uc + STR_LEN( noper );
2249 U32 state = 1; /* required init */
2250 U16 charid = 0; /* sanity init */
2251 U32 wordlen = 0; /* required init */
2253 if (OP(noper) == NOTHING) {
2254 regnode *noper_next= regnext(noper);
2255 if (noper_next != tail && OP(noper_next) == flags) {
2257 uc= (U8*)STRING(noper);
2258 e= uc + STR_LEN(noper);
2262 if (OP(noper) != NOTHING) {
2263 for ( ; uc < e ; uc += len ) {
2268 charid = trie->charmap[ uvc ];
2270 SV** const svpp = hv_fetch( widecharmap,
2277 charid=(U16)SvIV( *svpp );
2280 /* charid is now 0 if we dont know the char read, or
2281 * nonzero if we do */
2288 if ( !trie->states[ state ].trans.list ) {
2289 TRIE_LIST_NEW( state );
2292 check <= TRIE_LIST_USED( state );
2295 if ( TRIE_LIST_ITEM( state, check ).forid
2298 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2303 newstate = next_alloc++;
2304 prev_states[newstate] = state;
2305 TRIE_LIST_PUSH( state, charid, newstate );
2310 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2314 TRIE_HANDLE_WORD(state);
2316 } /* end second pass */
2318 /* next alloc is the NEXT state to be allocated */
2319 trie->statecount = next_alloc;
2320 trie->states = (reg_trie_state *)
2321 PerlMemShared_realloc( trie->states,
2323 * sizeof(reg_trie_state) );
2325 /* and now dump it out before we compress it */
2326 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2327 revcharmap, next_alloc,
2331 trie->trans = (reg_trie_trans *)
2332 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
2339 for( state=1 ; state < next_alloc ; state ++ ) {
2343 DEBUG_TRIE_COMPILE_MORE_r(
2344 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
2348 if (trie->states[state].trans.list) {
2349 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
2353 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2354 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
2355 if ( forid < minid ) {
2357 } else if ( forid > maxid ) {
2361 if ( transcount < tp + maxid - minid + 1) {
2363 trie->trans = (reg_trie_trans *)
2364 PerlMemShared_realloc( trie->trans,
2366 * sizeof(reg_trie_trans) );
2367 Zero( trie->trans + (transcount / 2),
2371 base = trie->uniquecharcount + tp - minid;
2372 if ( maxid == minid ) {
2374 for ( ; zp < tp ; zp++ ) {
2375 if ( ! trie->trans[ zp ].next ) {
2376 base = trie->uniquecharcount + zp - minid;
2377 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
2379 trie->trans[ zp ].check = state;
2385 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
2387 trie->trans[ tp ].check = state;
2392 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2393 const U32 tid = base
2394 - trie->uniquecharcount
2395 + TRIE_LIST_ITEM( state, idx ).forid;
2396 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
2398 trie->trans[ tid ].check = state;
2400 tp += ( maxid - minid + 1 );
2402 Safefree(trie->states[ state ].trans.list);
2405 DEBUG_TRIE_COMPILE_MORE_r(
2406 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
2409 trie->states[ state ].trans.base=base;
2411 trie->lasttrans = tp + 1;
2415 Second Pass -- Flat Table Representation.
2417 we dont use the 0 slot of either trans[] or states[] so we add 1 to
2418 each. We know that we will need Charcount+1 trans at most to store
2419 the data (one row per char at worst case) So we preallocate both
2420 structures assuming worst case.
2422 We then construct the trie using only the .next slots of the entry
2425 We use the .check field of the first entry of the node temporarily
2426 to make compression both faster and easier by keeping track of how
2427 many non zero fields are in the node.
2429 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
2432 There are two terms at use here: state as a TRIE_NODEIDX() which is
2433 a number representing the first entry of the node, and state as a
2434 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
2435 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
2436 if there are 2 entrys per node. eg:
2444 The table is internally in the right hand, idx form. However as we
2445 also have to deal with the states array which is indexed by nodenum
2446 we have to use TRIE_NODENUM() to convert.
2449 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
2450 "%*sCompiling trie using table compiler\n",
2451 (int)depth * 2 + 2, ""));
2453 trie->trans = (reg_trie_trans *)
2454 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
2455 * trie->uniquecharcount + 1,
2456 sizeof(reg_trie_trans) );
2457 trie->states = (reg_trie_state *)
2458 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2459 sizeof(reg_trie_state) );
2460 next_alloc = trie->uniquecharcount + 1;
2463 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2465 regnode *noper = NEXTOPER( cur );
2466 const U8 *uc = (U8*)STRING( noper );
2467 const U8 *e = uc + STR_LEN( noper );
2469 U32 state = 1; /* required init */
2471 U16 charid = 0; /* sanity init */
2472 U32 accept_state = 0; /* sanity init */
2474 U32 wordlen = 0; /* required init */
2476 if (OP(noper) == NOTHING) {
2477 regnode *noper_next= regnext(noper);
2478 if (noper_next != tail && OP(noper_next) == flags) {
2480 uc= (U8*)STRING(noper);
2481 e= uc + STR_LEN(noper);
2485 if ( OP(noper) != NOTHING ) {
2486 for ( ; uc < e ; uc += len ) {
2491 charid = trie->charmap[ uvc ];
2493 SV* const * const svpp = hv_fetch( widecharmap,
2497 charid = svpp ? (U16)SvIV(*svpp) : 0;
2501 if ( !trie->trans[ state + charid ].next ) {
2502 trie->trans[ state + charid ].next = next_alloc;
2503 trie->trans[ state ].check++;
2504 prev_states[TRIE_NODENUM(next_alloc)]
2505 = TRIE_NODENUM(state);
2506 next_alloc += trie->uniquecharcount;
2508 state = trie->trans[ state + charid ].next;
2510 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
2512 /* charid is now 0 if we dont know the char read, or
2513 * nonzero if we do */
2516 accept_state = TRIE_NODENUM( state );
2517 TRIE_HANDLE_WORD(accept_state);
2519 } /* end second pass */
2521 /* and now dump it out before we compress it */
2522 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
2524 next_alloc, depth+1));
2528 * Inplace compress the table.*
2530 For sparse data sets the table constructed by the trie algorithm will
2531 be mostly 0/FAIL transitions or to put it another way mostly empty.
