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_inline.h"
90 #include "invlist_inline.h"
91 #include "unicode_constants.h"
93 #define HAS_NONLATIN1_FOLD_CLOSURE(i) \
94 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
95 #define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
96 _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
97 #define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
98 #define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
101 #define STATIC static
104 /* this is a chain of data about sub patterns we are processing that
105 need to be handled separately/specially in study_chunk. Its so
106 we can simulate recursion without losing state. */
108 typedef struct scan_frame {
109 regnode *last_regnode; /* last node to process in this frame */
110 regnode *next_regnode; /* next node to process when last is reached */
111 U32 prev_recursed_depth;
112 I32 stopparen; /* what stopparen do we use */
113 U32 is_top_frame; /* what flags do we use? */
115 struct scan_frame *this_prev_frame; /* this previous frame */
116 struct scan_frame *prev_frame; /* previous frame */
117 struct scan_frame *next_frame; /* next frame */
120 /* Certain characters are output as a sequence with the first being a
122 #define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
125 struct RExC_state_t {
126 U32 flags; /* RXf_* are we folding, multilining? */
127 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
128 char *precomp; /* uncompiled string. */
129 char *precomp_end; /* pointer to end of uncompiled string. */
130 REGEXP *rx_sv; /* The SV that is the regexp. */
131 regexp *rx; /* perl core regexp structure */
132 regexp_internal *rxi; /* internal data for regexp object
134 char *start; /* Start of input for compile */
135 char *end; /* End of input for compile */
136 char *parse; /* Input-scan pointer. */
137 char *adjusted_start; /* 'start', adjusted. See code use */
138 STRLEN precomp_adj; /* an offset beyond precomp. See code use */
139 SSize_t whilem_seen; /* number of WHILEM in this expr */
140 regnode *emit_start; /* Start of emitted-code area */
141 regnode *emit_bound; /* First regnode outside of the
143 regnode *emit; /* Code-emit pointer; if = &emit_dummy,
144 implies compiling, so don't emit */
145 regnode_ssc emit_dummy; /* placeholder for emit to point to;
146 large enough for the largest
147 non-EXACTish node, so can use it as
149 I32 naughty; /* How bad is this pattern? */
150 I32 sawback; /* Did we see \1, ...? */
152 SSize_t size; /* Code size. */
153 I32 npar; /* Capture buffer count, (OPEN) plus
154 one. ("par" 0 is the whole
156 I32 nestroot; /* root parens we are in - used by
160 regnode **open_parens; /* pointers to open parens */
161 regnode **close_parens; /* pointers to close parens */
162 regnode *end_op; /* END node in program */
163 I32 utf8; /* whether the pattern is utf8 or not */
164 I32 orig_utf8; /* whether the pattern was originally in utf8 */
165 /* XXX use this for future optimisation of case
166 * where pattern must be upgraded to utf8. */
167 I32 uni_semantics; /* If a d charset modifier should use unicode
168 rules, even if the pattern is not in
170 HV *paren_names; /* Paren names */
172 regnode **recurse; /* Recurse regops */
173 I32 recurse_count; /* Number of recurse regops we have generated */
174 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
176 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
179 I32 override_recoding;
181 I32 recode_x_to_native;
183 I32 in_multi_char_class;
184 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
186 int code_index; /* next code_blocks[] slot */
187 SSize_t maxlen; /* mininum possible number of chars in string to match */
188 scan_frame *frame_head;
189 scan_frame *frame_last;
192 #ifdef ADD_TO_REGEXEC
193 char *starttry; /* -Dr: where regtry was called. */
194 #define RExC_starttry (pRExC_state->starttry)
196 SV *runtime_code_qr; /* qr with the runtime code blocks */
198 const char *lastparse;
200 AV *paren_name_list; /* idx -> name */
201 U32 study_chunk_recursed_count;
204 #define RExC_lastparse (pRExC_state->lastparse)
205 #define RExC_lastnum (pRExC_state->lastnum)
206 #define RExC_paren_name_list (pRExC_state->paren_name_list)
207 #define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
208 #define RExC_mysv (pRExC_state->mysv1)
209 #define RExC_mysv1 (pRExC_state->mysv1)
210 #define RExC_mysv2 (pRExC_state->mysv2)
213 bool seen_unfolded_sharp_s;
218 #define RExC_flags (pRExC_state->flags)
219 #define RExC_pm_flags (pRExC_state->pm_flags)
220 #define RExC_precomp (pRExC_state->precomp)
221 #define RExC_precomp_adj (pRExC_state->precomp_adj)
222 #define RExC_adjusted_start (pRExC_state->adjusted_start)
223 #define RExC_precomp_end (pRExC_state->precomp_end)
224 #define RExC_rx_sv (pRExC_state->rx_sv)
225 #define RExC_rx (pRExC_state->rx)
226 #define RExC_rxi (pRExC_state->rxi)
227 #define RExC_start (pRExC_state->start)
228 #define RExC_end (pRExC_state->end)
229 #define RExC_parse (pRExC_state->parse)
230 #define RExC_whilem_seen (pRExC_state->whilem_seen)
232 /* Set during the sizing pass when there is a LATIN SMALL LETTER SHARP S in any
233 * EXACTF node, hence was parsed under /di rules. If later in the parse,
234 * something forces the pattern into using /ui rules, the sharp s should be
235 * folded into the sequence 'ss', which takes up more space than previously
236 * calculated. This means that the sizing pass needs to be restarted. (The
237 * node also becomes an EXACTFU_SS.) For all other characters, an EXACTF node
238 * that gets converted to /ui (and EXACTFU) occupies the same amount of space,
239 * so there is no need to resize [perl #125990]. */
240 #define RExC_seen_unfolded_sharp_s (pRExC_state->seen_unfolded_sharp_s)
242 #ifdef RE_TRACK_PATTERN_OFFSETS
243 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the
246 #define RExC_emit (pRExC_state->emit)
247 #define RExC_emit_dummy (pRExC_state->emit_dummy)
248 #define RExC_emit_start (pRExC_state->emit_start)
249 #define RExC_emit_bound (pRExC_state->emit_bound)
250 #define RExC_sawback (pRExC_state->sawback)
251 #define RExC_seen (pRExC_state->seen)
252 #define RExC_size (pRExC_state->size)
253 #define RExC_maxlen (pRExC_state->maxlen)
254 #define RExC_npar (pRExC_state->npar)
255 #define RExC_nestroot (pRExC_state->nestroot)
256 #define RExC_extralen (pRExC_state->extralen)
257 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
258 #define RExC_utf8 (pRExC_state->utf8)
259 #define RExC_uni_semantics (pRExC_state->uni_semantics)
260 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
261 #define RExC_open_parens (pRExC_state->open_parens)
262 #define RExC_close_parens (pRExC_state->close_parens)
263 #define RExC_end_op (pRExC_state->end_op)
264 #define RExC_paren_names (pRExC_state->paren_names)
265 #define RExC_recurse (pRExC_state->recurse)
266 #define RExC_recurse_count (pRExC_state->recurse_count)
267 #define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
268 #define RExC_study_chunk_recursed_bytes \
269 (pRExC_state->study_chunk_recursed_bytes)
270 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
271 #define RExC_contains_locale (pRExC_state->contains_locale)
273 # define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
275 #define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
276 #define RExC_frame_head (pRExC_state->frame_head)
277 #define RExC_frame_last (pRExC_state->frame_last)
278 #define RExC_frame_count (pRExC_state->frame_count)
279 #define RExC_strict (pRExC_state->strict)
280 #define RExC_study_started (pRExC_state->study_started)
281 #define RExC_warn_text (pRExC_state->warn_text)
283 /* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
284 * a flag to disable back-off on the fixed/floating substrings - if it's
285 * a high complexity pattern we assume the benefit of avoiding a full match
286 * is worth the cost of checking for the substrings even if they rarely help.
288 #define RExC_naughty (pRExC_state->naughty)
289 #define TOO_NAUGHTY (10)
290 #define MARK_NAUGHTY(add) \
291 if (RExC_naughty < TOO_NAUGHTY) \
292 RExC_naughty += (add)
293 #define MARK_NAUGHTY_EXP(exp, add) \
294 if (RExC_naughty < TOO_NAUGHTY) \
295 RExC_naughty += RExC_naughty / (exp) + (add)
297 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
298 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
299 ((*s) == '{' && regcurly(s)))
302 * Flags to be passed up and down.
304 #define WORST 0 /* Worst case. */
305 #define HASWIDTH 0x01 /* Known to match non-null strings. */
307 /* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
308 * character. (There needs to be a case: in the switch statement in regexec.c
309 * for any node marked SIMPLE.) Note that this is not the same thing as
312 #define SPSTART 0x04 /* Starts with * or + */
313 #define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
314 #define TRYAGAIN 0x10 /* Weeded out a declaration. */
315 #define RESTART_PASS1 0x20 /* Need to restart sizing pass */
316 #define NEED_UTF8 0x40 /* In conjunction with RESTART_PASS1, need to
317 calcuate sizes as UTF-8 */
319 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
321 /* whether trie related optimizations are enabled */
322 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
323 #define TRIE_STUDY_OPT
324 #define FULL_TRIE_STUDY
330 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
331 #define PBITVAL(paren) (1 << ((paren) & 7))
332 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
333 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
334 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
336 #define REQUIRE_UTF8(flagp) STMT_START { \
339 *flagp = RESTART_PASS1|NEED_UTF8; \
344 /* Change from /d into /u rules, and restart the parse if we've already seen
345 * something whose size would increase as a result, by setting *flagp and
346 * returning 'restart_retval'. RExC_uni_semantics is a flag that indicates
347 * we've change to /u during the parse. */
348 #define REQUIRE_UNI_RULES(flagp, restart_retval) \
350 if (DEPENDS_SEMANTICS) { \
352 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \
353 RExC_uni_semantics = 1; \
354 if (RExC_seen_unfolded_sharp_s) { \
355 *flagp |= RESTART_PASS1; \
356 return restart_retval; \
361 /* This converts the named class defined in regcomp.h to its equivalent class
362 * number defined in handy.h. */
363 #define namedclass_to_classnum(class) ((int) ((class) / 2))
364 #define classnum_to_namedclass(classnum) ((classnum) * 2)
366 #define _invlist_union_complement_2nd(a, b, output) \
367 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
368 #define _invlist_intersection_complement_2nd(a, b, output) \
369 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
371 /* About scan_data_t.
373 During optimisation we recurse through the regexp program performing
374 various inplace (keyhole style) optimisations. In addition study_chunk
375 and scan_commit populate this data structure with information about
376 what strings MUST appear in the pattern. We look for the longest
377 string that must appear at a fixed location, and we look for the
378 longest string that may appear at a floating location. So for instance
383 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
384 strings (because they follow a .* construct). study_chunk will identify
385 both FOO and BAR as being the longest fixed and floating strings respectively.
387 The strings can be composites, for instance
391 will result in a composite fixed substring 'foo'.
393 For each string some basic information is maintained:
396 This is the position the string must appear at, or not before.
397 It also implicitly (when combined with minlenp) tells us how many
398 characters must match before the string we are searching for.
399 Likewise when combined with minlenp and the length of the string it
400 tells us how many characters must appear after the string we have
404 Only used for floating strings. This is the rightmost point that
405 the string can appear at. If set to SSize_t_MAX it indicates that the
406 string can occur infinitely far to the right.
407 For fixed strings, it is equal to min_offset.
410 A pointer to the minimum number of characters of the pattern that the
411 string was found inside. This is important as in the case of positive
412 lookahead or positive lookbehind we can have multiple patterns
417 The minimum length of the pattern overall is 3, the minimum length
418 of the lookahead part is 3, but the minimum length of the part that
419 will actually match is 1. So 'FOO's minimum length is 3, but the
420 minimum length for the F is 1. This is important as the minimum length
421 is used to determine offsets in front of and behind the string being
422 looked for. Since strings can be composites this is the length of the
423 pattern at the time it was committed with a scan_commit. Note that
424 the length is calculated by study_chunk, so that the minimum lengths
425 are not known until the full pattern has been compiled, thus the
426 pointer to the value.
430 In the case of lookbehind the string being searched for can be
431 offset past the start point of the final matching string.
432 If this value was just blithely removed from the min_offset it would
433 invalidate some of the calculations for how many chars must match
434 before or after (as they are derived from min_offset and minlen and
435 the length of the string being searched for).
436 When the final pattern is compiled and the data is moved from the
437 scan_data_t structure into the regexp structure the information
438 about lookbehind is factored in, with the information that would
439 have been lost precalculated in the end_shift field for the
442 The fields pos_min and pos_delta are used to store the minimum offset
443 and the delta to the maximum offset at the current point in the pattern.
447 struct scan_data_substrs {
448 SV *str; /* longest substring found in pattern */
449 SSize_t min_offset; /* earliest point in string it can appear */
450 SSize_t max_offset; /* latest point in string it can appear */
451 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
452 SSize_t lookbehind; /* is the pos of the string modified by LB */
453 I32 flags; /* per substring SF_* and SCF_* flags */
456 typedef struct scan_data_t {
457 /*I32 len_min; unused */
458 /*I32 len_delta; unused */
462 SSize_t last_end; /* min value, <0 unless valid. */
463 SSize_t last_start_min;
464 SSize_t last_start_max;
465 U8 cur_is_floating; /* whether the last_* values should be set as
466 * the next fixed (0) or floating (1)
469 /* [0] is longest fixed substring so far, [1] is longest float so far */
470 struct scan_data_substrs substrs[2];
472 I32 flags; /* common SF_* and SCF_* flags */
474 SSize_t *last_closep;
475 regnode_ssc *start_class;
479 * Forward declarations for pregcomp()'s friends.
482 static const scan_data_t zero_scan_data = {
483 0, 0, NULL, 0, 0, 0, 0,
485 { NULL, 0, 0, 0, 0, 0 },
486 { NULL, 0, 0, 0, 0, 0 },
493 #define SF_BEFORE_SEOL 0x0001
494 #define SF_BEFORE_MEOL 0x0002
495 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
497 #define SF_IS_INF 0x0040
498 #define SF_HAS_PAR 0x0080
499 #define SF_IN_PAR 0x0100
500 #define SF_HAS_EVAL 0x0200
503 /* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
504 * longest substring in the pattern. When it is not set the optimiser keeps
505 * track of position, but does not keep track of the actual strings seen,
507 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
510 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
511 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
512 * turned off because of the alternation (BRANCH). */
513 #define SCF_DO_SUBSTR 0x0400
515 #define SCF_DO_STCLASS_AND 0x0800
516 #define SCF_DO_STCLASS_OR 0x1000
517 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
518 #define SCF_WHILEM_VISITED_POS 0x2000
520 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
521 #define SCF_SEEN_ACCEPT 0x8000
522 #define SCF_TRIE_DOING_RESTUDY 0x10000
523 #define SCF_IN_DEFINE 0x20000
528 #define UTF cBOOL(RExC_utf8)
530 /* The enums for all these are ordered so things work out correctly */
531 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
532 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
533 == REGEX_DEPENDS_CHARSET)
534 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
535 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
536 >= REGEX_UNICODE_CHARSET)
537 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
538 == REGEX_ASCII_RESTRICTED_CHARSET)
539 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
540 >= REGEX_ASCII_RESTRICTED_CHARSET)
541 #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
542 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
544 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
546 /* For programs that want to be strictly Unicode compatible by dying if any
547 * attempt is made to match a non-Unicode code point against a Unicode
549 #define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
551 #define OOB_NAMEDCLASS -1
553 /* There is no code point that is out-of-bounds, so this is problematic. But
554 * its only current use is to initialize a variable that is always set before
556 #define OOB_UNICODE 0xDEADBEEF
558 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
561 /* length of regex to show in messages that don't mark a position within */
562 #define RegexLengthToShowInErrorMessages 127
565 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
566 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
567 * op/pragma/warn/regcomp.
569 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
570 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
572 #define REPORT_LOCATION " in regex; marked by " MARKER1 \
573 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
575 /* The code in this file in places uses one level of recursion with parsing
576 * rebased to an alternate string constructed by us in memory. This can take
577 * the form of something that is completely different from the input, or
578 * something that uses the input as part of the alternate. In the first case,
579 * there should be no possibility of an error, as we are in complete control of
580 * the alternate string. But in the second case we don't control the input
581 * portion, so there may be errors in that. Here's an example:
583 * is handled specially because \x{df} folds to a sequence of more than one
584 * character, 'ss'. What is done is to create and parse an alternate string,
585 * which looks like this:
586 * /(?:\x{DF}|[abc\x{DF}def])/ui
587 * where it uses the input unchanged in the middle of something it constructs,
588 * which is a branch for the DF outside the character class, and clustering
589 * parens around the whole thing. (It knows enough to skip the DF inside the
590 * class while in this substitute parse.) 'abc' and 'def' may have errors that
591 * need to be reported. The general situation looks like this:
594 * Input: ----------------------------------------------------
595 * Constructed: ---------------------------------------------------
598 * The input string sI..eI is the input pattern. The string sC..EC is the
599 * constructed substitute parse string. The portions sC..tC and eC..EC are
600 * constructed by us. The portion tC..eC is an exact duplicate of the input
601 * pattern tI..eI. In the diagram, these are vertically aligned. Suppose that
602 * while parsing, we find an error at xC. We want to display a message showing
603 * the real input string. Thus we need to find the point xI in it which
604 * corresponds to xC. xC >= tC, since the portion of the string sC..tC has
605 * been constructed by us, and so shouldn't have errors. We get:
607 * xI = sI + (tI - sI) + (xC - tC)
609 * and, the offset into sI is:
611 * (xI - sI) = (tI - sI) + (xC - tC)
613 * When the substitute is constructed, we save (tI -sI) as RExC_precomp_adj,
614 * and we save tC as RExC_adjusted_start.
616 * During normal processing of the input pattern, everything points to that,
617 * with RExC_precomp_adj set to 0, and RExC_adjusted_start set to sI.
620 #define tI_sI RExC_precomp_adj
621 #define tC RExC_adjusted_start
622 #define sC RExC_precomp
623 #define xI_offset(xC) ((IV) (tI_sI + (xC - tC)))
624 #define xI(xC) (sC + xI_offset(xC))
625 #define eC RExC_precomp_end
627 #define REPORT_LOCATION_ARGS(xC) \
629 (xI(xC) > eC) /* Don't run off end */ \
630 ? eC - sC /* Length before the <--HERE */ \
631 : ( __ASSERT_(xI_offset(xC) >= 0) xI_offset(xC) ), \
632 sC), /* The input pattern printed up to the <--HERE */ \
634 (xI(xC) > eC) ? 0 : eC - xI(xC), /* Length after <--HERE */ \
635 (xI(xC) > eC) ? eC : xI(xC)) /* pattern after <--HERE */
637 /* Used to point after bad bytes for an error message, but avoid skipping
638 * past a nul byte. */
639 #define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
642 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
643 * arg. Show regex, up to a maximum length. If it's too long, chop and add
646 #define _FAIL(code) STMT_START { \
647 const char *ellipses = ""; \
648 IV len = RExC_precomp_end - RExC_precomp; \
651 SAVEFREESV(RExC_rx_sv); \
652 if (len > RegexLengthToShowInErrorMessages) { \
653 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
654 len = RegexLengthToShowInErrorMessages - 10; \
660 #define FAIL(msg) _FAIL( \
661 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
662 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
664 #define FAIL2(msg,arg) _FAIL( \
665 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
666 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
669 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
671 #define Simple_vFAIL(m) STMT_START { \
672 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
673 m, REPORT_LOCATION_ARGS(RExC_parse)); \
677 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
679 #define vFAIL(m) STMT_START { \
681 SAVEFREESV(RExC_rx_sv); \
686 * Like Simple_vFAIL(), but accepts two arguments.
688 #define Simple_vFAIL2(m,a1) STMT_START { \
689 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
690 REPORT_LOCATION_ARGS(RExC_parse)); \
694 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
696 #define vFAIL2(m,a1) STMT_START { \
698 SAVEFREESV(RExC_rx_sv); \
699 Simple_vFAIL2(m, a1); \
704 * Like Simple_vFAIL(), but accepts three arguments.
706 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
707 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
708 REPORT_LOCATION_ARGS(RExC_parse)); \
712 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
714 #define vFAIL3(m,a1,a2) STMT_START { \
716 SAVEFREESV(RExC_rx_sv); \
717 Simple_vFAIL3(m, a1, a2); \
721 * Like Simple_vFAIL(), but accepts four arguments.
723 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
724 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
725 REPORT_LOCATION_ARGS(RExC_parse)); \
728 #define vFAIL4(m,a1,a2,a3) STMT_START { \
730 SAVEFREESV(RExC_rx_sv); \
731 Simple_vFAIL4(m, a1, a2, a3); \
734 /* A specialized version of vFAIL2 that works with UTF8f */
735 #define vFAIL2utf8f(m, a1) STMT_START { \
737 SAVEFREESV(RExC_rx_sv); \
738 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
739 REPORT_LOCATION_ARGS(RExC_parse)); \
742 #define vFAIL3utf8f(m, a1, a2) STMT_START { \
744 SAVEFREESV(RExC_rx_sv); \
745 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
746 REPORT_LOCATION_ARGS(RExC_parse)); \
749 /* These have asserts in them because of [perl #122671] Many warnings in
750 * regcomp.c can occur twice. If they get output in pass1 and later in that
751 * pass, the pattern has to be converted to UTF-8 and the pass restarted, they
752 * would get output again. So they should be output in pass2, and these
753 * asserts make sure new warnings follow that paradigm. */
755 /* m is not necessarily a "literal string", in this macro */
756 #define reg_warn_non_literal_string(loc, m) STMT_START { \
757 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
758 "%s" REPORT_LOCATION, \
759 m, REPORT_LOCATION_ARGS(loc)); \
762 #define ckWARNreg(loc,m) STMT_START { \
763 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
765 REPORT_LOCATION_ARGS(loc)); \
768 #define vWARN(loc, m) STMT_START { \
769 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
771 REPORT_LOCATION_ARGS(loc)); \
774 #define vWARN_dep(loc, m) STMT_START { \
775 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
777 REPORT_LOCATION_ARGS(loc)); \
780 #define ckWARNdep(loc,m) STMT_START { \
781 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
783 REPORT_LOCATION_ARGS(loc)); \
786 #define ckWARNregdep(loc,m) STMT_START { \
787 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
790 REPORT_LOCATION_ARGS(loc)); \
793 #define ckWARN2reg_d(loc,m, a1) STMT_START { \
794 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
796 a1, REPORT_LOCATION_ARGS(loc)); \
799 #define ckWARN2reg(loc, m, a1) STMT_START { \
800 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
802 a1, REPORT_LOCATION_ARGS(loc)); \
805 #define vWARN3(loc, m, a1, a2) STMT_START { \
806 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
808 a1, a2, REPORT_LOCATION_ARGS(loc)); \
811 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
812 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
815 REPORT_LOCATION_ARGS(loc)); \
818 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
819 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
822 REPORT_LOCATION_ARGS(loc)); \
825 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
826 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
829 REPORT_LOCATION_ARGS(loc)); \
832 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
833 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
836 REPORT_LOCATION_ARGS(loc)); \
839 /* Macros for recording node offsets. 20001227 mjd@plover.com
840 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
841 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
842 * Element 0 holds the number n.
843 * Position is 1 indexed.
845 #ifndef RE_TRACK_PATTERN_OFFSETS
846 #define Set_Node_Offset_To_R(node,byte)
847 #define Set_Node_Offset(node,byte)
848 #define Set_Cur_Node_Offset
849 #define Set_Node_Length_To_R(node,len)
850 #define Set_Node_Length(node,len)
851 #define Set_Node_Cur_Length(node,start)
852 #define Node_Offset(n)
853 #define Node_Length(n)
854 #define Set_Node_Offset_Length(node,offset,len)
855 #define ProgLen(ri) ri->u.proglen
856 #define SetProgLen(ri,x) ri->u.proglen = x
858 #define ProgLen(ri) ri->u.offsets[0]
859 #define SetProgLen(ri,x) ri->u.offsets[0] = x
860 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
862 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
863 __LINE__, (int)(node), (int)(byte))); \
865 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
868 RExC_offsets[2*(node)-1] = (byte); \
873 #define Set_Node_Offset(node,byte) \
874 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
875 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
877 #define Set_Node_Length_To_R(node,len) STMT_START { \
879 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
880 __LINE__, (int)(node), (int)(len))); \
882 Perl_croak(aTHX_ "value of node is %d in Length macro", \
885 RExC_offsets[2*(node)] = (len); \
890 #define Set_Node_Length(node,len) \
891 Set_Node_Length_To_R((node)-RExC_emit_start, len)
892 #define Set_Node_Cur_Length(node, start) \
893 Set_Node_Length(node, RExC_parse - start)
895 /* Get offsets and lengths */
896 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
897 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
899 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
900 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
901 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
905 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
906 #define EXPERIMENTAL_INPLACESCAN
907 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
911 Perl_re_printf(pTHX_ const char *fmt, ...)
915 PerlIO *f= Perl_debug_log;
916 PERL_ARGS_ASSERT_RE_PRINTF;
918 result = PerlIO_vprintf(f, fmt, ap);
924 Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
928 PerlIO *f= Perl_debug_log;
929 PERL_ARGS_ASSERT_RE_INDENTF;
931 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
932 result = PerlIO_vprintf(f, fmt, ap);
936 #endif /* DEBUGGING */
938 #define DEBUG_RExC_seen() \
939 DEBUG_OPTIMISE_MORE_r({ \
940 Perl_re_printf( aTHX_ "RExC_seen: "); \
942 if (RExC_seen & REG_ZERO_LEN_SEEN) \
943 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
945 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
946 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
948 if (RExC_seen & REG_GPOS_SEEN) \
949 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
951 if (RExC_seen & REG_RECURSE_SEEN) \
952 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
954 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
955 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
957 if (RExC_seen & REG_VERBARG_SEEN) \
958 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
960 if (RExC_seen & REG_CUTGROUP_SEEN) \
961 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
963 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
964 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
966 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
967 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
969 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
970 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
972 Perl_re_printf( aTHX_ "\n"); \
975 #define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
976 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
981 S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
982 const char *close_str)
987 Perl_re_printf( aTHX_ "%s", open_str);
988 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
989 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
990 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
991 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
992 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
993 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
994 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
995 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
996 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
997 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
998 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
999 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1000 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1001 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1002 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1003 Perl_re_printf( aTHX_ "%s", close_str);
1008 S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1009 U32 depth, int is_inf)
1011 GET_RE_DEBUG_FLAGS_DECL;
1013 DEBUG_OPTIMISE_MORE_r({
1016 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1020 (IV)data->pos_delta,
1024 S_debug_show_study_flags(aTHX_ data->flags," [","]");
1026 Perl_re_printf( aTHX_
1027 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1029 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1030 is_inf ? "INF " : ""
1033 if (data->last_found) {
1035 Perl_re_printf(aTHX_
1036 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1037 SvPVX_const(data->last_found),
1039 (IV)data->last_start_min,
1040 (IV)data->last_start_max
1043 for (i = 0; i < 2; i++) {
1044 Perl_re_printf(aTHX_
1045 " %s%s: '%s' @ %" IVdf "/%" IVdf,
1046 data->cur_is_floating == i ? "*" : "",
1047 i ? "Float" : "Fixed",
1048 SvPVX_const(data->substrs[i].str),
1049 (IV)data->substrs[i].min_offset,
1050 (IV)data->substrs[i].max_offset
1052 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1056 Perl_re_printf( aTHX_ "\n");
1062 S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1063 regnode *scan, U32 depth, U32 flags)
1065 GET_RE_DEBUG_FLAGS_DECL;
1072 Next = regnext(scan);
1073 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1074 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1077 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1078 Next ? (REG_NODE_NUM(Next)) : 0 );
1079 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1080 Perl_re_printf( aTHX_ "\n");
1085 # define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1086 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1088 # define DEBUG_PEEP(str, scan, depth, flags) \
1089 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1092 # define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1093 # define DEBUG_PEEP(str, scan, depth, flags) NOOP
1097 /* =========================================================
1098 * BEGIN edit_distance stuff.