2532 (Note that leaf nodes will not contain any transitions.)
2534 This algorithm compresses the tables by eliminating most such
2535 transitions, at the cost of a modest bit of extra work during lookup:
2537 - Each states[] entry contains a .base field which indicates the
2538 index in the state[] array wheres its transition data is stored.
2540 - If .base is 0 there are no valid transitions from that node.
2542 - If .base is nonzero then charid is added to it to find an entry in
2545 -If trans[states[state].base+charid].check!=state then the
2546 transition is taken to be a 0/Fail transition. Thus if there are fail
2547 transitions at the front of the node then the .base offset will point
2548 somewhere inside the previous nodes data (or maybe even into a node
2549 even earlier), but the .check field determines if the transition is
2553 The following process inplace converts the table to the compressed
2554 table: We first do not compress the root node 1,and mark all its
2555 .check pointers as 1 and set its .base pointer as 1 as well. This
2556 allows us to do a DFA construction from the compressed table later,
2557 and ensures that any .base pointers we calculate later are greater
2560 - We set 'pos' to indicate the first entry of the second node.
2562 - We then iterate over the columns of the node, finding the first and
2563 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
2564 and set the .check pointers accordingly, and advance pos
2565 appropriately and repreat for the next node. Note that when we copy
2566 the next pointers we have to convert them from the original
2567 NODEIDX form to NODENUM form as the former is not valid post
2570 - If a node has no transitions used we mark its base as 0 and do not
2571 advance the pos pointer.
2573 - If a node only has one transition we use a second pointer into the
2574 structure to fill in allocated fail transitions from other states.
2575 This pointer is independent of the main pointer and scans forward
2576 looking for null transitions that are allocated to a state. When it
2577 finds one it writes the single transition into the "hole". If the
2578 pointer doesnt find one the single transition is appended as normal.
2580 - Once compressed we can Renew/realloc the structures to release the
2583 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
2584 specifically Fig 3.47 and the associated pseudocode.
2588 const U32 laststate = TRIE_NODENUM( next_alloc );
2591 trie->statecount = laststate;
2593 for ( state = 1 ; state < laststate ; state++ ) {
2595 const U32 stateidx = TRIE_NODEIDX( state );
2596 const U32 o_used = trie->trans[ stateidx ].check;
2597 U32 used = trie->trans[ stateidx ].check;
2598 trie->trans[ stateidx ].check = 0;
2601 used && charid < trie->uniquecharcount;
2604 if ( flag || trie->trans[ stateidx + charid ].next ) {
2605 if ( trie->trans[ stateidx + charid ].next ) {
2607 for ( ; zp < pos ; zp++ ) {
2608 if ( ! trie->trans[ zp ].next ) {
2612 trie->states[ state ].trans.base
2614 + trie->uniquecharcount
2616 trie->trans[ zp ].next
2617 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
2619 trie->trans[ zp ].check = state;
2620 if ( ++zp > pos ) pos = zp;
2627 trie->states[ state ].trans.base
2628 = pos + trie->uniquecharcount - charid ;
2630 trie->trans[ pos ].next
2631 = SAFE_TRIE_NODENUM(
2632 trie->trans[ stateidx + charid ].next );
2633 trie->trans[ pos ].check = state;
2638 trie->lasttrans = pos + 1;
2639 trie->states = (reg_trie_state *)
2640 PerlMemShared_realloc( trie->states, laststate
2641 * sizeof(reg_trie_state) );
2642 DEBUG_TRIE_COMPILE_MORE_r(
2643 PerlIO_printf( Perl_debug_log,
2644 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
2645 (int)depth * 2 + 2,"",
2646 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
2650 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
2653 } /* end table compress */
2655 DEBUG_TRIE_COMPILE_MORE_r(
2656 PerlIO_printf(Perl_debug_log,
2657 "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
2658 (int)depth * 2 + 2, "",
2659 (UV)trie->statecount,
2660 (UV)trie->lasttrans)
2662 /* resize the trans array to remove unused space */
2663 trie->trans = (reg_trie_trans *)
2664 PerlMemShared_realloc( trie->trans, trie->lasttrans
2665 * sizeof(reg_trie_trans) );
2667 { /* Modify the program and insert the new TRIE node */
2668 U8 nodetype =(U8)(flags & 0xFF);
2672 regnode *optimize = NULL;
2673 #ifdef RE_TRACK_PATTERN_OFFSETS
2676 U32 mjd_nodelen = 0;
2677 #endif /* RE_TRACK_PATTERN_OFFSETS */
2678 #endif /* DEBUGGING */
2680 This means we convert either the first branch or the first Exact,
2681 depending on whether the thing following (in 'last') is a branch
2682 or not and whther first is the startbranch (ie is it a sub part of
2683 the alternation or is it the whole thing.)
2684 Assuming its a sub part we convert the EXACT otherwise we convert
2685 the whole branch sequence, including the first.
2687 /* Find the node we are going to overwrite */
2688 if ( first != startbranch || OP( last ) == BRANCH ) {
2689 /* branch sub-chain */
2690 NEXT_OFF( first ) = (U16)(last - first);
2691 #ifdef RE_TRACK_PATTERN_OFFSETS
2693 mjd_offset= Node_Offset((convert));
2694 mjd_nodelen= Node_Length((convert));
2697 /* whole branch chain */
2699 #ifdef RE_TRACK_PATTERN_OFFSETS
2702 const regnode *nop = NEXTOPER( convert );
2703 mjd_offset= Node_Offset((nop));
2704 mjd_nodelen= Node_Length((nop));
2708 PerlIO_printf(Perl_debug_log,
2709 "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2710 (int)depth * 2 + 2, "",
2711 (UV)mjd_offset, (UV)mjd_nodelen)
2714 /* But first we check to see if there is a common prefix we can
2715 split out as an EXACT and put in front of the TRIE node. */
2716 trie->startstate= 1;
2717 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2719 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2723 const U32 base = trie->states[ state ].trans.base;
2725 if ( trie->states[state].wordnum )
2728 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2729 if ( ( base + ofs >= trie->uniquecharcount ) &&
2730 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2731 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2733 if ( ++count > 1 ) {
2734 SV **tmp = av_fetch( revcharmap, ofs, 0);
2735 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2736 if ( state == 1 ) break;
2738 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2740 PerlIO_printf(Perl_debug_log,
2741 "%*sNew Start State=%"UVuf" Class: [",
2742 (int)depth * 2 + 2, "",
2745 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2746 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2748 TRIE_BITMAP_SET(trie,*ch);
2750 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2752 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2756 TRIE_BITMAP_SET(trie,*ch);
2758 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2759 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2765 SV **tmp = av_fetch( revcharmap, idx, 0);
2767 char *ch = SvPV( *tmp, len );
2769 SV *sv=sv_newmortal();
2770 PerlIO_printf( Perl_debug_log,
2771 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2772 (int)depth * 2 + 2, "",
2774 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2775 PL_colors[0], PL_colors[1],
2776 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2777 PERL_PV_ESCAPE_FIRSTCHAR
2782 OP( convert ) = nodetype;
2783 str=STRING(convert);
2786 STR_LEN(convert) += len;
2792 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2797 trie->prefixlen = (state-1);
2799 regnode *n = convert+NODE_SZ_STR(convert);
2800 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2801 trie->startstate = state;
2802 trie->minlen -= (state - 1);
2803 trie->maxlen -= (state - 1);
2805 /* At least the UNICOS C compiler choked on this
2806 * being argument to DEBUG_r(), so let's just have
2809 #ifdef PERL_EXT_RE_BUILD
2815 regnode *fix = convert;
2816 U32 word = trie->wordcount;
2818 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2819 while( ++fix < n ) {
2820 Set_Node_Offset_Length(fix, 0, 0);
2823 SV ** const tmp = av_fetch( trie_words, word, 0 );
2825 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2826 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2828 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2836 NEXT_OFF(convert) = (U16)(tail - convert);
2837 DEBUG_r(optimize= n);
2843 if ( trie->maxlen ) {
2844 NEXT_OFF( convert ) = (U16)(tail - convert);
2845 ARG_SET( convert, data_slot );
2846 /* Store the offset to the first unabsorbed branch in
2847 jump[0], which is otherwise unused by the jump logic.
2848 We use this when dumping a trie and during optimisation. */
2850 trie->jump[0] = (U16)(nextbranch - convert);
2852 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2853 * and there is a bitmap
2854 * and the first "jump target" node we found leaves enough room
2855 * then convert the TRIE node into a TRIEC node, with the bitmap
2856 * embedded inline in the opcode - this is hypothetically faster.
2858 if ( !trie->states[trie->startstate].wordnum
2860 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2862 OP( convert ) = TRIEC;
2863 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2864 PerlMemShared_free(trie->bitmap);
2867 OP( convert ) = TRIE;
2869 /* store the type in the flags */
2870 convert->flags = nodetype;
2874 + regarglen[ OP( convert ) ];
2876 /* XXX We really should free up the resource in trie now,
2877 as we won't use them - (which resources?) dmq */
2879 /* needed for dumping*/
2880 DEBUG_r(if (optimize) {
2881 regnode *opt = convert;
2883 while ( ++opt < optimize) {
2884 Set_Node_Offset_Length(opt,0,0);
2887 Try to clean up some of the debris left after the
2890 while( optimize < jumper ) {
2891 mjd_nodelen += Node_Length((optimize));
2892 OP( optimize ) = OPTIMIZED;
2893 Set_Node_Offset_Length(optimize,0,0);
2896 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2898 } /* end node insert */
2900 /* Finish populating the prev field of the wordinfo array. Walk back
2901 * from each accept state until we find another accept state, and if
2902 * so, point the first word's .prev field at the second word. If the
2903 * second already has a .prev field set, stop now. This will be the
2904 * case either if we've already processed that word's accept state,
2905 * or that state had multiple words, and the overspill words were
2906 * already linked up earlier.
2913 for (word=1; word <= trie->wordcount; word++) {
2915 if (trie->wordinfo[word].prev)
2917 state = trie->wordinfo[word].accept;
2919 state = prev_states[state];
2922 prev = trie->states[state].wordnum;
2926 trie->wordinfo[word].prev = prev;
2928 Safefree(prev_states);
2932 /* and now dump out the compressed format */
2933 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2935 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2937 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2938 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2940 SvREFCNT_dec_NN(revcharmap);
2944 : trie->startstate>1
2950 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2952 /* The Trie is constructed and compressed now so we can build a fail array if
2955 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and
2957 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi,
2961 We find the fail state for each state in the trie, this state is the longest
2962 proper suffix of the current state's 'word' that is also a proper prefix of
2963 another word in our trie. State 1 represents the word '' and is thus the
2964 default fail state. This allows the DFA not to have to restart after its
2965 tried and failed a word at a given point, it simply continues as though it
2966 had been matching the other word in the first place.
2968 'abcdgu'=~/abcdefg|cdgu/
2969 When we get to 'd' we are still matching the first word, we would encounter
2970 'g' which would fail, which would bring us to the state representing 'd' in
2971 the second word where we would try 'g' and succeed, proceeding to match
2974 /* add a fail transition */
2975 const U32 trie_offset = ARG(source);
2976 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2978 const U32 ucharcount = trie->uniquecharcount;
2979 const U32 numstates = trie->statecount;
2980 const U32 ubound = trie->lasttrans + ucharcount;
2984 U32 base = trie->states[ 1 ].trans.base;
2987 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T"));
2988 GET_RE_DEBUG_FLAGS_DECL;
2990 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2992 PERL_UNUSED_ARG(depth);
2996 ARG_SET( stclass, data_slot );
2997 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2998 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2999 aho->trie=trie_offset;
3000 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
3001 Copy( trie->states, aho->states, numstates, reg_trie_state );
3002 Newxz( q, numstates, U32);
3003 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
3006 /* initialize fail[0..1] to be 1 so that we always have
3007 a valid final fail state */
3008 fail[ 0 ] = fail[ 1 ] = 1;
3010 for ( charid = 0; charid < ucharcount ; charid++ ) {
3011 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
3013 q[ q_write ] = newstate;
3014 /* set to point at the root */
3015 fail[ q[ q_write++ ] ]=1;
3018 while ( q_read < q_write) {
3019 const U32 cur = q[ q_read++ % numstates ];
3020 base = trie->states[ cur ].trans.base;
3022 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
3023 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
3025 U32 fail_state = cur;
3028 fail_state = fail[ fail_state ];
3029 fail_base = aho->states[ fail_state ].trans.base;
3030 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
3032 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
3033 fail[ ch_state ] = fail_state;
3034 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
3036 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
3038 q[ q_write++ % numstates] = ch_state;
3042 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
3043 when we fail in state 1, this allows us to use the
3044 charclass scan to find a valid start char. This is based on the principle
3045 that theres a good chance the string being searched contains lots of stuff
3046 that cant be a start char.