1100 * This calculates how many single character changes of any type are needed to
1101 * transform a string into another one. It is taken from version 3.1 of
1103 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1106 /* Our unsorted dictionary linked list. */
1107 /* Note we use UVs, not chars. */
1112 struct dictionary* next;
1114 typedef struct dictionary item;
1117 PERL_STATIC_INLINE item*
1118 push(UV key,item* curr)
1121 Newxz(head, 1, item);
1129 PERL_STATIC_INLINE item*
1130 find(item* head, UV key)
1132 item* iterator = head;
1134 if (iterator->key == key){
1137 iterator = iterator->next;
1143 PERL_STATIC_INLINE item*
1144 uniquePush(item* head,UV key)
1146 item* iterator = head;
1149 if (iterator->key == key) {
1152 iterator = iterator->next;
1155 return push(key,head);
1158 PERL_STATIC_INLINE void
1159 dict_free(item* head)
1161 item* iterator = head;
1164 item* temp = iterator;
1165 iterator = iterator->next;
1172 /* End of Dictionary Stuff */
1174 /* All calculations/work are done here */
1176 S_edit_distance(const UV* src,
1178 const STRLEN x, /* length of src[] */
1179 const STRLEN y, /* length of tgt[] */
1180 const SSize_t maxDistance
1184 UV swapCount,swapScore,targetCharCount,i,j;
1186 UV score_ceil = x + y;
1188 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1190 /* intialize matrix start values */
1191 Newxz(scores, ( (x + 2) * (y + 2)), UV);
1192 scores[0] = score_ceil;
1193 scores[1 * (y + 2) + 0] = score_ceil;
1194 scores[0 * (y + 2) + 1] = score_ceil;
1195 scores[1 * (y + 2) + 1] = 0;
1196 head = uniquePush(uniquePush(head,src[0]),tgt[0]);
1201 for (i=1;i<=x;i++) {
1203 head = uniquePush(head,src[i]);
1204 scores[(i+1) * (y + 2) + 1] = i;
1205 scores[(i+1) * (y + 2) + 0] = score_ceil;
1208 for (j=1;j<=y;j++) {
1211 head = uniquePush(head,tgt[j]);
1212 scores[1 * (y + 2) + (j + 1)] = j;
1213 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1216 targetCharCount = find(head,tgt[j-1])->value;
1217 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1219 if (src[i-1] != tgt[j-1]){
1220 scores[(i+1) * (y + 2) + (j + 1)] = MIN(swapScore,(MIN(scores[i * (y + 2) + j], MIN(scores[(i+1) * (y + 2) + j], scores[i * (y + 2) + (j + 1)])) + 1));
1224 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1228 find(head,src[i-1])->value = i;
1232 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1235 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1239 /* END of edit_distance() stuff
1240 * ========================================================= */
1242 /* is c a control character for which we have a mnemonic? */
1243 #define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1246 S_cntrl_to_mnemonic(const U8 c)
1248 /* Returns the mnemonic string that represents character 'c', if one
1249 * exists; NULL otherwise. The only ones that exist for the purposes of
1250 * this routine are a few control characters */
1253 case '\a': return "\\a";
1254 case '\b': return "\\b";
1255 case ESC_NATIVE: return "\\e";
1256 case '\f': return "\\f";
1257 case '\n': return "\\n";
1258 case '\r': return "\\r";
1259 case '\t': return "\\t";
1265 /* Mark that we cannot extend a found fixed substring at this point.
1266 Update the longest found anchored substring or the longest found
1267 floating substrings if needed. */
1270 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1271 SSize_t *minlenp, int is_inf)
1273 const STRLEN l = CHR_SVLEN(data->last_found);
1274 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
1275 const STRLEN old_l = CHR_SVLEN(longest_sv);
1276 GET_RE_DEBUG_FLAGS_DECL;
1278 PERL_ARGS_ASSERT_SCAN_COMMIT;
1280 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
1281 const U8 i = data->cur_is_floating;
1282 SvSetMagicSV(longest_sv, data->last_found);
1283 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1286 data->substrs[0].max_offset = data->substrs[0].min_offset;
1288 data->substrs[1].max_offset = (l
1289 ? data->last_start_max
1290 : (data->pos_delta > SSize_t_MAX - data->pos_min
1292 : data->pos_min + data->pos_delta));
1294 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1295 data->substrs[1].max_offset = SSize_t_MAX;
1298 if (data->flags & SF_BEFORE_EOL)
1299 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1301 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1302 data->substrs[i].minlenp = minlenp;
1303 data->substrs[i].lookbehind = 0;
1306 SvCUR_set(data->last_found, 0);
1308 SV * const sv = data->last_found;
1309 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1310 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1315 data->last_end = -1;
1316 data->flags &= ~SF_BEFORE_EOL;
1317 DEBUG_STUDYDATA("commit", data, 0, is_inf);
1320 /* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1321 * list that describes which code points it matches */
1324 S_ssc_anything(pTHX_ regnode_ssc *ssc)
1326 /* Set the SSC 'ssc' to match an empty string or any code point */
1328 PERL_ARGS_ASSERT_SSC_ANYTHING;
1330 assert(is_ANYOF_SYNTHETIC(ssc));
1332 /* mortalize so won't leak */
1333 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
1334 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
1338 S_ssc_is_anything(const regnode_ssc *ssc)
1340 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1341 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1342 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1343 * in any way, so there's no point in using it */
1348 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
1350 assert(is_ANYOF_SYNTHETIC(ssc));
1352 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
1356 /* See if the list consists solely of the range 0 - Infinity */
1357 invlist_iterinit(ssc->invlist);
1358 ret = invlist_iternext(ssc->invlist, &start, &end)
1362 invlist_iterfinish(ssc->invlist);
1368 /* If e.g., both \w and \W are set, matches everything */
1369 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1371 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1372 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1382 S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
1384 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1385 * string, any code point, or any posix class under locale */
1387 PERL_ARGS_ASSERT_SSC_INIT;
1389 Zero(ssc, 1, regnode_ssc);
1390 set_ANYOF_SYNTHETIC(ssc);
1391 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
1394 /* If any portion of the regex is to operate under locale rules that aren't
1395 * fully known at compile time, initialization includes it. The reason
1396 * this isn't done for all regexes is that the optimizer was written under
1397 * the assumption that locale was all-or-nothing. Given the complexity and
1398 * lack of documentation in the optimizer, and that there are inadequate
1399 * test cases for locale, many parts of it may not work properly, it is
1400 * safest to avoid locale unless necessary. */
1401 if (RExC_contains_locale) {
1402 ANYOF_POSIXL_SETALL(ssc);
1405 ANYOF_POSIXL_ZERO(ssc);
1410 S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1411 const regnode_ssc *ssc)
1413 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1414 * to the list of code points matched, and locale posix classes; hence does
1415 * not check its flags) */
1420 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1422 assert(is_ANYOF_SYNTHETIC(ssc));
1424 invlist_iterinit(ssc->invlist);
1425 ret = invlist_iternext(ssc->invlist, &start, &end)
1429 invlist_iterfinish(ssc->invlist);
1435 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
1443 S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
1444 const regnode_charclass* const node)
1446 /* Returns a mortal inversion list defining which code points are matched
1447 * by 'node', which is of type ANYOF. Handles complementing the result if
1448 * appropriate. If some code points aren't knowable at this time, the
1449 * returned list must, and will, contain every code point that is a
1453 SV* only_utf8_locale_invlist = NULL;
1455 const U32 n = ARG(node);
1456 bool new_node_has_latin1 = FALSE;
1458 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1460 /* Look at the data structure created by S_set_ANYOF_arg() */
1461 if (n != ANYOF_ONLY_HAS_BITMAP) {
1462 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1463 AV * const av = MUTABLE_AV(SvRV(rv));
1464 SV **const ary = AvARRAY(av);
1465 assert(RExC_rxi->data->what[n] == 's');
1467 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
1468 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1])));
1470 else if (ary[0] && ary[0] != &PL_sv_undef) {
1472 /* Here, no compile-time swash, and there are things that won't be
1473 * known until runtime -- we have to assume it could be anything */
1474 invlist = sv_2mortal(_new_invlist(1));
1475 return _add_range_to_invlist(invlist, 0, UV_MAX);
1477 else if (ary[3] && ary[3] != &PL_sv_undef) {
1479 /* Here no compile-time swash, and no run-time only data. Use the
1480 * node's inversion list */
1481 invlist = sv_2mortal(invlist_clone(ary[3]));
1484 /* Get the code points valid only under UTF-8 locales */
1485 if ((ANYOF_FLAGS(node) & ANYOFL_FOLD)
1486 && ary[2] && ary[2] != &PL_sv_undef)
1488 only_utf8_locale_invlist = ary[2];
1493 invlist = sv_2mortal(_new_invlist(0));
1496 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1497 * code points, and an inversion list for the others, but if there are code
1498 * points that should match only conditionally on the target string being
1499 * UTF-8, those are placed in the inversion list, and not the bitmap.
1500 * Since there are circumstances under which they could match, they are
1501 * included in the SSC. But if the ANYOF node is to be inverted, we have
1502 * to exclude them here, so that when we invert below, the end result
1503 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1504 * have to do this here before we add the unconditionally matched code
1506 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1507 _invlist_intersection_complement_2nd(invlist,
1512 /* Add in the points from the bit map */
1513 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
1514 if (ANYOF_BITMAP_TEST(node, i)) {
1515 unsigned int start = i++;
1517 for (; i < NUM_ANYOF_CODE_POINTS && ANYOF_BITMAP_TEST(node, i); ++i) {
1520 invlist = _add_range_to_invlist(invlist, start, i-1);
1521 new_node_has_latin1 = TRUE;
1525 /* If this can match all upper Latin1 code points, have to add them
1526 * as well. But don't add them if inverting, as when that gets done below,
1527 * it would exclude all these characters, including the ones it shouldn't
1528 * that were added just above */
1529 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
1530 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1532 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1535 /* Similarly for these */
1536 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
1537 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
1540 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1541 _invlist_invert(invlist);
1543 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOFL_FOLD) {
1545 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1546 * locale. We can skip this if there are no 0-255 at all. */
1547 _invlist_union(invlist, PL_Latin1, &invlist);
1550 /* Similarly add the UTF-8 locale possible matches. These have to be
1551 * deferred until after the non-UTF-8 locale ones are taken care of just
1552 * above, or it leads to wrong results under ANYOF_INVERT */
1553 if (only_utf8_locale_invlist) {
1554 _invlist_union_maybe_complement_2nd(invlist,
1555 only_utf8_locale_invlist,
1556 ANYOF_FLAGS(node) & ANYOF_INVERT,
1563 /* These two functions currently do the exact same thing */
1564 #define ssc_init_zero ssc_init
1566 #define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1567 #define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1569 /* 'AND' a given class with another one. Can create false positives. 'ssc'
1570 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1571 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
1574 S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1575 const regnode_charclass *and_with)
1577 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1578 * another SSC or a regular ANYOF class. Can create false positives. */
1583 PERL_ARGS_ASSERT_SSC_AND;
1585 assert(is_ANYOF_SYNTHETIC(ssc));
1587 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1588 * the code point inversion list and just the relevant flags */
1589 if (is_ANYOF_SYNTHETIC(and_with)) {
1590 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
1591 anded_flags = ANYOF_FLAGS(and_with);
1593 /* XXX This is a kludge around what appears to be deficiencies in the
1594 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1595 * there are paths through the optimizer where it doesn't get weeded
1596 * out when it should. And if we don't make some extra provision for
1597 * it like the code just below, it doesn't get added when it should.