3048 fail[ 0 ] = fail[ 1 ] = 0;
3049 DEBUG_TRIE_COMPILE_r({
3050 PerlIO_printf(Perl_debug_log,
3051 "%*sStclass Failtable (%"UVuf" states): 0",
3052 (int)(depth * 2), "", (UV)numstates
3054 for( q_read=1; q_read<numstates; q_read++ ) {
3055 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
3057 PerlIO_printf(Perl_debug_log, "\n");
3060 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
3064 #define DEBUG_PEEP(str,scan,depth) \
3065 DEBUG_OPTIMISE_r({if (scan){ \
3066 SV * const mysv=sv_newmortal(); \
3067 regnode *Next = regnext(scan); \
3068 regprop(RExC_rx, mysv, scan); \
3069 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
3070 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
3071 Next ? (REG_NODE_NUM(Next)) : 0 ); \
3075 /* The below joins as many adjacent EXACTish nodes as possible into a single
3076 * one. The regop may be changed if the node(s) contain certain sequences that
3077 * require special handling. The joining is only done if:
3078 * 1) there is room in the current conglomerated node to entirely contain the
3080 * 2) they are the exact same node type
3082 * The adjacent nodes actually may be separated by NOTHING-kind nodes, and
3083 * these get optimized out
3085 * If a node is to match under /i (folded), the number of characters it matches
3086 * can be different than its character length if it contains a multi-character
3087 * fold. *min_subtract is set to the total delta of the input nodes.
3089 * And *has_exactf_sharp_s is set to indicate whether or not the node is EXACTF
3090 * and contains LATIN SMALL LETTER SHARP S
3092 * This is as good a place as any to discuss the design of handling these
3093 * multi-character fold sequences. It's been wrong in Perl for a very long
3094 * time. There are three code points in Unicode whose multi-character folds
3095 * were long ago discovered to mess things up. The previous designs for
3096 * dealing with these involved assigning a special node for them. This
3097 * approach doesn't always work, as evidenced by this example:
3098 * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches
3099 * Both sides fold to "sss", but if the pattern is parsed to create a node that
3100 * would match just the \xDF, it won't be able to handle the case where a
3101 * successful match would have to cross the node's boundary. The new approach
3102 * that hopefully generally solves the problem generates an EXACTFU_SS node
3103 * that is "sss" in this case.
3105 * It turns out that there are problems with all multi-character folds, and not
3106 * just these three. Now the code is general, for all such cases. The
3107 * approach taken is:
3108 * 1) This routine examines each EXACTFish node that could contain multi-
3109 * character folded sequences. Since a single character can fold into
3110 * such a sequence, the minimum match length for this node is less than
3111 * the number of characters in the node. This routine returns in
3112 * *min_subtract how much to subtract from the the actual length of the
3113 * string to get a real minimum match length; it is 0 if there are no
3114 * multi-char foldeds. This delta is used by the caller to adjust the min
3115 * length of the match, and the delta between min and max, so that the
3116 * optimizer doesn't reject these possibilities based on size constraints.
3117 * 2) For the sequence involving the Sharp s (\xDF), the node type EXACTFU_SS
3118 * is used for an EXACTFU node that contains at least one "ss" sequence in
3119 * it. For non-UTF-8 patterns and strings, this is the only case where
3120 * there is a possible fold length change. That means that a regular
3121 * EXACTFU node without UTF-8 involvement doesn't have to concern itself
3122 * with length changes, and so can be processed faster. regexec.c takes
3123 * advantage of this. Generally, an EXACTFish node that is in UTF-8 is
3124 * pre-folded by regcomp.c. This saves effort in regex matching.
3125 * However, the pre-folding isn't done for non-UTF8 patterns because the
3126 * fold of the MICRO SIGN requires UTF-8, and we don't want to slow things
3127 * down by forcing the pattern into UTF8 unless necessary. Also what
3128 * EXACTF and EXACTFL nodes fold to isn't known until runtime. The fold
3129 * possibilities for the non-UTF8 patterns are quite simple, except for
3130 * the sharp s. All the ones that don't involve a UTF-8 target string are
3131 * members of a fold-pair, and arrays are set up for all of them so that
3132 * the other member of the pair can be found quickly. Code elsewhere in
3133 * this file makes sure that in EXACTFU nodes, the sharp s gets folded to
3134 * 'ss', even if the pattern isn't UTF-8. This avoids the issues
3135 * described in the next item.
3136 * 3) A problem remains for the sharp s in EXACTF and EXACTFA nodes when the
3137 * pattern isn't in UTF-8. (BTW, there cannot be an EXACTF node with a
3138 * UTF-8 pattern.) An assumption that the optimizer part of regexec.c
3139 * (probably unwittingly, in Perl_regexec_flags()) makes is that a
3140 * character in the pattern corresponds to at most a single character in
3141 * the target string. (And I do mean character, and not byte here, unlike
3142 * other parts of the documentation that have never been updated to
3143 * account for multibyte Unicode.) sharp s in EXACTF nodes can match the
3144 * two character string 'ss'; in EXACTFA nodes it can match
3145 * "\x{17F}\x{17F}". These violate the assumption, and they are the only
3146 * instances where it is violated. I'm reluctant to try to change the
3147 * assumption, as the code involved is impenetrable to me (khw), so
3148 * instead the code here punts. This routine examines (when the pattern
3149 * isn't UTF-8) EXACTF and EXACTFA nodes for the sharp s, and returns a
3150 * boolean indicating whether or not the node contains a sharp s. When it
3151 * is true, the caller sets a flag that later causes the optimizer in this
3152 * file to not set values for the floating and fixed string lengths, and
3153 * thus avoids the optimizer code in regexec.c that makes the invalid
3154 * assumption. Thus, there is no optimization based on string lengths for
3155 * non-UTF8-pattern EXACTF and EXACTFA nodes that contain the sharp s.
3156 * (The reason the assumption is wrong only in these two cases is that all
3157 * other non-UTF-8 folds are 1-1; and, for UTF-8 patterns, we pre-fold all
3158 * other folds to their expanded versions. We can't prefold sharp s to
3159 * 'ss' in EXACTF nodes because we don't know at compile time if it
3160 * actually matches 'ss' or not. It will match iff the target string is
3161 * in UTF-8, unlike the EXACTFU nodes, where it always matches; and
3162 * EXACTFA and EXACTFL where it never does. In an EXACTFA node in a UTF-8
3163 * pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the problem;
3164 * but in a non-UTF8 pattern, folding it to that above-Latin1 string would
3165 * require the pattern to be forced into UTF-8, the overhead of which we
3168 * Similarly, the code that generates tries doesn't currently handle
3169 * not-already-folded multi-char folds, and it looks like a pain to change
3170 * that. Therefore, trie generation of EXACTFA nodes with the sharp s
3171 * doesn't work. Instead, such an EXACTFA is turned into a new regnode,
3172 * EXACTFA_NO_TRIE, which the trie code knows not to handle. Most people
3173 * using /iaa matching will be doing so almost entirely with ASCII
3174 * strings, so this should rarely be encountered in practice */
3176 #define JOIN_EXACT(scan,min_subtract,has_exactf_sharp_s, flags) \
3177 if (PL_regkind[OP(scan)] == EXACT) \
3178 join_exact(pRExC_state,(scan),(min_subtract),has_exactf_sharp_s, (flags),NULL,depth+1)
3181 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
3182 UV *min_subtract, bool *has_exactf_sharp_s,
3183 U32 flags,regnode *val, U32 depth)
3185 /* Merge several consecutive EXACTish nodes into one. */
3186 regnode *n = regnext(scan);
3188 regnode *next = scan + NODE_SZ_STR(scan);
3192 regnode *stop = scan;
3193 GET_RE_DEBUG_FLAGS_DECL;
3195 PERL_UNUSED_ARG(depth);
3198 PERL_ARGS_ASSERT_JOIN_EXACT;
3199 #ifndef EXPERIMENTAL_INPLACESCAN
3200 PERL_UNUSED_ARG(flags);
3201 PERL_UNUSED_ARG(val);
3203 DEBUG_PEEP("join",scan,depth);
3205 /* Look through the subsequent nodes in the chain. Skip NOTHING, merge
3206 * EXACT ones that are mergeable to the current one. */
3208 && (PL_regkind[OP(n)] == NOTHING
3209 || (stringok && OP(n) == OP(scan)))
3211 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX)
3214 if (OP(n) == TAIL || n > next)
3216 if (PL_regkind[OP(n)] == NOTHING) {
3217 DEBUG_PEEP("skip:",n,depth);
3218 NEXT_OFF(scan) += NEXT_OFF(n);
3219 next = n + NODE_STEP_REGNODE;
3226 else if (stringok) {
3227 const unsigned int oldl = STR_LEN(scan);
3228 regnode * const nnext = regnext(n);
3230 /* XXX I (khw) kind of doubt that this works on platforms where
3231 * U8_MAX is above 255 because of lots of other assumptions */
3232 /* Don't join if the sum can't fit into a single node */
3233 if (oldl + STR_LEN(n) > U8_MAX)
3236 DEBUG_PEEP("merg",n,depth);
3239 NEXT_OFF(scan) += NEXT_OFF(n);
3240 STR_LEN(scan) += STR_LEN(n);
3241 next = n + NODE_SZ_STR(n);
3242 /* Now we can overwrite *n : */
3243 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
3251 #ifdef EXPERIMENTAL_INPLACESCAN
3252 if (flags && !NEXT_OFF(n)) {
3253 DEBUG_PEEP("atch", val, depth);
3254 if (reg_off_by_arg[OP(n)]) {
3255 ARG_SET(n, val - n);
3258 NEXT_OFF(n) = val - n;
3266 *has_exactf_sharp_s = FALSE;
3268 /* Here, all the adjacent mergeable EXACTish nodes have been merged. We
3269 * can now analyze for sequences of problematic code points. (Prior to
3270 * this final joining, sequences could have been split over boundaries, and
3271 * hence missed). The sequences only happen in folding, hence for any
3272 * non-EXACT EXACTish node */
3273 if (OP(scan) != EXACT) {
3274 const U8 * const s0 = (U8*) STRING(scan);
3276 const U8 * const s_end = s0 + STR_LEN(scan);
3278 /* One pass is made over the node's string looking for all the
3279 * possibilities. to avoid some tests in the loop, there are two main
3280 * cases, for UTF-8 patterns (which can't have EXACTF nodes) and
3284 /* Examine the string for a multi-character fold sequence. UTF-8
3285 * patterns have all characters pre-folded by the time this code is
3287 while (s < s_end - 1) /* Can stop 1 before the end, as minimum
3288 length sequence we are looking for is 2 */
3290 int count = 0; /* How many characters in a multi-char fold */
3291 int len = is_MULTI_CHAR_FOLD_utf8_safe(s, s_end);
3292 if (! len) { /* Not a multi-char fold: get next char */
3297 /* Nodes with 'ss' require special handling, except for EXACTFL
3298 * and EXACTFA-ish for which there is no multi-char fold to
3300 if (len == 2 && *s == 's' && *(s+1) == 's'
3301 && OP(scan) != EXACTFL
3302 && OP(scan) != EXACTFA
3303 && OP(scan) != EXACTFA_NO_TRIE)
3306 OP(scan) = EXACTFU_SS;
3309 else { /* Here is a generic multi-char fold. */
3310 const U8* multi_end = s + len;
3312 /* Count how many characters in it. In the case of /l and
3313 * /aa, no folds which contain ASCII code points are
3314 * allowed, so check for those, and skip if found. (In
3315 * EXACTFL, no folds are allowed to any Latin1 code point,
3316 * not just ASCII. But there aren't any of these
3317 * currently, nor ever likely, so don't take the time to
3318 * test for them. The code that generates the
3319 * is_MULTI_foo() macros croaks should one actually get put
3320 * into Unicode .) */
3321 if (OP(scan) != EXACTFL
3322 && OP(scan) != EXACTFA
3323 && OP(scan) != EXACTFA_NO_TRIE)
3325 count = utf8_length(s, multi_end);
3329 while (s < multi_end) {
3332 goto next_iteration;
3342 /* The delta is how long the sequence is minus 1 (1 is how long
3343 * the character that folds to the sequence is) */
3344 *min_subtract += count - 1;
3348 else if (OP(scan) == EXACTFA) {
3350 /* Non-UTF-8 pattern, EXACTFA node. There can't be a multi-char
3351 * fold to the ASCII range (and there are no existing ones in the
3352 * upper latin1 range). But, as outlined in the comments preceding
3353 * this function, we need to flag any occurrences of the sharp s.