1598 * This solution is to add it only when AND'ing, which is here, and
1599 * only when what is being AND'ed is the pristine, original node
1600 * matching anything. Thus it is like adding it to ssc_anything() but
1601 * only when the result is to be AND'ed. Probably the same solution
1602 * could be adopted for the same problem we have with /l matching,
1603 * which is solved differently in S_ssc_init(), and that would lead to
1604 * fewer false positives than that solution has. But if this solution
1605 * creates bugs, the consequences are only that a warning isn't raised
1606 * that should be; while the consequences for having /l bugs is
1607 * incorrect matches */
1608 if (ssc_is_anything((regnode_ssc *)and_with)) {
1609 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
1613 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
1614 if (OP(and_with) == ANYOFD) {
1615 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1618 anded_flags = ANYOF_FLAGS(and_with)
1619 &( ANYOF_COMMON_FLAGS
1620 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1621 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
1622 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1624 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1629 ANYOF_FLAGS(ssc) &= anded_flags;
1631 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1632 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1633 * 'and_with' may be inverted. When not inverted, we have the situation of
1635 * (C1 | P1) & (C2 | P2)
1636 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1637 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1638 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1639 * <= ((C1 & C2) | P1 | P2)
1640 * Alternatively, the last few steps could be:
1641 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1642 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1643 * <= (C1 | C2 | (P1 & P2))
1644 * We favor the second approach if either P1 or P2 is non-empty. This is
1645 * because these components are a barrier to doing optimizations, as what
1646 * they match cannot be known until the moment of matching as they are
1647 * dependent on the current locale, 'AND"ing them likely will reduce or
1649 * But we can do better if we know that C1,P1 are in their initial state (a
1650 * frequent occurrence), each matching everything:
1651 * (<everything>) & (C2 | P2) = C2 | P2
1652 * Similarly, if C2,P2 are in their initial state (again a frequent
1653 * occurrence), the result is a no-op
1654 * (C1 | P1) & (<everything>) = C1 | P1
1657 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1658 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1659 * <= (C1 & ~C2) | (P1 & ~P2)
1662 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
1663 && ! is_ANYOF_SYNTHETIC(and_with))
1667 ssc_intersection(ssc,
1669 FALSE /* Has already been inverted */
1672 /* If either P1 or P2 is empty, the intersection will be also; can skip
1674 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
1675 ANYOF_POSIXL_ZERO(ssc);
1677 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1679 /* Note that the Posix class component P from 'and_with' actually
1681 * P = Pa | Pb | ... | Pn
1682 * where each component is one posix class, such as in [\w\s].
1684 * ~P = ~(Pa | Pb | ... | Pn)
1685 * = ~Pa & ~Pb & ... & ~Pn
1686 * <= ~Pa | ~Pb | ... | ~Pn
1687 * The last is something we can easily calculate, but unfortunately
1688 * is likely to have many false positives. We could do better
1689 * in some (but certainly not all) instances if two classes in
1690 * P have known relationships. For example
1691 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1693 * :lower: & :print: = :lower:
1694 * And similarly for classes that must be disjoint. For example,
1695 * since \s and \w can have no elements in common based on rules in
1696 * the POSIX standard,
1697 * \w & ^\S = nothing
1698 * Unfortunately, some vendor locales do not meet the Posix
1699 * standard, in particular almost everything by Microsoft.
1700 * The loop below just changes e.g., \w into \W and vice versa */
1702 regnode_charclass_posixl temp;
1703 int add = 1; /* To calculate the index of the complement */
1705 ANYOF_POSIXL_ZERO(&temp);
1706 for (i = 0; i < ANYOF_MAX; i++) {
1708 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1709 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
1711 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
1712 ANYOF_POSIXL_SET(&temp, i + add);
1714 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1716 ANYOF_POSIXL_AND(&temp, ssc);
1718 } /* else ssc already has no posixes */
1719 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1720 in its initial state */
1721 else if (! is_ANYOF_SYNTHETIC(and_with)
1722 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
1724 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1725 * copy it over 'ssc' */
1726 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
1727 if (is_ANYOF_SYNTHETIC(and_with)) {
1728 StructCopy(and_with, ssc, regnode_ssc);
1731 ssc->invlist = anded_cp_list;
1732 ANYOF_POSIXL_ZERO(ssc);
1733 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1734 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
1738 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
1739 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
1741 /* One or the other of P1, P2 is non-empty. */
1742 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1743 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1745 ssc_union(ssc, anded_cp_list, FALSE);
1747 else { /* P1 = P2 = empty */
1748 ssc_intersection(ssc, anded_cp_list, FALSE);
1754 S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1755 const regnode_charclass *or_with)
1757 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1758 * another SSC or a regular ANYOF class. Can create false positives if
1759 * 'or_with' is to be inverted. */
1764 PERL_ARGS_ASSERT_SSC_OR;
1766 assert(is_ANYOF_SYNTHETIC(ssc));
1768 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1769 * the code point inversion list and just the relevant flags */
1770 if (is_ANYOF_SYNTHETIC(or_with)) {
1771 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
1772 ored_flags = ANYOF_FLAGS(or_with);
1775 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
1776 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
1777 if (OP(or_with) != ANYOFD) {
1779 |= ANYOF_FLAGS(or_with)
1780 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1781 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
1782 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1784 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1789 ANYOF_FLAGS(ssc) |= ored_flags;
1791 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1792 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1793 * 'or_with' may be inverted. When not inverted, we have the simple
1794 * situation of computing:
1795 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1796 * If P1|P2 yields a situation with both a class and its complement are
1797 * set, like having both \w and \W, this matches all code points, and we
1798 * can delete these from the P component of the ssc going forward. XXX We
1799 * might be able to delete all the P components, but I (khw) am not certain
1800 * about this, and it is better to be safe.
1803 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1804 * <= (C1 | P1) | ~C2
1805 * <= (C1 | ~C2) | P1
1806 * (which results in actually simpler code than the non-inverted case)
1809 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
1810 && ! is_ANYOF_SYNTHETIC(or_with))
1812 /* We ignore P2, leaving P1 going forward */
1813 } /* else Not inverted */
1814 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
1815 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
1816 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1818 for (i = 0; i < ANYOF_MAX; i += 2) {
1819 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1821 ssc_match_all_cp(ssc);
1822 ANYOF_POSIXL_CLEAR(ssc, i);
1823 ANYOF_POSIXL_CLEAR(ssc, i+1);
1831 FALSE /* Already has been inverted */
1835 PERL_STATIC_INLINE void
1836 S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1838 PERL_ARGS_ASSERT_SSC_UNION;
1840 assert(is_ANYOF_SYNTHETIC(ssc));
1842 _invlist_union_maybe_complement_2nd(ssc->invlist,
1848 PERL_STATIC_INLINE void
1849 S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1851 const bool invert2nd)
1853 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1855 assert(is_ANYOF_SYNTHETIC(ssc));
1857 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1863 PERL_STATIC_INLINE void
1864 S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1866 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1868 assert(is_ANYOF_SYNTHETIC(ssc));
1870 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1873 PERL_STATIC_INLINE void
1874 S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1876 /* AND just the single code point 'cp' into the SSC 'ssc' */
1878 SV* cp_list = _new_invlist(2);
1880 PERL_ARGS_ASSERT_SSC_CP_AND;
1882 assert(is_ANYOF_SYNTHETIC(ssc));
1884 cp_list = add_cp_to_invlist(cp_list, cp);
1885 ssc_intersection(ssc, cp_list,
1886 FALSE /* Not inverted */
1888 SvREFCNT_dec_NN(cp_list);
1891 PERL_STATIC_INLINE void
1892 S_ssc_clear_locale(regnode_ssc *ssc)
1894 /* Set the SSC 'ssc' to not match any locale things */
1895 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1897 assert(is_ANYOF_SYNTHETIC(ssc));
1899 ANYOF_POSIXL_ZERO(ssc);
1900 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1903 #define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
1906 S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
1908 /* The synthetic start class is used to hopefully quickly winnow down
1909 * places where a pattern could start a match in the target string. If it
1910 * doesn't really narrow things down that much, there isn't much point to
1911 * having the overhead of using it. This function uses some very crude
1912 * heuristics to decide if to use the ssc or not.
1914 * It returns TRUE if 'ssc' rules out more than half what it considers to
1915 * be the "likely" possible matches, but of course it doesn't know what the
1916 * actual things being matched are going to be; these are only guesses
1918 * For /l matches, it assumes that the only likely matches are going to be
1919 * in the 0-255 range, uniformly distributed, so half of that is 127
1920 * For /a and /d matches, it assumes that the likely matches will be just
1921 * the ASCII range, so half of that is 63
1922 * For /u and there isn't anything matching above the Latin1 range, it
1923 * assumes that that is the only range likely to be matched, and uses
1924 * half that as the cut-off: 127. If anything matches above Latin1,
1925 * it assumes that all of Unicode could match (uniformly), except for
1926 * non-Unicode code points and things in the General Category "Other"
1927 * (unassigned, private use, surrogates, controls and formats). This
1928 * is a much large number. */
1930 U32 count = 0; /* Running total of number of code points matched by
1932 UV start, end; /* Start and end points of current range in inversion
1934 const U32 max_code_points = (LOC)
1936 : (( ! UNI_SEMANTICS
1937 || invlist_highest(ssc->invlist) < 256)
1940 const U32 max_match = max_code_points / 2;
1942 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
1944 invlist_iterinit(ssc->invlist);
1945 while (invlist_iternext(ssc->invlist, &start, &end)) {
1946 if (start >= max_code_points) {
1949 end = MIN(end, max_code_points - 1);
1950 count += end - start + 1;
1951 if (count >= max_match) {
1952 invlist_iterfinish(ssc->invlist);
1962 S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
1964 /* The inversion list in the SSC is marked mortal; now we need a more
1965 * permanent copy, which is stored the same way that is done in a regular
1966 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
1969 SV* invlist = invlist_clone(ssc->invlist);
1971 PERL_ARGS_ASSERT_SSC_FINALIZE;
1973 assert(is_ANYOF_SYNTHETIC(ssc));
1975 /* The code in this file assumes that all but these flags aren't relevant
1976 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
1977 * by the time we reach here */
1978 assert(! (ANYOF_FLAGS(ssc)
1979 & ~( ANYOF_COMMON_FLAGS
1980 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1981 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
1983 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
1985 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
1986 NULL, NULL, NULL, FALSE);
1988 /* Make sure is clone-safe */
1989 ssc->invlist = NULL;
1991 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1992 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
1995 if (RExC_contains_locale) {
1999 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
2002 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2003 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2004 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
2005 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2006 ? (TRIE_LIST_CUR( idx ) - 1) \
2012 dump_trie(trie,widecharmap,revcharmap)
2013 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2014 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
2016 These routines dump out a trie in a somewhat readable format.
2017 The _interim_ variants are used for debugging the interim
2018 tables that are used to generate the final compressed
2019 representation which is what dump_trie expects.
2021 Part of the reason for their existence is to provide a form
2022 of documentation as to how the different representations function.
2027 Dumps the final compressed table form of the trie to Perl_debug_log.
2028 Used for debugging make_trie().
2032 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2033 AV *revcharmap, U32 depth)
2036 SV *sv=sv_newmortal();
2037 int colwidth= widecharmap ? 6 : 4;
2039 GET_RE_DEBUG_FLAGS_DECL;
2041 PERL_ARGS_ASSERT_DUMP_TRIE;
2043 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
2044 depth+1, "Match","Base","Ofs" );
2046 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2047 SV ** const tmp = av_fetch( revcharmap, state, 0);
2049 Perl_re_printf( aTHX_ "%*s",
2051 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
2052 PL_colors[0], PL_colors[1],
2053 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2054 PERL_PV_ESCAPE_FIRSTCHAR
2059 Perl_re_printf( aTHX_ "\n");
2060 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
2062 for( state = 0 ; state < trie->uniquecharcount ; state++ )
2063 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2064 Perl_re_printf( aTHX_ "\n");
2066 for( state = 1 ; state < trie->statecount ; state++ ) {
2067 const U32 base = trie->states[ state ].trans.base;
2069 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
2071 if ( trie->states[ state ].wordnum ) {
2072 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
2074 Perl_re_printf( aTHX_ "%6s", "" );
2077 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
2082 while( ( base + ofs < trie->uniquecharcount ) ||
2083 ( base + ofs - trie->uniquecharcount < trie->lasttrans
2084 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2088 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
2090 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2091 if ( ( base + ofs >= trie->uniquecharcount )
2092 && ( base + ofs - trie->uniquecharcount
2094 && trie->trans[ base + ofs
2095 - trie->uniquecharcount ].check == state )
2097 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
2098 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2101 Perl_re_printf( aTHX_ "%*s",colwidth," ." );
2105 Perl_re_printf( aTHX_ "]");
2108 Perl_re_printf( aTHX_ "\n" );
2110 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
2112 for (word=1; word <= trie->wordcount; word++) {
2113 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2114 (int)word, (int)(trie->wordinfo[word].prev),
2115 (int)(trie->wordinfo[word].len));
2117 Perl_re_printf( aTHX_ "\n" );
2120 Dumps a fully constructed but uncompressed trie in list form.
2121 List tries normally only are used for construction when the number of
2122 possible chars (trie->uniquecharcount) is very high.