3354 * This character forbids trie formation (because of added
3357 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
3358 OP(scan) = EXACTFA_NO_TRIE;
3359 *has_exactf_sharp_s = TRUE;
3366 else if (OP(scan) != EXACTFL) {
3368 /* Non-UTF-8 pattern, not EXACTFA nor EXACTFL node. Look for the
3369 * multi-char folds that are all Latin1. (This code knows that
3370 * there are no current multi-char folds possible with EXACTFL,
3371 * relying on fold_grind.t to catch any errors if the very unlikely
3372 * event happens that some get added in future Unicode versions.)
3373 * As explained in the comments preceding this function, we look
3374 * also for the sharp s in EXACTF nodes; it can be in the final
3375 * position. Otherwise we can stop looking 1 byte earlier because
3376 * have to find at least two characters for a multi-fold */
3377 const U8* upper = (OP(scan) == EXACTF) ? s_end : s_end -1;
3380 int len = is_MULTI_CHAR_FOLD_latin1_safe(s, s_end);
3381 if (! len) { /* Not a multi-char fold. */
3382 if (*s == LATIN_SMALL_LETTER_SHARP_S && OP(scan) == EXACTF)
3384 *has_exactf_sharp_s = TRUE;
3391 && isARG2_lower_or_UPPER_ARG1('s', *s)
3392 && isARG2_lower_or_UPPER_ARG1('s', *(s+1)))
3395 /* EXACTF nodes need to know that the minimum length
3396 * changed so that a sharp s in the string can match this
3397 * ss in the pattern, but they remain EXACTF nodes, as they
3398 * won't match this unless the target string is is UTF-8,
3399 * which we don't know until runtime */
3400 if (OP(scan) != EXACTF) {
3401 OP(scan) = EXACTFU_SS;
3405 *min_subtract += len - 1;
3412 /* Allow dumping but overwriting the collection of skipped
3413 * ops and/or strings with fake optimized ops */
3414 n = scan + NODE_SZ_STR(scan);
3422 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
3426 /* REx optimizer. Converts nodes into quicker variants "in place".
3427 Finds fixed substrings. */
3429 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
3430 to the position after last scanned or to NULL. */
3432 #define INIT_AND_WITHP \
3433 assert(!and_withp); \
3434 Newx(and_withp,1, regnode_ssc); \
3435 SAVEFREEPV(and_withp)
3437 /* this is a chain of data about sub patterns we are processing that
3438 need to be handled separately/specially in study_chunk. Its so
3439 we can simulate recursion without losing state. */
3441 typedef struct scan_frame {
3442 regnode *last; /* last node to process in this frame */
3443 regnode *next; /* next node to process when last is reached */
3444 struct scan_frame *prev; /*previous frame*/
3445 U32 prev_recursed_depth;
3446 I32 stop; /* what stopparen do we use */
3450 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
3453 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
3454 SSize_t *minlenp, SSize_t *deltap,
3459 regnode_ssc *and_withp,
3460 U32 flags, U32 depth)
3461 /* scanp: Start here (read-write). */
3462 /* deltap: Write maxlen-minlen here. */
3463 /* last: Stop before this one. */
3464 /* data: string data about the pattern */
3465 /* stopparen: treat close N as END */
3466 /* recursed: which subroutines have we recursed into */
3467 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
3470 /* There must be at least this number of characters to match */
3473 regnode *scan = *scanp, *next;
3475 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
3476 int is_inf_internal = 0; /* The studied chunk is infinite */
3477 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
3478 scan_data_t data_fake;
3479 SV *re_trie_maxbuff = NULL;
3480 regnode *first_non_open = scan;
3481 SSize_t stopmin = SSize_t_MAX;
3482 scan_frame *frame = NULL;
3483 GET_RE_DEBUG_FLAGS_DECL;
3485 PERL_ARGS_ASSERT_STUDY_CHUNK;
3488 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
3491 while (first_non_open && OP(first_non_open) == OPEN)
3492 first_non_open=regnext(first_non_open);
3497 while ( scan && OP(scan) != END && scan < last ){
3498 UV min_subtract = 0; /* How mmany chars to subtract from the minimum
3499 node length to get a real minimum (because
3500 the folded version may be shorter) */
3501 bool has_exactf_sharp_s = FALSE;
3502 /* Peephole optimizer: */
3503 DEBUG_OPTIMISE_MORE_r(
3505 PerlIO_printf(Perl_debug_log,
3506 "%*sstudy_chunk stopparen=%ld depth=%lu recursed_depth=%lu ",
3507 ((int) depth*2), "", (long)stopparen,
3508 (unsigned long)depth, (unsigned long)recursed_depth);
3509 if (recursed_depth) {
3512 for ( j = 0 ; j < recursed_depth ; j++ ) {
3513 PerlIO_printf(Perl_debug_log,"[");
3514 for ( i = 0 ; i < (U32)RExC_npar ; i++ )
3515 PerlIO_printf(Perl_debug_log,"%d",
3516 PAREN_TEST(RExC_study_chunk_recursed +
3517 (j * RExC_study_chunk_recursed_bytes), i)
3520 PerlIO_printf(Perl_debug_log,"]");
3523 PerlIO_printf(Perl_debug_log,"\n");
3526 DEBUG_STUDYDATA("Peep:", data, depth);
3527 DEBUG_PEEP("Peep", scan, depth);
3530 /* Its not clear to khw or hv why this is done here, and not in the
3531 * clauses that deal with EXACT nodes. khw's guess is that it's
3532 * because of a previous design */
3533 JOIN_EXACT(scan,&min_subtract, &has_exactf_sharp_s, 0);
3535 /* Follow the next-chain of the current node and optimize
3536 away all the NOTHINGs from it. */
3537 if (OP(scan) != CURLYX) {
3538 const int max = (reg_off_by_arg[OP(scan)]
3540 /* I32 may be smaller than U16 on CRAYs! */
3541 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
3542 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
3546 /* Skip NOTHING and LONGJMP. */
3547 while ((n = regnext(n))
3548 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
3549 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
3550 && off + noff < max)
3552 if (reg_off_by_arg[OP(scan)])
3555 NEXT_OFF(scan) = off;
3560 /* The principal pseudo-switch. Cannot be a switch, since we
3561 look into several different things. */
3562 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
3563 || OP(scan) == IFTHEN) {
3564 next = regnext(scan);
3566 /* demq: the op(next)==code check is to see if we have
3567 * "branch-branch" AFAICT */
3569 if (OP(next) == code || code == IFTHEN) {
3570 /* NOTE - There is similar code to this block below for
3571 * handling TRIE nodes on a re-study. If you change stuff here
3572 * check there too. */
3573 SSize_t max1 = 0, min1 = SSize_t_MAX, num = 0;
3575 regnode * const startbranch=scan;
3577 if (flags & SCF_DO_SUBSTR)
3578 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge
3581 if (flags & SCF_DO_STCLASS)
3582 ssc_init_zero(pRExC_state, &accum);
3584 while (OP(scan) == code) {
3585 SSize_t deltanext, minnext, fake;
3587 regnode_ssc this_class;
3590 data_fake.flags = 0;
3592 data_fake.whilem_c = data->whilem_c;
3593 data_fake.last_closep = data->last_closep;
3596 data_fake.last_closep = &fake;
3598 data_fake.pos_delta = delta;
3599 next = regnext(scan);
3600 scan = NEXTOPER(scan);
3602 scan = NEXTOPER(scan);
3603 if (flags & SCF_DO_STCLASS) {
3604 ssc_init(pRExC_state, &this_class);
3605 data_fake.start_class = &this_class;
3606 f = SCF_DO_STCLASS_AND;
3608 if (flags & SCF_WHILEM_VISITED_POS)
3609 f |= SCF_WHILEM_VISITED_POS;
3611 /* we suppose the run is continuous, last=next...*/
3612 minnext = study_chunk(pRExC_state, &scan, minlenp,
3613 &deltanext, next, &data_fake, stopparen,
3614 recursed_depth, NULL, f,depth+1);
3617 if (deltanext == SSize_t_MAX) {
3618 is_inf = is_inf_internal = 1;
3620 } else if (max1 < minnext + deltanext)
3621 max1 = minnext + deltanext;
3623 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3625 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3626 if ( stopmin > minnext)
3627 stopmin = min + min1;
3628 flags &= ~SCF_DO_SUBSTR;
3630 data->flags |= SCF_SEEN_ACCEPT;
3633 if (data_fake.flags & SF_HAS_EVAL)
3634 data->flags |= SF_HAS_EVAL;
3635 data->whilem_c = data_fake.whilem_c;
3637 if (flags & SCF_DO_STCLASS)
3638 ssc_or(pRExC_state, &accum, &this_class);
3640 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
3642 if (flags & SCF_DO_SUBSTR) {
3643 data->pos_min += min1;
3644 if (data->pos_delta >= SSize_t_MAX - (max1 - min1))
3645 data->pos_delta = SSize_t_MAX;
3647 data->pos_delta += max1 - min1;
3648 if (max1 != min1 || is_inf)
3649 data->longest = &(data->longest_float);
3652 if (delta == SSize_t_MAX
3653 || SSize_t_MAX - delta - (max1 - min1) < 0)
3654 delta = SSize_t_MAX;
3656 delta += max1 - min1;
3657 if (flags & SCF_DO_STCLASS_OR) {
3658 ssc_or(pRExC_state, data->start_class, &accum);
3660 ssc_and(pRExC_state, data->start_class, and_withp);
3661 flags &= ~SCF_DO_STCLASS;
3664 else if (flags & SCF_DO_STCLASS_AND) {
3666 ssc_and(pRExC_state, data->start_class, &accum);
3667 flags &= ~SCF_DO_STCLASS;
3670 /* Switch to OR mode: cache the old value of
3671 * data->start_class */
3673 StructCopy(data->start_class, and_withp, regnode_ssc);
3674 flags &= ~SCF_DO_STCLASS_AND;
3675 StructCopy(&accum, data->start_class, regnode_ssc);
3676 flags |= SCF_DO_STCLASS_OR;
3680 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch )
3685 Assuming this was/is a branch we are dealing with: 'scan'
3686 now points at the item that follows the branch sequence,
3687 whatever it is. We now start at the beginning of the
3688 sequence and look for subsequences of
3694 which would be constructed from a pattern like
3697 If we can find such a subsequence we need to turn the first
3698 element into a trie and then add the subsequent branch exact
3699 strings to the trie.
3703 1. patterns where the whole set of branches can be
3706 2. patterns where only a subset can be converted.