2123 Used for debugging make_trie().
2126 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2127 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2131 SV *sv=sv_newmortal();
2132 int colwidth= widecharmap ? 6 : 4;
2133 GET_RE_DEBUG_FLAGS_DECL;
2135 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2137 /* print out the table precompression. */
2138 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
2140 Perl_re_indentf( aTHX_ "%s",
2141 depth+1, "------:-----+-----------------\n" );
2143 for( state=1 ; state < next_alloc ; state ++ ) {
2146 Perl_re_indentf( aTHX_ " %4" UVXf " :",
2147 depth+1, (UV)state );
2148 if ( ! trie->states[ state ].wordnum ) {
2149 Perl_re_printf( aTHX_ "%5s| ","");
2151 Perl_re_printf( aTHX_ "W%4x| ",
2152 trie->states[ state ].wordnum
2155 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
2156 SV ** const tmp = av_fetch( revcharmap,
2157 TRIE_LIST_ITEM(state,charid).forid, 0);
2159 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
2161 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2163 PL_colors[0], PL_colors[1],
2164 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2165 | PERL_PV_ESCAPE_FIRSTCHAR
2167 TRIE_LIST_ITEM(state,charid).forid,
2168 (UV)TRIE_LIST_ITEM(state,charid).newstate
2171 Perl_re_printf( aTHX_ "\n%*s| ",
2172 (int)((depth * 2) + 14), "");
2175 Perl_re_printf( aTHX_ "\n");
2180 Dumps a fully constructed but uncompressed trie in table form.
2181 This is the normal DFA style state transition table, with a few
2182 twists to facilitate compression later.
2183 Used for debugging make_trie().
2186 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2187 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2192 SV *sv=sv_newmortal();
2193 int colwidth= widecharmap ? 6 : 4;
2194 GET_RE_DEBUG_FLAGS_DECL;
2196 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
2199 print out the table precompression so that we can do a visual check
2200 that they are identical.
2203 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
2205 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2206 SV ** const tmp = av_fetch( revcharmap, charid, 0);
2208 Perl_re_printf( aTHX_ "%*s",
2210 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
2211 PL_colors[0], PL_colors[1],
2212 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2213 PERL_PV_ESCAPE_FIRSTCHAR
2219 Perl_re_printf( aTHX_ "\n");
2220 Perl_re_indentf( aTHX_ "State+-", depth+1 );
2222 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
2223 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
2226 Perl_re_printf( aTHX_ "\n" );
2228 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2230 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
2232 (UV)TRIE_NODENUM( state ) );
2234 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2235 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2237 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
2239 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
2241 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
2242 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
2243 (UV)trie->trans[ state ].check );
2245 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
2246 (UV)trie->trans[ state ].check,
2247 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2255 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2256 startbranch: the first branch in the whole branch sequence
2257 first : start branch of sequence of branch-exact nodes.
2258 May be the same as startbranch
2259 last : Thing following the last branch.
2260 May be the same as tail.
2261 tail : item following the branch sequence
2262 count : words in the sequence
2263 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
2264 depth : indent depth
2266 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
2268 A trie is an N'ary tree where the branches are determined by digital
2269 decomposition of the key. IE, at the root node you look up the 1st character and
2270 follow that branch repeat until you find the end of the branches. Nodes can be
2271 marked as "accepting" meaning they represent a complete word. Eg:
2275 would convert into the following structure. Numbers represent states, letters
2276 following numbers represent valid transitions on the letter from that state, if
2277 the number is in square brackets it represents an accepting state, otherwise it
2278 will be in parenthesis.
2280 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2284 (1) +-i->(6)-+-s->[7]
2286 +-s->(3)-+-h->(4)-+-e->[5]
2288 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2290 This shows that when matching against the string 'hers' we will begin at state 1
2291 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2292 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2293 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2294 single traverse. We store a mapping from accepting to state to which word was
2295 matched, and then when we have multiple possibilities we try to complete the
2296 rest of the regex in the order in which they occurred in the alternation.
2298 The only prior NFA like behaviour that would be changed by the TRIE support is
2299 the silent ignoring of duplicate alternations which are of the form:
2301 / (DUPE|DUPE) X? (?{ ... }) Y /x
2303 Thus EVAL blocks following a trie may be called a different number of times with
2304 and without the optimisation. With the optimisations dupes will be silently
2305 ignored. This inconsistent behaviour of EVAL type nodes is well established as
2306 the following demonstrates:
2308 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2310 which prints out 'word' three times, but
2312 'words'=~/(word|word|word)(?{ print $1 })S/
2314 which doesnt print it out at all. This is due to other optimisations kicking in.
2316 Example of what happens on a structural level:
2318 The regexp /(ac|ad|ab)+/ will produce the following debug output:
2320 1: CURLYM[1] {1,32767}(18)
2331 This would be optimizable with startbranch=5, first=5, last=16, tail=16
2332 and should turn into:
2334 1: CURLYM[1] {1,32767}(18)
2336 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2344 Cases where tail != last would be like /(?foo|bar)baz/:
2354 which would be optimizable with startbranch=1, first=1, last=7, tail=8
2355 and would end up looking like:
2358 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2365 d = uvchr_to_utf8_flags(d, uv, 0);
2367 is the recommended Unicode-aware way of saying
2372 #define TRIE_STORE_REVCHAR(val) \
2375 SV *zlopp = newSV(UTF8_MAXBYTES); \
2376 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
2377 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
2378 SvCUR_set(zlopp, kapow - flrbbbbb); \
2381 av_push(revcharmap, zlopp); \
2383 char ooooff = (char)val; \
2384 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2388 /* This gets the next character from the input, folding it if not already
2390 #define TRIE_READ_CHAR STMT_START { \
2393 /* if it is UTF then it is either already folded, or does not need \
2395 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
2397 else if (folder == PL_fold_latin1) { \
2398 /* This folder implies Unicode rules, which in the range expressible \
2399 * by not UTF is the lower case, with the two exceptions, one of \
2400 * which should have been taken care of before calling this */ \
2401 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2402 uvc = toLOWER_L1(*uc); \
2403 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2406 /* raw data, will be folded later if needed */ \
2414 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2415 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
2416 U32 ging = TRIE_LIST_LEN( state ) * 2; \
2417 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
2418 TRIE_LIST_LEN( state ) = ging; \
2420 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2421 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2422 TRIE_LIST_CUR( state )++; \
2425 #define TRIE_LIST_NEW(state) STMT_START { \
2426 Newxz( trie->states[ state ].trans.list, \
2427 4, reg_trie_trans_le ); \
2428 TRIE_LIST_CUR( state ) = 1; \
2429 TRIE_LIST_LEN( state ) = 4; \
2432 #define TRIE_HANDLE_WORD(state) STMT_START { \
2433 U16 dupe= trie->states[ state ].wordnum; \
2434 regnode * const noper_next = regnext( noper ); \
2437 /* store the word for dumping */ \
2439 if (OP(noper) != NOTHING) \
2440 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
2442 tmp = newSVpvn_utf8( "", 0, UTF ); \
2443 av_push( trie_words, tmp ); \
2447 trie->wordinfo[curword].prev = 0; \
2448 trie->wordinfo[curword].len = wordlen; \
2449 trie->wordinfo[curword].accept = state; \
2451 if ( noper_next < tail ) { \
2453 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2455 trie->jump[curword] = (U16)(noper_next - convert); \
2457 jumper = noper_next; \
2459 nextbranch= regnext(cur); \
2463 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2464 /* chain, so that when the bits of chain are later */\
2465 /* linked together, the dups appear in the chain */\
2466 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2467 trie->wordinfo[dupe].prev = curword; \
2469 /* we haven't inserted this word yet. */ \
2470 trie->states[ state ].wordnum = curword; \
2475 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2476 ( ( base + charid >= ucharcount \
2477 && base + charid < ubound \
2478 && state == trie->trans[ base - ucharcount + charid ].check \
2479 && trie->trans[ base - ucharcount + charid ].next ) \
2480 ? trie->trans[ base - ucharcount + charid ].next \
2481 : ( state==1 ? special : 0 ) \
2484 #define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2486 TRIE_BITMAP_SET(trie, uvc); \
2487 /* store the folded codepoint */ \
2489 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2492 /* store first byte of utf8 representation of */ \
2493 /* variant codepoints */ \
2494 if (! UVCHR_IS_INVARIANT(uvc)) { \
2495 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2500 #define MADE_JUMP_TRIE 2
2501 #define MADE_EXACT_TRIE 4
2504 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2505 regnode *first, regnode *last, regnode *tail,
2506 U32 word_count, U32 flags, U32 depth)
2508 /* first pass, loop through and scan words */
2509 reg_trie_data *trie;
2510 HV *widecharmap = NULL;
2511 AV *revcharmap = newAV();
2517 regnode *jumper = NULL;
2518 regnode *nextbranch = NULL;
2519 regnode *convert = NULL;
2520 U32 *prev_states; /* temp array mapping each state to previous one */
2521 /* we just use folder as a flag in utf8 */
2522 const U8 * folder = NULL;
2524 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2525 * which stands for one trie structure, one hash, optionally followed
2528 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2529 AV *trie_words = NULL;
2530 /* along with revcharmap, this only used during construction but both are
2531 * useful during debugging so we store them in the struct when debugging.
2534 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
2535 STRLEN trie_charcount=0;
2537 SV *re_trie_maxbuff;
2538 GET_RE_DEBUG_FLAGS_DECL;
2540 PERL_ARGS_ASSERT_MAKE_TRIE;
2542 PERL_UNUSED_ARG(depth);
2546 case EXACT: case EXACTL: break;
2550 case EXACTFLU8: folder = PL_fold_latin1; break;
2551 case EXACTF: folder = PL_fold; break;
2552 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
2555 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
2557 trie->startstate = 1;
2558 trie->wordcount = word_count;
2559 RExC_rxi->data->data[ data_slot ] = (void*)trie;
2560 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
2561 if (flags == EXACT || flags == EXACTL)
2562 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2563 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2564 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2567 trie_words = newAV();
2570 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2571 assert(re_trie_maxbuff);
2572 if (!SvIOK(re_trie_maxbuff)) {
2573 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2575 DEBUG_TRIE_COMPILE_r({
2576 Perl_re_indentf( aTHX_
2577 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2579 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
2580 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
2583 /* Find the node we are going to overwrite */
2584 if ( first == startbranch && OP( last ) != BRANCH ) {
2585 /* whole branch chain */
2588 /* branch sub-chain */
2589 convert = NEXTOPER( first );
2592 /* -- First loop and Setup --
2594 We first traverse the branches and scan each word to determine if it
2595 contains widechars, and how many unique chars there are, this is
2596 important as we have to build a table with at least as many columns as we
2599 We use an array of integers to represent the character codes 0..255
2600 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2601 the native representation of the character value as the key and IV's for
2604 *TODO* If we keep track of how many times each character is used we can
2605 remap the columns so that the table compression later on is more
2606 efficient in terms of memory by ensuring the most common value is in the
2607 middle and the least common are on the outside. IMO this would be better
2608 than a most to least common mapping as theres a decent chance the most
2609 common letter will share a node with the least common, meaning the node
2610 will not be compressible. With a middle is most common approach the worst
2611 case is when we have the least common nodes twice.