3708 In case 1 we can replace the whole set with a single regop
3709 for the trie. In case 2 we need to keep the start and end
3712 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
3713 becomes BRANCH TRIE; BRANCH X;
3715 There is an additional case, that being where there is a
3716 common prefix, which gets split out into an EXACT like node
3717 preceding the TRIE node.
3719 If x(1..n)==tail then we can do a simple trie, if not we make
3720 a "jump" trie, such that when we match the appropriate word
3721 we "jump" to the appropriate tail node. Essentially we turn
3722 a nested if into a case structure of sorts.
3727 if (!re_trie_maxbuff) {
3728 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
3729 if (!SvIOK(re_trie_maxbuff))
3730 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
3732 if ( SvIV(re_trie_maxbuff)>=0 ) {
3734 regnode *first = (regnode *)NULL;
3735 regnode *last = (regnode *)NULL;
3736 regnode *tail = scan;
3741 SV * const mysv = sv_newmortal(); /* for dumping */
3743 /* var tail is used because there may be a TAIL
3744 regop in the way. Ie, the exacts will point to the
3745 thing following the TAIL, but the last branch will
3746 point at the TAIL. So we advance tail. If we
3747 have nested (?:) we may have to move through several
3751 while ( OP( tail ) == TAIL ) {
3752 /* this is the TAIL generated by (?:) */
3753 tail = regnext( tail );
3757 DEBUG_TRIE_COMPILE_r({
3758 regprop(RExC_rx, mysv, tail );
3759 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
3760 (int)depth * 2 + 2, "",
3761 "Looking for TRIE'able sequences. Tail node is: ",
3762 SvPV_nolen_const( mysv )
3768 Step through the branches
3769 cur represents each branch,
3770 noper is the first thing to be matched as part
3772 noper_next is the regnext() of that node.
3774 We normally handle a case like this
3775 /FOO[xyz]|BAR[pqr]/ via a "jump trie" but we also
3776 support building with NOJUMPTRIE, which restricts
3777 the trie logic to structures like /FOO|BAR/.
3779 If noper is a trieable nodetype then the branch is
3780 a possible optimization target. If we are building
3781 under NOJUMPTRIE then we require that noper_next is
3782 the same as scan (our current position in the regex
3785 Once we have two or more consecutive such branches
3786 we can create a trie of the EXACT's contents and
3787 stitch it in place into the program.
3789 If the sequence represents all of the branches in
3790 the alternation we replace the entire thing with a
3793 Otherwise when it is a subsequence we need to
3794 stitch it in place and replace only the relevant
3795 branches. This means the first branch has to remain
3796 as it is used by the alternation logic, and its
3797 next pointer, and needs to be repointed at the item
3798 on the branch chain following the last branch we
3799 have optimized away.
3801 This could be either a BRANCH, in which case the
3802 subsequence is internal, or it could be the item
3803 following the branch sequence in which case the
3804 subsequence is at the end (which does not
3805 necessarily mean the first node is the start of the
3808 TRIE_TYPE(X) is a define which maps the optype to a
3812 ----------------+-----------
3816 EXACTFU_SS | EXACTFU
3821 #define TRIE_TYPE(X) ( ( NOTHING == (X) ) ? NOTHING : \
3822 ( EXACT == (X) ) ? EXACT : \
3823 ( EXACTFU == (X) || EXACTFU_SS == (X) ) ? EXACTFU : \
3824 ( EXACTFA == (X) ) ? EXACTFA : \
3827 /* dont use tail as the end marker for this traverse */
3828 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
3829 regnode * const noper = NEXTOPER( cur );
3830 U8 noper_type = OP( noper );
3831 U8 noper_trietype = TRIE_TYPE( noper_type );
3832 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
3833 regnode * const noper_next = regnext( noper );
3834 U8 noper_next_type = (noper_next && noper_next != tail) ? OP(noper_next) : 0;
3835 U8 noper_next_trietype = (noper_next && noper_next != tail) ? TRIE_TYPE( noper_next_type ) :0;
3838 DEBUG_TRIE_COMPILE_r({
3839 regprop(RExC_rx, mysv, cur);
3840 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
3841 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
3843 regprop(RExC_rx, mysv, noper);
3844 PerlIO_printf( Perl_debug_log, " -> %s",
3845 SvPV_nolen_const(mysv));
3848 regprop(RExC_rx, mysv, noper_next );
3849 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
3850 SvPV_nolen_const(mysv));
3852 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d,tt==%s,nt==%s,nnt==%s)\n",
3853 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur),
3854 PL_reg_name[trietype], PL_reg_name[noper_trietype], PL_reg_name[noper_next_trietype]
3858 /* Is noper a trieable nodetype that can be merged
3859 * with the current trie (if there is one)? */
3863 ( noper_trietype == NOTHING)
3864 || ( trietype == NOTHING )
3865 || ( trietype == noper_trietype )
3868 && noper_next == tail
3872 /* Handle mergable triable node Either we are
3873 * the first node in a new trieable sequence,
3874 * in which case we do some bookkeeping,
3875 * otherwise we update the end pointer. */
3878 if ( noper_trietype == NOTHING ) {
3879 #if !defined(DEBUGGING) && !defined(NOJUMPTRIE)
3880 regnode * const noper_next = regnext( noper );
3881 U8 noper_next_type = (noper_next && noper_next!=tail) ? OP(noper_next) : 0;
3882 U8 noper_next_trietype = noper_next_type ? TRIE_TYPE( noper_next_type ) :0;
3885 if ( noper_next_trietype ) {
3886 trietype = noper_next_trietype;
3887 } else if (noper_next_type) {
3888 /* a NOTHING regop is 1 regop wide.
3889 * We need at least two for a trie
3890 * so we can't merge this in */
3894 trietype = noper_trietype;
3897 if ( trietype == NOTHING )
3898 trietype = noper_trietype;
3903 } /* end handle mergable triable node */
3905 /* handle unmergable node -
3906 * noper may either be a triable node which can
3907 * not be tried together with the current trie,
3908 * or a non triable node */
3910 /* If last is set and trietype is not
https://perl5.git.perl.org / perl5.git / blob