2615 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2616 regnode *noper = NEXTOPER( cur );
2620 U32 wordlen = 0; /* required init */
2621 STRLEN minchars = 0;
2622 STRLEN maxchars = 0;
2623 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2626 if (OP(noper) == NOTHING) {
2627 /* skip past a NOTHING at the start of an alternation
2628 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2630 regnode *noper_next= regnext(noper);
2631 if (noper_next < tail)
2635 if ( noper < tail &&
2637 OP(noper) == flags ||
2640 OP(noper) == EXACTFU_SS
2644 uc= (U8*)STRING(noper);
2645 e= uc + STR_LEN(noper);
2652 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
2653 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2654 regardless of encoding */
2655 if (OP( noper ) == EXACTFU_SS) {
2656 /* false positives are ok, so just set this */
2657 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
2661 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2663 TRIE_CHARCOUNT(trie)++;
2666 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2667 * is in effect. Under /i, this character can match itself, or
2668 * anything that folds to it. If not under /i, it can match just
2669 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2670 * all fold to k, and all are single characters. But some folds
2671 * expand to more than one character, so for example LATIN SMALL
2672 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2673 * the string beginning at 'uc' is 'ffi', it could be matched by
2674 * three characters, or just by the one ligature character. (It
2675 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2676 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2677 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2678 * match.) The trie needs to know the minimum and maximum number
2679 * of characters that could match so that it can use size alone to
2680 * quickly reject many match attempts. The max is simple: it is
2681 * the number of folded characters in this branch (since a fold is
2682 * never shorter than what folds to it. */
2686 /* And the min is equal to the max if not under /i (indicated by
2687 * 'folder' being NULL), or there are no multi-character folds. If
2688 * there is a multi-character fold, the min is incremented just
2689 * once, for the character that folds to the sequence. Each
2690 * character in the sequence needs to be added to the list below of
2691 * characters in the trie, but we count only the first towards the
2692 * min number of characters needed. This is done through the
2693 * variable 'foldlen', which is returned by the macros that look
2694 * for these sequences as the number of bytes the sequence
2695 * occupies. Each time through the loop, we decrement 'foldlen' by
2696 * how many bytes the current char occupies. Only when it reaches
2697 * 0 do we increment 'minchars' or look for another multi-character
2699 if (folder == NULL) {
2702 else if (foldlen > 0) {
2703 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
2708 /* See if *uc is the beginning of a multi-character fold. If
2709 * so, we decrement the length remaining to look at, to account
2710 * for the current character this iteration. (We can use 'uc'
2711 * instead of the fold returned by TRIE_READ_CHAR because for
2712 * non-UTF, the latin1_safe macro is smart enough to account
2713 * for all the unfolded characters, and because for UTF, the
2714 * string will already have been folded earlier in the
2715 * compilation process */
2717 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2718 foldlen -= UTF8SKIP(uc);
2721 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2726 /* The current character (and any potential folds) should be added
2727 * to the possible matching characters for this position in this
2731 U8 folded= folder[ (U8) uvc ];
2732 if ( !trie->charmap[ folded ] ) {
2733 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2734 TRIE_STORE_REVCHAR( folded );
2737 if ( !trie->charmap[ uvc ] ) {
2738 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
2739 TRIE_STORE_REVCHAR( uvc );
2742 /* store the codepoint in the bitmap, and its folded
2744 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
2745 set_bit = 0; /* We've done our bit :-) */
2749 /* XXX We could come up with the list of code points that fold
2750 * to this using PL_utf8_foldclosures, except not for
2751 * multi-char folds, as there may be multiple combinations
2752 * there that could work, which needs to wait until runtime to
2753 * resolve (The comment about LIGATURE FFI above is such an
2758 widecharmap = newHV();
2760 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
2763 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
2765 if ( !SvTRUE( *svpp ) ) {
2766 sv_setiv( *svpp, ++trie->uniquecharcount );
2767 TRIE_STORE_REVCHAR(uvc);
2770 } /* end loop through characters in this branch of the trie */
2772 /* We take the min and max for this branch and combine to find the min
2773 * and max for all branches processed so far */
2774 if( cur == first ) {
2775 trie->minlen = minchars;
2776 trie->maxlen = maxchars;
2777 } else if (minchars < trie->minlen) {
2778 trie->minlen = minchars;
2779 } else if (maxchars > trie->maxlen) {
2780 trie->maxlen = maxchars;
2782 } /* end first pass */
2783 DEBUG_TRIE_COMPILE_r(
2784 Perl_re_indentf( aTHX_
2785 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2787 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
2788 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2789 (int)trie->minlen, (int)trie->maxlen )
2793 We now know what we are dealing with in terms of unique chars and
2794 string sizes so we can calculate how much memory a naive
2795 representation using a flat table will take. If it's over a reasonable
2796 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
2797 conservative but potentially much slower representation using an array
2800 At the end we convert both representations into the same compressed
2801 form that will be used in regexec.c for matching with. The latter
2802 is a form that cannot be used to construct with but has memory
2803 properties similar to the list form and access properties similar
2804 to the table form making it both suitable for fast searches and
2805 small enough that its feasable to store for the duration of a program.
2807 See the comment in the code where the compressed table is produced
2808 inplace from the flat tabe representation for an explanation of how
2809 the compression works.
2814 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2817 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2818 > SvIV(re_trie_maxbuff) )
2821 Second Pass -- Array Of Lists Representation
2823 Each state will be represented by a list of charid:state records
2824 (reg_trie_trans_le) the first such element holds the CUR and LEN
2825 points of the allocated array. (See defines above).
2827 We build the initial structure using the lists, and then convert
2828 it into the compressed table form which allows faster lookups
2829 (but cant be modified once converted).
2832 STRLEN transcount = 1;
2834 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
2837 trie->states = (reg_trie_state *)
2838 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2839 sizeof(reg_trie_state) );
2843 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2845 regnode *noper = NEXTOPER( cur );
2846 U32 state = 1; /* required init */
2847 U16 charid = 0; /* sanity init */
2848 U32 wordlen = 0; /* required init */
2850 if (OP(noper) == NOTHING) {
2851 regnode *noper_next= regnext(noper);
2852 if (noper_next < tail)
2856 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
2857 const U8 *uc= (U8*)STRING(noper);
2858 const U8 *e= uc + STR_LEN(noper);
2860 for ( ; uc < e ; uc += len ) {
2865 charid = trie->charmap[ uvc ];
2867 SV** const svpp = hv_fetch( widecharmap,
2874 charid=(U16)SvIV( *svpp );
2877 /* charid is now 0 if we dont know the char read, or
2878 * nonzero if we do */
2885 if ( !trie->states[ state ].trans.list ) {
2886 TRIE_LIST_NEW( state );
2889 check <= TRIE_LIST_USED( state );
2892 if ( TRIE_LIST_ITEM( state, check ).forid
2895 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2900 newstate = next_alloc++;
2901 prev_states[newstate] = state;
2902 TRIE_LIST_PUSH( state, charid, newstate );
2907 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
2911 TRIE_HANDLE_WORD(state);
2913 } /* end second pass */
2915 /* next alloc is the NEXT state to be allocated */
2916 trie->statecount = next_alloc;
2917 trie->states = (reg_trie_state *)
2918 PerlMemShared_realloc( trie->states,
2920 * sizeof(reg_trie_state) );
2922 /* and now dump it out before we compress it */
2923 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2924 revcharmap, next_alloc,
2928 trie->trans = (reg_trie_trans *)
2929 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
2936 for( state=1 ; state < next_alloc ; state ++ ) {
2940 DEBUG_TRIE_COMPILE_MORE_r(
2941 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
2945 if (trie->states[state].trans.list) {
2946 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
2950 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2951 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
2952 if ( forid < minid ) {
2954 } else if ( forid > maxid ) {
2958 if ( transcount < tp + maxid - minid + 1) {
2960 trie->trans = (reg_trie_trans *)
2961 PerlMemShared_realloc( trie->trans,
2963 * sizeof(reg_trie_trans) );
2964 Zero( trie->trans + (transcount / 2),
2968 base = trie->uniquecharcount + tp - minid;
2969 if ( maxid == minid ) {
2971 for ( ; zp < tp ; zp++ ) {
2972 if ( ! trie->trans[ zp ].next ) {
2973 base = trie->uniquecharcount + zp - minid;
2974 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
2976 trie->trans[ zp ].check = state;
2982 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
2984 trie->trans[ tp ].check = state;
2989 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2990 const U32 tid = base
2991 - trie->uniquecharcount
2992 + TRIE_LIST_ITEM( state, idx ).forid;
2993 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
2995 trie->trans[ tid ].check = state;
2997 tp += ( maxid - minid + 1 );
2999 Safefree(trie->states[ state ].trans.list);
3002 DEBUG_TRIE_COMPILE_MORE_r(
3003 Perl_re_printf( aTHX_ " base: %d\n",base);
3006 trie->states[ state ].trans.base=base;
3008 trie->lasttrans = tp + 1;
3012 Second Pass -- Flat Table Representation.
3014 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3015 each. We know that we will need Charcount+1 trans at most to store
3016 the data (one row per char at worst case) So we preallocate both
3017 structures assuming worst case.
3019 We then construct the trie using only the .next slots of the entry
3022 We use the .check field of the first entry of the node temporarily
3023 to make compression both faster and easier by keeping track of how
3024 many non zero fields are in the node.
3026 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3029 There are two terms at use here: state as a TRIE_NODEIDX() which is
3030 a number representing the first entry of the node, and state as a
3031 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3032 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3033 if there are 2 entrys per node. eg:
3041 The table is internally in the right hand, idx form. However as we
3042 also have to deal with the states array which is indexed by nodenum
3043 we have to use TRIE_NODENUM() to convert.
3046 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
3049 trie->trans = (reg_trie_trans *)
3050 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3051 * trie->uniquecharcount + 1,
3052 sizeof(reg_trie_trans) );
3053 trie->states = (reg_trie_state *)
3054 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3055 sizeof(reg_trie_state) );
3056 next_alloc = trie->uniquecharcount + 1;
3059 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3061 regnode *noper = NEXTOPER( cur );
3063 U32 state = 1; /* required init */
3065 U16 charid = 0; /* sanity init */
3066 U32 accept_state = 0; /* sanity init */
3068 U32 wordlen = 0; /* required init */
3070 if (OP(noper) == NOTHING) {
3071 regnode *noper_next= regnext(noper);
3072 if (noper_next < tail)
3076 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
3077 const U8 *uc= (U8*)STRING(noper);
3078 const U8 *e= uc + STR_LEN(noper);
3080 for ( ; uc < e ; uc += len ) {
3085 charid = trie->charmap[ uvc ];
3087 SV* const * const svpp = hv_fetch( widecharmap,
3091 charid = svpp ? (U16)SvIV(*svpp) : 0;
3095 if ( !trie->trans[ state + charid ].next ) {
3096 trie->trans[ state + charid ].next = next_alloc;
3097 trie->trans[ state ].check++;
3098 prev_states[TRIE_NODENUM(next_alloc)]
3099 = TRIE_NODENUM(state);
3100 next_alloc += trie->uniquecharcount;
3102 state = trie->trans[ state + charid ].next;
3104 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
3106 /* charid is now 0 if we dont know the char read, or
3107 * nonzero if we do */
3110 accept_state = TRIE_NODENUM( state );
3111 TRIE_HANDLE_WORD(accept_state);
3113 } /* end second pass */
3115 /* and now dump it out before we compress it */
3116 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3118 next_alloc, depth+1));
3122 * Inplace compress the table.*
3124 For sparse data sets the table constructed by the trie algorithm will
3125 be mostly 0/FAIL transitions or to put it another way mostly empty.
3126 (Note that leaf nodes will not contain any transitions.)
3128 This algorithm compresses the tables by eliminating most such
3129 transitions, at the cost of a modest bit of extra work during lookup:
3131 - Each states[] entry contains a .base field which indicates the
3132 index in the state[] array wheres its transition data is stored.
3134 - If .base is 0 there are no valid transitions from that node.
3136 - If .base is nonzero then charid is added to it to find an entry in
3139 -If trans[states[state].base+charid].check!=state then the
3140 transition is taken to be a 0/Fail transition. Thus if there are fail
3141 transitions at the front of the node then the .base offset will point
3142 somewhere inside the previous nodes data (or maybe even into a node
3143 even earlier), but the .check field determines if the transition is
3147 The following process inplace converts the table to the compressed
3148 table: We first do not compress the root node 1,and mark all its
3149 .check pointers as 1 and set its .base pointer as 1 as well. This
3150 allows us to do a DFA construction from the compressed table later,
3151 and ensures that any .base pointers we calculate later are greater
3154 - We set 'pos' to indicate the first entry of the second node.
3156 - We then iterate over the columns of the node, finding the first and
3157 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3158 and set the .check pointers accordingly, and advance pos
3159 appropriately and repreat for the next node. Note that when we copy
3160 the next pointers we have to convert them from the original
3161 NODEIDX form to NODENUM form as the former is not valid post
3164 - If a node has no transitions used we mark its base as 0 and do not
3165 advance the pos pointer.
3167 - If a node only has one transition we use a second pointer into the
3168 structure to fill in allocated fail transitions from other states.
3169 This pointer is independent of the main pointer and scans forward
3170 looking for null transitions that are allocated to a state. When it
3171 finds one it writes the single transition into the "hole". If the
3172 pointer doesnt find one the single transition is appended as normal.
3174 - Once compressed we can Renew/realloc the structures to release the
3177 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3178 specifically Fig 3.47 and the associated pseudocode.
3182 const U32 laststate = TRIE_NODENUM( next_alloc );
3185 trie->statecount = laststate;
3187 for ( state = 1 ; state < laststate ; state++ ) {
3189 const U32 stateidx = TRIE_NODEIDX( state );
3190 const U32 o_used = trie->trans[ stateidx ].check;
3191 U32 used = trie->trans[ stateidx ].check;
3192 trie->trans[ stateidx ].check = 0;
3195 used && charid < trie->uniquecharcount;
3198 if ( flag || trie->trans[ stateidx + charid ].next ) {
3199 if ( trie->trans[ stateidx + charid ].next ) {
3201 for ( ; zp < pos ; zp++ ) {
3202 if ( ! trie->trans[ zp ].next ) {
3206 trie->states[ state ].trans.base
3208 + trie->uniquecharcount
3210 trie->trans[ zp ].next
3211 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3213 trie->trans[ zp ].check = state;
3214 if ( ++zp > pos ) pos = zp;
3221 trie->states[ state ].trans.base
3222 = pos + trie->uniquecharcount - charid ;
3224 trie->trans[ pos ].next
3225 = SAFE_TRIE_NODENUM(
3226 trie->trans[ stateidx + charid ].next );
3227 trie->trans[ pos ].check = state;
3232 trie->lasttrans = pos + 1;
3233 trie->states = (reg_trie_state *)
3234 PerlMemShared_realloc( trie->states, laststate
3235 * sizeof(reg_trie_state) );
3236 DEBUG_TRIE_COMPILE_MORE_r(
3237 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
3239 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3243 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
3246 } /* end table compress */
3248 DEBUG_TRIE_COMPILE_MORE_r(
3249 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
3251 (UV)trie->statecount,
3252 (UV)trie->lasttrans)
3254 /* resize the trans array to remove unused space */
3255 trie->trans = (reg_trie_trans *)
3256 PerlMemShared_realloc( trie->trans, trie->lasttrans
3257 * sizeof(reg_trie_trans) );
3259 { /* Modify the program and insert the new TRIE node */
3260 U8 nodetype =(U8)(flags & 0xFF);
3264 regnode *optimize = NULL;
3265 #ifdef RE_TRACK_PATTERN_OFFSETS
3268 U32 mjd_nodelen = 0;
3269 #endif /* RE_TRACK_PATTERN_OFFSETS */
3270 #endif /* DEBUGGING */
3272 This means we convert either the first branch or the first Exact,
3273 depending on whether the thing following (in 'last') is a branch
3274 or not and whther first is the startbranch (ie is it a sub part of
3275 the alternation or is it the whole thing.)
3276 Assuming its a sub part we convert the EXACT otherwise we convert
3277 the whole branch sequence, including the first.
3279 /* Find the node we are going to overwrite */
3280 if ( first != startbranch || OP( last ) == BRANCH ) {
3281 /* branch sub-chain */
3282 NEXT_OFF( first ) = (U16)(last - first);
3283 #ifdef RE_TRACK_PATTERN_OFFSETS
3285 mjd_offset= Node_Offset((convert));
3286 mjd_nodelen= Node_Length((convert));
3289 /* whole branch chain */
3291 #ifdef RE_TRACK_PATTERN_OFFSETS
3294 const regnode *nop = NEXTOPER( convert );
3295 mjd_offset= Node_Offset((nop));
3296 mjd_nodelen= Node_Length((nop));
3300 Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n",
3302 (UV)mjd_offset, (UV)mjd_nodelen)
3305 /* But first we check to see if there is a common prefix we can
3306 split out as an EXACT and put in front of the TRIE node. */
3307 trie->startstate= 1;
3308 if ( trie->bitmap && !widecharmap && !trie->jump ) {
3309 /* we want to find the first state that has more than
3310 * one transition, if that state is not the first state
3311 * then we have a common prefix which we can remove.
3314 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
3316 I32 first_ofs = -1; /* keeps track of the ofs of the first
3317 transition, -1 means none */
3319 const U32 base = trie->states[ state ].trans.base;
3321 /* does this state terminate an alternation? */
3322 if ( trie->states[state].wordnum )
3325 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
3326 if ( ( base + ofs >= trie->uniquecharcount ) &&
3327 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
3328 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
3330 if ( ++count > 1 ) {
3331 /* we have more than one transition */
3334 /* if this is the first state there is no common prefix
3335 * to extract, so we can exit */
3336 if ( state == 1 ) break;
3337 tmp = av_fetch( revcharmap, ofs, 0);
3338 ch = (U8*)SvPV_nolen_const( *tmp );
3340 /* if we are on count 2 then we need to initialize the
3341 * bitmap, and store the previous char if there was one
3344 /* clear the bitmap */
3345 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
3347 Perl_re_indentf( aTHX_ "New Start State=%" UVuf " Class: [",
3350 if (first_ofs >= 0) {
3351 SV ** const tmp = av_fetch( revcharmap, first_ofs, 0);
3352 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
3354 TRIE_BITMAP_SET_FOLDED(trie,*ch,folder);
3356 Perl_re_printf( aTHX_ "%s", (char*)ch)
3360 /* store the current firstchar in the bitmap */
3361 TRIE_BITMAP_SET_FOLDED(trie,*ch,folder);
3362 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "%s", ch));
3368 /* This state has only one transition, its transition is part
3369 * of a common prefix - we need to concatenate the char it
3370 * represents to what we have so far. */
3371 SV **tmp = av_fetch( revcharmap, first_ofs, 0);
3373 char *ch = SvPV( *tmp, len );
3375 SV *sv=sv_newmortal();
3376 Perl_re_indentf( aTHX_ "Prefix State: %" UVuf " Ofs:%" UVuf " Char='%s'\n",
3378 (UV)state, (UV)first_ofs,
3379 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
3380 PL_colors[0], PL_colors[1],
3381 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
3382 PERL_PV_ESCAPE_FIRSTCHAR
3387 OP( convert ) = nodetype;
3388 str=STRING(convert);
3391 STR_LEN(convert) += len;
3397 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "]\n"));
3402 trie->prefixlen = (state-1);
3404 regnode *n = convert+NODE_SZ_STR(convert);
3405 NEXT_OFF(convert) = NODE_SZ_STR(convert);
3406 trie->startstate = state;
3407 trie->minlen -= (state - 1);
3408 trie->maxlen -= (state - 1);
3410 /* At least the UNICOS C compiler choked on this
3411 * being argument to DEBUG_r(), so let's just have
3414 #ifdef PERL_EXT_RE_BUILD
3420 regnode *fix = convert;
3421 U32 word = trie->wordcount;
3423 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
3424 while( ++fix < n ) {
3425 Set_Node_Offset_Length(fix, 0, 0);
3428 SV ** const tmp = av_fetch( trie_words, word, 0 );
3430 if ( STR_LEN(convert) <= SvCUR(*tmp) )
3431 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
3433 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
3441 NEXT_OFF(convert) = (U16)(tail - convert);
3442 DEBUG_r(optimize= n);
3448 if ( trie->maxlen ) {
3449 NEXT_OFF( convert ) = (U16)(tail - convert);
3450 ARG_SET( convert, data_slot );
3451 /* Store the offset to the first unabsorbed branch in
3452 jump[0], which is otherwise unused by the jump logic.
3453 We use this when dumping a trie and during optimisation. */
3455 trie->jump[0] = (U16)(nextbranch - convert);
3457 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
3458 * and there is a bitmap
3459 * and the first "jump target" node we found leaves enough room
3460 * then convert the TRIE node into a TRIEC node, with the bitmap
3461 * embedded inline in the opcode - this is hypothetically faster.
3463 if ( !trie->states[trie->startstate].wordnum
3465 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
3467 OP( convert ) = TRIEC;
3468 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
3469 PerlMemShared_free(trie->bitmap);
3472 OP( convert ) = TRIE;
3474 /* store the type in the flags */
3475 convert->flags = nodetype;
3479 + regarglen[ OP( convert ) ];
3481 /* XXX We really should free up the resource in trie now,
3482 as we won't use them - (which resources?) dmq */
3484 /* needed for dumping*/
3485 DEBUG_r(if (optimize) {
3486 regnode *opt = convert;
3488 while ( ++opt < optimize) {
3489 Set_Node_Offset_Length(opt,0,0);
3492 Try to clean up some of the debris left after the
3495 while( optimize < jumper ) {
3496 mjd_nodelen += Node_Length((optimize));
3497 OP( optimize ) = OPTIMIZED;
3498 Set_Node_Offset_Length(optimize,0,0);
3501 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
3503 } /* end node insert */
3505 /* Finish populating the prev field of the wordinfo array. Walk back
3506 * from each accept state until we find another accept state, and if
3507 * so, point the first word's .prev field at the second word. If the
3508 * second already has a .prev field set, stop now. This will be the
3509 * case either if we've already processed that word's accept state,
3510 * or that state had multiple words, and the overspill words were
3511 * already linked up earlier.
3518 for (word=1; word <= trie->wordcount; word++) {
3520 if (trie->wordinfo[word].prev)
3522 state = trie->wordinfo[word].accept;
3524 state = prev_states[state];
3527 prev = trie->states[state].wordnum;
3531 trie->wordinfo[word].prev = prev;
3533 Safefree(prev_states);
3537 /* and now dump out the compressed format */
3538 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
3540 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
3542 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
3543 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
3545 SvREFCNT_dec_NN(revcharmap);
3549 : trie->startstate>1
3555 S_construct_ahocorasick_from_trie(pTHX_ RExC_state_t *pRExC_state, regnode *source, U32 depth)
3557 /* The Trie is constructed and compressed now so we can build a fail array if
3560 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and
3562 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi,
3566 We find the fail state for each state in the trie, this state is the longest
3567 proper suffix of the current state's 'word' that is also a proper prefix of
3568 another word in our trie. State 1 represents the word '' and is thus the
3569 default fail state. This allows the DFA not to have to restart after its
3570 tried and failed a word at a given point, it simply continues as though it
3571 had been matching the other word in the first place.
3573 'abcdgu'=~/abcdefg|cdgu/
3574 When we get to 'd' we are still matching the first word, we would encounter
3575 'g' which would fail, which would bring us to the state representing 'd' in
3576 the second word where we would try 'g' and succeed, proceeding to match
3579 /* add a fail transition */
3580 const U32 trie_offset = ARG(source);
3581 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
3583 const U32 ucharcount = trie->uniquecharcount;
3584 const U32 numstates = trie->statecount;
3585 const U32 ubound = trie->lasttrans + ucharcount;
3589 U32 base = trie->states[ 1 ].trans.base;
3592 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T"));
3594 GET_RE_DEBUG_FLAGS_DECL;
3596 PERL_ARGS_ASSERT_CONSTRUCT_AHOCORASICK_FROM_TRIE;
3597 PERL_UNUSED_CONTEXT;
3599 PERL_UNUSED_ARG(depth);
3602 if ( OP(source) == TRIE ) {
3603 struct regnode_1 *op = (struct regnode_1 *)
3604 PerlMemShared_calloc(1, sizeof(struct regnode_1));
3605 StructCopy(source,op,struct regnode_1);
3606 stclass = (regnode *)op;
3608 struct regnode_charclass *op = (struct regnode_charclass *)
3609 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
3610 StructCopy(source,op,struct regnode_charclass);
3611 stclass = (regnode *)op;
3613 OP(stclass)+=2; /* convert the TRIE type to its AHO-CORASICK equivalent */
3615 ARG_SET( stclass, data_slot );
3616 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
3617 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3618 aho->trie=trie_offset;
3619 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
3620 Copy( trie->states, aho->states, numstates, reg_trie_state );
3621 Newxz( q, numstates, U32);
3622 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
3625 /* initialize fail[0..1] to be 1 so that we always have
3626 a valid final fail state */
3627 fail[ 0 ] = fail[ 1 ] = 1;
3629 for ( charid = 0; charid < ucharcount ; charid++ ) {
3630 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
3632 q[ q_write ] = newstate;
3633 /* set to point at the root */
3634 fail[ q[ q_write++ ] ]=1;
3637 while ( q_read < q_write) {
3638 const U32 cur = q[ q_read++ % numstates ];
3639 base = trie->states[ cur ].trans.base;
3641 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
3642 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
3644 U32 fail_state = cur;
3647 fail_state = fail[ fail_state ];
3648 fail_base = aho->states[ fail_state ].trans.base;
3649 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
3651 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
3652 fail[ ch_state ] = fail_state;
3653 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
3655 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
3657 q[ q_write++ % numstates] = ch_state;
3661 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
3662 when we fail in state 1, this allows us to use the
3663 charclass scan to find a valid start char. This is based on the principle
3664 that theres a good chance the string being searched contains lots of stuff
3665 that cant be a start char.
3667 fail[ 0 ] = fail[ 1 ] = 0;
3668 DEBUG_TRIE_COMPILE_r({
3669 Perl_re_indentf( aTHX_ "Stclass Failtable (%" UVuf " states): 0",
3670 depth, (UV)numstates
3672 for( q_read=1; q_read<numstates; q_read++ ) {
3673 Perl_re_printf( aTHX_ ", %" UVuf, (UV)fail[q_read]);
3675 Perl_re_printf( aTHX_ "\n");
3678 /*RExC_seen |= REG_TRIEDFA_SEEN;*/
3683 /* The below joins as many adjacent EXACTish nodes as possible into a single
3684 * one. The regop may be changed if the node(s) contain certain sequences that
3685 * require special handling. The joining is only done if:
3686 * 1) there is room in the current conglomerated node to entirely contain the
3688 * 2) they are the exact same node type
3690 * The adjacent nodes actually may be separated by NOTHING-kind nodes, and
3691 * these get optimized out
3693 * XXX khw thinks this should be enhanced to fill EXACT (at least) nodes as full
3694 * as possible, even if that means splitting an existing node so that its first
3695 * part is moved to the preceeding node. This would maximise the efficiency of
3696 * memEQ during matching. Elsewhere in this file, khw proposes splitting
3697 * EXACTFish nodes into portions that don't change under folding vs those that
3698 * do. Those portions that don't change may be the only things in the pattern that
3699 * could be used to find fixed and floating strings.
3701 * If a node is to match under /i (folded), the number of characters it matches
3702 * can be different than its character length if it contains a multi-character
3703 * fold. *min_subtract is set to the total delta number of characters of the
3706 * And *unfolded_multi_char is set to indicate whether or not the node contains
3707 * an unfolded multi-char fold. This happens when whether the fold is valid or
3708 * not won't be known until runtime; namely for EXACTF nodes that contain LATIN
3709 * SMALL LETTER SHARP S, as only if the target string being matched against
3710 * turns out to be UTF-8 is that fold valid; and also for EXACTFL nodes whose
3711 * folding rules depend on the locale in force at runtime. (Multi-char folds
3712 * whose components are all above the Latin1 range are not run-time locale
3713 * dependent, and have already been folded by the time this function is
3716 * This is as good a place as any to discuss the design of handling these
3717 * multi-character fold sequences. It's been wrong in Perl for a very long
3718 * time. There are three code points in Unicode whose multi-character folds
3719 * were long ago discovered to mess things up. The previous designs for
3720 * dealing with these involved assigning a special node for them. This
3721 * approach doesn't always work, as evidenced by this example:
3722 * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches
3723 * Both sides fold to "sss", but if the pattern is parsed to create a node that
3724 * would match just the \xDF, it won't be able to handle the case where a
3725 * successful match would have to cross the node's boundary. The new approach
3726 * that hopefully generally solves the problem generates an EXACTFU_SS node
3727 * that is "sss" in this case.
3729 * It turns out that there are problems with all multi-character folds, and not
3730 * just these three. Now the code is general, for all such cases. The
3731 * approach taken is:
3732 * 1) This routine examines each EXACTFish node that could contain multi-
3733 * character folded sequences. Since a single character can fold into
3734 * such a sequence, the minimum match length for this node is less than
3735 * the number of characters in the node. This routine returns in
3736 * *min_subtract how many characters to subtract from the the actual
3737 * length of the string to get a real minimum match length; it is 0 if
3738 * there are no multi-char foldeds. This delta is used by the caller to
3739 * adjust the min length of the match, and the delta between min and max,
3740 * so that the optimizer doesn't reject these possibilities based on size
3742 * 2) For the sequence involving the Sharp s (\xDF), the node type EXACTFU_SS
3743 * is used for an EXACTFU node that contains at least one "ss" sequence in
3744 * it. For non-UTF-8 patterns and strings, this is the only case where
3745 * there is a possible fold length change. That means that a regular
3746 * EXACTFU node without UTF-8 involvement doesn't have to concern itself
3747 * with length changes, and so can be processed faster. regexec.c takes
3748 * advantage of this. Generally, an EXACTFish node that is in UTF-8 is
3749 * pre-folded by regcomp.c (except EXACTFL, some of whose folds aren't
3750 * known until runtime). This saves effort in regex matching. However,
3751 * the pre-folding isn't done for non-UTF8 patterns because the fold of
3752 * the MICRO SIGN requires UTF-8, and we don't want to slow things down by
3753 * forcing the pattern into UTF8 unless necessary. Also what EXACTF (and,
3754 * again, EXACTFL) nodes fold to isn't known until runtime. The fold
3755 * possibilities for the non-UTF8 patterns are quite simple, except for
3756 * the sharp s. All the ones that don't involve a UTF-8 target string are
3757 * members of a fold-pair, and arrays are set up for all of them so that
3758 * the other member of the pair can be found quickly. Code elsewhere in
3759 * this file makes sure that in EXACTFU nodes, the sharp s gets folded to
3760 * 'ss', even if the pattern isn't UTF-8. This avoids the issues
3761 * described in the next item.
3762 * 3) A problem remains for unfolded multi-char folds. (These occur when the
3763 * validity of the fold won't be known until runtime, and so must remain
3764 * unfolded for now. This happens for the sharp s in EXACTF and EXACTFA
3765 * nodes when the pattern isn't in UTF-8. (Note, BTW, that there cannot
3766 * be an EXACTF node with a UTF-8 pattern.) They also occur for various
3767 * folds in EXACTFL nodes, regardless of the UTF-ness of the pattern.)
3768 * The reason this is a problem is that the optimizer part of regexec.c
3769 * (probably unwittingly, in Perl_regexec_flags()) makes an assumption
3770 * that a character in the pattern corresponds to at most a single
3771 * character in the target string. (And I do mean character, and not byte
3772 * here, unlike other parts of the documentation that have never been
3773 * updated to account for multibyte Unicode.) sharp s in EXACTF and
3774 * EXACTFL nodes can match the two character string 'ss'; in EXACTFA nodes
3775 * it can match "\x{17F}\x{17F}". These, along with other ones in EXACTFL
3776 * nodes, violate the assumption, and they are the only instances where it
3777 * is violated. I'm reluctant to try to change the assumption, as the
3778 * code involved is impenetrable to me (khw), so instead the code here
3779 * punts. This routine examines EXACTFL nodes, and (when the pattern
3780 * isn't UTF-8) EXACTF and EXACTFA for such unfolded folds, and returns a
3781 * boolean indicating whether or not the node contains such a fold. When
3782 * it is true, the caller sets a flag that later causes the optimizer in
3783 * this file to not set values for the floating and fixed string lengths,
3784 * and thus avoids the optimizer code in regexec.c that makes the invalid
3785 * assumption. Thus, there is no optimization based on string lengths for
3786 * EXACTFL nodes that contain these few folds, nor for non-UTF8-pattern
3787 * EXACTF and EXACTFA nodes that contain the sharp s. (The reason the
3788 * assumption is wrong only in these cases is that all other non-UTF-8
3789 * folds are 1-1; and, for UTF-8 patterns, we pre-fold all other folds to
3790 * their expanded versions. (Again, we can't prefold sharp s to 'ss' in
3791 * EXACTF nodes because we don't know at compile time if it actually
3792 * matches 'ss' or not. For EXACTF nodes it will match iff the target
3793 * string is in UTF-8. This is in contrast to EXACTFU nodes, where it
3794 * always matches; and EXACTFA where it never does. In an EXACTFA node in
3795 * a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the
3796 * problem; but in a non-UTF8 pattern, folding it to that above-Latin1
3797 * string would require the pattern to be forced into UTF-8, the overhead
3798 * of which we want to avoid. Similarly the unfolded multi-char folds in
3799 * EXACTFL nodes will match iff the locale at the time of match is a UTF-8
3802 * Similarly, the code that generates tries doesn't currently handle
3803 * not-already-folded multi-char folds, and it looks like a pain to change
3804 * that. Therefore, trie generation of EXACTFA nodes with the sharp s
3805 * doesn't work. Instead, such an EXACTFA is turned into a new regnode,
3806 * EXACTFA_NO_TRIE, which the trie code knows not to handle. Most people
3807 * using /iaa matching will be doing so almost entirely with ASCII
3808 * strings, so this should rarely be encountered in practice */
3810 #define JOIN_EXACT(scan,min_subtract,unfolded_multi_char, flags) \
3811 if (PL_regkind[OP(scan)] == EXACT) \
3812 join_exact(pRExC_state,(scan),(min_subtract),unfolded_multi_char, (flags),NULL,depth+1)
3815 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
3816 UV *min_subtract, bool *unfolded_multi_char,
3817 U32 flags,regnode *val, U32 depth)
3819 /* Merge several consecutive EXACTish nodes into one. */
3820 regnode *n = regnext(scan);
3822 regnode *next = scan + NODE_SZ_STR(scan);
3826 regnode *stop = scan;
3827 GET_RE_DEBUG_FLAGS_DECL;
3829 PERL_UNUSED_ARG(depth);
3832 PERL_ARGS_ASSERT_JOIN_EXACT;
3833 #ifndef EXPERIMENTAL_INPLACESCAN
3834 PERL_UNUSED_ARG(flags);
3835 PERL_UNUSED_ARG(val);
3837 DEBUG_PEEP("join", scan, depth, 0);
3839 /* Look through the subsequent nodes in the chain. Skip NOTHING, merge
3840 * EXACT ones that are mergeable to the current one. */
3842 && (PL_regkind[OP(n)] == NOTHING
3843 || (stringok && OP(n) == OP(scan)))
3845 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX)
3848 if (OP(n) == TAIL || n > next)
3850 if (PL_regkind[OP(n)] == NOTHING) {
3851 DEBUG_PEEP("skip:", n, depth, 0);
3852 NEXT_OFF(scan) += NEXT_OFF(n);
3853 next = n + NODE_STEP_REGNODE;
3860 else if (stringok) {
3861 const unsigned int oldl = STR_LEN(scan);
3862 regnode * const nnext = regnext(n);
3864 /* XXX I (khw) kind of doubt that this works on platforms (should
3865 * Perl ever run on one) where U8_MAX is above 255 because of lots
3866 * of other assumptions */
3867 /* Don't join if the sum can't fit into a single node */
3868 if (oldl + STR_LEN(n) > U8_MAX)
3871 DEBUG_PEEP("merg", n, depth, 0);
3874 NEXT_OFF(scan) += NEXT_OFF(n);
3875 STR_LEN(scan) += STR_LEN(n);
3876 next = n + NODE_SZ_STR(n);
3877 /* Now we can overwrite *n : */
3878 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
3886 #ifdef EXPERIMENTAL_INPLACESCAN
3887 if (flags && !NEXT_OFF(n)) {
3888 DEBUG_PEEP("atch", val, depth, 0);
3889 if (reg_off_by_arg[OP(n)]) {
3890 ARG_SET(n, val - n);
3893 NEXT_OFF(n) = val - n;
3901 *unfolded_multi_char = FALSE;
3903 /* Here, all the adjacent mergeable EXACTish nodes have been merged. We
3904 * can now analyze for sequences of problematic code points. (Prior to
3905 * this final joining, sequences could have been split over boundaries, and
3906 * hence missed). The sequences only happen in folding, hence for any
3907 * non-EXACT EXACTish node */
3908 if (OP(scan) != EXACT && OP(scan) != EXACTL) {
3909 U8* s0 = (U8*) STRING(scan);
3911 U8* s_end = s0 + STR_LEN(scan);
3913 int total_count_delta = 0; /* Total delta number of characters that
3914 multi-char folds expand to */
3916 /* One pass is made over the node's string looking for all the
3917 * possibilities. To avoid some tests in the loop, there are two main
3918 * cases, for UTF-8 patterns (which can't have EXACTF nodes) and
3923 if (OP(scan) == EXACTFL) {
3926 /* An EXACTFL node would already have been changed to another
3927 * node type unless there is at least one character in it that
3928 * is problematic; likely a character whose fold definition
3929 * won't be known until runtime, and so has yet to be folded.
3930 * For all but the UTF-8 locale, folds are 1-1 in length, but
3931 * to handle the UTF-8 case, we need to create a temporary
3932 * folded copy using UTF-8 locale rules in order to analyze it.
3933 * This is because our macros that look to see if a sequence is
3934 * a multi-char fold assume everything is folded (otherwise the
3935 * tests in those macros would be too complicated and slow).
3936 * Note that here, the non-problematic folds will have already
3937 * been done, so we can just copy such characters. We actually
3938 * don't completely fold the EXACTFL string. We skip the
3939 * unfolded multi-char folds, as that would just create work
3940 * below to figure out the size they already are */
3942 Newx(folded, UTF8_MAX_FOLD_CHAR_EXPAND * STR_LEN(scan) + 1, U8);
3945 STRLEN s_len = UTF8SKIP(s);
3946 if (! is_PROBLEMATIC_LOCALE_FOLD_utf8(s)) {
3947 Copy(s, d, s_len, U8);
3950 else if (is_FOLDS_TO_MULTI_utf8(s)) {
3951 *unfolded_multi_char = TRUE;
3952 Copy(s, d, s_len, U8);
3955 else if (isASCII(*s)) {
3956 *(d++) = toFOLD(*s);
3960 _toFOLD_utf8_flags(s, s_end, d, &len, FOLD_FLAGS_FULL);
3966 /* Point the remainder of the routine to look at our temporary
3970 } /* End of creating folded copy of EXACTFL string */
3972 /* Examine the string for a multi-character fold sequence. UTF-8
3973 * patterns have all characters pre-folded by the time this code is
3975 while (s < s_end - 1) /* Can stop 1 before the end, as minimum
3976 length sequence we are looking for is 2 */
3978 int count = 0; /* How many characters in a multi-char fold */
3979 int len = is_MULTI_CHAR_FOLD_utf8_safe(s, s_end);
3980 if (! len) { /* Not a multi-char fold: get next char */
3985 /* Nodes with 'ss' require special handling, except for
3986 * EXACTFA-ish for which there is no multi-char fold to this */
3987 if (len == 2 && *s == 's' && *(s+1) == 's'
3988 && OP(scan) != EXACTFA
3989 && OP(scan) != EXACTFA_NO_TRIE)
3992 if (OP(scan) != EXACTFL) {
3993 OP(scan) = EXACTFU_SS;
3997 else { /* Here is a generic multi-char fold. */
3998 U8* multi_end = s + len;
4000 /* Count how many characters are in it. In the case of
4001 * /aa, no folds which contain ASCII code points are
4002 * allowed, so check for those, and skip if found. */
4003 if (OP(scan) != EXACTFA && OP(scan) != EXACTFA_NO_TRIE) {
4004 count = utf8_length(s, multi_end);
4008 while (s < multi_end) {
4011 goto next_iteration;
4021 /* The delta is how long the sequence is minus 1 (1 is how long
4022 * the character that folds to the sequence is) */
https://perl5.git.perl.org / perl5.git / blob